/var/spool/slurmd/job335413/slurm_script: line 12: activate: No such file or directory
W0216 06:17:19.560000 3603613 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] 
W0216 06:17:19.560000 3603613 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] *****************************************
W0216 06:17:19.560000 3603613 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
W0216 06:17:19.560000 3603613 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] *****************************************
W0216 06:17:23.303000 3997540 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] 
W0216 06:17:23.303000 3997540 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] *****************************************
W0216 06:17:23.303000 3997540 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
W0216 06:17:23.303000 3997540 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] *****************************************
W0216 06:17:23.307000 2664794 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] 
W0216 06:17:23.307000 2664794 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] *****************************************
W0216 06:17:23.307000 2664794 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
W0216 06:17:23.307000 2664794 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] *****************************************
W0216 06:17:23.369000 3344412 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] 
W0216 06:17:23.369000 3344412 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] *****************************************
W0216 06:17:23.369000 3344412 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
W0216 06:17:23.369000 3344412 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] *****************************************
W0216 06:17:23.370000 2133652 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] 
W0216 06:17:23.370000 2133652 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] *****************************************
W0216 06:17:23.370000 2133652 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
W0216 06:17:23.370000 2133652 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] *****************************************
W0216 06:17:23.426000 2201849 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] 
W0216 06:17:23.426000 2201849 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] *****************************************
W0216 06:17:23.426000 2201849 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
W0216 06:17:23.426000 2201849 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] *****************************************
W0216 06:17:23.493000 2062756 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] 
W0216 06:17:23.493000 2062756 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] *****************************************
W0216 06:17:23.493000 2062756 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
W0216 06:17:23.493000 2062756 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] *****************************************
W0216 06:17:23.501000 3343136 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] 
W0216 06:17:23.501000 3343136 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] *****************************************
W0216 06:17:23.501000 3343136 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
W0216 06:17:23.501000 3343136 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] *****************************************
W0216 06:17:23.536000 3262384 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] 
W0216 06:17:23.536000 3262384 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] *****************************************
W0216 06:17:23.536000 3262384 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
W0216 06:17:23.536000 3262384 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] *****************************************
W0216 06:17:23.537000 3772629 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] 
W0216 06:17:23.537000 3772629 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] *****************************************
W0216 06:17:23.537000 3772629 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
W0216 06:17:23.537000 3772629 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] *****************************************
W0216 06:17:23.583000 3261191 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] 
W0216 06:17:23.583000 3261191 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] *****************************************
W0216 06:17:23.583000 3261191 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
W0216 06:17:23.583000 3261191 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] *****************************************
W0216 06:17:23.584000 4003734 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] 
W0216 06:17:23.584000 4003734 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] *****************************************
W0216 06:17:23.584000 4003734 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
W0216 06:17:23.584000 4003734 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] *****************************************
W0216 06:17:23.621000 3781647 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] 
W0216 06:17:23.621000 3781647 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] *****************************************
W0216 06:17:23.621000 3781647 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
W0216 06:17:23.621000 3781647 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] *****************************************
W0216 06:17:23.849000 3791965 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] 
W0216 06:17:23.849000 3791965 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] *****************************************
W0216 06:17:23.849000 3791965 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
W0216 06:17:23.849000 3791965 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] *****************************************
W0216 06:17:24.080000 4000190 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] 
W0216 06:17:24.080000 4000190 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] *****************************************
W0216 06:17:24.080000 4000190 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
W0216 06:17:24.080000 4000190 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] *****************************************
W0216 06:17:26.564000 3335470 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] 
W0216 06:17:26.564000 3335470 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] *****************************************
W0216 06:17:26.564000 3335470 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
W0216 06:17:26.564000 3335470 .local/lib/python3.10/site-packages/torch/distributed/run.py:793] *****************************************
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.

Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.

Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.

Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.

Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json

Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,

Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}


Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.

Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json

Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}


Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.

Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}


Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}


Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}


Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.

Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.

Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json

Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.

Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.

Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.

Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.

Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.

Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.35it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.15it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.05it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:02,  1.84it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.10it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.04it/s]
Loading checkpoint shards:  20%|██   
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.

Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  3.09it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.32it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.39it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.51it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.29it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.36it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.76it/s]
Loading checkpoint shards:  2Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.

Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:02,  1.92it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  3.12it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  3.30it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  3.34it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  3.19it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  3.09it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  3.29it/s]
Loading checkpoint shards:  2Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.

Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.49it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.64it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.65it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.73it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.34it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.14it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.34it/s]
Loading checkpoint shards:  2
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.30it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.

Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.30it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.13it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.42it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.54it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:02,  1.63it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.09it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.14it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:02,  1.89it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.15it/s]
Lloading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}


Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.

Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  3.67it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  3.63it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  3.11it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:02,  1.92it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.58it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  3.06it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  3.05it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  3.16it/s]
Loading checkpoinInstantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.

Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.19it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.47it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:02,  1.97it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.17it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:03,  1.24it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.28it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.19it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.15it/s]
Loading checkpoin
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:02,  1.70it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  3.01it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.95it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.64it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.66it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.30it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.26it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.96it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:01,  2.66it/s]
L
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.40it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.84it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.79it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.84it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  3.39it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.00it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.30it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.30it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.57it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.64it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.60it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.31it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.76it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.23it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.38it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.13it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.21it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.73it/s]
L
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.03it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:02,  1.96it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:02,  1.83it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:02,  1.54it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:02,  1.70it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  3.09it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:02,  1.80it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:02,  1.68it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:01,  2.99it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.03it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:01,  2.79it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:01,  2.68it/s]
Loading checkpointloading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

     | 1/5 [00:00<00:02,  1.79it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:02,  1.88it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.24it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.13it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.17it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.03it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:01,  2.98it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:01,  2.71it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:01,  2.68it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:01,  2.75it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.59it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.57it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.53it/s]
Loading loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
oading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.17it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.43it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.53it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:01,  2.57it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:01,  2.98it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:01,  2.96it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:01,  2.75it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.69it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.51it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.74it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:01<00:00,  3.14it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.31it/s]
Loading checkpoint shards:  60%|�t shards:  40%|████      | 2/5 [00:00<00:01,  2.99it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:01,  3.00it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:01<00:01,  2.11it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.11it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.13it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:01,  2.90it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.01it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.00it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.49it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.74it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.54it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.64it/s]
Loading checkpoint shards:  60%|██████   0%|██        | 1/5 [00:00<00:01,  2.53it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.73it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.40it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.27it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.43it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.14it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.49it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.18it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.27it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.97it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.68it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.75it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,oading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.47it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.28it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.33it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.40it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.40it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.09it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.72it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:01<00:00,  3.24it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.88it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.74it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  4.20it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.61it/s]
Loading checkpoint shards:  60%|�t shards:  40%|████      | 2/5 [00:00<00:00,  4.01it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.92it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.58it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.34it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:01,  2.77it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.94it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.61it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.61it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  4.15it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  4.11it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  4.08it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  4.32it/s]
Loading checkpoint shards:  60%|██████   
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.16it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  3.54it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  3.37it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  3.26it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  3.04it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  3.06it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  3.08it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  3.17it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  4.02it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:01,  2.85it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.92it/s]
Loading checkpoint shards:  40%|�00,  3.25it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  4.00it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.58it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.30it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.13it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:01,  2.79it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.06it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.67it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.86it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  4.11it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.87it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.54it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.63it/s]
Loading check0%|██        | 1/5 [00:00<00:01,  3.28it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  4.06it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.98it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  4.09it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  4.06it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.63it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.52it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:01,  2.65it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.58it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  4.06it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  4.19it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  4.05it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,oading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.44it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.21it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.41it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.21it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.05it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:01,  2.97it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.01it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.84it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.99it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.71it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.86it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.65it/s]
Loading checkpoint shards:  60%|�0%|██        | 1/5 [00:00<00:01,  2.60it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.18it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.44it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.22it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.30it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.16it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:01,  2.88it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.30it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.01it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.50it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.32it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.60it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00, shards:  40%|████      | 2/5 [00:00<00:01,  2.47it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.69it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:01,  2.74it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:01,  2.59it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.47it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.59it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:01<00:00,  3.41it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.43it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:01<00:00,  3.29it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.92it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:01<00:00,  3.15it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:01<00:00,  2.98it/s]
Loading checkpoint shards:  80%|██
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.17it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.95it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  3.04it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.97it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.97it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  3.14it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.58it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.91it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  3.27it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.72it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.49it/s]
Loading checkpoint shards:checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.58it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.43it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:01<00:00,  3.22it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:01<00:00,  3.20it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:01<00:00,  3.27it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.93it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.80it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.77it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.79it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.69it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.52it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00  3.89it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.79it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.79it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.80it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:01<00:00,  3.14it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:00<00:00,  4.17it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:00<00:00,  4.09it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:00<00:00,  4.10it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  4.03it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.96it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.97it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.48it/s]
Loading checkpoint shards:  80%|█████�  3.78it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.80it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.56it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.57it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.63it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  4.16it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.94it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  4.03it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.82it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.69it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.76it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.62it/s]
Loading checkpoint shards:  80%|█████�:00,  3.51it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.56it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.25it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.78it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.


Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.33it/s]All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.75it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.22it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.69it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
��█████    | 3/5 [00:00<00:00,  3.71it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.70it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.51it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:00<00:00,  4.46it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.61it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  4.08it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.90it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.80it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.86it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.85it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.73it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.74it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.34it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.23it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.73it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.

Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.35it/s]loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
 | 3/5 [00:00<00:00,  3.67it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:01<00:00,  3.25it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.84it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.83it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:00<00:00,  4.27it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:00<00:00,  4.17it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:00<00:00,  4.17it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:00<00:00,  4.55it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.85it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.54it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.95it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.93it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.57it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.27it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
 | 3/5 [00:00<00:00,  3.54it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.37it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:01<00:00,  2.70it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.43it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.81it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  4.05it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  4.07it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.91it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.79it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.64it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.14it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.65it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.23it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.66it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
██████  | 4/5 [00:01<00:00,  3.78it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.75it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.65it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.50it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.57it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.97it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.48it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.37it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.22it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.52it/s]

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.45it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.04it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.50it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.22it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.48it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.27it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.41it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.91it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.44it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.84it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.19it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.65it/s]
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.79it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.45it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
��█████    | 3/5 [00:00<00:00,  3.57it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.47it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:01<00:00,  3.29it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.87it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.71it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  4.01it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.83it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.51it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.62it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.76it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.63it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.39it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.79it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.31it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.75it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
  3.30it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.42it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.33it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.43it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.43it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.68it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.60it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.63it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.54it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.64it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.70it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.68it/s]
Loading checkpoint shards:  80%|█████�
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.99it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.39it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}


Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.17it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.81it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.10it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.65it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.20it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.86it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.31it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.85it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
point shards:  60%|██████    | 3/5 [00:00<00:00,  3.29it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.42it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.92it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  4.05it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  4.06it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:00<00:00,  4.12it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.79it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.81it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.58it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.65it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.31it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.85it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.40it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.04it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.11it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.52it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.42it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.06it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

��█████    | 3/5 [00:00<00:00,  3.38it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.35it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.43it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  4.09it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.98it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  4.03it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  4.06it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.88it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.66it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.61it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.67it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.36it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.87it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.44it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.01it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

�██  | 4/5 [00:01<00:00,  3.73it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.30it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.84it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.44it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.22it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.94it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.50it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}


Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.38it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.95it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.33it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.96it/s]
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.42it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.08it/s]
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.89it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.54it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

�██  | 4/5 [00:01<00:00,  3.86it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.48it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.23it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.03it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.58it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.24it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.83it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
�███      | 2/5 [00:00<00:00,  3.81it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.84it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.63it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.91it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.72it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  4.16it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  4.13it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  4.18it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.76it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:01<00:00,  3.14it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.84it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  4.33it/s]
Loading checkpoint shards:  60%|██████    | 3/
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.73it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.22it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}


Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.03it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.65it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.92it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.52it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}


Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.05it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.75it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.93it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.43it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.86it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.37it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.78it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.32it/s]

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.39it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.20it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.51it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  2.84it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.15it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.96it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.86it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.48it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.


Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.72it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.14it/s]
All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.80it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.27it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.02it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.80it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.96it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.60it/s]
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}


Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.95it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.55it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.00it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.73it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.99it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.64it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.89it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.43it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}


Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.03it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.76it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}


Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.04it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.71it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.12it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.70it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.95it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.77it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.79it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.36it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.81it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.36it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}


Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.67it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.26it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.88it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.85it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}


Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.22it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.09it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
  40%|████      | 2/5 [00:00<00:00,  3.57it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.58it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.70it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.67it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  4.02it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.75it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  4.18it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.72it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.87it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.95it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.80it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  4.43it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.18it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.73it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.


Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  4.08it/s]All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.14it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.95it/s]
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}


Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.12it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.75it/s]
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}


Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.14it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.92it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}


Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.13it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.92it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}


Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.93it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.54it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.12it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.92it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}


Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.08it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.90it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.86it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.53it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}


Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.85it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.55it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.95it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.76it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.74it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.29it/s]

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.01it/s]All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.92it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.92it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.54it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.85it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.47it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
�██  | 4/5 [00:01<00:00,  3.62it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.07it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.67it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.98it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.59it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}


Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.89it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.57it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading file chat_template.jinja from cache at None
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.81it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.43it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.86it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.60it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.90it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.59it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.86it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.63it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.87it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.64it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
5 [00:00<00:00,  3.91it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:00<00:00,  4.24it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:00<00:00,  4.25it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:00<00:00,  4.30it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.88it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:00<00:00,  4.52it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.47it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.97it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.55it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.30it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.56it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.26it/s]
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.58it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.29it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file added_tokens.json from cache at None
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
loading file special_tokens_map.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading file chat_template.jinja from cache at None
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None

Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  4.01it/s]loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file added_tokens.json from cache at None
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file added_tokens.json from cache at None
loading file chat_template.jinja from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.75it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.42it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}


Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.28it/s]loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}


Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.76it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.37it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.16it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.94it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.01it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.90it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json

Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.07it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  3.60it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  3.25it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  3.27it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  3.58it/s]loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}


Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.03it/s]Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json

Loading checkpoint shards:  80%|████████  | 4/5 [00:00<00:00,  4.46it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:00<00:00,  4.64it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  4.02it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  4.06it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.83it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.90it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  4.17it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.70it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.33it/s]
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
loading file chat_template.jinja from cache at None
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.

Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.24it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.02it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.76it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.49it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.18it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.95it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.21it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.01it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.45it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.13it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.96it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.78it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.02it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.85it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None

Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
PyTorch: setting up devices
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json

Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  3.48it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.31it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  3.22it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  4.33it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  4.32it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.88it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  4.24it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.75it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.77it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.49it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  4.64it/s]
Loading checkpoint shards:  60%|██████    | 3/5loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
 [00:00<00:00,  4.64it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  4.27it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  3.16it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  4.58it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.90it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.93it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.75it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.45it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),

Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.27it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.99it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  3.34it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  3.11it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  3.32it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.87it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.94it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  3.73it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.40it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.55it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  4.08it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.85it/s]
Lo	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc

Loading checkpoint shards:  80%|████████  | 4/5 [00:00<00:00,  4.60it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:00<00:00,  4.60it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:00<00:00,  4.76it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:00<00:00,  4.33it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.66it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.96it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.86it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.64it/s]
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}


Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.87it/s]loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.86it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.63it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.96it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.68it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

ading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.75it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.58it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.47it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  4.31it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.67it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.38it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.62it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.33it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc

Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  4.41it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  4.22it/s]loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.

Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.87it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.36it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.08it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.04it/s]loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.91it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}


Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.77it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.96it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.91it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.85it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  4.56it/s]loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}


Loading checkpoint shards:  80%|████████  | 4/5 [00:00<00:00,  4.53it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:00<00:00,  4.35it/s]loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
loading file added_tokens.json from cache at None
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
loading file special_tokens_map.json from cache at None
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

loading file chat_template.jinja from cache at None
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}


Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.89it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  4.18it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  4.16it/s]loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc

Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.85it/s]loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc

Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.97it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:00<00:00,  4.62it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.64it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.42it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.44it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.22it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.36it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.06it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}


Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.28it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.99it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}


Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.02it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.84it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}


Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.08it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.95it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}


Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.83it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.62it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.06it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.89it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None

Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading configuration file config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/config.json
You are using a model of type qwen2_5_vl to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.
loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
Model config LlavaQwenConfig {
  "architectures": [
    "Qwen2_5_VLForConditionalGeneration"
  ],
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151645,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "image_token_id": 151655,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "max_position_embeddings": 128000,
  "max_window_layers": 28,
  "model_type": "llava_qwen",
  "num_attention_heads": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "mrope_section": [
      16,
      24,
      24
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 32768,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0.dev0",
  "use_cache": true,
  "use_sliding_window": false,
  "video_token_id": 151656,
  "vision_config": {
    "hidden_size": 1280,
    "in_chans": 3,
    "model_type": "qwen2_5_vl",
    "spatial_patch_size": 14,
    "tokens_per_second": 2
  },
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vision_token_id": 151654,
  "vocab_size": 152064
}

loading weights file model.safetensors from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/model.safetensors.index.json
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Instantiating LlavaQwenForCausalLM model under default dtype torch.bfloat16.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151645
}

Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.bfloat16.
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.

Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.23it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:01,  2.82it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.92it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  3.03it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  3.12it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  3.24it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  3.11it/s]
Loading checkpoint shardsProcessor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
:  20%|██        | 1/5 [00:00<00:01,  3.06it/s]
Loading checkpoint shards:  20%|██        | 1/5 [00:00<00:01,  2.91it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.85it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  4.04it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.50it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:01<00:00,  3.09it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.85it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.67it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.62it/s]
Loading checkpoint shards:  40%|████      | 2/5 [00:00<00:00,  3.52it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  4.31it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  4.46it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  4.13it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.75it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.88it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.30it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.85it/s]
Loading checkpoint shards:  60%|██████    | 3/5 [00:00<00:00,  3.94it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:00<00:00,  4.58it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:00<00:00,  4.69it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:00<00:00,  4.43it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  4.32it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.88it/s]
Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  4.01it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.59it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.27it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.

Loading checkpoint shards:  80%|████████  | 4/5 [00:01<00:00,  3.95it/s]loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}


Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.84it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.45it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.

Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.94it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.57it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}


Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.73it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.38it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}


Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.66it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.22it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}


Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.15it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.88it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}


Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.15it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.96it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}


Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  4.10it/s]
Loading checkpoint shards: 100%|██████████| 5/5 [00:01<00:00,  3.91it/s]
All model checkpoint weights were used when initializing LlavaQwenForCausalLM.

All the weights of LlavaQwenForCausalLM were initialized from the model checkpoint at Qwen/Qwen2.5-VL-7B-Instruct.
If your task is similar to the task the model of the checkpoint was trained on, you can already use LlavaQwenForCausalLM for predictions without further training.
loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file generation_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/generation_config.json
Generate config GenerationConfig {
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 151643,
  "do_sample": true,
  "eos_token_id": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "repetition_penalty": 1.05,
  "temperature": 0.1,
  "top_k": 1,
  "top_p": 0.001
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading configuration file preprocessor_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/preprocessor_config.json
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
Image processor Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
loading file vocab.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/vocab.json
loading file merges.txt from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/merges.txt
loading file tokenizer.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer.json
loading file added_tokens.json from cache at None
loading file special_tokens_map.json from cache at None
loading file tokenizer_config.json from cache at /fsx_0/user/zhaojiang/models/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/6e6556e8ce728c7b3e438d75ebf04ec93403dc19/tokenizer_config.json
loading file chat_template.jinja from cache at None
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
Processor Qwen2_5_VLProcessor:
- image_processor: Qwen2VLImageProcessor {
  "do_convert_rgb": true,
  "do_normalize": true,
  "do_rescale": true,
  "do_resize": true,
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_processor_type": "Qwen2VLImageProcessor",
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "max_pixels": 12845056,
  "merge_size": 2,
  "min_pixels": 3136,
  "patch_size": 14,
  "processor_class": "Qwen2_5_VLProcessor",
  "resample": 3,
  "rescale_factor": 0.00392156862745098,
  "size": {
    "longest_edge": 12845056,
    "shortest_edge": 3136
  },
  "temporal_patch_size": 2
}

- tokenizer: Qwen2TokenizerFast(name_or_path='Qwen/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={
	151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),
	151657: AddedToken("<tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151658: AddedToken("</tool_call>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
	151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),
}
)

{
  "processor_class": "Qwen2_5_VLProcessor"
}

You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding dimension will be 151668. This might induce some performance reduction as *Tensor Cores* will not be available. For more details about this, or help on choosing the correct value for resizing, refer to this guide: https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank74]: Traceback (most recent call last):
[rank74]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank74]:     train(attn_implementation="flash_attention_2")
[rank74]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank74]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank74]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank74]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank74]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank74]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
[rank74]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank74]:     self.load_model()
[rank74]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank74]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank74]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank74]:     self.model = _build_vision_tower(**self.config)
[rank74]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank74]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank74]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank74]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank74]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank74]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank74]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank74]:     with _open_file_like(f, "rb") as opened_file:
[rank74]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank74]:     return _open_file(name_or_buffer, mode)
[rank74]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank74]:     super().__init__(open(name, mode))
[rank74]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank73]: Traceback (most recent call last):
[rank73]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank73]:     train(attn_implementation="flash_attention_2")
[rank73]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank73]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank73]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank73]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank73]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank73]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank73]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank73]:     self.load_model()
[rank73]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank73]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank73]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank73]:     self.model = _build_vision_tower(**self.config)
[rank73]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank73]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank73]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank73]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank73]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank73]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank73]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank73]:     with _open_file_like(f, "rb") as opened_file:
[rank73]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank73]:     return _open_file(name_or_buffer, mode)
[rank73]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank73]:     super().__init__(open(name, mode))
[rank73]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank77]: Traceback (most recent call last):
[rank77]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank77]:     train(attn_implementation="flash_attention_2")
[rank77]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank77]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank77]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank77]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank77]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank77]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank77]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank77]:     self.load_model()
[rank77]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank77]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank77]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank77]:     self.model = _build_vision_tower(**self.config)
[rank77]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank77]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank77]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank77]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank77]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank77]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank77]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank77]:     with _open_file_like(f, "rb") as opened_file:
[rank77]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank77]:     return _open_file(name_or_buffer, mode)
[rank77]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank77]:     super().__init__(open(name, mode))
[rank77]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank82]: Traceback (most recent call last):
[rank82]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank82]:     train(attn_implementation="flash_attention_2")
[rank82]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank82]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank82]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank82]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank82]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank82]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank82]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank82]:     self.load_model()
[rank82]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank82]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank82]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank82]:     self.model = _build_vision_tower(**self.config)
[rank82]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank82]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank82]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank82]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank82]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank82]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank82]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank82]:     with _open_file_like(f, "rb") as opened_file:
[rank82]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank82]:     return _open_file(name_or_buffer, mode)
[rank82]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank82]:     super().__init__(open(name, mode))
[rank82]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank127]: Traceback (most recent call last):
[rank127]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank127]:     train(attn_implementation="flash_attention_2")
[rank127]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank127]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank127]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank127]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank127]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank127]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank127]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank127]:     self.load_model()
[rank127]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank127]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank127]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank127]:     self.model = _build_vision_tower(**self.config)
[rank127]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank127]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank127]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank127]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank127]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank127]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank127]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank127]:     with _open_file_like(f, "rb") as opened_file:
[rank127]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank127]:     return _open_file(name_or_buffer, mode)
[rank127]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank127]:     super().__init__(open(name, mode))
[rank127]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank17]: Traceback (most recent call last):
[rank17]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank17]:     train(attn_implementation="flash_attention_2")
[rank17]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank17]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank17]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank17]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank17]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank17]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank17]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank17]:     self.load_model()
[rank17]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank17]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank17]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank17]:     self.model = _build_vision_tower(**self.config)
[rank17]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank17]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank17]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank17]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank17]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank17]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank17]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank17]:     with _open_file_like(f, "rb") as opened_file:
[rank17]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank17]:     return _open_file(name_or_buffer, mode)
[rank17]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank17]:     super().__init__(open(name, mode))
[rank17]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank16]: Traceback (most recent call last):
[rank16]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank16]:     train(attn_implementation="flash_attention_2")
[rank16]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank16]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank16]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank16]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank16]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank16]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
[rank16]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank16]:     self.load_model()
[rank16]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank16]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank16]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank16]:     self.model = _build_vision_tower(**self.config)
[rank16]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank16]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank16]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank16]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank16]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank16]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank16]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank16]:     with _open_file_like(f, "rb") as opened_file:
[rank16]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank16]:     return _open_file(name_or_buffer, mode)
[rank16]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank16]:     super().__init__(open(name, mode))
[rank16]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank123]: Traceback (most recent call last):
[rank123]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank123]:     train(attn_implementation="flash_attention_2")
[rank123]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank123]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank123]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank123]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank123]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank123]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank123]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank123]:     self.load_model()
[rank123]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank123]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank123]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank123]:     self.model = _build_vision_tower(**self.config)
[rank123]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank123]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank123]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank123]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank123]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank123]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank123]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank123]:     with _open_file_like(f, "rb") as opened_file:
[rank123]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank123]:     return _open_file(name_or_buffer, mode)
[rank123]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank123]:     super().__init__(open(name, mode))
[rank123]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank46]: Traceback (most recent call last):
[rank46]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank46]:     train(attn_implementation="flash_attention_2")
[rank46]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank46]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank46]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank46]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank46]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank46]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
[rank46]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank46]:     self.load_model()
[rank46]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank46]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank46]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank46]:     self.model = _build_vision_tower(**self.config)
[rank46]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank46]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank46]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank46]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank46]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank46]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank46]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank46]:     with _open_file_like(f, "rb") as opened_file:
[rank46]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank46]:     return _open_file(name_or_buffer, mode)
[rank46]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank46]:     super().__init__(open(name, mode))
[rank46]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank45]: Traceback (most recent call last):
[rank45]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank45]:     train(attn_implementation="flash_attention_2")
[rank45]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank45]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank45]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank45]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank45]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank45]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank45]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank45]:     self.load_model()
[rank45]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank45]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank45]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank45]:     self.model = _build_vision_tower(**self.config)
[rank45]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank45]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank45]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank45]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank45]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank45]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank45]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank45]:     with _open_file_like(f, "rb") as opened_file:
[rank45]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank45]:     return _open_file(name_or_buffer, mode)
[rank45]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank45]:     super().__init__(open(name, mode))
[rank45]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank18]: Traceback (most recent call last):
[rank18]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank18]:     train(attn_implementation="flash_attention_2")
[rank18]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank18]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank18]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank18]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank18]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank18]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank18]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank18]:     self.load_model()
[rank18]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank18]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank18]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank18]:     self.model = _build_vision_tower(**self.config)
[rank18]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank18]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank18]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank18]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank18]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank18]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank18]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank18]:     with _open_file_like(f, "rb") as opened_file:
[rank18]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank18]:     return _open_file(name_or_buffer, mode)
[rank18]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank18]:     super().__init__(open(name, mode))
[rank18]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank36]: Traceback (most recent call last):
[rank36]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank36]:     train(attn_implementation="flash_attention_2")
[rank36]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank36]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank36]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank36]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank36]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank36]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank36]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank36]:     self.load_model()
[rank36]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank36]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank36]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank36]:     self.model = _build_vision_tower(**self.config)
[rank36]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank36]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank36]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank36]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank36]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank36]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank36]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank36]:     with _open_file_like(f, "rb") as opened_file:
[rank36]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank36]:     return _open_file(name_or_buffer, mode)
[rank36]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank36]:     super().__init__(open(name, mode))
[rank36]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank107]: Traceback (most recent call last):
[rank107]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank107]:     train(attn_implementation="flash_attention_2")
[rank107]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank107]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank107]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank107]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank107]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank107]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank107]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank107]:     self.load_model()
[rank107]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank107]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank107]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank107]:     self.model = _build_vision_tower(**self.config)
[rank107]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank107]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank107]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank107]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank107]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank107]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank107]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank107]:     with _open_file_like(f, "rb") as opened_file:
[rank107]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank107]:     return _open_file(name_or_buffer, mode)
[rank107]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank107]:     super().__init__(open(name, mode))
[rank107]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank122]: Traceback (most recent call last):
[rank122]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank122]:     train(attn_implementation="flash_attention_2")
[rank122]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank122]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank122]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank122]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank122]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank122]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
[rank122]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank122]:     self.load_model()
[rank122]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank122]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank122]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank122]:     self.model = _build_vision_tower(**self.config)
[rank122]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank122]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank122]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank122]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank122]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank122]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank122]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank122]:     with _open_file_like(f, "rb") as opened_file:
[rank122]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank122]:     return _open_file(name_or_buffer, mode)
[rank122]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank122]:     super().__init__(open(name, mode))
[rank122]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank81]: Traceback (most recent call last):
[rank81]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank81]:     train(attn_implementation="flash_attention_2")
[rank81]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank81]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank81]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank81]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank81]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank81]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank81]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank81]:     self.load_model()
[rank81]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank81]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank81]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank81]:     self.model = _build_vision_tower(**self.config)
[rank81]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank81]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank81]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank81]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank81]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank81]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank81]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank81]:     with _open_file_like(f, "rb") as opened_file:
[rank81]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank81]:     return _open_file(name_or_buffer, mode)
[rank81]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank81]:     super().__init__(open(name, mode))
[rank81]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank47]: Traceback (most recent call last):
[rank47]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank47]:     train(attn_implementation="flash_attention_2")
[rank47]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank47]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank47]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank47]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank47]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank47]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank47]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank47]:     self.load_model()
[rank47]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank47]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank47]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank47]:     self.model = _build_vision_tower(**self.config)
[rank47]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank47]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank47]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank47]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank47]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank47]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank47]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank47]:     with _open_file_like(f, "rb") as opened_file:
[rank47]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank47]:     return _open_file(name_or_buffer, mode)
[rank47]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank47]:     super().__init__(open(name, mode))
[rank47]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank19]: Traceback (most recent call last):
[rank19]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank19]:     train(attn_implementation="flash_attention_2")
[rank19]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank19]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank19]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank19]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank19]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank19]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank19]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank19]:     self.load_model()
[rank19]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank19]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank19]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank19]:     self.model = _build_vision_tower(**self.config)
[rank19]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank19]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
[rank19]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank19]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank19]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank19]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank19]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank19]:     with _open_file_like(f, "rb") as opened_file:
[rank19]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank19]:     return _open_file(name_or_buffer, mode)
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank19]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank19]:     super().__init__(open(name, mode))
[rank19]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank23]: Traceback (most recent call last):
[rank23]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank23]:     train(attn_implementation="flash_attention_2")
[rank23]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank23]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank23]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank23]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank23]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank23]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank23]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank23]:     self.load_model()
[rank23]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank23]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank23]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank23]:     self.model = _build_vision_tower(**self.config)
[rank23]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank23]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank23]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank23]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank23]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank23]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank23]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank23]:     with _open_file_like(f, "rb") as opened_file:
[rank23]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank23]:     return _open_file(name_or_buffer, mode)
[rank23]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank23]:     super().__init__(open(name, mode))
[rank23]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank108]: Traceback (most recent call last):
[rank108]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank108]:     train(attn_implementation="flash_attention_2")
[rank108]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank108]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank108]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank108]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank108]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank108]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank108]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank108]:     self.load_model()
[rank108]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank108]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank108]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank108]:     self.model = _build_vision_tower(**self.config)
[rank108]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank108]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank108]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank108]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank108]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank108]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank108]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank108]:     with _open_file_like(f, "rb") as opened_file:
[rank108]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank108]:     return _open_file(name_or_buffer, mode)
[rank108]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank108]:     super().__init__(open(name, mode))
[rank108]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank21]: Traceback (most recent call last):
[rank21]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank21]:     train(attn_implementation="flash_attention_2")
[rank21]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank21]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank21]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank21]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank21]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank21]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank21]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank21]:     self.load_model()
[rank21]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank21]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank21]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank21]:     self.model = _build_vision_tower(**self.config)
[rank21]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank21]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank21]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank21]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank21]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank21]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank21]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank21]:     with _open_file_like(f, "rb") as opened_file:
[rank21]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank21]:     return _open_file(name_or_buffer, mode)
[rank21]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank21]:     super().__init__(open(name, mode))
[rank21]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank104]: Traceback (most recent call last):
[rank104]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank104]:     train(attn_implementation="flash_attention_2")
[rank104]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank104]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank104]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank104]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank104]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank104]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank104]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank104]:     self.load_model()
[rank104]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank104]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank104]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank104]:     self.model = _build_vision_tower(**self.config)
[rank104]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank104]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank104]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank104]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank104]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank104]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank104]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank104]:     with _open_file_like(f, "rb") as opened_file:
[rank104]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank104]:     return _open_file(name_or_buffer, mode)
[rank104]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank104]:     super().__init__(open(name, mode))
[rank104]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank109]: Traceback (most recent call last):
[rank109]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank109]:     train(attn_implementation="flash_attention_2")
[rank109]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank109]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank109]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank109]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank109]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank109]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank109]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank109]:     self.load_model()
[rank109]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank109]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank109]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank109]:     self.model = _build_vision_tower(**self.config)
[rank109]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank109]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank109]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank109]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank109]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank109]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank109]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank109]:     with _open_file_like(f, "rb") as opened_file:
[rank109]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank109]:     return _open_file(name_or_buffer, mode)
[rank109]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank109]:     super().__init__(open(name, mode))
[rank109]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank110]: Traceback (most recent call last):
[rank110]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank110]:     train(attn_implementation="flash_attention_2")
[rank110]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank110]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank110]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank110]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank110]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank110]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank110]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank110]:     self.load_model()
[rank110]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank110]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank110]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank110]:     self.model = _build_vision_tower(**self.config)
[rank110]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank110]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank20]: Traceback (most recent call last):
[rank20]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank20]:     train(attn_implementation="flash_attention_2")
[rank20]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank20]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank20]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank20]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank20]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank20]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank110]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank110]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank110]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank110]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank110]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank110]:     with _open_file_like(f, "rb") as opened_file:
[rank110]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank110]:     return _open_file(name_or_buffer, mode)
[rank20]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank20]:     self.load_model()
[rank20]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank20]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank20]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank20]:     self.model = _build_vision_tower(**self.config)
[rank20]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank20]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank110]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank110]:     super().__init__(open(name, mode))
[rank110]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank66]: Traceback (most recent call last):
[rank66]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank66]:     train(attn_implementation="flash_attention_2")
[rank66]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank66]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank66]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank66]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank66]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank66]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank20]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank20]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank20]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank20]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank20]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank20]:     with _open_file_like(f, "rb") as opened_file:
[rank20]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank20]:     return _open_file(name_or_buffer, mode)
[rank66]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank66]:     self.load_model()
[rank66]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank66]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank66]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank66]:     self.model = _build_vision_tower(**self.config)
[rank66]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank66]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank20]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank20]:     super().__init__(open(name, mode))
[rank20]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank66]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank66]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank66]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank66]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank66]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank66]:     with _open_file_like(f, "rb") as opened_file:
[rank66]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank66]:     return _open_file(name_or_buffer, mode)
[rank66]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank66]:     super().__init__(open(name, mode))
[rank66]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank87]: Traceback (most recent call last):
[rank87]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank87]:     train(attn_implementation="flash_attention_2")
[rank87]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank87]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank87]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank87]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank87]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank87]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank87]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank87]:     self.load_model()
[rank87]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank87]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank87]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank87]:     self.model = _build_vision_tower(**self.config)
[rank87]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank87]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank87]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank87]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank87]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank87]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank87]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank87]:     with _open_file_like(f, "rb") as opened_file:
[rank87]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank87]:     return _open_file(name_or_buffer, mode)
[rank87]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank87]:     super().__init__(open(name, mode))
[rank87]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank79]: Traceback (most recent call last):
[rank79]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank79]:     train(attn_implementation="flash_attention_2")
[rank79]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank79]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank79]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank79]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank79]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank79]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank79]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank79]:     self.load_model()
[rank79]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank79]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank79]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank79]:     self.model = _build_vision_tower(**self.config)
[rank79]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank79]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank79]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank79]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank79]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank79]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank79]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank79]:     with _open_file_like(f, "rb") as opened_file:
[rank79]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank79]:     return _open_file(name_or_buffer, mode)
[rank79]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank79]:     super().__init__(open(name, mode))
[rank79]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank119]: Traceback (most recent call last):
[rank119]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank119]:     train(attn_implementation="flash_attention_2")
[rank119]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank119]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank119]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank119]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank119]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank119]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank119]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank119]:     self.load_model()
[rank119]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank119]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank119]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank119]:     self.model = _build_vision_tower(**self.config)
[rank119]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank119]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank119]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank119]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank119]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank119]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank119]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank119]:     with _open_file_like(f, "rb") as opened_file:
[rank119]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank119]:     return _open_file(name_or_buffer, mode)
[rank119]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank119]:     super().__init__(open(name, mode))
[rank119]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank32]: Traceback (most recent call last):
[rank32]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank32]:     train(attn_implementation="flash_attention_2")
[rank32]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank32]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank32]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank32]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank32]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank32]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank32]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank32]:     self.load_model()
[rank32]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank32]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank32]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank32]:     self.model = _build_vision_tower(**self.config)
[rank32]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank32]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank32]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank32]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank32]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank32]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank32]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank32]:     with _open_file_like(f, "rb") as opened_file:
[rank32]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank32]:     return _open_file(name_or_buffer, mode)
[rank32]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank32]:     super().__init__(open(name, mode))
[rank32]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank84]: Traceback (most recent call last):
[rank84]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank84]:     train(attn_implementation="flash_attention_2")
[rank84]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank84]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank84]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank84]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank84]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank84]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank84]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank84]:     self.load_model()
[rank84]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank84]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank84]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank84]:     self.model = _build_vision_tower(**self.config)
[rank84]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank84]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank84]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank84]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank84]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank84]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank84]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank84]:     with _open_file_like(f, "rb") as opened_file:
[rank84]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank84]:     return _open_file(name_or_buffer, mode)
[rank84]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank84]:     super().__init__(open(name, mode))
[rank84]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank85]: Traceback (most recent call last):
[rank85]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank85]:     train(attn_implementation="flash_attention_2")
[rank85]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank85]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank85]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank85]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank85]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank85]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank85]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank85]:     self.load_model()
[rank85]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank85]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank85]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank85]:     self.model = _build_vision_tower(**self.config)
[rank85]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank85]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank85]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank85]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank85]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank85]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank85]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank85]:     with _open_file_like(f, "rb") as opened_file:
[rank85]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank85]:     return _open_file(name_or_buffer, mode)
[rank85]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank85]:     super().__init__(open(name, mode))
[rank85]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank41]: Traceback (most recent call last):
[rank41]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank41]:     train(attn_implementation="flash_attention_2")
[rank41]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank41]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank41]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank41]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank41]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank41]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank41]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank41]:     self.load_model()
[rank41]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank41]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank41]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank41]:     self.model = _build_vision_tower(**self.config)
[rank41]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank41]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank41]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank41]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank41]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank41]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank41]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank41]:     with _open_file_like(f, "rb") as opened_file:
[rank41]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank41]:     return _open_file(name_or_buffer, mode)
[rank41]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank41]:     super().__init__(open(name, mode))
[rank41]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank44]: Traceback (most recent call last):
[rank44]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank44]:     train(attn_implementation="flash_attention_2")
[rank44]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank44]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank44]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank44]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank44]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank44]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank44]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank44]:     self.load_model()
[rank44]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank44]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank44]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank44]:     self.model = _build_vision_tower(**self.config)
[rank44]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank44]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank44]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank44]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank44]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank44]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank44]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank44]:     with _open_file_like(f, "rb") as opened_file:
[rank44]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank44]:     return _open_file(name_or_buffer, mode)
[rank44]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank44]:     super().__init__(open(name, mode))
[rank44]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank88]: Traceback (most recent call last):
[rank88]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank88]:     train(attn_implementation="flash_attention_2")
[rank88]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank88]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank88]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank88]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank88]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank88]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank126]: Traceback (most recent call last):
[rank126]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank126]:     train(attn_implementation="flash_attention_2")
[rank126]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank126]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank126]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank126]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank126]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank126]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank88]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank88]:     self.load_model()
[rank88]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank88]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank88]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank88]:     self.model = _build_vision_tower(**self.config)
[rank88]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank88]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank126]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank126]:     self.load_model()
[rank126]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank126]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank126]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank126]:     self.model = _build_vision_tower(**self.config)
[rank126]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank126]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank88]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank88]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank88]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank88]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank88]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank88]:     with _open_file_like(f, "rb") as opened_file:
[rank88]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank88]:     return _open_file(name_or_buffer, mode)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
[rank126]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank126]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank126]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank126]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank126]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank126]:     with _open_file_like(f, "rb") as opened_file:
[rank126]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank126]:     return _open_file(name_or_buffer, mode)
[rank88]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank88]:     super().__init__(open(name, mode))
[rank88]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank94]: Traceback (most recent call last):
[rank94]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank94]:     train(attn_implementation="flash_attention_2")
[rank94]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank94]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank94]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank126]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank126]:     super().__init__(open(name, mode))
[rank126]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank94]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank94]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank94]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank94]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank94]:     self.load_model()
[rank94]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank94]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank94]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank94]:     self.model = _build_vision_tower(**self.config)
[rank94]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank94]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank94]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank94]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank94]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank94]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank94]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank94]:     with _open_file_like(f, "rb") as opened_file:
[rank94]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank94]:     return _open_file(name_or_buffer, mode)
[rank94]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank94]:     super().__init__(open(name, mode))
[rank94]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank91]: Traceback (most recent call last):
[rank91]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank91]:     train(attn_implementation="flash_attention_2")
[rank91]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank91]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank91]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank91]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank91]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank91]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank91]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank91]:     self.load_model()
[rank91]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank91]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank91]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank91]:     self.model = _build_vision_tower(**self.config)
[rank91]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank91]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank91]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank91]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank91]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank91]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank91]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank91]:     with _open_file_like(f, "rb") as opened_file:
[rank91]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank91]:     return _open_file(name_or_buffer, mode)
[rank91]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank91]:     super().__init__(open(name, mode))
[rank91]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank64]: Traceback (most recent call last):
[rank64]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank64]:     train(attn_implementation="flash_attention_2")
[rank64]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank64]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank64]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank64]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank64]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank64]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank64]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank64]:     self.load_model()
[rank64]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank64]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank64]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank64]:     self.model = _build_vision_tower(**self.config)
[rank64]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank64]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
[rank64]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank64]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank64]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank64]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank64]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank64]:     with _open_file_like(f, "rb") as opened_file:
[rank64]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank64]:     return _open_file(name_or_buffer, mode)
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank64]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank64]:     super().__init__(open(name, mode))
[rank64]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank71]: Traceback (most recent call last):
[rank71]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank71]:     train(attn_implementation="flash_attention_2")
[rank71]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank71]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank71]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank71]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank71]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank71]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank71]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank71]:     self.load_model()
[rank71]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank71]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank71]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank71]:     self.model = _build_vision_tower(**self.config)
[rank71]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank71]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank71]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank71]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank71]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank71]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank71]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank71]:     with _open_file_like(f, "rb") as opened_file:
[rank71]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank71]:     return _open_file(name_or_buffer, mode)
[rank71]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank71]:     super().__init__(open(name, mode))
[rank71]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank113]: Traceback (most recent call last):
[rank113]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank113]:     train(attn_implementation="flash_attention_2")
[rank113]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank113]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank113]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank113]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank113]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank113]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank113]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank113]:     self.load_model()
[rank113]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank113]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank113]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank113]:     self.model = _build_vision_tower(**self.config)
[rank113]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank113]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
[rank113]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank113]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank113]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank113]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank113]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank113]:     with _open_file_like(f, "rb") as opened_file:
[rank113]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank113]:     return _open_file(name_or_buffer, mode)
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank113]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank113]:     super().__init__(open(name, mode))
[rank113]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank92]: Traceback (most recent call last):
[rank92]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank92]:     train(attn_implementation="flash_attention_2")
[rank92]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank92]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank92]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank92]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank92]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank92]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank92]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank92]:     self.load_model()
[rank92]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank92]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank92]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank92]:     self.model = _build_vision_tower(**self.config)
[rank92]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank92]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank92]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank92]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank92]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank92]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank92]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank92]:     with _open_file_like(f, "rb") as opened_file:
[rank92]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank92]:     return _open_file(name_or_buffer, mode)
[rank92]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank92]:     super().__init__(open(name, mode))
[rank92]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank51]: Traceback (most recent call last):
[rank51]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank51]:     train(attn_implementation="flash_attention_2")
[rank51]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank51]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank51]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank51]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank51]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank51]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank51]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank51]:     self.load_model()
[rank51]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank51]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank51]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank51]:     self.model = _build_vision_tower(**self.config)
[rank51]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank51]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank51]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank51]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank51]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank51]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank51]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank51]:     with _open_file_like(f, "rb") as opened_file:
[rank51]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank51]:     return _open_file(name_or_buffer, mode)
[rank51]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank51]:     super().__init__(open(name, mode))
[rank51]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank52]: Traceback (most recent call last):
[rank52]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank52]:     train(attn_implementation="flash_attention_2")
[rank52]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank52]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank52]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank52]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank52]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank52]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank52]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank52]:     self.load_model()
[rank52]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank52]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank52]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank52]:     self.model = _build_vision_tower(**self.config)
[rank52]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank52]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank52]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank52]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank52]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank52]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank52]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank52]:     with _open_file_like(f, "rb") as opened_file:
[rank52]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank52]:     return _open_file(name_or_buffer, mode)
[rank52]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank52]:     super().__init__(open(name, mode))
[rank52]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank115]: Traceback (most recent call last):
[rank115]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank115]:     train(attn_implementation="flash_attention_2")
[rank115]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank115]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank115]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank115]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank115]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank115]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank115]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank115]:     self.load_model()
[rank115]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank115]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank115]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank115]:     self.model = _build_vision_tower(**self.config)
[rank115]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank115]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank115]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank115]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank115]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank115]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank115]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank115]:     with _open_file_like(f, "rb") as opened_file:
[rank115]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank115]:     return _open_file(name_or_buffer, mode)
[rank115]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank115]:     super().__init__(open(name, mode))
[rank115]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank112]: Traceback (most recent call last):
[rank112]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank112]:     train(attn_implementation="flash_attention_2")
[rank112]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank112]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank112]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank112]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank112]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank112]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank112]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank112]:     self.load_model()
[rank112]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank112]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank112]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank112]:     self.model = _build_vision_tower(**self.config)
[rank112]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank112]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank112]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank112]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank112]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank112]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank112]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank112]:     with _open_file_like(f, "rb") as opened_file:
[rank112]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank112]:     return _open_file(name_or_buffer, mode)
[rank112]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank112]:     super().__init__(open(name, mode))
[rank112]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank14]: Traceback (most recent call last):
[rank14]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank14]:     train(attn_implementation="flash_attention_2")
[rank14]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank14]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank14]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank14]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank14]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank14]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank14]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank14]:     self.load_model()
[rank14]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank14]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank14]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank14]:     self.model = _build_vision_tower(**self.config)
[rank14]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank14]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank14]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank14]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank14]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank14]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank14]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank14]:     with _open_file_like(f, "rb") as opened_file:
[rank14]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank14]:     return _open_file(name_or_buffer, mode)
[rank14]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank14]:     super().__init__(open(name, mode))
[rank14]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank78]: Traceback (most recent call last):
[rank78]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank78]:     train(attn_implementation="flash_attention_2")
[rank78]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank78]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank78]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank78]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank78]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank78]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank78]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank78]:     self.load_model()
[rank78]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank78]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank78]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank78]:     self.model = _build_vision_tower(**self.config)
[rank78]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank78]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank78]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank78]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank78]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank78]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank78]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank78]:     with _open_file_like(f, "rb") as opened_file:
[rank78]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank78]:     return _open_file(name_or_buffer, mode)
[rank78]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank78]:     super().__init__(open(name, mode))
[rank78]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank80]: Traceback (most recent call last):
[rank80]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank80]:     train(attn_implementation="flash_attention_2")
[rank80]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank80]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank80]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank80]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank80]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank80]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank80]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank80]:     self.load_model()
[rank80]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank80]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank80]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank80]:     self.model = _build_vision_tower(**self.config)
[rank80]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank80]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank80]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank80]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank80]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank80]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank80]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank80]:     with _open_file_like(f, "rb") as opened_file:
[rank80]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank80]:     return _open_file(name_or_buffer, mode)
[rank80]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank80]:     super().__init__(open(name, mode))
[rank80]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank58]: Traceback (most recent call last):
[rank58]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank58]:     train(attn_implementation="flash_attention_2")
[rank58]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank58]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank58]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank58]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank58]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank58]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank58]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank58]:     self.load_model()
[rank58]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank58]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank58]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank58]:     self.model = _build_vision_tower(**self.config)
[rank58]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank58]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank58]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank58]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank58]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank58]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank58]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank58]:     with _open_file_like(f, "rb") as opened_file:
[rank58]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank58]:     return _open_file(name_or_buffer, mode)
[rank58]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank58]:     super().__init__(open(name, mode))
[rank58]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank56]: Traceback (most recent call last):
[rank56]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank56]:     train(attn_implementation="flash_attention_2")
[rank56]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank56]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank56]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank56]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank56]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank56]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank56]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank56]:     self.load_model()
[rank56]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank56]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank56]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank56]:     self.model = _build_vision_tower(**self.config)
[rank56]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank56]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank56]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank56]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank56]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank56]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank56]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank56]:     with _open_file_like(f, "rb") as opened_file:
[rank56]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank56]:     return _open_file(name_or_buffer, mode)
[rank56]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank56]:     super().__init__(open(name, mode))
[rank56]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank53]: Traceback (most recent call last):
[rank53]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank53]:     train(attn_implementation="flash_attention_2")
[rank53]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank53]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank53]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank53]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank53]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank53]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank53]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank53]:     self.load_model()
[rank53]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank53]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank53]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank53]:     self.model = _build_vision_tower(**self.config)
[rank53]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank53]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank53]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank53]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank53]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank53]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank53]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank53]:     with _open_file_like(f, "rb") as opened_file:
[rank53]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank53]:     return _open_file(name_or_buffer, mode)
[rank53]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank53]:     super().__init__(open(name, mode))
[rank53]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank116]: Traceback (most recent call last):
[rank116]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank116]:     train(attn_implementation="flash_attention_2")
[rank116]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank116]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank116]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank116]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank116]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank116]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank116]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank116]:     self.load_model()
[rank116]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank116]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank116]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank116]:     self.model = _build_vision_tower(**self.config)
[rank116]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank116]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank116]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank116]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank116]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank116]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank116]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank116]:     with _open_file_like(f, "rb") as opened_file:
[rank116]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank116]:     return _open_file(name_or_buffer, mode)
[rank116]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank116]:     super().__init__(open(name, mode))
[rank116]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank50]: Traceback (most recent call last):
[rank50]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank50]:     train(attn_implementation="flash_attention_2")
[rank50]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank50]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank50]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank50]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank50]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank50]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank50]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank50]:     self.load_model()
[rank50]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank50]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank50]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank50]:     self.model = _build_vision_tower(**self.config)
[rank50]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank50]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank50]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank50]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank50]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank50]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank50]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank50]:     with _open_file_like(f, "rb") as opened_file:
[rank50]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank50]:     return _open_file(name_or_buffer, mode)
[rank50]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank50]:     super().__init__(open(name, mode))
[rank50]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank120]: Traceback (most recent call last):
[rank120]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank120]:     train(attn_implementation="flash_attention_2")
[rank120]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank120]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank120]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank120]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank120]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank120]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank120]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank120]:     self.load_model()
[rank120]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank120]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank120]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank120]:     self.model = _build_vision_tower(**self.config)
[rank120]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank120]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank120]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank120]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank120]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank120]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank120]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank120]:     with _open_file_like(f, "rb") as opened_file:
[rank120]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank120]:     return _open_file(name_or_buffer, mode)
[rank120]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank120]:     super().__init__(open(name, mode))
[rank120]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank72]: Traceback (most recent call last):
[rank72]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank72]:     train(attn_implementation="flash_attention_2")
[rank72]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank72]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank72]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank72]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank72]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank72]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank72]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank72]:     self.load_model()
[rank72]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank72]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank72]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank72]:     self.model = _build_vision_tower(**self.config)
[rank72]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank72]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank72]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank72]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank72]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank72]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank72]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank72]:     with _open_file_like(f, "rb") as opened_file:
[rank72]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank72]:     return _open_file(name_or_buffer, mode)
[rank72]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank72]:     super().__init__(open(name, mode))
[rank72]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank65]: Traceback (most recent call last):
[rank65]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank65]:     train(attn_implementation="flash_attention_2")
[rank65]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank65]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank65]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank65]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank65]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank65]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank65]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank65]:     self.load_model()
[rank65]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank65]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank65]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank65]:     self.model = _build_vision_tower(**self.config)
[rank65]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank65]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank65]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank65]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank65]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank65]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank65]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank65]:     with _open_file_like(f, "rb") as opened_file:
[rank65]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank65]:     return _open_file(name_or_buffer, mode)
[rank65]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank65]:     super().__init__(open(name, mode))
[rank65]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank100]: Traceback (most recent call last):
[rank100]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank100]:     train(attn_implementation="flash_attention_2")
[rank100]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank100]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank100]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank100]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank100]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank100]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank100]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank100]:     self.load_model()
[rank100]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank100]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank100]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank100]:     self.model = _build_vision_tower(**self.config)
[rank100]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank100]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank100]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank100]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank100]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank100]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank100]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank100]:     with _open_file_like(f, "rb") as opened_file:
[rank100]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank100]:     return _open_file(name_or_buffer, mode)
[rank100]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank100]:     super().__init__(open(name, mode))
[rank100]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank96]: Traceback (most recent call last):
[rank96]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank96]:     train(attn_implementation="flash_attention_2")
[rank96]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank96]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank96]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank96]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank96]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank96]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank96]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank96]:     self.load_model()
[rank96]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank96]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank96]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank96]:     self.model = _build_vision_tower(**self.config)
[rank96]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank96]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank96]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank96]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank96]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank96]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank96]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank96]:     with _open_file_like(f, "rb") as opened_file:
[rank96]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank96]:     return _open_file(name_or_buffer, mode)
[rank96]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank96]:     super().__init__(open(name, mode))
[rank96]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
[rank103]: Traceback (most recent call last):
[rank103]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank103]:     train(attn_implementation="flash_attention_2")
[rank103]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank103]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank103]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank103]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank103]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank103]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank103]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank103]:     self.load_model()
[rank103]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank103]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank103]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank103]:     self.model = _build_vision_tower(**self.config)
[rank103]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank103]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank103]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank103]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank103]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank103]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank103]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank103]:     with _open_file_like(f, "rb") as opened_file:
[rank103]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank103]:     return _open_file(name_or_buffer, mode)
[rank103]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank103]:     super().__init__(open(name, mode))
[rank103]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank42]: Traceback (most recent call last):
[rank42]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank42]:     train(attn_implementation="flash_attention_2")
[rank42]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank42]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank42]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank42]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank42]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank42]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank42]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank42]:     self.load_model()
[rank42]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank42]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank42]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank42]:     self.model = _build_vision_tower(**self.config)
[rank42]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank42]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank42]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank42]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank42]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank42]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank42]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank42]:     with _open_file_like(f, "rb") as opened_file:
[rank42]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank42]:     return _open_file(name_or_buffer, mode)
[rank42]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank42]:     super().__init__(open(name, mode))
[rank42]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank111]: Traceback (most recent call last):
[rank111]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank111]:     train(attn_implementation="flash_attention_2")
[rank111]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank111]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank111]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank111]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank111]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank111]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank111]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank111]:     self.load_model()
[rank111]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank111]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank111]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank111]:     self.model = _build_vision_tower(**self.config)
[rank111]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank111]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank111]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank111]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank111]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank111]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank111]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank111]:     with _open_file_like(f, "rb") as opened_file:
[rank111]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank111]:     return _open_file(name_or_buffer, mode)
[rank111]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank111]:     super().__init__(open(name, mode))
[rank111]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank70]: Traceback (most recent call last):
[rank70]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank70]:     train(attn_implementation="flash_attention_2")
[rank70]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank70]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank70]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank70]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank70]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank70]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank70]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank70]:     self.load_model()
[rank70]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank70]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank70]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank70]:     self.model = _build_vision_tower(**self.config)
[rank70]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank70]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank70]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank70]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank70]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank70]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank70]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank70]:     with _open_file_like(f, "rb") as opened_file:
[rank70]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank70]:     return _open_file(name_or_buffer, mode)
[rank70]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank70]:     super().__init__(open(name, mode))
[rank70]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank60]: Traceback (most recent call last):
[rank60]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank60]:     train(attn_implementation="flash_attention_2")
[rank60]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank60]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank60]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank60]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank60]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank60]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank60]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank60]:     self.load_model()
[rank60]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank60]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank60]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank60]:     self.model = _build_vision_tower(**self.config)
[rank60]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank60]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank60]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank60]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank60]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank60]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank60]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank60]:     with _open_file_like(f, "rb") as opened_file:
[rank60]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank60]:     return _open_file(name_or_buffer, mode)
[rank60]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank60]:     super().__init__(open(name, mode))
[rank60]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank105]: Traceback (most recent call last):
[rank105]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank105]:     train(attn_implementation="flash_attention_2")
[rank105]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank105]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank105]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank105]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank105]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank105]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank105]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank105]:     self.load_model()
[rank105]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank105]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank105]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank105]:     self.model = _build_vision_tower(**self.config)
[rank105]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank105]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank105]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank105]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank105]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank105]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank105]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank105]:     with _open_file_like(f, "rb") as opened_file:
[rank105]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank105]:     return _open_file(name_or_buffer, mode)
[rank105]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank105]:     super().__init__(open(name, mode))
[rank105]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank49]: Traceback (most recent call last):
[rank49]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank49]:     train(attn_implementation="flash_attention_2")
[rank49]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank49]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank49]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank49]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank49]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank49]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank49]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank49]:     self.load_model()
[rank49]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank49]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank49]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank49]:     self.model = _build_vision_tower(**self.config)
[rank49]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank49]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank49]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank49]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank49]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank49]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank49]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank49]:     with _open_file_like(f, "rb") as opened_file:
[rank49]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank49]:     return _open_file(name_or_buffer, mode)
[rank49]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank49]:     super().__init__(open(name, mode))
[rank49]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank33]: Traceback (most recent call last):
[rank33]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank33]:     train(attn_implementation="flash_attention_2")
[rank33]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank33]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank33]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank33]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank33]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank33]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank33]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank33]:     self.load_model()
[rank33]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank33]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank33]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank33]:     self.model = _build_vision_tower(**self.config)
[rank33]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank33]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank33]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank33]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank33]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank33]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank33]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank33]:     with _open_file_like(f, "rb") as opened_file:
[rank33]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank33]:     return _open_file(name_or_buffer, mode)
[rank33]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank33]:     super().__init__(open(name, mode))
[rank33]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank37]: Traceback (most recent call last):
[rank37]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank37]:     train(attn_implementation="flash_attention_2")
[rank37]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank37]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank37]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank37]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank37]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank37]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank37]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank37]:     self.load_model()
[rank37]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank37]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank37]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank37]:     self.model = _build_vision_tower(**self.config)
[rank37]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank37]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank37]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank37]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank37]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank37]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank37]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank37]:     with _open_file_like(f, "rb") as opened_file:
[rank37]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank37]:     return _open_file(name_or_buffer, mode)
[rank37]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank37]:     super().__init__(open(name, mode))
[rank37]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank98]: Traceback (most recent call last):
[rank98]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank98]:     train(attn_implementation="flash_attention_2")
[rank98]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank98]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank98]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank98]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank98]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank98]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank98]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank98]:     self.load_model()
[rank98]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank98]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank98]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank98]:     self.model = _build_vision_tower(**self.config)
[rank98]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank98]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank98]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank98]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank98]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank98]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank98]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank98]:     with _open_file_like(f, "rb") as opened_file:
[rank98]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank98]:     return _open_file(name_or_buffer, mode)
[rank98]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank98]:     super().__init__(open(name, mode))
[rank98]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank75]: Traceback (most recent call last):
[rank75]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank75]:     train(attn_implementation="flash_attention_2")
[rank75]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank75]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank75]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank75]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank75]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank75]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank75]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank75]:     self.load_model()
[rank75]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank75]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank75]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank75]:     self.model = _build_vision_tower(**self.config)
[rank75]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank75]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank75]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank75]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank75]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank75]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank75]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank75]:     with _open_file_like(f, "rb") as opened_file:
[rank75]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank75]:     return _open_file(name_or_buffer, mode)
[rank75]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank75]:     super().__init__(open(name, mode))
[rank75]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank86]: Traceback (most recent call last):
[rank86]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank86]:     train(attn_implementation="flash_attention_2")
[rank86]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank86]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank86]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank86]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank86]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank86]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank86]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank86]:     self.load_model()
[rank86]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank86]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank86]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank86]:     self.model = _build_vision_tower(**self.config)
[rank86]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank86]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank86]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank86]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank86]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank86]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank86]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank86]:     with _open_file_like(f, "rb") as opened_file:
[rank86]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank86]:     return _open_file(name_or_buffer, mode)
[rank86]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank86]:     super().__init__(open(name, mode))
[rank86]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank95]: Traceback (most recent call last):
[rank95]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank95]:     train(attn_implementation="flash_attention_2")
[rank95]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank95]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank95]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank95]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank95]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank95]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank95]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank95]:     self.load_model()
[rank95]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank95]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank95]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank95]:     self.model = _build_vision_tower(**self.config)
[rank95]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank95]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank95]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank95]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank95]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank95]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank95]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank95]:     with _open_file_like(f, "rb") as opened_file:
[rank95]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank95]:     return _open_file(name_or_buffer, mode)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
[rank95]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank95]:     super().__init__(open(name, mode))
[rank95]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
[rank54]: Traceback (most recent call last):
[rank54]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank54]:     train(attn_implementation="flash_attention_2")
[rank54]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank54]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank54]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank54]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank54]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank54]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank54]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank54]:     self.load_model()
[rank54]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank54]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank54]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank54]:     self.model = _build_vision_tower(**self.config)
[rank54]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank54]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank54]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank54]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank54]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank54]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank54]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank54]:     with _open_file_like(f, "rb") as opened_file:
[rank54]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank54]:     return _open_file(name_or_buffer, mode)
[rank54]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank54]:     super().__init__(open(name, mode))
[rank54]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank102]: Traceback (most recent call last):
[rank102]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank102]:     train(attn_implementation="flash_attention_2")
[rank102]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank102]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank102]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank102]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank102]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank102]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank102]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank102]:     self.load_model()
[rank102]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank102]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank102]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank102]:     self.model = _build_vision_tower(**self.config)
[rank102]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank102]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank102]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank102]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank102]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank102]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank102]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank102]:     with _open_file_like(f, "rb") as opened_file:
[rank102]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank102]:     return _open_file(name_or_buffer, mode)
[rank102]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank102]:     super().__init__(open(name, mode))
[rank102]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank9]: Traceback (most recent call last):
[rank9]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank9]:     train(attn_implementation="flash_attention_2")
[rank9]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank9]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank9]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank9]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank9]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank9]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank9]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank9]:     self.load_model()
[rank9]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank9]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank9]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank9]:     self.model = _build_vision_tower(**self.config)
[rank9]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank9]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank9]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank9]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank9]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank9]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank9]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank9]:     with _open_file_like(f, "rb") as opened_file:
[rank9]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank9]:     return _open_file(name_or_buffer, mode)
[rank9]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank9]:     super().__init__(open(name, mode))
[rank9]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank121]: Traceback (most recent call last):
[rank121]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank121]:     train(attn_implementation="flash_attention_2")
[rank121]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank121]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank121]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank121]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank121]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank121]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank121]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank121]:     self.load_model()
[rank121]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank121]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank121]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank121]:     self.model = _build_vision_tower(**self.config)
[rank121]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank121]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank121]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank121]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank121]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank121]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank121]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank121]:     with _open_file_like(f, "rb") as opened_file:
[rank121]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank121]:     return _open_file(name_or_buffer, mode)
[rank121]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank121]:     super().__init__(open(name, mode))
[rank121]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank68]: Traceback (most recent call last):
[rank68]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank68]:     train(attn_implementation="flash_attention_2")
[rank68]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank68]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank68]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank68]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank68]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank68]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank68]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank68]:     self.load_model()
[rank68]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank68]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank68]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank68]:     self.model = _build_vision_tower(**self.config)
[rank68]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank68]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank68]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank68]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank68]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank68]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank68]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank68]:     with _open_file_like(f, "rb") as opened_file:
[rank68]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank68]:     return _open_file(name_or_buffer, mode)
[rank68]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank68]:     super().__init__(open(name, mode))
[rank68]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank22]: Traceback (most recent call last):
[rank22]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank22]:     train(attn_implementation="flash_attention_2")
[rank22]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank22]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank22]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank22]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank22]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank22]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank22]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank22]:     self.load_model()
[rank22]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank22]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank22]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank22]:     self.model = _build_vision_tower(**self.config)
[rank22]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank22]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank22]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank22]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank22]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank22]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank22]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank22]:     with _open_file_like(f, "rb") as opened_file:
[rank22]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank22]:     return _open_file(name_or_buffer, mode)
[rank22]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank22]:     super().__init__(open(name, mode))
[rank22]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank63]: Traceback (most recent call last):
[rank63]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank63]:     train(attn_implementation="flash_attention_2")
[rank63]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank63]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank63]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank63]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank63]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank63]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank63]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank63]:     self.load_model()
[rank63]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank63]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank63]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank63]:     self.model = _build_vision_tower(**self.config)
[rank63]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank63]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank63]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank63]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank63]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank63]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank63]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank63]:     with _open_file_like(f, "rb") as opened_file:
[rank63]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank63]:     return _open_file(name_or_buffer, mode)
[rank63]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank63]:     super().__init__(open(name, mode))
[rank63]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank117]: Traceback (most recent call last):
[rank117]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank117]:     train(attn_implementation="flash_attention_2")
[rank117]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank117]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank117]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank117]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank117]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank117]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank117]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank117]:     self.load_model()
[rank117]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank117]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank117]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank117]:     self.model = _build_vision_tower(**self.config)
[rank117]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank117]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank117]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank117]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank117]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank117]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank117]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank117]:     with _open_file_like(f, "rb") as opened_file:
[rank117]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank117]:     return _open_file(name_or_buffer, mode)
[rank117]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank117]:     super().__init__(open(name, mode))
[rank117]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank61]: Traceback (most recent call last):
[rank61]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank61]:     train(attn_implementation="flash_attention_2")
[rank61]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank61]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank61]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank61]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank61]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank61]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank101]: Traceback (most recent call last):
[rank101]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank101]:     train(attn_implementation="flash_attention_2")
[rank101]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank101]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank101]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank101]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank101]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank101]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank101]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank101]:     self.load_model()
[rank101]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank101]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank101]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank101]:     self.model = _build_vision_tower(**self.config)
[rank101]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank101]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank61]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank61]:     self.load_model()
[rank61]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank61]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank61]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank61]:     self.model = _build_vision_tower(**self.config)
[rank61]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank61]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank101]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank101]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank101]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank101]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank101]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank101]:     with _open_file_like(f, "rb") as opened_file:
[rank101]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank101]:     return _open_file(name_or_buffer, mode)
[rank61]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank61]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank61]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank61]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank61]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank61]:     with _open_file_like(f, "rb") as opened_file:
[rank61]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank61]:     return _open_file(name_or_buffer, mode)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
[rank101]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank101]:     super().__init__(open(name, mode))
[rank101]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank61]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank61]:     super().__init__(open(name, mode))
[rank61]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank83]: Traceback (most recent call last):
[rank83]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank83]:     train(attn_implementation="flash_attention_2")
[rank83]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank83]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank83]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank83]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank83]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank83]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank83]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank83]:     self.load_model()
[rank83]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank83]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank83]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank83]:     self.model = _build_vision_tower(**self.config)
[rank83]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank83]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank83]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank83]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank83]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank83]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank83]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank83]:     with _open_file_like(f, "rb") as opened_file:
[rank83]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank83]:     return _open_file(name_or_buffer, mode)
[rank83]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank83]:     super().__init__(open(name, mode))
[rank83]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank27]: Traceback (most recent call last):
[rank27]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank27]:     train(attn_implementation="flash_attention_2")
[rank27]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank27]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank27]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank27]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank27]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank27]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank27]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank27]:     self.load_model()
[rank27]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank27]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank27]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank27]:     self.model = _build_vision_tower(**self.config)
[rank27]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank27]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank27]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank27]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank27]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank27]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank27]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank27]:     with _open_file_like(f, "rb") as opened_file:
[rank27]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank27]:     return _open_file(name_or_buffer, mode)
[rank27]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank27]:     super().__init__(open(name, mode))
[rank27]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank25]: Traceback (most recent call last):
[rank25]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank25]:     train(attn_implementation="flash_attention_2")
[rank25]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank25]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank25]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank25]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank25]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank25]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank25]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank25]:     self.load_model()
[rank25]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank25]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank25]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank25]:     self.model = _build_vision_tower(**self.config)
[rank25]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank25]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank25]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank25]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank25]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank25]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank25]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank25]:     with _open_file_like(f, "rb") as opened_file:
[rank25]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank25]:     return _open_file(name_or_buffer, mode)
[rank25]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank25]:     super().__init__(open(name, mode))
[rank25]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank76]: Traceback (most recent call last):
[rank76]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank76]:     train(attn_implementation="flash_attention_2")
[rank76]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank76]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank76]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank76]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank76]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank76]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank76]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank76]:     self.load_model()
[rank76]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank76]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank76]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank76]:     self.model = _build_vision_tower(**self.config)
[rank76]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank76]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank76]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank76]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank76]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank76]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank76]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank76]:     with _open_file_like(f, "rb") as opened_file:
[rank76]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank76]:     return _open_file(name_or_buffer, mode)
[rank76]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank76]:     super().__init__(open(name, mode))
[rank76]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank11]: Traceback (most recent call last):
[rank11]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank11]:     train(attn_implementation="flash_attention_2")
[rank11]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank11]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank11]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank11]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank11]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank11]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank11]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank11]:     self.load_model()
[rank11]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank11]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank11]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank11]:     self.model = _build_vision_tower(**self.config)
[rank11]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank11]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank26]: Traceback (most recent call last):
[rank26]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank26]:     train(attn_implementation="flash_attention_2")
[rank26]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank26]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank26]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank26]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank26]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank26]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank11]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank11]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank11]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank11]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank11]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank11]:     with _open_file_like(f, "rb") as opened_file:
[rank11]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank11]:     return _open_file(name_or_buffer, mode)
[rank26]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank26]:     self.load_model()
[rank26]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank26]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank26]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank26]:     self.model = _build_vision_tower(**self.config)
[rank26]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank26]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank11]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank11]:     super().__init__(open(name, mode))
[rank11]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank26]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank26]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank26]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank26]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank26]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank26]:     with _open_file_like(f, "rb") as opened_file:
[rank26]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank26]:     return _open_file(name_or_buffer, mode)
[rank26]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank26]:     super().__init__(open(name, mode))
[rank26]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank29]: Traceback (most recent call last):
[rank29]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank29]:     train(attn_implementation="flash_attention_2")
[rank29]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank29]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank29]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank29]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank29]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank29]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank29]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank29]:     self.load_model()
[rank29]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank29]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank29]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank29]:     self.model = _build_vision_tower(**self.config)
[rank29]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank29]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank29]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank29]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank29]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank29]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank29]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank29]:     with _open_file_like(f, "rb") as opened_file:
[rank29]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank29]:     return _open_file(name_or_buffer, mode)
[rank29]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank29]:     super().__init__(open(name, mode))
[rank29]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank43]: Traceback (most recent call last):
[rank43]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank43]:     train(attn_implementation="flash_attention_2")
[rank43]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank43]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank43]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank43]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank43]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank43]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank43]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank43]:     self.load_model()
[rank43]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank43]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank43]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank43]:     self.model = _build_vision_tower(**self.config)
[rank43]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank43]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank43]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank43]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank43]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank43]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank43]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank43]:     with _open_file_like(f, "rb") as opened_file:
[rank43]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank43]:     return _open_file(name_or_buffer, mode)
[rank43]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank43]:     super().__init__(open(name, mode))
[rank43]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank124]: Traceback (most recent call last):
[rank124]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank124]:     train(attn_implementation="flash_attention_2")
[rank124]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank124]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank124]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank124]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank124]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank124]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank124]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank124]:     self.load_model()
[rank124]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank124]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank124]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank124]:     self.model = _build_vision_tower(**self.config)
[rank124]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank124]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank124]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank124]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank124]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank124]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank124]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank124]:     with _open_file_like(f, "rb") as opened_file:
[rank124]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank124]:     return _open_file(name_or_buffer, mode)
[rank124]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank124]:     super().__init__(open(name, mode))
[rank124]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank30]: Traceback (most recent call last):
[rank30]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank30]:     train(attn_implementation="flash_attention_2")
[rank30]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank30]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank30]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank30]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank30]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank30]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank30]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank30]:     self.load_model()
[rank30]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank30]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank30]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank30]:     self.model = _build_vision_tower(**self.config)
[rank30]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank30]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank30]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank30]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank30]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank30]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank30]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank30]:     with _open_file_like(f, "rb") as opened_file:
[rank30]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank30]:     return _open_file(name_or_buffer, mode)
[rank30]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank30]:     super().__init__(open(name, mode))
[rank30]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank40]: Traceback (most recent call last):
[rank40]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank40]:     train(attn_implementation="flash_attention_2")
[rank40]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank40]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank40]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank40]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank40]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank40]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank40]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank40]:     self.load_model()
[rank40]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank40]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank40]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank40]:     self.model = _build_vision_tower(**self.config)
[rank40]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank40]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank40]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank40]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank40]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank40]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank40]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank40]:     with _open_file_like(f, "rb") as opened_file:
[rank40]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank40]:     return _open_file(name_or_buffer, mode)
[rank40]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank40]:     super().__init__(open(name, mode))
[rank40]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank62]: Traceback (most recent call last):
[rank62]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank62]:     train(attn_implementation="flash_attention_2")
[rank62]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank62]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank62]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank62]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank62]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank62]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank62]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank62]:     self.load_model()
[rank62]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank62]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank62]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank62]:     self.model = _build_vision_tower(**self.config)
[rank62]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank62]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank62]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank62]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank62]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank62]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank62]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank62]:     with _open_file_like(f, "rb") as opened_file:
[rank62]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank62]:     return _open_file(name_or_buffer, mode)
[rank62]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank62]:     super().__init__(open(name, mode))
[rank62]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank114]: Traceback (most recent call last):
[rank114]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank114]:     train(attn_implementation="flash_attention_2")
[rank114]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank114]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank114]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank114]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank114]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank114]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank34]: Traceback (most recent call last):
[rank34]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank34]:     train(attn_implementation="flash_attention_2")
[rank34]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank34]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank34]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank34]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank34]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank34]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank114]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank114]:     self.load_model()
[rank114]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank114]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank114]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank114]:     self.model = _build_vision_tower(**self.config)
[rank114]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank114]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank34]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank34]:     self.load_model()
[rank34]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank34]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank34]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank34]:     self.model = _build_vision_tower(**self.config)
[rank34]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank34]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank114]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank114]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank114]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank114]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank114]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank114]:     with _open_file_like(f, "rb") as opened_file:
[rank114]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank114]:     return _open_file(name_or_buffer, mode)
[rank34]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank34]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank34]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank34]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank34]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank34]:     with _open_file_like(f, "rb") as opened_file:
[rank34]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank34]:     return _open_file(name_or_buffer, mode)
[rank114]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank114]:     super().__init__(open(name, mode))
[rank114]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank34]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank34]:     super().__init__(open(name, mode))
[rank34]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank24]: Traceback (most recent call last):
[rank24]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank24]:     train(attn_implementation="flash_attention_2")
[rank24]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank24]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank24]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank24]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank24]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank24]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank24]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank24]:     self.load_model()
[rank24]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank24]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank24]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank24]:     self.model = _build_vision_tower(**self.config)
[rank24]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank24]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank35]: Traceback (most recent call last):
[rank35]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank35]:     train(attn_implementation="flash_attention_2")
[rank35]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank35]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank35]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank35]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank35]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank35]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank24]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank24]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank24]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank24]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank24]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank24]:     with _open_file_like(f, "rb") as opened_file:
[rank24]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank24]:     return _open_file(name_or_buffer, mode)
[rank24]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank24]:     super().__init__(open(name, mode))
[rank24]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank35]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank35]:     self.load_model()
[rank35]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank35]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank35]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank35]:     self.model = _build_vision_tower(**self.config)
[rank35]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank35]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank35]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank35]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank35]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank35]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank35]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank35]:     with _open_file_like(f, "rb") as opened_file:
[rank35]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank35]:     return _open_file(name_or_buffer, mode)
[rank35]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank35]:     super().__init__(open(name, mode))
[rank35]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank28]: Traceback (most recent call last):
[rank28]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank28]:     train(attn_implementation="flash_attention_2")
[rank28]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank28]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank28]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank28]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank28]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank28]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank28]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank28]:     self.load_model()
[rank28]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank28]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank28]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank28]:     self.model = _build_vision_tower(**self.config)
[rank28]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank28]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank28]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank28]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank28]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank28]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank28]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank28]:     with _open_file_like(f, "rb") as opened_file:
[rank28]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank28]:     return _open_file(name_or_buffer, mode)
[rank28]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank28]:     super().__init__(open(name, mode))
[rank28]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank15]: Traceback (most recent call last):
[rank15]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank15]:     train(attn_implementation="flash_attention_2")
[rank15]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank15]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank15]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank15]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank15]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank15]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank15]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank15]:     self.load_model()
[rank15]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank15]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank15]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank15]:     self.model = _build_vision_tower(**self.config)
[rank15]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank15]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank15]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank15]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank15]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank15]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank15]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank15]:     with _open_file_like(f, "rb") as opened_file:
[rank15]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank15]:     return _open_file(name_or_buffer, mode)
[rank15]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank15]:     super().__init__(open(name, mode))
[rank15]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank48]: Traceback (most recent call last):
[rank48]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank48]:     train(attn_implementation="flash_attention_2")
[rank48]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank48]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank48]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank48]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank48]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank48]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank48]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank48]:     self.load_model()
[rank48]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank48]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank48]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank48]:     self.model = _build_vision_tower(**self.config)
[rank48]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank48]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank48]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank48]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank48]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank48]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank48]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank48]:     with _open_file_like(f, "rb") as opened_file:
[rank48]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank48]:     return _open_file(name_or_buffer, mode)
[rank48]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank48]:     super().__init__(open(name, mode))
[rank48]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank90]: Traceback (most recent call last):
[rank90]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank90]:     train(attn_implementation="flash_attention_2")
[rank90]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank90]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank90]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank90]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank90]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank90]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank90]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank90]:     self.load_model()
[rank90]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank90]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank90]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank90]:     self.model = _build_vision_tower(**self.config)
[rank90]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank90]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank90]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank90]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank90]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank90]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank90]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank90]:     with _open_file_like(f, "rb") as opened_file:
[rank90]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank90]:     return _open_file(name_or_buffer, mode)
[rank90]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank90]:     super().__init__(open(name, mode))
[rank90]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank93]: Traceback (most recent call last):
[rank93]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank93]:     train(attn_implementation="flash_attention_2")
[rank93]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank93]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank93]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank93]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank93]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank93]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank93]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank93]:     self.load_model()
[rank93]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank93]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank93]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank93]:     self.model = _build_vision_tower(**self.config)
[rank93]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank93]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank93]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank93]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank93]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank93]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank93]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank93]:     with _open_file_like(f, "rb") as opened_file:
[rank93]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank93]:     return _open_file(name_or_buffer, mode)
[rank93]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank93]:     super().__init__(open(name, mode))
[rank93]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank57]: Traceback (most recent call last):
[rank57]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank57]:     train(attn_implementation="flash_attention_2")
[rank57]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank57]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank57]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank57]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank57]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank57]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank57]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank57]:     self.load_model()
[rank57]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank57]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank57]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank57]:     self.model = _build_vision_tower(**self.config)
[rank57]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank57]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank57]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank57]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank57]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank57]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank57]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank57]:     with _open_file_like(f, "rb") as opened_file:
[rank57]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank57]:     return _open_file(name_or_buffer, mode)
[rank57]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank57]:     super().__init__(open(name, mode))
[rank57]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank55]: Traceback (most recent call last):
[rank55]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank55]:     train(attn_implementation="flash_attention_2")
[rank55]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank55]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank55]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank55]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank55]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank55]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank55]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank55]:     self.load_model()
[rank55]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank55]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank55]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank55]:     self.model = _build_vision_tower(**self.config)
[rank55]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank55]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank55]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank55]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank55]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank55]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank55]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank55]:     with _open_file_like(f, "rb") as opened_file:
[rank55]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank55]:     return _open_file(name_or_buffer, mode)
[rank55]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank55]:     super().__init__(open(name, mode))
[rank55]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank97]: Traceback (most recent call last):
[rank97]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank97]:     train(attn_implementation="flash_attention_2")
[rank97]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank97]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank97]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank97]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank97]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank97]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank97]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank97]:     self.load_model()
[rank97]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank97]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank97]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank97]:     self.model = _build_vision_tower(**self.config)
[rank97]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank97]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank97]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank97]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank97]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank97]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank97]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank97]:     with _open_file_like(f, "rb") as opened_file:
[rank97]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank97]:     return _open_file(name_or_buffer, mode)
[rank97]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank97]:     super().__init__(open(name, mode))
[rank97]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank59]: Traceback (most recent call last):
[rank59]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank59]:     train(attn_implementation="flash_attention_2")
[rank59]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank59]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank59]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank59]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank59]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank59]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank59]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank59]:     self.load_model()
[rank59]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank59]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank59]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank59]:     self.model = _build_vision_tower(**self.config)
[rank59]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank59]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank59]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank59]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank59]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank59]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank59]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank59]:     with _open_file_like(f, "rb") as opened_file:
[rank59]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank59]:     return _open_file(name_or_buffer, mode)
[rank59]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank59]:     super().__init__(open(name, mode))
[rank59]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank89]: Traceback (most recent call last):
[rank89]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank89]:     train(attn_implementation="flash_attention_2")
[rank89]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank89]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank89]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank89]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank89]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank89]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank89]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank89]:     self.load_model()
[rank89]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank89]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank89]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank89]:     self.model = _build_vision_tower(**self.config)
[rank89]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank89]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank89]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank89]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank89]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank89]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank89]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank89]:     with _open_file_like(f, "rb") as opened_file:
[rank89]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank89]:     return _open_file(name_or_buffer, mode)
[rank89]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank89]:     super().__init__(open(name, mode))
[rank89]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank39]: Traceback (most recent call last):
[rank39]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank39]:     train(attn_implementation="flash_attention_2")
[rank39]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank39]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank39]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank39]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank39]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank39]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank39]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank39]:     self.load_model()
[rank39]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank39]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank39]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank39]:     self.model = _build_vision_tower(**self.config)
[rank39]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank39]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank39]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank39]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank39]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank39]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank39]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank39]:     with _open_file_like(f, "rb") as opened_file:
[rank39]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank39]:     return _open_file(name_or_buffer, mode)
[rank39]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank39]:     super().__init__(open(name, mode))
[rank39]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank38]: Traceback (most recent call last):
[rank38]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank38]:     train(attn_implementation="flash_attention_2")
[rank38]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank38]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank38]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank38]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank38]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank38]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank38]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank38]:     self.load_model()
[rank38]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank38]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank38]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank38]:     self.model = _build_vision_tower(**self.config)
[rank38]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank38]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank38]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank38]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank38]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank38]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank38]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank38]:     with _open_file_like(f, "rb") as opened_file:
[rank38]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank38]:     return _open_file(name_or_buffer, mode)
[rank38]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank38]:     super().__init__(open(name, mode))
[rank38]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank13]: Traceback (most recent call last):
[rank13]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank13]:     train(attn_implementation="flash_attention_2")
[rank13]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank13]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank13]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank13]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank13]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank13]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank13]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank13]:     self.load_model()
[rank13]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank13]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank13]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank13]:     self.model = _build_vision_tower(**self.config)
[rank13]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank13]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank13]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank13]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank13]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank13]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank13]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank13]:     with _open_file_like(f, "rb") as opened_file:
[rank13]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank13]:     return _open_file(name_or_buffer, mode)
[rank13]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank13]:     super().__init__(open(name, mode))
[rank13]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank106]: Traceback (most recent call last):
[rank106]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank106]:     train(attn_implementation="flash_attention_2")
[rank106]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank106]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank106]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank106]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank106]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank106]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank106]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank106]:     self.load_model()
[rank106]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank106]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank106]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank106]:     self.model = _build_vision_tower(**self.config)
[rank106]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank106]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank99]: Traceback (most recent call last):
[rank99]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank99]:     train(attn_implementation="flash_attention_2")
[rank99]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank99]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank99]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank99]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank99]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank99]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank106]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank106]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank106]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank106]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank106]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank106]:     with _open_file_like(f, "rb") as opened_file:
[rank106]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank106]:     return _open_file(name_or_buffer, mode)
[rank99]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank99]:     self.load_model()
[rank99]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank99]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank99]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank99]:     self.model = _build_vision_tower(**self.config)
[rank99]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank99]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank106]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank106]:     super().__init__(open(name, mode))
[rank106]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank99]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank99]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank99]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank99]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank99]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank99]:     with _open_file_like(f, "rb") as opened_file:
[rank99]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank99]:     return _open_file(name_or_buffer, mode)
[rank99]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank99]:     super().__init__(open(name, mode))
[rank99]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank125]: Traceback (most recent call last):
[rank125]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank125]:     train(attn_implementation="flash_attention_2")
[rank125]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank125]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank125]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank125]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank125]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank125]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank125]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank125]:     self.load_model()
[rank125]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank125]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank125]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank125]:     self.model = _build_vision_tower(**self.config)
[rank125]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank125]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank125]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank125]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank125]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank125]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank125]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank125]:     with _open_file_like(f, "rb") as opened_file:
[rank125]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank125]:     return _open_file(name_or_buffer, mode)
[rank125]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank125]:     super().__init__(open(name, mode))
[rank125]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank118]: Traceback (most recent call last):
[rank118]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank118]:     train(attn_implementation="flash_attention_2")
[rank118]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank118]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank118]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank118]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank118]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank118]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank118]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank118]:     self.load_model()
[rank118]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank118]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank118]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank118]:     self.model = _build_vision_tower(**self.config)
[rank118]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank118]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank118]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank118]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank118]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank118]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank118]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank118]:     with _open_file_like(f, "rb") as opened_file:
[rank118]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank118]:     return _open_file(name_or_buffer, mode)
[rank118]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank118]:     super().__init__(open(name, mode))
[rank118]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank31]: Traceback (most recent call last):
[rank31]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank31]:     train(attn_implementation="flash_attention_2")
[rank31]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank31]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank31]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank31]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank31]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank31]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank31]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank31]:     self.load_model()
[rank31]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank31]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank31]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank31]:     self.model = _build_vision_tower(**self.config)
[rank31]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank31]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank31]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank31]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank31]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank31]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank31]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank31]:     with _open_file_like(f, "rb") as opened_file:
[rank31]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank31]:     return _open_file(name_or_buffer, mode)
[rank31]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank31]:     super().__init__(open(name, mode))
[rank31]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank10]: Traceback (most recent call last):
[rank10]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank10]:     train(attn_implementation="flash_attention_2")
[rank10]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank10]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank10]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank10]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank10]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank10]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank10]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank10]:     self.load_model()
[rank10]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank10]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank10]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank10]:     self.model = _build_vision_tower(**self.config)
[rank10]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank10]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank10]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank10]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank10]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank10]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank10]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank10]:     with _open_file_like(f, "rb") as opened_file:
[rank10]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank10]:     return _open_file(name_or_buffer, mode)
[rank10]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank10]:     super().__init__(open(name, mode))
[rank10]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank6]: Traceback (most recent call last):
[rank6]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank6]:     train(attn_implementation="flash_attention_2")
[rank6]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank6]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank6]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank6]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank6]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank6]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank6]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank6]:     self.load_model()
[rank6]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank6]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank6]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank6]:     self.model = _build_vision_tower(**self.config)
[rank6]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank6]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank6]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank6]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank6]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank6]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank6]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank6]:     with _open_file_like(f, "rb") as opened_file:
[rank6]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank6]:     return _open_file(name_or_buffer, mode)
[rank6]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank6]:     super().__init__(open(name, mode))
[rank6]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank4]: Traceback (most recent call last):
[rank4]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank4]:     train(attn_implementation="flash_attention_2")
[rank4]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank4]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank4]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank4]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank4]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank4]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank4]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank4]:     self.load_model()
[rank4]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank4]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank4]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank4]:     self.model = _build_vision_tower(**self.config)
[rank4]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank4]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank4]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank4]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank4]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank4]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank4]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank4]:     with _open_file_like(f, "rb") as opened_file:
[rank4]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank4]:     return _open_file(name_or_buffer, mode)
[rank4]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank4]:     super().__init__(open(name, mode))
[rank4]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank5]: Traceback (most recent call last):
[rank5]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank5]:     train(attn_implementation="flash_attention_2")
[rank5]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank5]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank5]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank5]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank5]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank5]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank5]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank5]:     self.load_model()
[rank5]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank5]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank5]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank5]:     self.model = _build_vision_tower(**self.config)
[rank5]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank5]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank5]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank5]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank5]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank5]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank5]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank5]:     with _open_file_like(f, "rb") as opened_file:
[rank5]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank5]:     return _open_file(name_or_buffer, mode)
[rank5]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank5]:     super().__init__(open(name, mode))
[rank5]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank2]: Traceback (most recent call last):
[rank2]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank2]:     train(attn_implementation="flash_attention_2")
[rank2]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank2]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank2]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank2]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank2]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank2]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank2]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank2]:     self.load_model()
[rank2]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank2]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank2]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank2]:     self.model = _build_vision_tower(**self.config)
[rank2]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank2]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank2]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank2]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank2]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank2]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank2]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank2]:     with _open_file_like(f, "rb") as opened_file:
[rank2]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank2]:     return _open_file(name_or_buffer, mode)
[rank2]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank2]:     super().__init__(open(name, mode))
[rank2]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank3]: Traceback (most recent call last):
[rank3]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank3]:     train(attn_implementation="flash_attention_2")
[rank3]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank3]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank3]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank3]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank3]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank3]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank3]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank3]:     self.load_model()
[rank3]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank3]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank3]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank3]:     self.model = _build_vision_tower(**self.config)
[rank3]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank3]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank3]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank3]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank3]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank3]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank3]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank3]:     with _open_file_like(f, "rb") as opened_file:
[rank3]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank3]:     return _open_file(name_or_buffer, mode)
[rank3]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank3]:     super().__init__(open(name, mode))
[rank3]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank0]: Traceback (most recent call last):
[rank0]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank0]:     train(attn_implementation="flash_attention_2")
[rank0]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank0]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank0]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank0]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank0]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank0]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank0]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank0]:     self.load_model()
[rank0]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank0]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank0]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank0]:     self.model = _build_vision_tower(**self.config)
[rank0]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank0]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank0]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank0]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank0]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank0]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank0]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank0]:     with _open_file_like(f, "rb") as opened_file:
[rank0]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank0]:     return _open_file(name_or_buffer, mode)
[rank0]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank0]:     super().__init__(open(name, mode))
[rank0]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank7]: Traceback (most recent call last):
[rank7]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank7]:     train(attn_implementation="flash_attention_2")
[rank7]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank7]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank7]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank7]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank7]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank7]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank7]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank7]:     self.load_model()
[rank7]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank7]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank7]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank7]:     self.model = _build_vision_tower(**self.config)
[rank7]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank7]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank7]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank7]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank7]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank7]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank7]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank7]:     with _open_file_like(f, "rb") as opened_file:
[rank7]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank7]:     return _open_file(name_or_buffer, mode)
[rank7]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank7]:     super().__init__(open(name, mode))
[rank7]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank1]: Traceback (most recent call last):
[rank1]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank1]:     train(attn_implementation="flash_attention_2")
[rank1]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank1]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank1]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank1]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank1]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank1]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank1]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank1]:     self.load_model()
[rank1]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank1]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank1]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank1]:     self.model = _build_vision_tower(**self.config)
[rank1]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank1]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank1]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank1]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank1]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank1]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank1]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank1]:     with _open_file_like(f, "rb") as opened_file:
[rank1]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank1]:     return _open_file(name_or_buffer, mode)
[rank1]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank1]:     super().__init__(open(name, mode))
[rank1]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank8]: Traceback (most recent call last):
[rank8]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank8]:     train(attn_implementation="flash_attention_2")
[rank8]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank8]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank8]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank8]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank8]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank8]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank8]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank8]:     self.load_model()
[rank8]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank8]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank8]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank8]:     self.model = _build_vision_tower(**self.config)
[rank8]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank8]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank8]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank8]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank8]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank8]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank8]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank8]:     with _open_file_like(f, "rb") as opened_file:
[rank8]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank8]:     return _open_file(name_or_buffer, mode)
[rank8]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank8]:     super().__init__(open(name, mode))
[rank8]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank67]: Traceback (most recent call last):
[rank67]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank67]:     train(attn_implementation="flash_attention_2")
[rank67]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank67]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank67]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank67]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank67]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank67]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank67]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank67]:     self.load_model()
[rank67]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank67]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank67]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank67]:     self.model = _build_vision_tower(**self.config)
[rank67]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank67]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank67]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank67]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank67]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank67]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank67]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank67]:     with _open_file_like(f, "rb") as opened_file:
[rank67]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank67]:     return _open_file(name_or_buffer, mode)
[rank67]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank67]:     super().__init__(open(name, mode))
[rank67]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank12]: Traceback (most recent call last):
[rank12]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank12]:     train(attn_implementation="flash_attention_2")
[rank12]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank12]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank12]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank12]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank12]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank12]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank12]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank12]:     self.load_model()
[rank12]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank12]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank12]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank12]:     self.model = _build_vision_tower(**self.config)
[rank12]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank12]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank12]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank12]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank12]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank12]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank12]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank12]:     with _open_file_like(f, "rb") as opened_file:
[rank12]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank12]:     return _open_file(name_or_buffer, mode)
[rank12]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank12]:     super().__init__(open(name, mode))
[rank12]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py:622: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank69]: Traceback (most recent call last):
[rank69]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train_mem.py", line 4, in <module>
[rank69]:     train(attn_implementation="flash_attention_2")
[rank69]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/train/train.py", line 1531, in train
[rank69]:     model.get_model().initialize_vision_modules(model_args=model_args, fsdp=training_args.fsdp)
[rank69]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/llava_arch.py", line 91, in initialize_vision_modules
[rank69]:     gen_vision_tower = build_gen_vision_tower(model_args)
[rank69]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/builder.py", line 43, in build_gen_vision_tower
[rank69]:     return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
[rank69]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 23, in __init__
[rank69]:     self.load_model()
[rank69]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_clip_encoder.py", line 38, in load_model
[rank69]:     self.vision_tower = EVAEncoderWrapper(self.vision_tower_pretrained, self.config)
[rank69]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 748, in __init__
[rank69]:     self.model = _build_vision_tower(**self.config)
[rank69]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 734, in _build_vision_tower
[rank69]:     state_dict = load_clip_visual_state_dict(vision_tower_path)
[rank69]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 645, in load_clip_visual_state_dict
[rank69]:     state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
[rank69]:   File "/opt/hpcaas/.mounts/fs-036153e63d56f4dc2/home/zhaojiang/interleaved-llava/llava/model/multimodal_encoder/eva_clip/eva_vit.py", line 622, in load_state_dict
[rank69]:     checkpoint = torch.load(checkpoint_path, map_location=map_location)
[rank69]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load
[rank69]:     with _open_file_like(f, "rb") as opened_file:
[rank69]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like
[rank69]:     return _open_file(name_or_buffer, mode)
[rank69]:   File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__
[rank69]:     super().__init__(open(name, mode))
[rank69]: FileNotFoundError: [Errno 2] No such file or directory: '/fsx-project/zhiyangxu1/projects/checkpoints/eva_clip_vision_tower/pytorch_model.bin'
[rank64]:[W216 06:19:23.390825192 ProcessGroupNCCL.cpp:1250] Warning: WARNING: process group has NOT been destroyed before we destruct ProcessGroupNCCL. On normal program exit, the application should call destroy_process_group to ensure that any pending NCCL operations have finished in this process. In rare cases this process can exit before this point and block the progress of another member of the process group. This constraint has always been present,  but this warning has only been added since PyTorch 2.4 (function operator())
[rank32]:[W216 06:19:24.743215061 ProcessGroupNCCL.cpp:1250] Warning: WARNING: process group has NOT been destroyed before we destruct ProcessGroupNCCL. On normal program exit, the application should call destroy_process_group to ensure that any pending NCCL operations have finished in this process. In rare cases this process can exit before this point and block the progress of another member of the process group. This constraint has always been present,  but this warning has only been added since PyTorch 2.4 (function operator())
[rank120]:[W216 06:19:24.043178547 ProcessGroupNCCL.cpp:1250] Warning: WARNING: process group has NOT been destroyed before we destruct ProcessGroupNCCL. On normal program exit, the application should call destroy_process_group to ensure that any pending NCCL operations have finished in this process. In rare cases this process can exit before this point and block the progress of another member of the process group. This constraint has always been present,  but this warning has only been added since PyTorch 2.4 (function operator())
[rank72]:[W216 06:19:24.535622088 ProcessGroupNCCL.cpp:1250] Warning: WARNING: process group has NOT been destroyed before we destruct ProcessGroupNCCL. On normal program exit, the application should call destroy_process_group to ensure that any pending NCCL operations have finished in this process. In rare cases this process can exit before this point and block the progress of another member of the process group. This constraint has always been present,  but this warning has only been added since PyTorch 2.4 (function operator())
[rank24]:[W216 06:19:24.365476101 ProcessGroupNCCL.cpp:1250] Warning: WARNING: process group has NOT been destroyed before we destruct ProcessGroupNCCL. On normal program exit, the application should call destroy_process_group to ensure that any pending NCCL operations have finished in this process. In rare cases this process can exit before this point and block the progress of another member of the process group. This constraint has always been present,  but this warning has only been added since PyTorch 2.4 (function operator())
[rank112]:[W216 06:19:24.312215837 ProcessGroupNCCL.cpp:1250] Warning: WARNING: process group has NOT been destroyed before we destruct ProcessGroupNCCL. On normal program exit, the application should call destroy_process_group to ensure that any pending NCCL operations have finished in this process. In rare cases this process can exit before this point and block the progress of another member of the process group. This constraint has always been present,  but this warning has only been added since PyTorch 2.4 (function operator())
[rank56]:[W216 06:19:24.033090198 ProcessGroupNCCL.cpp:1250] Warning: WARNING: process group has NOT been destroyed before we destruct ProcessGroupNCCL. On normal program exit, the application should call destroy_process_group to ensure that any pending NCCL operations have finished in this process. In rare cases this process can exit before this point and block the progress of another member of the process group. This constraint has always been present,  but this warning has only been added since PyTorch 2.4 (function operator())
[rank16]:[W216 06:19:24.418549166 ProcessGroupNCCL.cpp:1250] Warning: WARNING: process group has NOT been destroyed before we destruct ProcessGroupNCCL. On normal program exit, the application should call destroy_process_group to ensure that any pending NCCL operations have finished in this process. In rare cases this process can exit before this point and block the progress of another member of the process group. This constraint has always been present,  but this warning has only been added since PyTorch 2.4 (function operator())
[rank104]:[W216 06:19:24.090529068 ProcessGroupNCCL.cpp:1250] Warning: WARNING: process group has NOT been destroyed before we destruct ProcessGroupNCCL. On normal program exit, the application should call destroy_process_group to ensure that any pending NCCL operations have finished in this process. In rare cases this process can exit before this point and block the progress of another member of the process group. This constraint has always been present,  but this warning has only been added since PyTorch 2.4 (function operator())
[rank80]:[W216 06:19:24.347997549 ProcessGroupNCCL.cpp:1250] Warning: WARNING: process group has NOT been destroyed before we destruct ProcessGroupNCCL. On normal program exit, the application should call destroy_process_group to ensure that any pending NCCL operations have finished in this process. In rare cases this process can exit before this point and block the progress of another member of the process group. This constraint has always been present,  but this warning has only been added since PyTorch 2.4 (function operator())
[rank40]:[W216 06:19:24.336615537 ProcessGroupNCCL.cpp:1250] Warning: WARNING: process group has NOT been destroyed before we destruct ProcessGroupNCCL. On normal program exit, the application should call destroy_process_group to ensure that any pending NCCL operations have finished in this process. In rare cases this process can exit before this point and block the progress of another member of the process group. This constraint has always been present,  but this warning has only been added since PyTorch 2.4 (function operator())
[rank48]:[W216 06:19:24.614614735 ProcessGroupNCCL.cpp:1250] Warning: WARNING: process group has NOT been destroyed before we destruct ProcessGroupNCCL. On normal program exit, the application should call destroy_process_group to ensure that any pending NCCL operations have finished in this process. In rare cases this process can exit before this point and block the progress of another member of the process group. This constraint has always been present,  but this warning has only been added since PyTorch 2.4 (function operator())
[rank88]:[W216 06:19:24.944224723 ProcessGroupNCCL.cpp:1250] Warning: WARNING: process group has NOT been destroyed before we destruct ProcessGroupNCCL. On normal program exit, the application should call destroy_process_group to ensure that any pending NCCL operations have finished in this process. In rare cases this process can exit before this point and block the progress of another member of the process group. This constraint has always been present,  but this warning has only been added since PyTorch 2.4 (function operator())
[rank96]:[W216 06:19:24.392506101 ProcessGroupNCCL.cpp:1250] Warning: WARNING: process group has NOT been destroyed before we destruct ProcessGroupNCCL. On normal program exit, the application should call destroy_process_group to ensure that any pending NCCL operations have finished in this process. In rare cases this process can exit before this point and block the progress of another member of the process group. This constraint has always been present,  but this warning has only been added since PyTorch 2.4 (function operator())
[rank8]:[W216 06:19:25.649468459 ProcessGroupNCCL.cpp:1250] Warning: WARNING: process group has NOT been destroyed before we destruct ProcessGroupNCCL. On normal program exit, the application should call destroy_process_group to ensure that any pending NCCL operations have finished in this process. In rare cases this process can exit before this point and block the progress of another member of the process group. This constraint has always been present,  but this warning has only been added since PyTorch 2.4 (function operator())
W0216 06:19:25.322000 2201849 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 2201975 closing signal SIGTERM
W0216 06:19:25.323000 2201849 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 2201976 closing signal SIGTERM
W0216 06:19:25.324000 2201849 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 2201978 closing signal SIGTERM
W0216 06:19:25.324000 2201849 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 2201979 closing signal SIGTERM
W0216 06:19:25.325000 2201849 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 2201981 closing signal SIGTERM
W0216 06:19:25.325000 2201849 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 2201984 closing signal SIGTERM
[rank0]:[W216 06:19:25.123885358 ProcessGroupNCCL.cpp:1250] Warning: WARNING: process group has NOT been destroyed before we destruct ProcessGroupNCCL. On normal program exit, the application should call destroy_process_group to ensure that any pending NCCL operations have finished in this process. In rare cases this process can exit before this point and block the progress of another member of the process group. This constraint has always been present,  but this warning has only been added since PyTorch 2.4 (function operator())
W0216 06:19:25.431000 2062756 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 2062885 closing signal SIGTERM
W0216 06:19:25.432000 2062756 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 2062886 closing signal SIGTERM
W0216 06:19:25.433000 2062756 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 2062887 closing signal SIGTERM
W0216 06:19:25.433000 2062756 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 2062888 closing signal SIGTERM
W0216 06:19:25.434000 2062756 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 2062889 closing signal SIGTERM
W0216 06:19:25.434000 2062756 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 2062890 closing signal SIGTERM
W0216 06:19:25.434000 2062756 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 2062892 closing signal SIGTERM
W0216 06:19:25.632000 2133652 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 2133778 closing signal SIGTERM
W0216 06:19:25.633000 2133652 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 2133779 closing signal SIGTERM
W0216 06:19:25.633000 2133652 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 2133780 closing signal SIGTERM
W0216 06:19:25.633000 2133652 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 2133781 closing signal SIGTERM
W0216 06:19:25.634000 2133652 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 2133782 closing signal SIGTERM
W0216 06:19:25.634000 2133652 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 2133783 closing signal SIGTERM
W0216 06:19:25.634000 2133652 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 2133784 closing signal SIGTERM
W0216 06:19:25.644000 3781647 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3782417 closing signal SIGTERM
W0216 06:19:25.644000 3781647 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3782418 closing signal SIGTERM
W0216 06:19:25.645000 3781647 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3782419 closing signal SIGTERM
W0216 06:19:25.645000 3781647 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3782420 closing signal SIGTERM
W0216 06:19:25.645000 3781647 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3782422 closing signal SIGTERM
W0216 06:19:25.646000 3781647 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3782423 closing signal SIGTERM
W0216 06:19:25.646000 3781647 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3782424 closing signal SIGTERM
W0216 06:19:25.732000 4003734 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 4003863 closing signal SIGTERM
W0216 06:19:25.733000 4003734 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 4003864 closing signal SIGTERM
W0216 06:19:25.733000 4003734 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 4003865 closing signal SIGTERM
W0216 06:19:25.733000 4003734 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 4003866 closing signal SIGTERM
W0216 06:19:25.734000 4003734 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 4003867 closing signal SIGTERM
W0216 06:19:25.734000 4003734 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 4003868 closing signal SIGTERM
W0216 06:19:25.734000 4003734 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 4003869 closing signal SIGTERM
W0216 06:19:25.745000 3344412 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3344545 closing signal SIGTERM
W0216 06:19:25.746000 3344412 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3344546 closing signal SIGTERM
W0216 06:19:25.746000 3344412 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3344547 closing signal SIGTERM
W0216 06:19:25.747000 3344412 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3344548 closing signal SIGTERM
W0216 06:19:25.747000 3344412 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3344549 closing signal SIGTERM
W0216 06:19:25.747000 3344412 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3344550 closing signal SIGTERM
W0216 06:19:25.748000 3344412 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3344551 closing signal SIGTERM
W0216 06:19:25.828000 2664794 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 2664925 closing signal SIGTERM
W0216 06:19:25.829000 2664794 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 2664926 closing signal SIGTERM
W0216 06:19:25.829000 2664794 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 2664927 closing signal SIGTERM
W0216 06:19:25.829000 2664794 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 2664928 closing signal SIGTERM
W0216 06:19:25.830000 2664794 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 2664929 closing signal SIGTERM
W0216 06:19:25.830000 2664794 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 2664930 closing signal SIGTERM
W0216 06:19:25.831000 2664794 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 2664931 closing signal SIGTERM
W0216 06:19:25.836000 3335470 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3337099 closing signal SIGTERM
W0216 06:19:25.837000 3335470 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3337100 closing signal SIGTERM
W0216 06:19:25.838000 3335470 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3337101 closing signal SIGTERM
W0216 06:19:25.838000 3335470 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3337102 closing signal SIGTERM
W0216 06:19:25.839000 3335470 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3337103 closing signal SIGTERM
W0216 06:19:25.839000 3335470 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3337104 closing signal SIGTERM
W0216 06:19:25.840000 3335470 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3337105 closing signal SIGTERM
W0216 06:19:26.034000 3791965 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3792184 closing signal SIGTERM
W0216 06:19:26.034000 3791965 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3792185 closing signal SIGTERM
W0216 06:19:26.035000 3791965 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3792186 closing signal SIGTERM
W0216 06:19:26.035000 3791965 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3792187 closing signal SIGTERM
W0216 06:19:26.036000 3791965 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3792188 closing signal SIGTERM
W0216 06:19:26.036000 3791965 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3792189 closing signal SIGTERM
W0216 06:19:26.037000 3791965 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3792190 closing signal SIGTERM
W0216 06:19:26.236000 4000190 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 4000944 closing signal SIGTERM
W0216 06:19:26.237000 4000190 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 4000945 closing signal SIGTERM
W0216 06:19:26.237000 4000190 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 4000946 closing signal SIGTERM
W0216 06:19:26.238000 4000190 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 4000947 closing signal SIGTERM
W0216 06:19:26.238000 4000190 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 4000948 closing signal SIGTERM
W0216 06:19:26.238000 4000190 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 4000949 closing signal SIGTERM
W0216 06:19:26.239000 4000190 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 4000951 closing signal SIGTERM
W0216 06:19:26.435000 3262384 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3263050 closing signal SIGTERM
W0216 06:19:26.436000 3262384 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3263051 closing signal SIGTERM
W0216 06:19:26.437000 3262384 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3263052 closing signal SIGTERM
W0216 06:19:26.437000 3262384 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3263054 closing signal SIGTERM
W0216 06:19:26.437000 3262384 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3263055 closing signal SIGTERM
W0216 06:19:26.438000 3262384 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3263056 closing signal SIGTERM
W0216 06:19:26.438000 3262384 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3263057 closing signal SIGTERM
W0216 06:19:26.441000 3343136 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3343331 closing signal SIGTERM
W0216 06:19:26.441000 3343136 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3343332 closing signal SIGTERM
W0216 06:19:26.442000 3343136 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3343334 closing signal SIGTERM
W0216 06:19:26.442000 3343136 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3343335 closing signal SIGTERM
W0216 06:19:26.442000 3343136 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3343337 closing signal SIGTERM
W0216 06:19:26.443000 3343136 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3343338 closing signal SIGTERM
W0216 06:19:26.459000 3603613 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3603750 closing signal SIGTERM
W0216 06:19:26.460000 3603613 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3603751 closing signal SIGTERM
W0216 06:19:26.460000 3603613 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3603752 closing signal SIGTERM
W0216 06:19:26.460000 3603613 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3603753 closing signal SIGTERM
W0216 06:19:26.460000 3603613 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3603754 closing signal SIGTERM
W0216 06:19:26.461000 3603613 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3603755 closing signal SIGTERM
W0216 06:19:26.461000 3603613 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3603756 closing signal SIGTERM
W0216 06:19:26.664000 3997540 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3997673 closing signal SIGTERM
W0216 06:19:26.664000 3997540 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3997674 closing signal SIGTERM
W0216 06:19:26.665000 3997540 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3997675 closing signal SIGTERM
W0216 06:19:26.665000 3997540 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3997676 closing signal SIGTERM
W0216 06:19:26.665000 3997540 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3997677 closing signal SIGTERM
W0216 06:19:26.665000 3997540 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3997678 closing signal SIGTERM
W0216 06:19:26.666000 3997540 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3997679 closing signal SIGTERM
W0216 06:19:26.952000 3772629 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3773382 closing signal SIGTERM
W0216 06:19:26.953000 3772629 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3773383 closing signal SIGTERM
W0216 06:19:26.953000 3772629 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3773384 closing signal SIGTERM
W0216 06:19:26.954000 3772629 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3773385 closing signal SIGTERM
W0216 06:19:26.954000 3772629 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3773386 closing signal SIGTERM
W0216 06:19:26.954000 3772629 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3773387 closing signal SIGTERM
W0216 06:19:26.955000 3772629 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3773388 closing signal SIGTERM
W0216 06:19:27.140000 3261191 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3261322 closing signal SIGTERM
W0216 06:19:27.140000 3261191 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3261325 closing signal SIGTERM
W0216 06:19:27.141000 3261191 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3261326 closing signal SIGTERM
W0216 06:19:27.141000 3261191 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3261327 closing signal SIGTERM
W0216 06:19:27.141000 3261191 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3261328 closing signal SIGTERM
W0216 06:19:27.142000 3261191 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3261329 closing signal SIGTERM
E0216 06:19:27.524000 2664794 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:869] failed (exitcode: 1) local_rank: 7 (pid: 2664932) of binary: /usr/bin/python3.10
Traceback (most recent call last):
  File "/home/zhaojiang/.local/bin/torchrun", line 8, in <module>
    sys.exit(main())
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 355, in wrapper
    return f(*args, **kwargs)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/run.py", line 919, in main
    run(args)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/run.py", line 910, in run
    elastic_launch(
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 138, in __call__
    return launch_agent(self._config, self._entrypoint, list(args))
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 269, in launch_agent
    raise ChildFailedError(
torch.distributed.elastic.multiprocessing.errors.ChildFailedError: 
============================================================
llava/train/train_mem.py FAILED
------------------------------------------------------------
Failures:
  <NO_OTHER_FAILURES>
------------------------------------------------------------
Root Cause (first observed failure):
[0]:
  time      : 2025-02-16_06:19:25
  host      : h100-st-p548xlarge-52.ar-ai-use2.hpcaas
  rank      : 23 (local_rank: 7)
  exitcode  : 1 (pid: 2664932)
  error_file: <N/A>
  traceback : To enable traceback see: https://pytorch.org/docs/stable/elastic/errors.html
============================================================
E0216 06:19:27.679000 2133652 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:869] failed (exitcode: 1) local_rank: 7 (pid: 2133785) of binary: /usr/bin/python3.10
E0216 06:19:27.692000 3344412 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:869] failed (exitcode: 1) local_rank: 0 (pid: 3344544) of binary: /usr/bin/python3.10
Traceback (most recent call last):
  File "/home/zhaojiang/.local/bin/torchrun", line 8, in <module>
    sys.exit(main())
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 355, in wrapper
    return f(*args, **kwargs)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/run.py", line 919, in main
    run(args)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/run.py", line 910, in run
    elastic_launch(
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 138, in __call__
    return launch_agent(self._config, self._entrypoint, list(args))
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 269, in launch_agent
    raise ChildFailedError(
torch.distributed.elastic.multiprocessing.errors.ChildFailedError: 
============================================================
llava/train/train_mem.py FAILED
------------------------------------------------------------
Failures:
  <NO_OTHER_FAILURES>
------------------------------------------------------------
Root Cause (first observed failure):
[0]:
  time      : 2025-02-16_06:19:25
  host      : h100-st-p548xlarge-179.ar-ai-use2.hpcaas
  rank      : 111 (local_rank: 7)
  exitcode  : 1 (pid: 2133785)
  error_file: <N/A>
  traceback : To enable traceback see: https://pytorch.org/docs/stable/elastic/errors.html
============================================================
Traceback (most recent call last):
  File "/home/zhaojiang/.local/bin/torchrun", line 8, in <module>
    sys.exit(main())
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 355, in wrapper
    return f(*args, **kwargs)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/run.py", line 919, in main
    run(args)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/run.py", line 910, in run
    elastic_launch(
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 138, in __call__
    return launch_agent(self._config, self._entrypoint, list(args))
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 269, in launch_agent
    raise ChildFailedError(
torch.distributed.elastic.multiprocessing.errors.ChildFailedError: 
============================================================
llava/train/train_mem.py FAILED
------------------------------------------------------------
Failures:
  <NO_OTHER_FAILURES>
------------------------------------------------------------
Root Cause (first observed failure):
[0]:
  time      : 2025-02-16_06:19:25
  host      : h100-st-p548xlarge-74.ar-ai-use2.hpcaas
  rank      : 32 (local_rank: 0)
  exitcode  : 1 (pid: 3344544)
  error_file: <N/A>
  traceback : To enable traceback see: https://pytorch.org/docs/stable/elastic/errors.html
============================================================
E0216 06:19:27.770000 2201849 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:869] failed (exitcode: 1) local_rank: 2 (pid: 2201977) of binary: /usr/bin/python3.10
E0216 06:19:27.771000 3335470 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:869] failed (exitcode: 1) local_rank: 0 (pid: 3337098) of binary: /usr/bin/python3.10
Traceback (most recent call last):
  File "/home/zhaojiang/.local/bin/torchrun", line 8, in <module>
    sys.exit(main())
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 355, in wrapper
    return f(*args, **kwargs)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/run.py", line 919, in main
Traceback (most recent call last):
  File "/home/zhaojiang/.local/bin/torchrun", line 8, in <module>
    sys.exit(main())
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 355, in wrapper
    run(args)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/run.py", line 910, in run
    return f(*args, **kwargs)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/run.py", line 919, in main
    elastic_launch(
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 138, in __call__
    return launch_agent(self._config, self._entrypoint, list(args))
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 269, in launch_agent
    run(args)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/run.py", line 910, in run
    raise ChildFailedError(
torch.distributed.elastic.multiprocessing.errors.ChildFailedError: 
============================================================
llava/train/train_mem.py FAILED
------------------------------------------------------------
Failures:
[1]:
  time      : 2025-02-16_06:19:25
  host      : h100-st-p548xlarge-165.ar-ai-use2.hpcaas
  rank      : 77 (local_rank: 5)
  exitcode  : 1 (pid: 2201980)
  error_file: <N/A>
  traceback : To enable traceback see: https://pytorch.org/docs/stable/elastic/errors.html
------------------------------------------------------------
Root Cause (first observed failure):
[0]:
  time      : 2025-02-16_06:19:25
  host      : h100-st-p548xlarge-165.ar-ai-use2.hpcaas
  rank      : 74 (local_rank: 2)
  exitcode  : 1 (pid: 2201977)
  error_file: <N/A>
  traceback : To enable traceback see: https://pytorch.org/docs/stable/elastic/errors.html
============================================================
    elastic_launch(
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 138, in __call__
    return launch_agent(self._config, self._entrypoint, list(args))
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 269, in launch_agent
    raise ChildFailedError(
torch.distributed.elastic.multiprocessing.errors.ChildFailedError: 
============================================================
llava/train/train_mem.py FAILED
------------------------------------------------------------
Failures:
  <NO_OTHER_FAILURES>
------------------------------------------------------------
Root Cause (first observed failure):
[0]:
  time      : 2025-02-16_06:19:25
  host      : h100-st-p548xlarge-141.ar-ai-use2.hpcaas
  rank      : 64 (local_rank: 0)
  exitcode  : 1 (pid: 3337098)
  error_file: <N/A>
  traceback : To enable traceback see: https://pytorch.org/docs/stable/elastic/errors.html
============================================================
E0216 06:19:27.861000 4000190 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:869] failed (exitcode: 1) local_rank: 6 (pid: 4000950) of binary: /usr/bin/python3.10
srun: error: h100-st-p548xlarge-52: task 2: Exited with exit code 1
srun: Terminating StepId=335413.0
Traceback (most recent call last):
  File "/home/zhaojiang/.local/bin/torchrun", line 8, in <module>
    sys.exit(main())
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 355, in wrapper
    return f(*args, **kwargs)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/run.py", line 919, in main
    run(args)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/run.py", line 910, in run
    elastic_launch(
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 138, in __call__
    return launch_agent(self._config, self._entrypoint, list(args))
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 269, in launch_agent
    raise ChildFailedError(
torch.distributed.elastic.multiprocessing.errors.ChildFailedError: 
============================================================
llava/train/train_mem.py FAILED
------------------------------------------------------------
Failures:
  <NO_OTHER_FAILURES>
------------------------------------------------------------
Root Cause (first observed failure):
[0]:
  time      : 2025-02-16_06:19:26
  host      : h100-st-p548xlarge-206.ar-ai-use2.hpcaas
  rank      : 126 (local_rank: 6)
  exitcode  : 1 (pid: 4000950)
  error_file: <N/A>
  traceback : To enable traceback see: https://pytorch.org/docs/stable/elastic/errors.html
============================================================
slurmstepd: error: *** STEP 335413.0 ON h100-st-p548xlarge-41 CANCELLED AT 2025-02-16T06:19:27 ***
W0216 06:19:27.893000 3343136 .local/lib/python3.10/site-packages/torch/distributed/elastic/agent/server/api.py:704] Received Signals.SIGTERM death signal, shutting down workers
W0216 06:19:27.893000 3603613 .local/lib/python3.10/site-packages/torch/distributed/elastic/agent/server/api.py:704] Received Signals.SIGTERM death signal, shutting down workers
W0216 06:19:27.894000 3343136 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3343332 closing signal SIGTERM
W0216 06:19:27.894000 3791965 .local/lib/python3.10/site-packages/torch/distributed/elastic/agent/server/api.py:704] Received Signals.SIGTERM death signal, shutting down workers
W0216 06:19:27.894000 3772629 .local/lib/python3.10/site-packages/torch/distributed/elastic/agent/server/api.py:704] Received Signals.SIGTERM death signal, shutting down workers
W0216 06:19:27.893000 4003734 .local/lib/python3.10/site-packages/torch/distributed/elastic/agent/server/api.py:704] Received Signals.SIGTERM death signal, shutting down workers
W0216 06:19:27.894000 3603613 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3603751 closing signal SIGTERM
W0216 06:19:27.894000 3781647 .local/lib/python3.10/site-packages/torch/distributed/elastic/agent/server/api.py:704] Received Signals.SIGTERM death signal, shutting down workers
W0216 06:19:27.894000 2062756 .local/lib/python3.10/site-packages/torch/distributed/elastic/agent/server/api.py:704] Received Signals.SIGTERM death signal, shutting down workers
W0216 06:19:27.894000 3791965 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3792186 closing signal SIGTERM
W0216 06:19:27.894000 3262384 .local/lib/python3.10/site-packages/torch/distributed/elastic/agent/server/api.py:704] Received Signals.SIGTERM death signal, shutting down workers
W0216 06:19:27.894000 3772629 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3773382 closing signal SIGTERM
W0216 06:19:27.894000 3997540 .local/lib/python3.10/site-packages/torch/distributed/elastic/agent/server/api.py:704] Received Signals.SIGTERM death signal, shutting down workers
W0216 06:19:27.894000 4003734 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 4003869 closing signal SIGTERM
W0216 06:19:27.894000 3603613 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3603752 closing signal SIGTERM
W0216 06:19:27.894000 3781647 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3782423 closing signal SIGTERM
W0216 06:19:27.895000 3997540 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3997675 closing signal SIGTERM
W0216 06:19:27.894000 3262384 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3263052 closing signal SIGTERM
W0216 06:19:27.895000 3772629 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3773384 closing signal SIGTERM
W0216 06:19:27.895000 3781647 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3782424 closing signal SIGTERM
W0216 06:19:27.895000 3772629 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3773385 closing signal SIGTERM
W0216 06:19:27.895000 3997540 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3997678 closing signal SIGTERM
W0216 06:19:27.895000 3262384 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3263057 closing signal SIGTERM
W0216 06:19:27.895000 3772629 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3773386 closing signal SIGTERM
W0216 06:19:27.895000 3997540 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3997679 closing signal SIGTERM
W0216 06:19:27.896000 3772629 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3773387 closing signal SIGTERM
W0216 06:19:27.896000 3772629 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3773388 closing signal SIGTERM
W0216 06:19:27.919000 3261191 .local/lib/python3.10/site-packages/torch/distributed/elastic/agent/server/api.py:704] Received Signals.SIGTERM death signal, shutting down workers
W0216 06:19:27.919000 3261191 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3261325 closing signal SIGTERM
W0216 06:19:27.920000 3261191 .local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 3261327 closing signal SIGTERM
Traceback (most recent call last):
  File "/home/zhaojiang/.local/bin/torchrun", line 8, in <module>
    sys.exit(main())
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 355, in wrapper
    return f(*args, **kwargs)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/run.py", line 919, in main
    run(args)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/run.py", line 910, in run
    elastic_launch(
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 138, in __call__
    return launch_agent(self._config, self._entrypoint, list(args))
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 260, in launch_agent
    result = agent.run()
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/metrics/api.py", line 137, in wrapper
    result = f(*args, **kwargs)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/agent/server/api.py", line 696, in run
    result = self._invoke_run(role)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/agent/server/api.py", line 856, in _invoke_run
    run_result = self._monitor_workers(self._worker_group)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/metrics/api.py", line 137, in wrapper
    result = f(*args, **kwargs)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/agent/server/local_elastic_agent.py", line 387, in _monitor_workers
    result = self._pcontext.wait(0)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py", line 531, in wait
    return self._poll()
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py", line 861, in _poll
    self.close()  # terminate all running procs
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py", line 572, in close
    self._close(death_sig=death_sig, timeout=timeout)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py", line 909, in _close
    handler.proc.wait(time_to_wait)
  File "/usr/lib/python3.10/subprocess.py", line 1209, in wait
    return self._wait(timeout=timeout)
  File "/usr/lib/python3.10/subprocess.py", line 1953, in _wait
    time.sleep(delay)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py", line 84, in _terminate_process_handler
    raise SignalException(f"Process {os.getpid()} got signal: {sigval}", sigval=sigval)
torch.distributed.elastic.multiprocessing.api.SignalException: Process 2062756 got signal: 15
Traceback (most recent call last):
  File "/home/zhaojiang/.local/bin/torchrun", line 8, in <module>
    sys.exit(main())
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 355, in wrapper
    return f(*args, **kwargs)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/run.py", line 919, in main
    run(args)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/run.py", line 910, in run
    elastic_launch(
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 138, in __call__
    return launch_agent(self._config, self._entrypoint, list(args))
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 260, in launch_agent
    result = agent.run()
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/metrics/api.py", line 137, in wrapper
    result = f(*args, **kwargs)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/agent/server/api.py", line 696, in run
    result = self._invoke_run(role)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/agent/server/api.py", line 856, in _invoke_run
    run_result = self._monitor_workers(self._worker_group)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/metrics/api.py", line 137, in wrapper
    result = f(*args, **kwargs)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/agent/server/local_elastic_agent.py", line 387, in _monitor_workers
    result = self._pcontext.wait(0)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py", line 531, in wait
    return self._poll()
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py", line 861, in _poll
    self.close()  # terminate all running procs
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py", line 572, in close
    self._close(death_sig=death_sig, timeout=timeout)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py", line 909, in _close
    handler.proc.wait(time_to_wait)
  File "/usr/lib/python3.10/subprocess.py", line 1209, in wait
    return self._wait(timeout=timeout)
  File "/usr/lib/python3.10/subprocess.py", line 1953, in _wait
    time.sleep(delay)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py", line 84, in _terminate_process_handler
    raise SignalException(f"Process {os.getpid()} got signal: {sigval}", sigval=sigval)
torch.distributed.elastic.multiprocessing.api.SignalException: Process 3791965 got signal: 15
srun: error: h100-st-p548xlarge-74: task 4: Terminated
srun: error: h100-st-p548xlarge-206: task 15: Terminated
srun: error: h100-st-p548xlarge-165: task 9: Terminated
srun: error: h100-st-p548xlarge-179: task 13: Terminated
srun: error: h100-st-p548xlarge-141: task 8: Terminated
Traceback (most recent call last):
  File "/home/zhaojiang/.local/bin/torchrun", line 8, in <module>
    sys.exit(main())
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 355, in wrapper
    return f(*args, **kwargs)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/run.py", line 919, in main
    run(args)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/run.py", line 910, in run
    elastic_launch(
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 138, in __call__
    return launch_agent(self._config, self._entrypoint, list(args))
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 260, in launch_agent
    result = agent.run()
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/metrics/api.py", line 137, in wrapper
    result = f(*args, **kwargs)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/agent/server/api.py", line 696, in run
    result = self._invoke_run(role)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/agent/server/api.py", line 856, in _invoke_run
    run_result = self._monitor_workers(self._worker_group)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/metrics/api.py", line 137, in wrapper
Traceback (most recent call last):
  File "/home/zhaojiang/.local/bin/torchrun", line 8, in <module>
    result = f(*args, **kwargs)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/agent/server/local_elastic_agent.py", line 387, in _monitor_workers
    sys.exit(main())
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 355, in wrapper
    result = self._pcontext.wait(0)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py", line 531, in wait
    return f(*args, **kwargs)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/run.py", line 919, in main
    return self._poll()
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py", line 861, in _poll
    self.close()  # terminate all running procs
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py", line 572, in close
    run(args)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/run.py", line 910, in run
    self._close(death_sig=death_sig, timeout=timeout)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py", line 909, in _close
    handler.proc.wait(time_to_wait)
  File "/usr/lib/python3.10/subprocess.py", line 1209, in wait
    elastic_launch(
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 138, in __call__
    return self._wait(timeout=timeout)
  File "/usr/lib/python3.10/subprocess.py", line 1953, in _wait
    time.sleep(delay)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py", line 84, in _terminate_process_handler
    raise SignalException(f"Process {os.getpid()} got signal: {sigval}", sigval=sigval)
torch.distributed.elastic.multiprocessing.api.SignalException: Process 3343136 got signal: 15
    return launch_agent(self._config, self._entrypoint, list(args))
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 260, in launch_agent
    result = agent.run()
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/metrics/api.py", line 137, in wrapper
    result = f(*args, **kwargs)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/agent/server/api.py", line 696, in run
    result = self._invoke_run(role)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/agent/server/api.py", line 856, in _invoke_run
    run_result = self._monitor_workers(self._worker_group)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/metrics/api.py", line 137, in wrapper
    result = f(*args, **kwargs)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/agent/server/local_elastic_agent.py", line 387, in _monitor_workers
    result = self._pcontext.wait(0)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py", line 531, in wait
    return self._poll()
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py", line 861, in _poll
    self.close()  # terminate all running procs
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py", line 572, in close
    self._close(death_sig=death_sig, timeout=timeout)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py", line 909, in _close
    handler.proc.wait(time_to_wait)
  File "/usr/lib/python3.10/subprocess.py", line 1209, in wait
    return self._wait(timeout=timeout)
  File "/usr/lib/python3.10/subprocess.py", line 1953, in _wait
    time.sleep(delay)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py", line 84, in _terminate_process_handler
    raise SignalException(f"Process {os.getpid()} got signal: {sigval}", sigval=sigval)
torch.distributed.elastic.multiprocessing.api.SignalException: Process 4003734 got signal: 15
Traceback (most recent call last):
  File "/home/zhaojiang/.local/bin/torchrun", line 8, in <module>
    sys.exit(main())
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 355, in wrapper
    return f(*args, **kwargs)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/run.py", line 919, in main
    run(args)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/run.py", line 910, in run
    elastic_launch(
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 138, in __call__
    return launch_agent(self._config, self._entrypoint, list(args))
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 260, in launch_agent
    result = agent.run()
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/metrics/api.py", line 137, in wrapper
    result = f(*args, **kwargs)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/agent/server/api.py", line 696, in run
    result = self._invoke_run(role)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/agent/server/api.py", line 856, in _invoke_run
    run_result = self._monitor_workers(self._worker_group)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/metrics/api.py", line 137, in wrapper
    result = f(*args, **kwargs)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/agent/server/local_elastic_agent.py", line 387, in _monitor_workers
    result = self._pcontext.wait(0)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py", line 531, in wait
    return self._poll()
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py", line 861, in _poll
    self.close()  # terminate all running procs
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py", line 572, in close
    self._close(death_sig=death_sig, timeout=timeout)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py", line 909, in _close
    handler.proc.wait(time_to_wait)
  File "/usr/lib/python3.10/subprocess.py", line 1209, in wait
    return self._wait(timeout=timeout)
  File "/usr/lib/python3.10/subprocess.py", line 1953, in _wait
    time.sleep(delay)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py", line 84, in _terminate_process_handler
    raise SignalException(f"Process {os.getpid()} got signal: {sigval}", sigval=sigval)
torch.distributed.elastic.multiprocessing.api.SignalException: Process 3781647 got signal: 15
Traceback (most recent call last):
  File "/home/zhaojiang/.local/bin/torchrun", line 8, in <module>
    sys.exit(main())
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 355, in wrapper
    return f(*args, **kwargs)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/run.py", line 919, in main
    run(args)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/run.py", line 910, in run
    elastic_launch(
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 138, in __call__
    return launch_agent(self._config, self._entrypoint, list(args))
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 260, in launch_agent
    result = agent.run()
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/metrics/api.py", line 137, in wrapper
    result = f(*args, **kwargs)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/agent/server/api.py", line 696, in run
    result = self._invoke_run(role)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/agent/server/api.py", line 856, in _invoke_run
    run_result = self._monitor_workers(self._worker_group)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/metrics/api.py", line 137, in wrapper
    result = f(*args, **kwargs)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/agent/server/local_elastic_agent.py", line 387, in _monitor_workers
    result = self._pcontext.wait(0)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py", line 531, in wait
    return self._poll()
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py", line 861, in _poll
    self.close()  # terminate all running procs
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py", line 572, in close
    self._close(death_sig=death_sig, timeout=timeout)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py", line 909, in _close
    handler.proc.wait(time_to_wait)
  File "/usr/lib/python3.10/subprocess.py", line 1209, in wait
    return self._wait(timeout=timeout)
  File "/usr/lib/python3.10/subprocess.py", line 1953, in _wait
    time.sleep(delay)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py", line 84, in _terminate_process_handler
    raise SignalException(f"Process {os.getpid()} got signal: {sigval}", sigval=sigval)
torch.distributed.elastic.multiprocessing.api.SignalException: Process 3262384 got signal: 15
srun: error: h100-st-p548xlarge-169: task 10: Exited with exit code 1
srun: error: h100-st-p548xlarge-42: task 1: Exited with exit code 1
Traceback (most recent call last):
  File "/home/zhaojiang/.local/bin/torchrun", line 8, in <module>
    sys.exit(main())
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 355, in wrapper
    return f(*args, **kwargs)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/run.py", line 919, in main
    run(args)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/run.py", line 910, in run
    elastic_launch(
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 138, in __call__
    return launch_agent(self._config, self._entrypoint, list(args))
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 260, in launch_agent
    result = agent.run()
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/metrics/api.py", line 137, in wrapper
    result = f(*args, **kwargs)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/agent/server/api.py", line 696, in run
    result = self._invoke_run(role)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/agent/server/api.py", line 856, in _invoke_run
    run_result = self._monitor_workers(self._worker_group)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/metrics/api.py", line 137, in wrapper
    result = f(*args, **kwargs)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/agent/server/local_elastic_agent.py", line 387, in _monitor_workers
    result = self._pcontext.wait(0)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py", line 531, in wait
    return self._poll()
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py", line 861, in _poll
    self.close()  # terminate all running procs
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py", line 572, in close
    self._close(death_sig=death_sig, timeout=timeout)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py", line 909, in _close
    handler.proc.wait(time_to_wait)
  File "/usr/lib/python3.10/subprocess.py", line 1209, in wait
    return self._wait(timeout=timeout)
  File "/usr/lib/python3.10/subprocess.py", line 1953, in _wait
    time.sleep(delay)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py", line 84, in _terminate_process_handler
    raise SignalException(f"Process {os.getpid()} got signal: {sigval}", sigval=sigval)
torch.distributed.elastic.multiprocessing.api.SignalException: Process 3603613 got signal: 15
srun: error: h100-st-p548xlarge-75: task 5: Exited with exit code 1
srun: error: h100-st-p548xlarge-183: task 14: Exited with exit code 1
srun: error: h100-st-p548xlarge-170: task 11: Exited with exit code 1
Traceback (most recent call last):
  File "/home/zhaojiang/.local/bin/torchrun", line 8, in <module>
    sys.exit(main())
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 355, in wrapper
    return f(*args, **kwargs)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/run.py", line 919, in main
    run(args)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/run.py", line 910, in run
    elastic_launch(
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 138, in __call__
    return launch_agent(self._config, self._entrypoint, list(args))
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 260, in launch_agent
    result = agent.run()
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/metrics/api.py", line 137, in wrapper
    result = f(*args, **kwargs)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/agent/server/api.py", line 696, in run
    result = self._invoke_run(role)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/agent/server/api.py", line 856, in _invoke_run
    run_result = self._monitor_workers(self._worker_group)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/metrics/api.py", line 137, in wrapper
    result = f(*args, **kwargs)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/agent/server/local_elastic_agent.py", line 387, in _monitor_workers
    result = self._pcontext.wait(0)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py", line 531, in wait
    return self._poll()
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py", line 861, in _poll
    self.close()  # terminate all running procs
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py", line 572, in close
    self._close(death_sig=death_sig, timeout=timeout)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py", line 909, in _close
    handler.proc.wait(time_to_wait)
  File "/usr/lib/python3.10/subprocess.py", line 1209, in wait
    return self._wait(timeout=timeout)
  File "/usr/lib/python3.10/subprocess.py", line 1953, in _wait
    time.sleep(delay)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py", line 84, in _terminate_process_handler
    raise SignalException(f"Process {os.getpid()} got signal: {sigval}", sigval=sigval)
torch.distributed.elastic.multiprocessing.api.SignalException: Process 3997540 got signal: 15
srun: error: h100-st-p548xlarge-65: task 3: Exited with exit code 1
srun: error: h100-st-p548xlarge-113: task 7: Exited with exit code 1
srun: error: h100-st-p548xlarge-112: task 6: Exited with exit code 1
Traceback (most recent call last):
  File "/home/zhaojiang/.local/bin/torchrun", line 8, in <module>
    sys.exit(main())
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 355, in wrapper
    return f(*args, **kwargs)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/run.py", line 919, in main
    run(args)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/run.py", line 910, in run
    elastic_launch(
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 138, in __call__
    return launch_agent(self._config, self._entrypoint, list(args))
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 260, in launch_agent
    result = agent.run()
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/metrics/api.py", line 137, in wrapper
    result = f(*args, **kwargs)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/agent/server/api.py", line 696, in run
    result = self._invoke_run(role)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/agent/server/api.py", line 856, in _invoke_run
    run_result = self._monitor_workers(self._worker_group)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/metrics/api.py", line 137, in wrapper
    result = f(*args, **kwargs)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/agent/server/local_elastic_agent.py", line 387, in _monitor_workers
    result = self._pcontext.wait(0)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py", line 531, in wait
    return self._poll()
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py", line 861, in _poll
    self.close()  # terminate all running procs
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py", line 572, in close
    self._close(death_sig=death_sig, timeout=timeout)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py", line 909, in _close
    handler.proc.wait(time_to_wait)
  File "/usr/lib/python3.10/subprocess.py", line 1209, in wait
    return self._wait(timeout=timeout)
  File "/usr/lib/python3.10/subprocess.py", line 1953, in _wait
    time.sleep(delay)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py", line 84, in _terminate_process_handler
    raise SignalException(f"Process {os.getpid()} got signal: {sigval}", sigval=sigval)
torch.distributed.elastic.multiprocessing.api.SignalException: Process 3261191 got signal: 15
srun: error: h100-st-p548xlarge-41: task 0: Exited with exit code 1
Traceback (most recent call last):
  File "/home/zhaojiang/.local/bin/torchrun", line 8, in <module>
    sys.exit(main())
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 355, in wrapper
    return f(*args, **kwargs)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/run.py", line 919, in main
    run(args)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/run.py", line 910, in run
    elastic_launch(
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 138, in __call__
    return launch_agent(self._config, self._entrypoint, list(args))
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 260, in launch_agent
    result = agent.run()
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/metrics/api.py", line 137, in wrapper
    result = f(*args, **kwargs)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/agent/server/api.py", line 696, in run
    result = self._invoke_run(role)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/agent/server/api.py", line 856, in _invoke_run
    run_result = self._monitor_workers(self._worker_group)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/metrics/api.py", line 137, in wrapper
    result = f(*args, **kwargs)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/agent/server/local_elastic_agent.py", line 387, in _monitor_workers
    result = self._pcontext.wait(0)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py", line 531, in wait
    return self._poll()
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py", line 861, in _poll
    self.close()  # terminate all running procs
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py", line 572, in close
    self._close(death_sig=death_sig, timeout=timeout)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py", line 909, in _close
    handler.proc.wait(time_to_wait)
  File "/usr/lib/python3.10/subprocess.py", line 1209, in wait
    return self._wait(timeout=timeout)
  File "/usr/lib/python3.10/subprocess.py", line 1953, in _wait
    time.sleep(delay)
  File "/home/zhaojiang/.local/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/api.py", line 84, in _terminate_process_handler
    raise SignalException(f"Process {os.getpid()} got signal: {sigval}", sigval=sigval)
torch.distributed.elastic.multiprocessing.api.SignalException: Process 3772629 got signal: 15
srun: error: h100-st-p548xlarge-172: task 12: Exited with exit code 1
srun: Force Terminated StepId=335413.0
pretrain.sh: 82: python: not found