Synchronizing local compiler cache.

.gitattributes

@@ -2374,3 +2374,37 @@ neuronxcc-2.13.66.0+6dfecc895/MODULE_24b374d43be11ae153c9+2c2d707e/model.neff fi
 neuronxcc-2.13.66.0+6dfecc895/MODULE_ed08451c8d838fd56480+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.13.66.0+6dfecc895/MODULE_00c44e05161052cccaa0+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.13.66.0+6dfecc895/MODULE_e60e150057b9b6b21173+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.66.0+6dfecc895/MODULE_0198baf5b6a9d47329e3+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.66.0+6dfecc895/MODULE_09c6476126adb91dbd05+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.66.0+6dfecc895/MODULE_3ee2119c36456c7be776+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.66.0+6dfecc895/MODULE_451211c384fc26575d88+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.66.0+6dfecc895/MODULE_a930aaa3b948dbec7f26+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.66.0+6dfecc895/MODULE_bc523953f1f486abaf54+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.66.0+6dfecc895/MODULE_bce8999563afe6e1af7e+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.66.0+6dfecc895/MODULE_c5913bb260db80d4e0f7+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.66.0+6dfecc895/MODULE_c6268311797d0ffe8898+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.66.0+6dfecc895/MODULE_c92ea0d1669b390840ac+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.66.0+6dfecc895/MODULE_e67f6c821525bd111b17+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.66.0+6dfecc895/MODULE_e6b09d368e3c27ab7a76+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.68.0+6dfecc895/MODULE_0406d654284cbd36a965+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.68.0+6dfecc895/MODULE_14a4877b2bf020e4c4f7+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.68.0+6dfecc895/MODULE_21d7ca83fa40eaf22249+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.68.0+6dfecc895/MODULE_277942fefc2990a108db+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.68.0+6dfecc895/MODULE_308d19f6ceb008fe81c6+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.68.0+6dfecc895/MODULE_41686c6d7dab61752a4b+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.68.0+6dfecc895/MODULE_42adf97202125e822891+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.68.0+6dfecc895/MODULE_4bbdb8b4a9b111f388c0+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.68.0+6dfecc895/MODULE_4e02f6a89a1b9523cdb6+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.68.0+6dfecc895/MODULE_65cd20f6ba12834d6aee+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.68.0+6dfecc895/MODULE_7df0e1eba454893a7358+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.68.0+6dfecc895/MODULE_9fb4f17da3ced15711f2+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.68.0+6dfecc895/MODULE_b521a6c9af84469e7c0b+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.68.0+6dfecc895/MODULE_d9a0688e74add49a38ab+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.68.0+6dfecc895/MODULE_dcbcfc977b79c7f1df4e+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.68.0+6dfecc895/MODULE_de41be3e18f84c316d01+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.68.0+6dfecc895/MODULE_e08070f3a58b9fa8575f+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.68.0+6dfecc895/MODULE_eea3cf695e9cf086ac91+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.68.0+6dfecc895/MODULE_f1f512eea34f52b7f8d9+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.68.0+6dfecc895/MODULE_f38d6a7d178ff69e2bb3+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.68.0+6dfecc895/MODULE_f7d991bf6082f2b30284+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.68.0+6dfecc895/MODULE_faad03c30fe250362f50+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text

neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.22.dev0/inference/bloom/hf-internal-testing/tiny-random-BloomForCausalLM/f9a5439ee67f962284df.json

@@ -0,0 +1 @@
+{"apply_residual_connection_post_layernorm": false, "architectures": ["BloomForCausalLM"], "attention_dropout": 0.1, "bos_token_id": 1, "dtype": "float32", "eos_token_id": 2, "gradient_checkpointing": false, "hidden_dropout": 0.1, "hidden_size": 32, "id2label": {"0": "LABEL_0", "1": "LABEL_1", "2": "LABEL_2"}, "initializer_range": 0.02, "is_decoder": true, "label2id": {"LABEL_0": 0, "LABEL_1": 1, "LABEL_2": 2}, "layer_norm_epsilon": 1e-05, "model_type": "bloom", "n_head": 4, "n_layer": 5, "n_positions": 512, "neuron": {"auto_cast_type": "fp32", "batch_size": 2, "checkpoint_id": "hf-internal-testing/tiny-random-BloomForCausalLM", "checkpoint_revision": "92b07e9b7b4f986fa7c54e2ac3b9201b4ba5212e", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 2, "sequence_length": 512, "task": "text-generation"}, "pad_token_id": 3, "pretraining_tp": 1, "seq_length": 7, "slow_but_exact": true, "torch_dtype": "float32", "type_vocab_size": 16, "use_cache": true, "vocab_size": 1024}

neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.22.dev0/inference/gpt2/gpt2/65cb8c493bb52d550958.json

@@ -0,0 +1 @@
+{"activation_function": "gelu_new", "architectures": ["GPT2LMHeadModel"], "attn_pdrop": 0.1, "bos_token_id": 50256, "embd_pdrop": 0.1, "eos_token_id": 50256, "initializer_range": 0.02, "layer_norm_epsilon": 1e-05, "model_type": "gpt2", "n_ctx": 1024, "n_embd": 768, "n_head": 12, "n_inner": null, "n_layer": 12, "n_positions": 1024, "neuron": {"auto_cast_type": "fp32", "batch_size": 2, "checkpoint_id": "gpt2", "checkpoint_revision": "607a30d783dfa663caf39e06633721c8d4cfcd7e", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 2, "sequence_length": 1024, "task": "text-generation"}, "reorder_and_upcast_attn": false, "resid_pdrop": 0.1, "scale_attn_by_inverse_layer_idx": false, "scale_attn_weights": true, "summary_activation": null, "summary_first_dropout": 0.1, "summary_proj_to_labels": true, "summary_type": "cls_index", "summary_use_proj": true, "task_specific_params": {"text-generation": {"do_sample": true, "max_length": 50}}, "use_cache": true, "vocab_size": 50257}

neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.22.dev0/inference/gpt2/hf-internal-testing/tiny-random-gpt2/2713954a7f357d88b849.json

@@ -0,0 +1 @@
+{"activation_function": "gelu_new", "attention_probs_dropout_prob": 0.1, "attn_pdrop": 0.1, "bos_token_id": 98, "embd_pdrop": 0.1, "eos_token_id": 98, "gradient_checkpointing": false, "hidden_act": "gelu", "hidden_dropout_prob": 0.1, "initializer_range": 0.02, "intermediate_size": 37, "layer_norm_epsilon": 1e-05, "model_type": "gpt2", "n_ctx": 512, "n_embd": 32, "n_head": 4, "n_inner": null, "n_layer": 5, "n_positions": 512, "neuron": {"auto_cast_type": "fp32", "batch_size": 2, "checkpoint_id": "hf-internal-testing/tiny-random-gpt2", "checkpoint_revision": "71034c5d8bde858ff824298bdedc65515b97d2b9", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 2, "sequence_length": 512, "task": "text-generation"}, "pad_token_id": 98, "reorder_and_upcast_attn": false, "resid_pdrop": 0.1, "scale_attn_by_inverse_layer_idx": false, "scale_attn_weights": true, "summary_activation": null, "summary_first_dropout": 0.1, "summary_proj_to_labels": true, "summary_type": "cls_index", "summary_use_proj": true, "type_vocab_size": 16, "use_cache": true, "vocab_size": 1000}

neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.22.dev0/inference/llama/NousResearch/Llama-2-7b-chat-hf/8d31ca1db6445239b6cc.json

@@ -0,0 +1 @@
+{"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 1, "eos_token_id": 2, "hidden_act": "silu", "hidden_size": 4096, "initializer_range": 0.02, "intermediate_size": 11008, "max_position_embeddings": 4096, "model_type": "llama", "neuron": {"auto_cast_type": "f16", "batch_size": 4, "checkpoint_id": "NousResearch/Llama-2-7b-chat-hf", "checkpoint_revision": "37892f30c23786c0d5367d80481fa0d9fba93cf8", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 2, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 32, "pad_token_id": 0, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 10000.0, "tie_word_embeddings": false, "torch_dtype": "float16", "use_cache": true, "vocab_size": 32000}

neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.22.dev0/inference/llama/NousResearch/Llama-2-7b-chat-hf/f6a36646c9688804bab7.json

@@ -0,0 +1 @@
+{"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 1, "eos_token_id": 2, "hidden_act": "silu", "hidden_size": 4096, "initializer_range": 0.02, "intermediate_size": 11008, "max_position_embeddings": 4096, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 2, "checkpoint_id": "NousResearch/Llama-2-7b-chat-hf", "checkpoint_revision": "37892f30c23786c0d5367d80481fa0d9fba93cf8", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 2, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 32, "pad_token_id": 0, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 10000.0, "tie_word_embeddings": false, "torch_dtype": "float16", "use_cache": true, "vocab_size": 32000}

neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.22.dev0/inference/llama/dacorvo/tiny-random-llama/fb4b47ea0eea68e736b2.json

@@ -0,0 +1 @@
+{"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 1, "eos_token_id": 2, "hidden_act": "silu", "hidden_size": 128, "initializer_range": 0.02, "intermediate_size": 256, "max_position_embeddings": 512, "model_type": "llama", "neuron": {"auto_cast_type": "fp32", "batch_size": 2, "checkpoint_id": "dacorvo/tiny-random-llama", "checkpoint_revision": "7fdafd2fe6a2d31c6abb72ae60db606d8bb23196", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 2, "sequence_length": 512, "task": "text-generation"}, "num_attention_heads": 1, "num_hidden_layers": 1, "num_key_value_heads": 1, "pretraining_tp": 1, "rms_norm_eps": 1e-06, "rope_scaling": null, "rope_theta": 10000.0, "tie_word_embeddings": false, "torch_dtype": "float32", "use_cache": true, "vocab_size": 32000}

neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.22.dev0/inference/llama/meta-llama/Meta-Llama-3-8B-Instruct/8991f74549273593691d.json

@@ -0,0 +1 @@
+{"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 128000, "eos_token_id": 128001, "hidden_act": "silu", "hidden_size": 4096, "initializer_range": 0.02, "intermediate_size": 14336, "max_position_embeddings": 8192, "model_type": "llama", "neuron": {"auto_cast_type": "bf16", "batch_size": 4, "checkpoint_id": "meta-llama/Meta-Llama-3-8B-Instruct", "checkpoint_revision": "339ce92d052f002cdbac4a4bd551d1c61dd8345e", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 2, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 8, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 500000.0, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 128256}

neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.22.dev0/inference/llama/meta-llama/Meta-Llama-3-8B/638108a35a53ccc460bd.json

@@ -0,0 +1 @@
+{"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 128000, "eos_token_id": 128001, "hidden_act": "silu", "hidden_size": 4096, "initializer_range": 0.02, "intermediate_size": 14336, "max_position_embeddings": 8192, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 1, "checkpoint_id": "meta-llama/Meta-Llama-3-8B", "checkpoint_revision": "1460c22666392e470910ce3d44ffeb2ab7dbd4df", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 8, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 500000.0, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 128256}

neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.22.dev0/inference/llama/meta-llama/Meta-Llama-3-8B/75d01b56a778419c897d.json

@@ -0,0 +1 @@
+{"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 128000, "eos_token_id": 128001, "hidden_act": "silu", "hidden_size": 4096, "initializer_range": 0.02, "intermediate_size": 14336, "max_position_embeddings": 8192, "model_type": "llama", "neuron": {"auto_cast_type": "bf16", "batch_size": 4, "checkpoint_id": "meta-llama/Meta-Llama-3-8B", "checkpoint_revision": "cd892e8f4da1043d4b01d5ea182a2e8412bf658f", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 2, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 8, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 500000.0, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 128256}

neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.22.dev0/inference/llama/meta-llama/Meta-Llama-3-8B/77bf56a610a467c3b01c.json

@@ -0,0 +1 @@
+{"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 128000, "eos_token_id": 128001, "hidden_act": "silu", "hidden_size": 4096, "initializer_range": 0.02, "intermediate_size": 14336, "max_position_embeddings": 8192, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 4, "checkpoint_id": "meta-llama/Meta-Llama-3-8B", "checkpoint_revision": "1460c22666392e470910ce3d44ffeb2ab7dbd4df", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 8, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 500000.0, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 128256}

neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.22.dev0/inference/llama/meta-llama/Meta-Llama-3-8B/7cf5159a98ef6aa54442.json

@@ -0,0 +1 @@
+{"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 128000, "eos_token_id": 128001, "hidden_act": "silu", "hidden_size": 4096, "initializer_range": 0.02, "intermediate_size": 14336, "max_position_embeddings": 8192, "model_type": "llama", "neuron": {"auto_cast_type": "bf16", "batch_size": 4, "checkpoint_id": "meta-llama/Meta-Llama-3-8B", "checkpoint_revision": "561487d18c41c76bcb5fc6cfb73a324982f04f47", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 2, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 8, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 500000.0, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 128256}

neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.22.dev0/inference/llama/meta-llama/Meta-Llama-3-8B/97351cfb2a7ab3fe894d.json

@@ -0,0 +1 @@
+{"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 128000, "eos_token_id": 128001, "hidden_act": "silu", "hidden_size": 4096, "initializer_range": 0.02, "intermediate_size": 14336, "max_position_embeddings": 8192, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 8, "checkpoint_id": "meta-llama/Meta-Llama-3-8B", "checkpoint_revision": "1460c22666392e470910ce3d44ffeb2ab7dbd4df", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 8, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 500000.0, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 128256}

neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.22.dev0/inference/llama/princeton-nlp/Sheared-LLaMA-1.3B/f1bd1033a11a1478af9e.json

@@ -0,0 +1 @@
+{"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 1, "eos_token_id": 2, "hidden_act": "silu", "hidden_size": 2048, "initializer_range": 0.02, "intermediate_size": 5504, "max_position_embeddings": 4096, "model_type": "llama", "neuron": {"auto_cast_type": "fp32", "batch_size": 4, "checkpoint_id": "princeton-nlp/Sheared-LLaMA-1.3B", "checkpoint_revision": "a4b76938edbf571ea7d7d9904861cbdca08809b4", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 2, "sequence_length": 1024, "task": "text-generation"}, "num_attention_heads": 16, "num_hidden_layers": 24, "num_key_value_heads": 16, "pad_token_id": 0, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 10000.0, "tie_word_embeddings": false, "torch_dtype": "float32", "use_cache": true, "vocab_size": 32000}

neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.22.dev0/inference/mistral/HuggingFaceH4/zephyr-7b-beta/f4459480a285f358164c.json

@@ -0,0 +1 @@
+{"architectures": ["MistralForCausalLM"], "attention_dropout": 0.0, "bos_token_id": 1, "eos_token_id": 2, "hidden_act": "silu", "hidden_size": 4096, "initializer_range": 0.02, "intermediate_size": 14336, "max_position_embeddings": 32768, "model_type": "mistral", "neuron": {"auto_cast_type": "bf16", "batch_size": 4, "checkpoint_id": "HuggingFaceH4/zephyr-7b-beta", "checkpoint_revision": "b70e0c9a2d9e14bd1e812d3c398e5f313e93b473", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 2, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 8, "pad_token_id": 2, "rms_norm_eps": 1e-05, "rope_theta": 10000.0, "sliding_window": 4096, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 32000}

neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.22.dev0/inference/mistral/NucleusOrg/Nucleus-1B-alpha-1/e189713e28900a6fd483.json

@@ -0,0 +1 @@
+{"architectures": ["MistralForCausalLM"], "attention_dropout": 0.0, "bos_token_id": 1, "eos_token_id": 2, "hidden_act": "silu", "hidden_size": 4096, "initializer_range": 0.02, "intermediate_size": 14336, "max_position_embeddings": 32768, "model_type": "mistral", "neuron": {"auto_cast_type": "fp16", "batch_size": 4, "checkpoint_id": "NucleusOrg/Nucleus-1B-alpha-1", "checkpoint_revision": "f484a4533292b1ba4ba29b9afb3bb07eef813024", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 2, "sequence_length": 2048, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 4, "num_key_value_heads": 8, "rms_norm_eps": 1e-05, "rope_theta": 10000.0, "sliding_window": 4096, "tie_word_embeddings": false, "torch_dtype": "float16", "use_cache": true, "vocab_size": 32000}

neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.22.dev0/inference/mistral/dacorvo/tiny-random-MistralForCausalLM/8ab25062c2316baf852b.json

@@ -0,0 +1 @@
+{"architectures": ["MistralForCausalLM"], "attention_dropout": 0.0, "attention_probs_dropout_prob": 0.1, "bos_token_id": 1, "eos_token_id": 2, "hidden_act": "gelu", "hidden_dropout_prob": 0.1, "hidden_size": 32, "initializer_range": 0.02, "intermediate_size": 37, "is_decoder": true, "max_position_embeddings": 512, "model_type": "mistral", "neuron": {"auto_cast_type": "fp32", "batch_size": 2, "checkpoint_id": "dacorvo/tiny-random-MistralForCausalLM", "checkpoint_revision": "81d453e3c8985649e9ee3d4c9378461029d1c73a", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 2, "sequence_length": 512, "task": "text-generation"}, "num_attention_heads": 4, "num_hidden_layers": 2, "num_key_value_heads": 2, "pad_token_id": 0, "rms_norm_eps": 1e-06, "rope_theta": 10000.0, "sliding_window": 4096, "tie_word_embeddings": false, "torch_dtype": "float32", "type_vocab_size": 16, "use_cache": true, "vocab_size": 32000}

neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.22.dev0/inference/mistral/mistralai/Mistral-7b-Instruct-v0.2/8ff7681fcc40de857709.json

@@ -0,0 +1 @@
+{"architectures": ["MistralForCausalLM"], "attention_dropout": 0.0, "bos_token_id": 1, "eos_token_id": 2, "hidden_act": "silu", "hidden_size": 4096, "initializer_range": 0.02, "intermediate_size": 14336, "max_position_embeddings": 32768, "model_type": "mistral", "neuron": {"auto_cast_type": "bf16", "batch_size": 4, "checkpoint_id": "mistralai/Mistral-7b-Instruct-v0.2", "checkpoint_revision": "41b61a33a2483885c981aa79e0df6b32407ed873", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 2, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 8, "rms_norm_eps": 1e-05, "rope_theta": 1000000.0, "sliding_window": null, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 32000}

neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.22.dev0/inference/mistral/optimum/mistral-1.1b-testing/3f196a5a5beadf2af838.json

@@ -0,0 +1 @@
+{"architectures": ["MistralForCausalLM"], "attention_dropout": 0.0, "bos_token_id": 1, "eos_token_id": 2, "hidden_act": "silu", "hidden_size": 2048, "initializer_range": 0.02, "intermediate_size": 5632, "max_position_embeddings": 32768, "model_type": "mistral", "neuron": {"auto_cast_type": "bf16", "batch_size": 4, "checkpoint_id": "optimum/mistral-1.1b-testing", "checkpoint_revision": "ce03bc8d47dbd2c173ff65f3a8de1325ba724195", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 2, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 22, "num_key_value_heads": 4, "rms_norm_eps": 1e-05, "rope_theta": 1000000.0, "sliding_window": null, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 32000}

neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.22.dev0/inference/mistral/optimum/mistral-en-1.1B-testing/f5764c6ffc3e69951b77.json

@@ -0,0 +1 @@
+{"architectures": ["MistralForCausalLM"], "attention_dropout": 0.0, "bos_token_id": 0, "eos_token_id": 0, "hidden_act": "silu", "hidden_size": 2048, "initializer_range": 0.02, "intermediate_size": 5632, "max_position_embeddings": 4096, "model_type": "mistral", "neuron": {"auto_cast_type": "fp16", "batch_size": 4, "checkpoint_id": "optimum/mistral-en-1.1B-testing", "checkpoint_revision": "df7d850906b071dbb69cf11af1033d7cd735227d", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 2, "sequence_length": 2048, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 22, "num_key_value_heads": 4, "rms_norm_eps": 1e-05, "rope_theta": 10000.0, "sliding_window": 1024, "tie_word_embeddings": false, "torch_dtype": "float16", "use_cache": true, "vocab_size": 50257}

neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.22.dev0/inference/mixtral/dacorvo/Mixtral-tiny/79f811350e822243d689.json

@@ -0,0 +1 @@
+{"architectures": ["MixtralForCausalLM"], "attention_dropout": 0.0, "bos_token_id": 1, "eos_token_id": 2, "hidden_act": "silu", "hidden_size": 1024, "initializer_range": 0.02, "intermediate_size": 3584, "max_position_embeddings": 1024, "model_type": "mixtral", "neuron": {"auto_cast_type": "fp32", "batch_size": 2, "checkpoint_id": "dacorvo/Mixtral-tiny", "checkpoint_revision": "c557ba205ddff6ea911f4719e0d543d6c08356b6", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 2, "sequence_length": 1024, "task": "text-generation"}, "num_attention_heads": 32, "num_experts_per_tok": 2, "num_hidden_layers": 2, "num_key_value_heads": 8, "num_local_experts": 8, "output_router_logits": false, "rms_norm_eps": 1e-05, "rope_theta": 10000.0, "router_aux_loss_coef": 0.001, "sliding_window": 4096, "tie_word_embeddings": false, "torch_dtype": "float16", "use_cache": true, "vocab_size": 32000}

neuronxcc-2.13.66.0+6dfecc895/MODULE_0198baf5b6a9d47329e3+2c2d707e/compile_flags.txt

@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none

neuronxcc-2.13.66.0+6dfecc895/MODULE_0198baf5b6a9d47329e3+2c2d707e/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:21f10a90b8eb3545cb1849d75d5f4227f652e3fc4826f1aa2e1112f09f38ddb4
+size 399354

neuronxcc-2.13.66.0+6dfecc895/MODULE_0198baf5b6a9d47329e3+2c2d707e/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:680c226d5e7cd2b7059861da5126b02030d8d54e3d57efe65205936873c64f72
+size 3472384

neuronxcc-2.13.66.0+6dfecc895/MODULE_07c6ee519ad94eb45d54+2c2d707e/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:57d62212cd9debb3cb766f31772a487c30f2417304a42c00ac785d2adb0dd99b
+size 318265

neuronxcc-2.13.66.0+6dfecc895/MODULE_09c6476126adb91dbd05+2c2d707e/compile_flags.txt

@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none

neuronxcc-2.13.66.0+6dfecc895/MODULE_09c6476126adb91dbd05+2c2d707e/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:16b777e099301b690719a557eb532bbdad47ff8d4b79e6f97ef7854c9de5bf7d
+size 92525501

neuronxcc-2.13.66.0+6dfecc895/MODULE_09c6476126adb91dbd05+2c2d707e/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d6ed37cc04dba91e8756c7a5173b90eb096502c710e001fe62ad9f16281cf205
+size 15975424

neuronxcc-2.13.66.0+6dfecc895/MODULE_0bf1f29da445ebf60048+2c2d707e/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:763f990132267f20f64afac5317285725acfe7111bb62e62bf78d607040d2c5d
+size 318265

neuronxcc-2.13.66.0+6dfecc895/MODULE_148dac32be618613b227+2c2d707e/compile_flags.txt

@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none

neuronxcc-2.13.66.0+6dfecc895/MODULE_148dac32be618613b227+2c2d707e/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1c4e81b2e25941934b99f260bc124c1e515509714cc424fbfd814be8afaf944f
+size 54726

neuronxcc-2.13.66.0+6dfecc895/MODULE_181e4ab821fae74c8d8c+2c2d707e/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7961ff15f8c34781560e05c9cca11fb358bade475df00af00432c805e8afb836
+size 132432

neuronxcc-2.13.66.0+6dfecc895/MODULE_24756fabbe2c5c540c77+2c2d707e/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9aea8582686b388717121ba61324610d50e68b310eb20a03675d9a37ca389459
+size 132432

neuronxcc-2.13.66.0+6dfecc895/MODULE_2479adeb0b632872414f+2c2d707e/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a8652a9d43fe6eb1f68cf74df9ef3a6b4a45d8f7b1b6486d97c51cb1bb7dc351
+size 252757

neuronxcc-2.13.66.0+6dfecc895/MODULE_25974973e3d2fc72e8b8+2c2d707e/compile_flags.txt

@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none

neuronxcc-2.13.66.0+6dfecc895/MODULE_25974973e3d2fc72e8b8+2c2d707e/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f3c4b477b2a8040294370b2d5d6f4186c72a0ed03f206aba7a1484e5023e0e46
+size 547596

neuronxcc-2.13.66.0+6dfecc895/MODULE_25c861423631880c47aa+2c2d707e/compile_flags.txt

@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none

neuronxcc-2.13.66.0+6dfecc895/MODULE_25c861423631880c47aa+2c2d707e/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b2ed85c046af30a5bc40c86d39ff42e5ddb1376250abef1ee998bb6e9d5d6334
+size 131100

neuronxcc-2.13.66.0+6dfecc895/MODULE_268875679604c63831dd+2c2d707e/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1fec7929ff63675cadde424a2159cc9d1bbd8d584bee5d61373f210c7014ec81
+size 306522

neuronxcc-2.13.66.0+6dfecc895/MODULE_26b214595cb99db2c0a0+2c2d707e/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3b619ce935dcc8ffd729ee2e0cd4606bd0d638305213af101473a007f75b6cf8
+size 537236624

neuronxcc-2.13.66.0+6dfecc895/MODULE_28d6d8fbd91b3f09888b+2c2d707e/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:746a258a8134699614c4fde0c4dad684a0374d80b694e58289367ff607a61fa6
+size 382074

neuronxcc-2.13.66.0+6dfecc895/MODULE_2fd09cf93328cbea3db8+2c2d707e/compile_flags.txt

@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none

neuronxcc-2.13.66.0+6dfecc895/MODULE_2fd09cf93328cbea3db8+2c2d707e/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:26a700902df7abe922d60d5418e2282c93a9d9e264e11ebaceb81b5004dbd59b
+size 54726

neuronxcc-2.13.66.0+6dfecc895/MODULE_3a7efdd19953e1bd6973+2c2d707e/compile_flags.txt

@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none

neuronxcc-2.13.66.0+6dfecc895/MODULE_3a7efdd19953e1bd6973+2c2d707e/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:24b0bfafc1eb02a5f08f16f45a679483b09f0404eab5ecc8add4ec3575ed509b
+size 131100