Synchronizing local compiler cache.

.gitattributes

@@ -3305,3 +3305,67 @@ neuronxcc-2.14.182.0+a56cbff7/MODULE_e9cd7823552b4052fd1f+2c2d707e/model.neff fi
 neuronxcc-2.14.182.0+a56cbff7/MODULE_f06eb185da899bd94560+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.14.182.0+a56cbff7/MODULE_f5782d94847d0a80b7d3+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.14.182.0+a56cbff7/MODULE_fdf13076ef71fca7878d+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.66.0+6dfecc895/MODULE_09034987b24d3adb04f4+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.66.0+6dfecc895/MODULE_0b74f0af54c254808190+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.66.0+6dfecc895/MODULE_0ec1ce0dd4ecc89fd3f2+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.66.0+6dfecc895/MODULE_12c92c11656e4c9928c7+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.66.0+6dfecc895/MODULE_380e69bb361b2c799f94+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.66.0+6dfecc895/MODULE_5cc9f3ece84db9a21349+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.66.0+6dfecc895/MODULE_6fe3dc017860f855cfeb+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.66.0+6dfecc895/MODULE_85a3010c3ec608dc0ef9+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.66.0+6dfecc895/MODULE_98bf5aa58afcc3ef746a+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.66.0+6dfecc895/MODULE_c032f3ce1f0a9ce7a769+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.66.0+6dfecc895/MODULE_c29e8ee00ad4d5fde006+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.66.0+6dfecc895/MODULE_ea2d51d326cf73dd732a+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.182.0+a56cbff7/MODULE_0995e3cf3c95acbc080c+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.182.0+a56cbff7/MODULE_11550df6dc05abc55422+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.182.0+a56cbff7/MODULE_129437d50f9340395d1c+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.182.0+a56cbff7/MODULE_219b52bfdf96bbc7b9db+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.182.0+a56cbff7/MODULE_24d8c4822406a0219cd7+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.182.0+a56cbff7/MODULE_27369fceaad24943ec9c+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.182.0+a56cbff7/MODULE_2850c14927137504ac23+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.182.0+a56cbff7/MODULE_29c67a51cf5c7d49994e+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.182.0+a56cbff7/MODULE_2d31dc896f33ed50bbc8+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.182.0+a56cbff7/MODULE_2de5f419f2fb6a78e37e+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.182.0+a56cbff7/MODULE_2edebf65164c1ec47f2b+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.182.0+a56cbff7/MODULE_36a69d286f888310a682+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.182.0+a56cbff7/MODULE_3d34278099de7736f95b+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.182.0+a56cbff7/MODULE_41984e951d747a860bac+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.182.0+a56cbff7/MODULE_450eaa019ecdda36f50f+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.182.0+a56cbff7/MODULE_470cef8beb6c17ada734+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.182.0+a56cbff7/MODULE_47ae1cbd97906ea14f66+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.182.0+a56cbff7/MODULE_5b019cdb85abccfe6b9f+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.182.0+a56cbff7/MODULE_5cadde67746d46e15b46+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.182.0+a56cbff7/MODULE_5f78bf0249224a3e0467+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.182.0+a56cbff7/MODULE_622540f250033e85bb06+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.182.0+a56cbff7/MODULE_66b3de65924b58935421+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.182.0+a56cbff7/MODULE_6fdac1956f2fe3816b59+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.182.0+a56cbff7/MODULE_70be9db8dcb2a93e1664+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.182.0+a56cbff7/MODULE_7b73f74f6655901fbba8+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.182.0+a56cbff7/MODULE_7d77e08a63c0c2a54263+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.182.0+a56cbff7/MODULE_7f8cdb9a4fcdd08ca89c+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.182.0+a56cbff7/MODULE_826aa17cf5c3730e263f+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.182.0+a56cbff7/MODULE_8ab10a70e73f4acc4b04+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.182.0+a56cbff7/MODULE_8e39f79a52a697d72a1a+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.182.0+a56cbff7/MODULE_9461ae320814547a00ce+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.182.0+a56cbff7/MODULE_94949a656354223ab507+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.182.0+a56cbff7/MODULE_9a70af68e65c3effb418+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.182.0+a56cbff7/MODULE_a6e90bb89fa61a08a960+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.182.0+a56cbff7/MODULE_aec7343e8b85f40733ce+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.182.0+a56cbff7/MODULE_afe34429c874b0f3b9c2+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.182.0+a56cbff7/MODULE_b5a1aa152b76159ec599+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.182.0+a56cbff7/MODULE_b9152c82fc40d259dbfc+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.182.0+a56cbff7/MODULE_bb85abc2ea59d8484968+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.182.0+a56cbff7/MODULE_be3a896f9dc3fa427c56+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.182.0+a56cbff7/MODULE_c402c13c9ae71bda563d+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.182.0+a56cbff7/MODULE_c4aab030aa76b3127fcc+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.182.0+a56cbff7/MODULE_ce4178f973cf597733e1+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.182.0+a56cbff7/MODULE_d03a90d450c6d1a0e03a+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.182.0+a56cbff7/MODULE_d0a3b34918aac1823ec2+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.182.0+a56cbff7/MODULE_d98becba0e8337353210+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.182.0+a56cbff7/MODULE_ded3aedf2b580dae490a+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.182.0+a56cbff7/MODULE_df0734d88c01ec309810+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.182.0+a56cbff7/MODULE_e4fd2fca5d14cb667818+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.182.0+a56cbff7/MODULE_fabe9023fbb212236b7c+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.182.0+a56cbff7/MODULE_fb6e5d0a11a21c13138c+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.182.0+a56cbff7/MODULE_fcd85d066cb013a59267+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text

neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.23/inference/llama/meta-llama/Meta-Llama-3-8B/ae1a0500a28baa42566d.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, "mlp_bias": false, "model_type": "llama", "neuron": {"auto_cast_type": "fp32", "batch_size": 1, "checkpoint_id": "meta-llama/Meta-Llama-3-8B", "checkpoint_revision": "62bd457b6fe961a42a631306577e622c83876cb6", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 24, "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/0.0.24.dev0/inference/bloom/hf-internal-testing/tiny-random-BloomForCausalLM/4efe4ec09e4490308cc0.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": 1, "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": 100, "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/0.0.24.dev0/inference/bloom/hf-internal-testing/tiny-random-BloomForCausalLM/9d5f94b71f6ac3455612.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": "fp16", "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": 100, "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/0.0.24.dev0/inference/bloom/hf-internal-testing/tiny-random-BloomForCausalLM/b9db5e6da88e9f07ee3e.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": "fp16", "batch_size": 1, "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": 100, "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/0.0.24.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/0.0.24.dev0/inference/gpt2/gpt2/d5e93094d604b84cb59a.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": "fp16", "batch_size": 4, "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/0.0.24.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/0.0.24.dev0/inference/gpt2/hf-internal-testing/tiny-random-gpt2/7e574905d10c185da6d3.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": "fp16", "batch_size": 1, "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": 100, "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/0.0.24.dev0/inference/gpt2/hf-internal-testing/tiny-random-gpt2/b99d0747aa4e9af275fb.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": 1, "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": 100, "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/0.0.24.dev0/inference/gpt2/hf-internal-testing/tiny-random-gpt2/e62dd30300cbb6a57da5.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": "fp16", "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": 100, "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/0.0.24.dev0/inference/llama/HuggingFaceTB/cosmo-1b/00271b4df65a1d15b20b.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": 8192, "max_position_embeddings": 2048, "mlp_bias": false, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 4, "checkpoint_id": "HuggingFaceTB/cosmo-1b", "checkpoint_revision": "0d5e341cfe835dffc81b6186f9715c094889f8ce", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 2, "sequence_length": 2048, "task": "text-generation"}, "num_attention_heads": 16, "num_hidden_layers": 24, "num_key_value_heads": 16, "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/0.0.24.dev0/inference/llama/dacorvo/tiny-random-llama/323db011f6520e1e3511.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, "mlp_bias": false, "model_type": "llama", "neuron": {"auto_cast_type": "fp32", "batch_size": 1, "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": 100, "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/0.0.24.dev0/inference/llama/dacorvo/tiny-random-llama/77e0a7441f673619e605.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, "mlp_bias": false, "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/0.0.24.dev0/inference/llama/dacorvo/tiny-random-llama/a2f4c269e6888edaa9f9.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, "mlp_bias": false, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "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": 100, "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/0.0.24.dev0/inference/llama/dacorvo/tiny-random-llama/d6cc8048cc71199a8785.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, "mlp_bias": false, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 1, "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": 100, "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/0.0.24.dev0/inference/llama/meta-llama/Meta-Llama-3-8B/d927181f49778556a1aa.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, "mlp_bias": false, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 32, "checkpoint_id": "meta-llama/Meta-Llama-3-8B", "checkpoint_revision": "62bd457b6fe961a42a631306577e622c83876cb6", "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/0.0.24.dev0/inference/mistral/dacorvo/tiny-random-MistralForCausalLM/52fbcfd7fedd024371f9.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": "fp16", "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": 100, "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/0.0.24.dev0/inference/mistral/dacorvo/tiny-random-MistralForCausalLM/61da37a594d2621c84f3.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": "fp16", "batch_size": 1, "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": 100, "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/0.0.24.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/0.0.24.dev0/inference/mistral/dacorvo/tiny-random-MistralForCausalLM/8adb053370d3399fa36d.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": 1, "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": 100, "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/0.0.24.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/0.0.24.dev0/inference/mixtral/dacorvo/Mixtral-tiny/43944cbea36886b7c858.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": 1, "checkpoint_id": "dacorvo/Mixtral-tiny", "checkpoint_revision": "c557ba205ddff6ea911f4719e0d543d6c08356b6", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 2, "sequence_length": 100, "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, "router_jitter_noise": 0.0, "sliding_window": 4096, "tie_word_embeddings": false, "torch_dtype": "float16", "use_cache": true, "vocab_size": 32000}

neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.24.dev0/inference/mixtral/dacorvo/Mixtral-tiny/52680b2cdc233c828caf.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, "router_jitter_noise": 0.0, "sliding_window": 4096, "tie_word_embeddings": false, "torch_dtype": "float16", "use_cache": true, "vocab_size": 32000}

neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.24.dev0/inference/mixtral/dacorvo/Mixtral-tiny/57b16e8fab66604f85eb.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": "fp16", "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": 100, "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, "router_jitter_noise": 0.0, "sliding_window": 4096, "tie_word_embeddings": false, "torch_dtype": "float16", "use_cache": true, "vocab_size": 32000}

neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.24.dev0/inference/mixtral/dacorvo/Mixtral-tiny/f6fbd647258f2c561a81.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": "fp16", "batch_size": 1, "checkpoint_id": "dacorvo/Mixtral-tiny", "checkpoint_revision": "c557ba205ddff6ea911f4719e0d543d6c08356b6", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 2, "sequence_length": 100, "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, "router_jitter_noise": 0.0, "sliding_window": 4096, "tie_word_embeddings": false, "torch_dtype": "float16", "use_cache": true, "vocab_size": 32000}

neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.24.dev0/inference/opt/hf-internal-testing/tiny-random-OPTForCausalLM/765507817dbb6125de6c.json

@@ -0,0 +1 @@
+{"_remove_final_layer_norm": false, "activation_function": "relu", "architectures": ["OPTForCausalLM"], "attention_dropout": 0.1, "bos_token_id": 2, "do_layer_norm_before": true, "dropout": 0.1, "embed_dim": 16, "enable_bias": true, "eos_token_id": 2, "ffn_dim": 4, "hidden_size": 16, "init_std": 0.02, "is_decoder": true, "layer_norm_elementwise_affine": true, "layerdrop": 0.0, "max_position_embeddings": 100, "model_type": "opt", "neuron": {"auto_cast_type": "fp16", "batch_size": 2, "checkpoint_id": "hf-internal-testing/tiny-random-OPTForCausalLM", "checkpoint_revision": "190d1f4fc0011d2eaeaa05282e0fbd2445e4b11f", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 2, "sequence_length": 100, "task": "text-generation"}, "num_attention_heads": 4, "num_hidden_layers": 5, "pad_token_id": 1, "torch_dtype": "float32", "use_cache": true, "vocab_size": 50265, "word_embed_proj_dim": 16}

neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.24.dev0/inference/opt/hf-internal-testing/tiny-random-OPTForCausalLM/7f2d5e84740d788567e2.json

@@ -0,0 +1 @@
+{"_remove_final_layer_norm": false, "activation_function": "relu", "architectures": ["OPTForCausalLM"], "attention_dropout": 0.1, "bos_token_id": 2, "do_layer_norm_before": true, "dropout": 0.1, "embed_dim": 16, "enable_bias": true, "eos_token_id": 2, "ffn_dim": 4, "hidden_size": 16, "init_std": 0.02, "is_decoder": true, "layer_norm_elementwise_affine": true, "layerdrop": 0.0, "max_position_embeddings": 100, "model_type": "opt", "neuron": {"auto_cast_type": "fp32", "batch_size": 1, "checkpoint_id": "hf-internal-testing/tiny-random-OPTForCausalLM", "checkpoint_revision": "190d1f4fc0011d2eaeaa05282e0fbd2445e4b11f", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 2, "sequence_length": 100, "task": "text-generation"}, "num_attention_heads": 4, "num_hidden_layers": 5, "pad_token_id": 1, "torch_dtype": "float32", "use_cache": true, "vocab_size": 50265, "word_embed_proj_dim": 16}

neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.24.dev0/inference/opt/hf-internal-testing/tiny-random-OPTForCausalLM/a8f94bb3f4eb93f8bd2e.json

@@ -0,0 +1 @@
+{"_remove_final_layer_norm": false, "activation_function": "relu", "architectures": ["OPTForCausalLM"], "attention_dropout": 0.1, "bos_token_id": 2, "do_layer_norm_before": true, "dropout": 0.1, "embed_dim": 16, "enable_bias": true, "eos_token_id": 2, "ffn_dim": 4, "hidden_size": 16, "init_std": 0.02, "is_decoder": true, "layer_norm_elementwise_affine": true, "layerdrop": 0.0, "max_position_embeddings": 100, "model_type": "opt", "neuron": {"auto_cast_type": "fp16", "batch_size": 1, "checkpoint_id": "hf-internal-testing/tiny-random-OPTForCausalLM", "checkpoint_revision": "190d1f4fc0011d2eaeaa05282e0fbd2445e4b11f", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 2, "sequence_length": 100, "task": "text-generation"}, "num_attention_heads": 4, "num_hidden_layers": 5, "pad_token_id": 1, "torch_dtype": "float32", "use_cache": true, "vocab_size": 50265, "word_embed_proj_dim": 16}

neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.24.dev0/inference/opt/hf-internal-testing/tiny-random-OPTForCausalLM/f3adfe5a9c79b5a36fd7.json

@@ -0,0 +1 @@
+{"_remove_final_layer_norm": false, "activation_function": "relu", "architectures": ["OPTForCausalLM"], "attention_dropout": 0.1, "bos_token_id": 2, "do_layer_norm_before": true, "dropout": 0.1, "embed_dim": 16, "enable_bias": true, "eos_token_id": 2, "ffn_dim": 4, "hidden_size": 16, "init_std": 0.02, "is_decoder": true, "layer_norm_elementwise_affine": true, "layerdrop": 0.0, "max_position_embeddings": 100, "model_type": "opt", "neuron": {"auto_cast_type": "fp32", "batch_size": 2, "checkpoint_id": "hf-internal-testing/tiny-random-OPTForCausalLM", "checkpoint_revision": "190d1f4fc0011d2eaeaa05282e0fbd2445e4b11f", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 2, "sequence_length": 100, "task": "text-generation"}, "num_attention_heads": 4, "num_hidden_layers": 5, "pad_token_id": 1, "torch_dtype": "float32", "use_cache": true, "vocab_size": 50265, "word_embed_proj_dim": 16}

neuronxcc-2.13.66.0+6dfecc895/MODULE_09034987b24d3adb04f4+2c2d707e/compile_flags.txt

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

neuronxcc-2.13.66.0+6dfecc895/MODULE_09034987b24d3adb04f4+2c2d707e/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1a3c13b8fb71a9f99ac65f9f309b4424fbf0d756b3e546c8990721e8c642aa23
+size 293240

neuronxcc-2.13.66.0+6dfecc895/MODULE_09034987b24d3adb04f4+2c2d707e/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:442cd9b6052e23026338ab554937a9cdfb3b0f321c6d2d74b42903618ee3f195
+size 2376704

neuronxcc-2.13.66.0+6dfecc895/MODULE_0b74f0af54c254808190+2c2d707e/compile_flags.txt

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

neuronxcc-2.13.66.0+6dfecc895/MODULE_0b74f0af54c254808190+2c2d707e/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:caee485b9859bca997571a361fdb35091abec5a0a1d35a2d5d3a8e3d65c0f12c
+size 293338

neuronxcc-2.13.66.0+6dfecc895/MODULE_0b74f0af54c254808190+2c2d707e/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:133c31de6951395f0533c8053293f1a736d5feae37b5dfa3cf75af7ac935cf0e
+size 1721344

neuronxcc-2.13.66.0+6dfecc895/MODULE_0ec1ce0dd4ecc89fd3f2+2c2d707e/compile_flags.txt

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

neuronxcc-2.13.66.0+6dfecc895/MODULE_0ec1ce0dd4ecc89fd3f2+2c2d707e/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9402e6cf477e164d9e09be9c5a993acdb59983e7e1741dc0db97951e2e553a2f
+size 293338

neuronxcc-2.13.66.0+6dfecc895/MODULE_0ec1ce0dd4ecc89fd3f2+2c2d707e/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:14e9632a360ae079272ac1dc32ece7b7826784a75dd714c6d220cfae28b9457d
+size 1741824

neuronxcc-2.13.66.0+6dfecc895/MODULE_12c92c11656e4c9928c7+2c2d707e/compile_flags.txt

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

neuronxcc-2.13.66.0+6dfecc895/MODULE_12c92c11656e4c9928c7+2c2d707e/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6d27bc0402e9f2ac7661cf42125a2cf6ff61967bbeb9ea70421f5f0ac3ac05da
+size 293338

neuronxcc-2.13.66.0+6dfecc895/MODULE_12c92c11656e4c9928c7+2c2d707e/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:eb15af32ab68480aa74eaa350ec905bd48966f036cfaa89c263cd1e8dca2d088
+size 1690624

neuronxcc-2.13.66.0+6dfecc895/MODULE_380e69bb361b2c799f94+2c2d707e/compile_flags.txt

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

neuronxcc-2.13.66.0+6dfecc895/MODULE_380e69bb361b2c799f94+2c2d707e/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:53256eecf15f3ae6682cbae959fbffc2d8565d723a13e8670ed3a99348a86bb9
+size 293240

neuronxcc-2.13.66.0+6dfecc895/MODULE_380e69bb361b2c799f94+2c2d707e/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6dd6d10fb1bb25ddeab9f055dbdc92016669afb6d7acd5d8e18ec582ea12b007
+size 5796864

neuronxcc-2.13.66.0+6dfecc895/MODULE_5cc9f3ece84db9a21349+2c2d707e/compile_flags.txt

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

neuronxcc-2.13.66.0+6dfecc895/MODULE_5cc9f3ece84db9a21349+2c2d707e/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5fbcc3e830c892400e165771935f2a23905fcf62574d209f054291a25e79cb1a
+size 293240

neuronxcc-2.13.66.0+6dfecc895/MODULE_5cc9f3ece84db9a21349+2c2d707e/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9b5848e70d5bb362ef96b8145f1f9801cba38b70a2b8e9ba9e3a0188f1f01559
+size 11019264

neuronxcc-2.13.66.0+6dfecc895/MODULE_60c3805041c4f7fd047d+2c2d707e/compile_flags.txt

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

neuronxcc-2.13.66.0+6dfecc895/MODULE_60c3805041c4f7fd047d+2c2d707e/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0ebbf59e5aec04f063c4903ca155126e739f7ca42c2d24bc297b590f82416dd2
+size 293338