[INFO|2025-05-29 20:54:01] configuration_utils.py:696 >> loading configuration file config.json from cache at /home/kiho/.cache/huggingface/hub/models--meta-llama--Llama-3.1-8B-Instruct/snapshots/0e9e39f249a16976918f6564b8830bc894c89659/config.json [INFO|2025-05-29 20:54:01] configuration_utils.py:768 >> Model config LlamaConfig { "_name_or_path": "meta-llama/Llama-3.1-8B-Instruct", "architectures": [ "LlamaForCausalLM" ], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 128000, "eos_token_id": [ 128001, 128008, 128009 ], "head_dim": 128, "hidden_act": "silu", "hidden_size": 4096, "initializer_range": 0.02, "intermediate_size": 14336, "max_position_embeddings": 131072, "mlp_bias": false, "model_type": "llama", "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 8, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": { "factor": 8.0, "high_freq_factor": 4.0, "low_freq_factor": 1.0, "original_max_position_embeddings": 8192, "rope_type": "llama3" }, "rope_theta": 500000.0, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "transformers_version": "4.48.2", "use_cache": true, "vocab_size": 128256 } [INFO|2025-05-29 20:54:01] tokenization_utils_base.py:2034 >> loading file tokenizer.json from cache at /home/kiho/.cache/huggingface/hub/models--meta-llama--Llama-3.1-8B-Instruct/snapshots/0e9e39f249a16976918f6564b8830bc894c89659/tokenizer.json [INFO|2025-05-29 20:54:01] tokenization_utils_base.py:2034 >> loading file tokenizer.model from cache at None [INFO|2025-05-29 20:54:01] tokenization_utils_base.py:2034 >> loading file added_tokens.json from cache at None [INFO|2025-05-29 20:54:01] tokenization_utils_base.py:2034 >> loading file special_tokens_map.json from cache at /home/kiho/.cache/huggingface/hub/models--meta-llama--Llama-3.1-8B-Instruct/snapshots/0e9e39f249a16976918f6564b8830bc894c89659/special_tokens_map.json [INFO|2025-05-29 20:54:01] tokenization_utils_base.py:2034 >> loading file tokenizer_config.json from cache at /home/kiho/.cache/huggingface/hub/models--meta-llama--Llama-3.1-8B-Instruct/snapshots/0e9e39f249a16976918f6564b8830bc894c89659/tokenizer_config.json [INFO|2025-05-29 20:54:01] tokenization_utils_base.py:2034 >> loading file chat_template.jinja from cache at None [INFO|2025-05-29 20:54:01] tokenization_utils_base.py:2304 >> Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained. [INFO|2025-05-29 20:54:01] logging.py:157 >> Add pad token: <|eot_id|> [INFO|2025-05-29 20:54:01] logging.py:157 >> Add <|eot_id|>,<|eom_id|> to stop words. [INFO|2025-05-29 20:54:01] logging.py:157 >> Loading dataset Codes_query_filtered_330k_ns_over8_1.json... [INFO|2025-05-29 20:54:13] configuration_utils.py:696 >> loading configuration file config.json from cache at /home/kiho/.cache/huggingface/hub/models--meta-llama--Llama-3.1-8B-Instruct/snapshots/0e9e39f249a16976918f6564b8830bc894c89659/config.json [INFO|2025-05-29 20:54:13] configuration_utils.py:768 >> Model config LlamaConfig { "_name_or_path": "meta-llama/Llama-3.1-8B-Instruct", "architectures": [ "LlamaForCausalLM" ], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 128000, "eos_token_id": [ 128001, 128008, 128009 ], "head_dim": 128, "hidden_act": "silu", "hidden_size": 4096, "initializer_range": 0.02, "intermediate_size": 14336, "max_position_embeddings": 131072, "mlp_bias": false, "model_type": "llama", "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 8, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": { "factor": 8.0, "high_freq_factor": 4.0, "low_freq_factor": 1.0, "original_max_position_embeddings": 8192, "rope_type": "llama3" }, "rope_theta": 500000.0, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "transformers_version": "4.48.2", "use_cache": true, "vocab_size": 128256 } [WARNING|2025-05-29 20:54:13] logging.py:162 >> Input length is smaller than max length. Consider increase input length. [INFO|2025-05-29 20:54:13] logging.py:157 >> Using llama3 scaling strategy and setting scaling factor to 1.0. [INFO|2025-05-29 20:54:13] logging.py:157 >> Using block diagonal attention for sequence packing without cross-attention. [INFO|2025-05-29 20:54:13] logging.py:157 >> Liger kernel has been applied to the model. [INFO|2025-05-29 20:54:13] modeling_utils.py:3904 >> loading weights file model.safetensors from cache at /home/kiho/.cache/huggingface/hub/models--meta-llama--Llama-3.1-8B-Instruct/snapshots/0e9e39f249a16976918f6564b8830bc894c89659/model.safetensors.index.json [INFO|2025-05-29 20:54:13] modeling_utils.py:1582 >> Instantiating LlamaForCausalLM model under default dtype torch.bfloat16. [INFO|2025-05-29 20:54:13] configuration_utils.py:1140 >> Generate config GenerationConfig { "bos_token_id": 128000, "eos_token_id": [ 128001, 128008, 128009 ] } [INFO|2025-05-29 20:54:17] modeling_utils.py:4888 >> All model checkpoint weights were used when initializing LlamaForCausalLM. [INFO|2025-05-29 20:54:17] modeling_utils.py:4896 >> All the weights of LlamaForCausalLM were initialized from the model checkpoint at meta-llama/Llama-3.1-8B-Instruct. If your task is similar to the task the model of the checkpoint was trained on, you can already use LlamaForCausalLM for predictions without further training. [INFO|2025-05-29 20:54:17] configuration_utils.py:1095 >> loading configuration file generation_config.json from cache at /home/kiho/.cache/huggingface/hub/models--meta-llama--Llama-3.1-8B-Instruct/snapshots/0e9e39f249a16976918f6564b8830bc894c89659/generation_config.json [INFO|2025-05-29 20:54:17] configuration_utils.py:1140 >> Generate config GenerationConfig { "bos_token_id": 128000, "do_sample": true, "eos_token_id": [ 128001, 128008, 128009 ], "temperature": 0.6, "top_p": 0.9 } [INFO|2025-05-29 20:54:18] logging.py:157 >> Gradient checkpointing enabled. [INFO|2025-05-29 20:54:18] logging.py:157 >> Using torch SDPA for faster training and inference. [INFO|2025-05-29 20:54:18] logging.py:157 >> Upcasting trainable params to float32. [INFO|2025-05-29 20:54:18] logging.py:157 >> Fine-tuning method: Freeze [INFO|2025-05-29 20:54:18] logging.py:157 >> Set trainable layers: .30.,.31. [INFO|2025-05-29 20:54:18] logging.py:157 >> trainable params: 436,224,000 || all params: 8,030,261,248 || trainable%: 5.4323 [INFO|2025-05-29 20:54:18] trainer.py:741 >> Using auto half precision backend [INFO|2025-05-29 20:54:18] logging.py:157 >> Found linear modules: o_proj,down_proj,k_proj,q_proj,up_proj,gate_proj,v_proj [INFO|2025-05-29 20:54:18] logging.py:157 >> Using APOLLO optimizer with args: {'rank': 256, 'proj': 'random', 'proj_type': 'std', 'update_proj_gap': 200, 'scale': 1, 'scale_type': 'channel', 'scale_front': False}. [INFO|2025-05-29 20:54:18] trainer.py:2369 >> ***** Running training ***** [INFO|2025-05-29 20:54:18] trainer.py:2370 >> Num examples = 4,433 [INFO|2025-05-29 20:54:18] trainer.py:2371 >> Num Epochs = 1 [INFO|2025-05-29 20:54:18] trainer.py:2372 >> Instantaneous batch size per device = 16 [INFO|2025-05-29 20:54:18] trainer.py:2375 >> Total train batch size (w. parallel, distributed & accumulation) = 512 [INFO|2025-05-29 20:54:18] trainer.py:2376 >> Gradient Accumulation steps = 8 [INFO|2025-05-29 20:54:18] trainer.py:2377 >> Total optimization steps = 8 [INFO|2025-05-29 20:54:18] trainer.py:2378 >> Number of trainable parameters = 436,224,000 [INFO|2025-05-29 20:56:12] logging.py:157 >> {'loss': 1.1291, 'learning_rate': 4.8097e-05, 'epoch': 0.11, 'throughput': 18586.20} [INFO|2025-05-29 20:57:58] logging.py:157 >> {'loss': 1.1246, 'learning_rate': 4.2678e-05, 'epoch': 0.23, 'throughput': 19137.96} [INFO|2025-05-29 20:59:46] logging.py:157 >> {'loss': 1.0721, 'learning_rate': 3.4567e-05, 'epoch': 0.34, 'throughput': 19271.44} [INFO|2025-05-29 21:01:32] logging.py:157 >> {'loss': 1.0703, 'learning_rate': 2.5000e-05, 'epoch': 0.46, 'throughput': 19369.21} [INFO|2025-05-29 21:03:19] logging.py:157 >> {'loss': 1.0601, 'learning_rate': 1.5433e-05, 'epoch': 0.57, 'throughput': 19438.28} [INFO|2025-05-29 21:05:05] logging.py:157 >> {'loss': 1.0241, 'learning_rate': 7.3223e-06, 'epoch': 0.69, 'throughput': 19487.62} [INFO|2025-05-29 21:06:51] logging.py:157 >> {'loss': 1.0462, 'learning_rate': 1.9030e-06, 'epoch': 0.80, 'throughput': 19522.08} [INFO|2025-05-29 21:08:38] logging.py:157 >> {'loss': 1.0376, 'learning_rate': 0.0000e+00, 'epoch': 0.91, 'throughput': 19546.66} [INFO|2025-05-29 21:08:38] trainer.py:3910 >> Saving model checkpoint to saves/Llama-3.1-8B-Instruct/freeze/llama_nsx_8_1/checkpoint-8 [INFO|2025-05-29 21:08:38] configuration_utils.py:420 >> Configuration saved in saves/Llama-3.1-8B-Instruct/freeze/llama_nsx_8_1/checkpoint-8/config.json [INFO|2025-05-29 21:08:38] configuration_utils.py:909 >> Configuration saved in saves/Llama-3.1-8B-Instruct/freeze/llama_nsx_8_1/checkpoint-8/generation_config.json [INFO|2025-05-29 21:09:01] modeling_utils.py:2996 >> The model is bigger than the maximum size per checkpoint (5GB) and is going to be split in 4 checkpoint shards. You can find where each parameters has been saved in the index located at saves/Llama-3.1-8B-Instruct/freeze/llama_nsx_8_1/checkpoint-8/model.safetensors.index.json. [INFO|2025-05-29 21:09:01] tokenization_utils_base.py:2491 >> tokenizer config file saved in saves/Llama-3.1-8B-Instruct/freeze/llama_nsx_8_1/checkpoint-8/tokenizer_config.json [INFO|2025-05-29 21:09:01] tokenization_utils_base.py:2500 >> Special tokens file saved in saves/Llama-3.1-8B-Instruct/freeze/llama_nsx_8_1/checkpoint-8/special_tokens_map.json [INFO|2025-05-29 21:09:02] trainer.py:2643 >> Training completed. Do not forget to share your model on huggingface.co/models =) [INFO|2025-05-29 21:09:02] trainer.py:3910 >> Saving model checkpoint to saves/Llama-3.1-8B-Instruct/freeze/llama_nsx_8_1 [INFO|2025-05-29 21:09:02] configuration_utils.py:420 >> Configuration saved in saves/Llama-3.1-8B-Instruct/freeze/llama_nsx_8_1/config.json [INFO|2025-05-29 21:09:02] configuration_utils.py:909 >> Configuration saved in saves/Llama-3.1-8B-Instruct/freeze/llama_nsx_8_1/generation_config.json [INFO|2025-05-29 21:09:26] modeling_utils.py:2996 >> The model is bigger than the maximum size per checkpoint (5GB) and is going to be split in 4 checkpoint shards. You can find where each parameters has been saved in the index located at saves/Llama-3.1-8B-Instruct/freeze/llama_nsx_8_1/model.safetensors.index.json. [INFO|2025-05-29 21:09:26] tokenization_utils_base.py:2491 >> tokenizer config file saved in saves/Llama-3.1-8B-Instruct/freeze/llama_nsx_8_1/tokenizer_config.json [INFO|2025-05-29 21:09:26] tokenization_utils_base.py:2500 >> Special tokens file saved in saves/Llama-3.1-8B-Instruct/freeze/llama_nsx_8_1/special_tokens_map.json [WARNING|2025-05-29 21:09:26] logging.py:162 >> No metric eval_loss to plot. [WARNING|2025-05-29 21:09:26] logging.py:162 >> No metric eval_accuracy to plot. [INFO|2025-05-29 21:09:26] modelcard.py:449 >> Dropping the following result as it does not have all the necessary fields: {'task': {'name': 'Causal Language Modeling', 'type': 'text-generation'}}