diff --git a/.gitattributes b/.gitattributes
index 873ca819f6da26daec2115d7e41c6fac84d73014..6197056e14598be32aead0614d4885c723efa368 100644
--- a/.gitattributes
+++ b/.gitattributes
@@ -6488,3 +6488,15 @@ neuronxcc-2.16.303.0+d9f03cda/MODULE_efc674ceb57a1f7010a5+7ac4dbae/model.neff fi
 neuronxcc-2.16.303.0+d9f03cda/MODULE_ff0a9145bffbede97cde+7ac4dbae/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.16.303.0+d9f03cda/MODULE_5e72d9266a7fb87f15cb+7ac4dbae/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.16.303.0+d9f03cda/MODULE_97b24281cd74002e4729+7ac4dbae/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.143.0+e39249ad/MODULE_0e92acc32b557ad5547d+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.143.0+e39249ad/MODULE_39a9ccce9adcabc94675+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.143.0+e39249ad/MODULE_3c33fc8c5b8139f55998+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.143.0+e39249ad/MODULE_402e15278f41a32c430e+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.143.0+e39249ad/MODULE_6550f3db82ff02b1c1ff+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.143.0+e39249ad/MODULE_67a9118947e020080e96+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.143.0+e39249ad/MODULE_7dbd99330688237c55ff+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.143.0+e39249ad/MODULE_aa4954701eb403167936+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.143.0+e39249ad/MODULE_aefce72d1e0612585993+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.143.0+e39249ad/MODULE_b0acdacec2caf4887452+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.143.0+e39249ad/MODULE_c7dd940fbcfda75b0522+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.143.0+e39249ad/MODULE_e1bacfe9616dab23b889+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
diff --git a/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27.dev0/inference/qwen2/Qwen/Qwen2.5-Coder-32B-Instruct/39b69bbd54bed005e329.json b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27.dev0/inference/qwen2/Qwen/Qwen2.5-Coder-32B-Instruct/39b69bbd54bed005e329.json
new file mode 100644
index 0000000000000000000000000000000000000000..4c06bff95333034656c2409a8635ea6f0b647a6c
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27.dev0/inference/qwen2/Qwen/Qwen2.5-Coder-32B-Instruct/39b69bbd54bed005e329.json
@@ -0,0 +1 @@
+{"architectures": ["Qwen2ForCausalLM"], "attention_dropout": 0.0, "bos_token_id": 151643, "eos_token_id": 151645, "hidden_act": "silu", "hidden_size": 5120, "initializer_range": 0.02, "intermediate_size": 27648, "max_position_embeddings": 32768, "max_window_layers": 64, "model_type": "qwen2", "neuron": {"auto_cast_type": "bf16", "batch_size": 8, "checkpoint_id": "Qwen/Qwen2.5-Coder-32B-Instruct", "checkpoint_revision": "b47205940b83b5b484577359f71ee7b88472df67", "compiler_type": "neuronx-cc", "compiler_version": "2.15.143.0+e39249ad", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 40, "num_hidden_layers": 64, "num_key_value_heads": 8, "rms_norm_eps": 1e-06, "rope_theta": 1000000.0, "sliding_window": null, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "use_sliding_window": false, "vocab_size": 152064}
\ No newline at end of file
diff --git a/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27.dev0/inference/qwen2/Qwen/Qwen2.5-Coder-7B-Instruct/c19875bdff42cd555d4c.json b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27.dev0/inference/qwen2/Qwen/Qwen2.5-Coder-7B-Instruct/c19875bdff42cd555d4c.json
new file mode 100644
index 0000000000000000000000000000000000000000..f8426bc1d6194a4e7fc513aab03814c5747a72c7
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27.dev0/inference/qwen2/Qwen/Qwen2.5-Coder-7B-Instruct/c19875bdff42cd555d4c.json
@@ -0,0 +1 @@
+{"architectures": ["Qwen2ForCausalLM"], "attention_dropout": 0.0, "bos_token_id": 151643, "eos_token_id": 151645, "hidden_act": "silu", "hidden_size": 3584, "initializer_range": 0.02, "intermediate_size": 18944, "max_position_embeddings": 32768, "max_window_layers": 28, "model_type": "qwen2", "neuron": {"auto_cast_type": "bf16", "batch_size": 1, "checkpoint_id": "Qwen/Qwen2.5-Coder-7B-Instruct", "checkpoint_revision": "0eb6b1ed2d0c4306bc637d09ecef51e59d3dfe05", "compiler_type": "neuronx-cc", "compiler_version": "2.15.143.0+e39249ad", "num_cores": 2, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 28, "num_hidden_layers": 28, "num_key_value_heads": 4, "rms_norm_eps": 1e-06, "rope_theta": 1000000.0, "sliding_window": null, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "use_sliding_window": false, "vocab_size": 152064}
\ No newline at end of file
diff --git a/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/bloom/hf-internal-testing/tiny-random-BloomForCausalLM/4538babd01182da34f41.json b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/bloom/hf-internal-testing/tiny-random-BloomForCausalLM/4538babd01182da34f41.json
new file mode 100644
index 0000000000000000000000000000000000000000..0edf582da3baa5f67bdad5a603a8f3c4091982a8
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/bloom/hf-internal-testing/tiny-random-BloomForCausalLM/4538babd01182da34f41.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": "2.15.143.0+e39249ad", "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}
\ No newline at end of file
diff --git a/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/bloom/hf-internal-testing/tiny-random-BloomForCausalLM/b40a794eff5ea470980f.json b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/bloom/hf-internal-testing/tiny-random-BloomForCausalLM/b40a794eff5ea470980f.json
new file mode 100644
index 0000000000000000000000000000000000000000..72d178a4db1b98fc95b254cda8aecc40131bbc3f
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/bloom/hf-internal-testing/tiny-random-BloomForCausalLM/b40a794eff5ea470980f.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": "2.15.143.0+e39249ad", "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}
\ No newline at end of file
diff --git a/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/bloom/hf-internal-testing/tiny-random-BloomForCausalLM/cdbdbf737272e98ae974.json b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/bloom/hf-internal-testing/tiny-random-BloomForCausalLM/cdbdbf737272e98ae974.json
new file mode 100644
index 0000000000000000000000000000000000000000..24747b0b77b8027c274056ce35833b97539a8f05
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/bloom/hf-internal-testing/tiny-random-BloomForCausalLM/cdbdbf737272e98ae974.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": "2.15.143.0+e39249ad", "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}
\ No newline at end of file
diff --git a/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/gpt2/gpt2/aadac442fccfdd730bb5.json b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/gpt2/gpt2/aadac442fccfdd730bb5.json
new file mode 100644
index 0000000000000000000000000000000000000000..60376d7ece85795e4beeb4a28074f2021203851e
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/gpt2/gpt2/aadac442fccfdd730bb5.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": "2.15.143.0+e39249ad", "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}
\ No newline at end of file
diff --git a/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/gpt2/hf-internal-testing/tiny-random-gpt2/1f8668afaa87da7a4bc8.json b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/gpt2/hf-internal-testing/tiny-random-gpt2/1f8668afaa87da7a4bc8.json
new file mode 100644
index 0000000000000000000000000000000000000000..dd27b9604b37b41ece5856e007aa85e6cb09cc78
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/gpt2/hf-internal-testing/tiny-random-gpt2/1f8668afaa87da7a4bc8.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": "2.15.143.0+e39249ad", "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}
\ No newline at end of file
diff --git a/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/gpt2/hf-internal-testing/tiny-random-gpt2/5e4f40c305b67f8eb70c.json b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/gpt2/hf-internal-testing/tiny-random-gpt2/5e4f40c305b67f8eb70c.json
new file mode 100644
index 0000000000000000000000000000000000000000..e94d4708497a0e95ef96aeaacf239655f2d4043c
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/gpt2/hf-internal-testing/tiny-random-gpt2/5e4f40c305b67f8eb70c.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": "2.15.143.0+e39249ad", "num_cores": 24, "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}
\ No newline at end of file
diff --git a/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/gpt2/hf-internal-testing/tiny-random-gpt2/af0ef68e4ec57e23cab5.json b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/gpt2/hf-internal-testing/tiny-random-gpt2/af0ef68e4ec57e23cab5.json
new file mode 100644
index 0000000000000000000000000000000000000000..0ce482be44dc883e415a01a727c656571c6c5c22
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/gpt2/hf-internal-testing/tiny-random-gpt2/af0ef68e4ec57e23cab5.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": "2.15.143.0+e39249ad", "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}
\ No newline at end of file
diff --git a/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/gpt2/hf-internal-testing/tiny-random-gpt2/e0c5534b610fbfb22b6e.json b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/gpt2/hf-internal-testing/tiny-random-gpt2/e0c5534b610fbfb22b6e.json
new file mode 100644
index 0000000000000000000000000000000000000000..7dff80657856caa5237430211041b7ac8ff21b97
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/gpt2/hf-internal-testing/tiny-random-gpt2/e0c5534b610fbfb22b6e.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": "2.15.143.0+e39249ad", "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}
\ No newline at end of file
diff --git a/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/llama/NousResearch/Hermes-2-Theta-Llama-3-8B/ff84bf8571014fc7a409.json b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/llama/NousResearch/Hermes-2-Theta-Llama-3-8B/ff84bf8571014fc7a409.json
new file mode 100644
index 0000000000000000000000000000000000000000..40f7c452c0af66e777529ac07f51db1cc2bf889a
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/llama/NousResearch/Hermes-2-Theta-Llama-3-8B/ff84bf8571014fc7a409.json
@@ -0,0 +1 @@
+{"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 128000, "eos_token_id": 128003, "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": 4, "checkpoint_id": "NousResearch/Hermes-2-Theta-Llama-3-8B", "checkpoint_revision": "57a73110702e7b05ba3f39fef36297454c680725", "compiler_type": "neuronx-cc", "compiler_version": "2.15.143.0+e39249ad", "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}
\ No newline at end of file
diff --git a/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/llama/llamafactory/tiny-random-Llama-3/1ef3929e28ba91cc3cc6.json b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/llama/llamafactory/tiny-random-Llama-3/1ef3929e28ba91cc3cc6.json
new file mode 100644
index 0000000000000000000000000000000000000000..d92a2f5db737ad00eada7555cc1fd4db0d1acf9c
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/llama/llamafactory/tiny-random-Llama-3/1ef3929e28ba91cc3cc6.json
@@ -0,0 +1 @@
+{"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 128000, "eos_token_id": 128009, "hidden_act": "silu", "hidden_size": 16, "initializer_range": 0.02, "intermediate_size": 64, "max_position_embeddings": 4096, "mlp_bias": false, "model_type": "llama", "neuron": {"auto_cast_type": "fp32", "batch_size": 1, "checkpoint_id": "llamafactory/tiny-random-Llama-3", "checkpoint_revision": "2d9d8c1112e9cd4b0d66bb612e09be7da7997b18", "compiler_type": "neuronx-cc", "compiler_version": "2.15.143.0+e39249ad", "num_cores": 2, "sequence_length": 100, "task": "text-generation"}, "num_attention_heads": 4, "num_hidden_layers": 2, "num_key_value_heads": 4, "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": 128256}
\ No newline at end of file
diff --git a/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/llama/llamafactory/tiny-random-Llama-3/71ec0a49c3535c618d03.json b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/llama/llamafactory/tiny-random-Llama-3/71ec0a49c3535c618d03.json
new file mode 100644
index 0000000000000000000000000000000000000000..eb1e6463feb73aa0fbfa4eb395a69ceda99ed687
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/llama/llamafactory/tiny-random-Llama-3/71ec0a49c3535c618d03.json
@@ -0,0 +1 @@
+{"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 128000, "eos_token_id": 128009, "hidden_act": "silu", "hidden_size": 16, "initializer_range": 0.02, "intermediate_size": 64, "max_position_embeddings": 4096, "mlp_bias": false, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 2, "checkpoint_id": "llamafactory/tiny-random-Llama-3", "checkpoint_revision": "2d9d8c1112e9cd4b0d66bb612e09be7da7997b18", "compiler_type": "neuronx-cc", "compiler_version": "2.15.143.0+e39249ad", "num_cores": 2, "sequence_length": 100, "task": "text-generation"}, "num_attention_heads": 4, "num_hidden_layers": 2, "num_key_value_heads": 4, "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": 128256}
\ No newline at end of file
diff --git a/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/llama/llamafactory/tiny-random-Llama-3/eb3f26aa23141bff2292.json b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/llama/llamafactory/tiny-random-Llama-3/eb3f26aa23141bff2292.json
new file mode 100644
index 0000000000000000000000000000000000000000..93ecdb5926159f947f00d538ebf66070d99f7981
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/llama/llamafactory/tiny-random-Llama-3/eb3f26aa23141bff2292.json
@@ -0,0 +1 @@
+{"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 128000, "eos_token_id": 128009, "hidden_act": "silu", "hidden_size": 16, "initializer_range": 0.02, "intermediate_size": 64, "max_position_embeddings": 4096, "mlp_bias": false, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 1, "checkpoint_id": "llamafactory/tiny-random-Llama-3", "checkpoint_revision": "2d9d8c1112e9cd4b0d66bb612e09be7da7997b18", "compiler_type": "neuronx-cc", "compiler_version": "2.15.143.0+e39249ad", "num_cores": 2, "sequence_length": 100, "task": "text-generation"}, "num_attention_heads": 4, "num_hidden_layers": 2, "num_key_value_heads": 4, "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": 128256}
\ No newline at end of file
diff --git a/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/mistral/dacorvo/tiny-random-MistralForCausalLM/7d2df9c19098ee676fa8.json b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/mistral/dacorvo/tiny-random-MistralForCausalLM/7d2df9c19098ee676fa8.json
new file mode 100644
index 0000000000000000000000000000000000000000..0284c7f6c10d8cc00a4a24cca798adf7578ad973
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/mistral/dacorvo/tiny-random-MistralForCausalLM/7d2df9c19098ee676fa8.json
@@ -0,0 +1 @@
+{"architectures": ["MistralForCausalLM"], "attention_dropout": 0.0, "attention_probs_dropout_prob": 0.1, "bos_token_id": 1, "eos_token_id": 2, "head_dim": 8, "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": "2.15.143.0+e39249ad", "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}
\ No newline at end of file
diff --git a/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/mistral/dacorvo/tiny-random-MistralForCausalLM/ac0287246ef4575f8b78.json b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/mistral/dacorvo/tiny-random-MistralForCausalLM/ac0287246ef4575f8b78.json
new file mode 100644
index 0000000000000000000000000000000000000000..2b1de7cc9162ae3aa404ea7f09222d05fb7cc0d2
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/mistral/dacorvo/tiny-random-MistralForCausalLM/ac0287246ef4575f8b78.json
@@ -0,0 +1 @@
+{"architectures": ["MistralForCausalLM"], "attention_dropout": 0.0, "attention_probs_dropout_prob": 0.1, "bos_token_id": 1, "eos_token_id": 2, "head_dim": 8, "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": "2.15.143.0+e39249ad", "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}
\ No newline at end of file
diff --git a/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/mistral/dacorvo/tiny-random-MistralForCausalLM/af0f1b79454574c744eb.json b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/mistral/dacorvo/tiny-random-MistralForCausalLM/af0f1b79454574c744eb.json
new file mode 100644
index 0000000000000000000000000000000000000000..5d5fdd0b9b330d098a44a0e833f04180f9c50a31
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/mistral/dacorvo/tiny-random-MistralForCausalLM/af0f1b79454574c744eb.json
@@ -0,0 +1 @@
+{"architectures": ["MistralForCausalLM"], "attention_dropout": 0.0, "attention_probs_dropout_prob": 0.1, "bos_token_id": 1, "eos_token_id": 2, "head_dim": 8, "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": "2.15.143.0+e39249ad", "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}
\ No newline at end of file
diff --git a/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/mistral/optimum/mistral-1.1b-testing/37aed4e6734aa8542ee8.json b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/mistral/optimum/mistral-1.1b-testing/37aed4e6734aa8542ee8.json
new file mode 100644
index 0000000000000000000000000000000000000000..db563a3b80b128ea6b36043faf9b23b3bfa93efc
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/mistral/optimum/mistral-1.1b-testing/37aed4e6734aa8542ee8.json
@@ -0,0 +1 @@
+{"architectures": ["MistralForCausalLM"], "attention_dropout": 0.0, "bos_token_id": 1, "eos_token_id": 2, "head_dim": 64, "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": "2.15.143.0+e39249ad", "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}
\ No newline at end of file
diff --git a/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/mixtral/dacorvo/Mixtral-tiny/8dde518ab8e435c4c830.json b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/mixtral/dacorvo/Mixtral-tiny/8dde518ab8e435c4c830.json
new file mode 100644
index 0000000000000000000000000000000000000000..346841b87db6b189239455938250b4b0e294f4fc
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/mixtral/dacorvo/Mixtral-tiny/8dde518ab8e435c4c830.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": "2.15.143.0+e39249ad", "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}
\ No newline at end of file
diff --git a/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/mixtral/dacorvo/Mixtral-tiny/9ac32cae9a1e9b72b9ed.json b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/mixtral/dacorvo/Mixtral-tiny/9ac32cae9a1e9b72b9ed.json
new file mode 100644
index 0000000000000000000000000000000000000000..5c9147706731c7389fc94af73531aa61732c3b35
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/mixtral/dacorvo/Mixtral-tiny/9ac32cae9a1e9b72b9ed.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": "2.15.143.0+e39249ad", "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}
\ No newline at end of file
diff --git a/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/mixtral/dacorvo/Mixtral-tiny/d777c911e955212a0132.json b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/mixtral/dacorvo/Mixtral-tiny/d777c911e955212a0132.json
new file mode 100644
index 0000000000000000000000000000000000000000..afd3e08f6cca6df12ddbde8d5a32ba0098f5a03f
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/mixtral/dacorvo/Mixtral-tiny/d777c911e955212a0132.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": "2.15.143.0+e39249ad", "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}
\ No newline at end of file
diff --git a/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/mixtral/dacorvo/Mixtral-tiny/fc0751b75e6a6a09570f.json b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/mixtral/dacorvo/Mixtral-tiny/fc0751b75e6a6a09570f.json
new file mode 100644
index 0000000000000000000000000000000000000000..2469b6e3fa5c4effdf9c7cbbe40eb4a1b768b073
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/mixtral/dacorvo/Mixtral-tiny/fc0751b75e6a6a09570f.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": 4, "checkpoint_id": "dacorvo/Mixtral-tiny", "checkpoint_revision": "c557ba205ddff6ea911f4719e0d543d6c08356b6", "compiler_type": "neuronx-cc", "compiler_version": "2.15.143.0+e39249ad", "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}
\ No newline at end of file
diff --git a/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/opt/hf-internal-testing/tiny-random-OPTForCausalLM/93d7843b7abe79348924.json b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/opt/hf-internal-testing/tiny-random-OPTForCausalLM/93d7843b7abe79348924.json
new file mode 100644
index 0000000000000000000000000000000000000000..2a54f028916913440407924bec90fd3f7a288c3c
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/opt/hf-internal-testing/tiny-random-OPTForCausalLM/93d7843b7abe79348924.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": "3f3f2ee1e499cb7ad89b877068684fdc9d9513c3", "compiler_type": "neuronx-cc", "compiler_version": "2.15.143.0+e39249ad", "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}
\ No newline at end of file
diff --git a/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/opt/hf-internal-testing/tiny-random-OPTForCausalLM/b392ca0eec3368da72ec.json b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/opt/hf-internal-testing/tiny-random-OPTForCausalLM/b392ca0eec3368da72ec.json
new file mode 100644
index 0000000000000000000000000000000000000000..b04d0e42f426d51c555f3561b5364f219d4aeff6
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/opt/hf-internal-testing/tiny-random-OPTForCausalLM/b392ca0eec3368da72ec.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": "3f3f2ee1e499cb7ad89b877068684fdc9d9513c3", "compiler_type": "neuronx-cc", "compiler_version": "2.15.143.0+e39249ad", "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}
\ No newline at end of file
diff --git a/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/opt/hf-internal-testing/tiny-random-OPTForCausalLM/f829b04255e2c75cd7c8.json b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/opt/hf-internal-testing/tiny-random-OPTForCausalLM/f829b04255e2c75cd7c8.json
new file mode 100644
index 0000000000000000000000000000000000000000..4534d9a43560929841fe4f40ab4d9fbe6adbaa23
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/opt/hf-internal-testing/tiny-random-OPTForCausalLM/f829b04255e2c75cd7c8.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": "3f3f2ee1e499cb7ad89b877068684fdc9d9513c3", "compiler_type": "neuronx-cc", "compiler_version": "2.15.143.0+e39249ad", "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}
\ No newline at end of file
diff --git a/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/qwen2/Qwen/Qwen2.5-0.5B/f404b6bcd5e4c946681b.json b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/qwen2/Qwen/Qwen2.5-0.5B/f404b6bcd5e4c946681b.json
new file mode 100644
index 0000000000000000000000000000000000000000..8d5618993a837a816dd5b293513c097be1903815
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27/inference/qwen2/Qwen/Qwen2.5-0.5B/f404b6bcd5e4c946681b.json
@@ -0,0 +1 @@
+{"architectures": ["Qwen2ForCausalLM"], "attention_dropout": 0.0, "bos_token_id": 151643, "eos_token_id": 151643, "hidden_act": "silu", "hidden_size": 896, "initializer_range": 0.02, "intermediate_size": 4864, "max_position_embeddings": 32768, "max_window_layers": 24, "model_type": "qwen2", "neuron": {"auto_cast_type": "fp16", "batch_size": 4, "checkpoint_id": "Qwen/Qwen2.5-0.5B", "checkpoint_revision": "060db6499f32faf8b98477b0a26969ef7d8b9987", "compiler_type": "neuronx-cc", "compiler_version": "2.15.143.0+e39249ad", "num_cores": 2, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 14, "num_hidden_layers": 24, "num_key_value_heads": 2, "rms_norm_eps": 1e-06, "rope_theta": 1000000.0, "sliding_window": null, "tie_word_embeddings": true, "torch_dtype": "bfloat16", "use_cache": true, "use_mrope": false, "use_sliding_window": false, "vocab_size": 151936}
\ No newline at end of file
diff --git a/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.28.dev0/inference/granite/ibm-granite/granite-3.0-8b-instruct/1438e8ced52a00df5a5e.json b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.28.dev0/inference/granite/ibm-granite/granite-3.0-8b-instruct/1438e8ced52a00df5a5e.json
new file mode 100644
index 0000000000000000000000000000000000000000..999b4d92df53133ecf1d83333b726e9c204fc7b9
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.28.dev0/inference/granite/ibm-granite/granite-3.0-8b-instruct/1438e8ced52a00df5a5e.json
@@ -0,0 +1 @@
+{"architectures": ["GraniteForCausalLM"], "attention_bias": false, "attention_dropout": 0.1, "attention_multiplier": 0.0078125, "bos_token_id": 0, "embedding_multiplier": 12.0, "eos_token_id": 0, "hidden_act": "silu", "hidden_size": 4096, "initializer_range": 0.02, "intermediate_size": 12800, "logits_scaling": 16.0, "max_position_embeddings": 4096, "mlp_bias": false, "model_type": "granite", "neuron": {"auto_cast_type": "bf16", "batch_size": 1, "checkpoint_id": "ibm-granite/granite-3.0-8b-instruct", "checkpoint_revision": "8fe1e202a17f7763bd0af471253e00cc846d1c05", "compiler_type": "neuronx-cc", "compiler_version": "2.15.143.0+e39249ad", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 40, "num_key_value_heads": 8, "pad_token_id": 0, "residual_multiplier": 0.22, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 10000, "tie_word_embeddings": true, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 49155}
\ No newline at end of file
diff --git a/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.28.dev0/inference/qwen2/Qwen/Qwen2.5-7B-Instruct/febd5daac4e51196a9ec.json b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.28.dev0/inference/qwen2/Qwen/Qwen2.5-7B-Instruct/febd5daac4e51196a9ec.json
new file mode 100644
index 0000000000000000000000000000000000000000..67e747668ae8727839a3aaba709a35fedd26f66e
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.28.dev0/inference/qwen2/Qwen/Qwen2.5-7B-Instruct/febd5daac4e51196a9ec.json
@@ -0,0 +1 @@
+{"architectures": ["Qwen2ForCausalLM"], "attention_dropout": 0.0, "bos_token_id": 151643, "eos_token_id": 151645, "hidden_act": "silu", "hidden_size": 3584, "initializer_range": 0.02, "intermediate_size": 18944, "max_position_embeddings": 32768, "max_window_layers": 28, "model_type": "qwen2", "neuron": {"auto_cast_type": "bf16", "batch_size": 1, "checkpoint_id": "Qwen/Qwen2.5-7B-Instruct", "checkpoint_revision": "bb46c15ee4bb56c5b63245ef50fd7637234d6f75", "compiler_type": "neuronx-cc", "compiler_version": "2.15.143.0+e39249ad", "num_cores": 2, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 28, "num_hidden_layers": 28, "num_key_value_heads": 4, "rms_norm_eps": 1e-06, "rope_theta": 1000000.0, "sliding_window": null, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "use_sliding_window": false, "vocab_size": 152064}
\ No newline at end of file
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_0e92acc32b557ad5547d+39f12043/compile_flags.json b/neuronxcc-2.15.143.0+e39249ad/MODULE_0e92acc32b557ad5547d+39f12043/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..146361c3d3be9576d184ec08a57699cb27736489
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/MODULE_0e92acc32b557ad5547d+39f12043/compile_flags.json
@@ -0,0 +1 @@
+["--model-type=transformer", "--auto-cast=none", "--execute-repetition=1"]
\ No newline at end of file
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_0e92acc32b557ad5547d+39f12043/model.done b/neuronxcc-2.15.143.0+e39249ad/MODULE_0e92acc32b557ad5547d+39f12043/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_0e92acc32b557ad5547d+39f12043/model.hlo_module.pb b/neuronxcc-2.15.143.0+e39249ad/MODULE_0e92acc32b557ad5547d+39f12043/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..f1151a45ebb7dbbbd84ff1bbff0d78ae7fd7f686
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/MODULE_0e92acc32b557ad5547d+39f12043/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2514b1562518240f392dc23f26eed4690632f195ab3fdf22ce55b08f9fa8b84d
+size 327080
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_0e92acc32b557ad5547d+39f12043/model.neff b/neuronxcc-2.15.143.0+e39249ad/MODULE_0e92acc32b557ad5547d+39f12043/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..4be8fce3e925e60ec6302af3921f4dd42156aaa5
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/MODULE_0e92acc32b557ad5547d+39f12043/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cd5c03370d27a0f2139658bebe2adacf802aa58a577f20d1e5942e82663d673b
+size 3277824
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_39a9ccce9adcabc94675+39f12043/compile_flags.json b/neuronxcc-2.15.143.0+e39249ad/MODULE_39a9ccce9adcabc94675+39f12043/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..146361c3d3be9576d184ec08a57699cb27736489
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/MODULE_39a9ccce9adcabc94675+39f12043/compile_flags.json
@@ -0,0 +1 @@
+["--model-type=transformer", "--auto-cast=none", "--execute-repetition=1"]
\ No newline at end of file
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_39a9ccce9adcabc94675+39f12043/model.done b/neuronxcc-2.15.143.0+e39249ad/MODULE_39a9ccce9adcabc94675+39f12043/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_39a9ccce9adcabc94675+39f12043/model.hlo_module.pb b/neuronxcc-2.15.143.0+e39249ad/MODULE_39a9ccce9adcabc94675+39f12043/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..7b8eeb1a6fac919d76abfe273230fb4e07b88542
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/MODULE_39a9ccce9adcabc94675+39f12043/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:66139a9078f78d3463139b25024ac289e96ce1e865112ae0e4244da6d84cb53a
+size 420857
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_39a9ccce9adcabc94675+39f12043/model.neff b/neuronxcc-2.15.143.0+e39249ad/MODULE_39a9ccce9adcabc94675+39f12043/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..f51124892439248972693fc22e0e2519e62ef0c0
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/MODULE_39a9ccce9adcabc94675+39f12043/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c84f867b50085e0c6914a24ad8f4ce02588f36ca004b3c67f84b8bb18b008aeb
+size 2898944
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_3c33fc8c5b8139f55998+39f12043/compile_flags.json b/neuronxcc-2.15.143.0+e39249ad/MODULE_3c33fc8c5b8139f55998+39f12043/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..146361c3d3be9576d184ec08a57699cb27736489
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/MODULE_3c33fc8c5b8139f55998+39f12043/compile_flags.json
@@ -0,0 +1 @@
+["--model-type=transformer", "--auto-cast=none", "--execute-repetition=1"]
\ No newline at end of file
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_3c33fc8c5b8139f55998+39f12043/model.done b/neuronxcc-2.15.143.0+e39249ad/MODULE_3c33fc8c5b8139f55998+39f12043/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_3c33fc8c5b8139f55998+39f12043/model.hlo_module.pb b/neuronxcc-2.15.143.0+e39249ad/MODULE_3c33fc8c5b8139f55998+39f12043/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..e26a0bfa303a065caca71ac3eabebf46a2c7f18b
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/MODULE_3c33fc8c5b8139f55998+39f12043/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b1b1ae225c4153d533ac9f21418f6dc18c680a20eb8ee5e6eba4c7d9b525c855
+size 420857
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_3c33fc8c5b8139f55998+39f12043/model.neff b/neuronxcc-2.15.143.0+e39249ad/MODULE_3c33fc8c5b8139f55998+39f12043/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..9ec47fea9d3968a613135c9051225f8e7d668d3f
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/MODULE_3c33fc8c5b8139f55998+39f12043/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:781c61aef015d69019a5b102df834d1c89f065a0b8aeefc2da8c25d93e856e43
+size 3011584
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_402e15278f41a32c430e+39f12043/compile_flags.json b/neuronxcc-2.15.143.0+e39249ad/MODULE_402e15278f41a32c430e+39f12043/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..146361c3d3be9576d184ec08a57699cb27736489
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/MODULE_402e15278f41a32c430e+39f12043/compile_flags.json
@@ -0,0 +1 @@
+["--model-type=transformer", "--auto-cast=none", "--execute-repetition=1"]
\ No newline at end of file
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_402e15278f41a32c430e+39f12043/model.done b/neuronxcc-2.15.143.0+e39249ad/MODULE_402e15278f41a32c430e+39f12043/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_402e15278f41a32c430e+39f12043/model.hlo_module.pb b/neuronxcc-2.15.143.0+e39249ad/MODULE_402e15278f41a32c430e+39f12043/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..89a5f80ba25399bbfb7e3a8e3fc8a7bf28aefed7
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/MODULE_402e15278f41a32c430e+39f12043/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:85b2908092b6828bd169dc87b156620f30b188c6b8aae2f980fc407902cc1bc9
+size 420857
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_402e15278f41a32c430e+39f12043/model.neff b/neuronxcc-2.15.143.0+e39249ad/MODULE_402e15278f41a32c430e+39f12043/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..6fe6f5d7478bd588eb997608e77bd60d33f5c1bc
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/MODULE_402e15278f41a32c430e+39f12043/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e06270f172345507c41126fa7d0dbb6f03994659f231c87a5a876e2f326ccf1e
+size 2765824
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_6550f3db82ff02b1c1ff+39f12043/compile_flags.json b/neuronxcc-2.15.143.0+e39249ad/MODULE_6550f3db82ff02b1c1ff+39f12043/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..146361c3d3be9576d184ec08a57699cb27736489
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/MODULE_6550f3db82ff02b1c1ff+39f12043/compile_flags.json
@@ -0,0 +1 @@
+["--model-type=transformer", "--auto-cast=none", "--execute-repetition=1"]
\ No newline at end of file
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_6550f3db82ff02b1c1ff+39f12043/model.done b/neuronxcc-2.15.143.0+e39249ad/MODULE_6550f3db82ff02b1c1ff+39f12043/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_6550f3db82ff02b1c1ff+39f12043/model.hlo_module.pb b/neuronxcc-2.15.143.0+e39249ad/MODULE_6550f3db82ff02b1c1ff+39f12043/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..1b7c8df84dbb3e2e3dce1d199ac815fddd80c6e1
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/MODULE_6550f3db82ff02b1c1ff+39f12043/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:11ae7d1bd42a59bd70b349ee9db316ab2c76c50bdded4b7a2e54b78d40c0e8c1
+size 420857
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_6550f3db82ff02b1c1ff+39f12043/model.neff b/neuronxcc-2.15.143.0+e39249ad/MODULE_6550f3db82ff02b1c1ff+39f12043/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..63d97d63ee9774d7a3d8c22a45c02ad65f34ceee
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/MODULE_6550f3db82ff02b1c1ff+39f12043/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:de9d486c03967c75b66a525b123a6376d3484869a1b525e0e833133243924423
+size 2776064
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_67a9118947e020080e96+39f12043/compile_flags.json b/neuronxcc-2.15.143.0+e39249ad/MODULE_67a9118947e020080e96+39f12043/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..146361c3d3be9576d184ec08a57699cb27736489
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/MODULE_67a9118947e020080e96+39f12043/compile_flags.json
@@ -0,0 +1 @@
+["--model-type=transformer", "--auto-cast=none", "--execute-repetition=1"]
\ No newline at end of file
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_67a9118947e020080e96+39f12043/model.done b/neuronxcc-2.15.143.0+e39249ad/MODULE_67a9118947e020080e96+39f12043/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_67a9118947e020080e96+39f12043/model.hlo_module.pb b/neuronxcc-2.15.143.0+e39249ad/MODULE_67a9118947e020080e96+39f12043/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..2315861fc8bffed539cd20bba0b07308d99b45ec
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/MODULE_67a9118947e020080e96+39f12043/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cb6de45faca3f425b709bb6e8c2990ae325ef7cbda8431c04af4838386c25b90
+size 327080
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_67a9118947e020080e96+39f12043/model.neff b/neuronxcc-2.15.143.0+e39249ad/MODULE_67a9118947e020080e96+39f12043/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..3c9d4fd80c69e5a2f15a9448a0dd2c2d7e833b99
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/MODULE_67a9118947e020080e96+39f12043/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6609bcb561b93b46a45ae928cad55c6fbf6c532dffb64619e2f5ce26282b6667
+size 2939904
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_7dbd99330688237c55ff+39f12043/compile_flags.json b/neuronxcc-2.15.143.0+e39249ad/MODULE_7dbd99330688237c55ff+39f12043/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..146361c3d3be9576d184ec08a57699cb27736489
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/MODULE_7dbd99330688237c55ff+39f12043/compile_flags.json
@@ -0,0 +1 @@
+["--model-type=transformer", "--auto-cast=none", "--execute-repetition=1"]
\ No newline at end of file
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_7dbd99330688237c55ff+39f12043/model.done b/neuronxcc-2.15.143.0+e39249ad/MODULE_7dbd99330688237c55ff+39f12043/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_7dbd99330688237c55ff+39f12043/model.hlo_module.pb b/neuronxcc-2.15.143.0+e39249ad/MODULE_7dbd99330688237c55ff+39f12043/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..52b4072998eedd97075103cd4fe2dfc48e2b3bc2
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/MODULE_7dbd99330688237c55ff+39f12043/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:eac64fc49416b6148ba7bdb435e10bbeec794a6d423ad15fe1a777e8a76adc1d
+size 322569
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_7dbd99330688237c55ff+39f12043/model.neff b/neuronxcc-2.15.143.0+e39249ad/MODULE_7dbd99330688237c55ff+39f12043/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..a9eed3b8ad28360e6a3054c7846cc8e2ecc0555e
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/MODULE_7dbd99330688237c55ff+39f12043/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1dc22ba8c11134aaab40acdbeaf80da2fcb4659c6a57860f4beb5f4d3bc6fbab
+size 2929664
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_82cbf543db4557dd4363+39f12043/compile_flags.json b/neuronxcc-2.15.143.0+e39249ad/MODULE_82cbf543db4557dd4363+39f12043/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..146361c3d3be9576d184ec08a57699cb27736489
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/MODULE_82cbf543db4557dd4363+39f12043/compile_flags.json
@@ -0,0 +1 @@
+["--model-type=transformer", "--auto-cast=none", "--execute-repetition=1"]
\ No newline at end of file
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_82cbf543db4557dd4363+39f12043/model.done b/neuronxcc-2.15.143.0+e39249ad/MODULE_82cbf543db4557dd4363+39f12043/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_82cbf543db4557dd4363+39f12043/model.hlo_module.pb b/neuronxcc-2.15.143.0+e39249ad/MODULE_82cbf543db4557dd4363+39f12043/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..a7af26fdf16f27dce49903782aba0c35f7c1714a
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/MODULE_82cbf543db4557dd4363+39f12043/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:69e614bde40f7fbb55e8d79b068eaf2fdea5fbc223f9ae109444cc7aa366427f
+size 52462
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_82cbf543db4557dd4363+39f12043/model.neff b/neuronxcc-2.15.143.0+e39249ad/MODULE_82cbf543db4557dd4363+39f12043/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..d0741ac37a55addcf16d604c9ae7aa63b303937e
Binary files /dev/null and b/neuronxcc-2.15.143.0+e39249ad/MODULE_82cbf543db4557dd4363+39f12043/model.neff differ
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_aa4954701eb403167936+39f12043/compile_flags.json b/neuronxcc-2.15.143.0+e39249ad/MODULE_aa4954701eb403167936+39f12043/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..146361c3d3be9576d184ec08a57699cb27736489
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/MODULE_aa4954701eb403167936+39f12043/compile_flags.json
@@ -0,0 +1 @@
+["--model-type=transformer", "--auto-cast=none", "--execute-repetition=1"]
\ No newline at end of file
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_aa4954701eb403167936+39f12043/model.done b/neuronxcc-2.15.143.0+e39249ad/MODULE_aa4954701eb403167936+39f12043/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_aa4954701eb403167936+39f12043/model.hlo_module.pb b/neuronxcc-2.15.143.0+e39249ad/MODULE_aa4954701eb403167936+39f12043/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..70feec64e63635f86ec6ca4f741976f982cc7fd0
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/MODULE_aa4954701eb403167936+39f12043/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a2bcea3e588d878e444be361535a6e9d56a25f457f486001b1ec058e08de8499
+size 420857
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_aa4954701eb403167936+39f12043/model.neff b/neuronxcc-2.15.143.0+e39249ad/MODULE_aa4954701eb403167936+39f12043/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..95583a41b63f1be5d31893b7a41d91674573d660
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/MODULE_aa4954701eb403167936+39f12043/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:29849a3ab4927b114fc6911e61019a624113ec15a2f1451b0d90e78fb52e82e9
+size 2827264
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_aefce72d1e0612585993+39f12043/compile_flags.json b/neuronxcc-2.15.143.0+e39249ad/MODULE_aefce72d1e0612585993+39f12043/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..146361c3d3be9576d184ec08a57699cb27736489
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/MODULE_aefce72d1e0612585993+39f12043/compile_flags.json
@@ -0,0 +1 @@
+["--model-type=transformer", "--auto-cast=none", "--execute-repetition=1"]
\ No newline at end of file
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_aefce72d1e0612585993+39f12043/model.done b/neuronxcc-2.15.143.0+e39249ad/MODULE_aefce72d1e0612585993+39f12043/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_aefce72d1e0612585993+39f12043/model.hlo_module.pb b/neuronxcc-2.15.143.0+e39249ad/MODULE_aefce72d1e0612585993+39f12043/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..d0738ea6fa25f9b0caa2ff9d73aedfe3fb9bb656
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/MODULE_aefce72d1e0612585993+39f12043/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a64e74fcfefb34b7e8185bc4677c2c4dc1aa8a4af86c8154a19a9049c9f51efc
+size 327080
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_aefce72d1e0612585993+39f12043/model.neff b/neuronxcc-2.15.143.0+e39249ad/MODULE_aefce72d1e0612585993+39f12043/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..5eb391fbb111d9ab1a8d6eac846fc1d2024e451c
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/MODULE_aefce72d1e0612585993+39f12043/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:20558c3b236d16339e11032b145efa7395d88494c4ba887420b80a3f85300d9a
+size 14930944
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_b0acdacec2caf4887452+39f12043/compile_flags.json b/neuronxcc-2.15.143.0+e39249ad/MODULE_b0acdacec2caf4887452+39f12043/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..146361c3d3be9576d184ec08a57699cb27736489
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/MODULE_b0acdacec2caf4887452+39f12043/compile_flags.json
@@ -0,0 +1 @@
+["--model-type=transformer", "--auto-cast=none", "--execute-repetition=1"]
\ No newline at end of file
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_b0acdacec2caf4887452+39f12043/model.done b/neuronxcc-2.15.143.0+e39249ad/MODULE_b0acdacec2caf4887452+39f12043/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_b0acdacec2caf4887452+39f12043/model.hlo_module.pb b/neuronxcc-2.15.143.0+e39249ad/MODULE_b0acdacec2caf4887452+39f12043/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..beb383cd6e5d05c8c46d306970173ceba2db32e1
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/MODULE_b0acdacec2caf4887452+39f12043/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6842534cad9f0e98c4978ee6047c49e8b50d1d138acc8aa5af2586f591565b73
+size 327080
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_b0acdacec2caf4887452+39f12043/model.neff b/neuronxcc-2.15.143.0+e39249ad/MODULE_b0acdacec2caf4887452+39f12043/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..9bd15a6e82dfba1f854570c9722840334cfd10c5
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/MODULE_b0acdacec2caf4887452+39f12043/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2e65dfd819a00164d688cdd03aece0bfbe8a98b9b7ad7fa1af2b332c86b94315
+size 4107264
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_c7dd940fbcfda75b0522+39f12043/compile_flags.json b/neuronxcc-2.15.143.0+e39249ad/MODULE_c7dd940fbcfda75b0522+39f12043/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..146361c3d3be9576d184ec08a57699cb27736489
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/MODULE_c7dd940fbcfda75b0522+39f12043/compile_flags.json
@@ -0,0 +1 @@
+["--model-type=transformer", "--auto-cast=none", "--execute-repetition=1"]
\ No newline at end of file
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_c7dd940fbcfda75b0522+39f12043/model.done b/neuronxcc-2.15.143.0+e39249ad/MODULE_c7dd940fbcfda75b0522+39f12043/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_c7dd940fbcfda75b0522+39f12043/model.hlo_module.pb b/neuronxcc-2.15.143.0+e39249ad/MODULE_c7dd940fbcfda75b0522+39f12043/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..800c6c92e876ae65f22e917b660928f00370d926
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/MODULE_c7dd940fbcfda75b0522+39f12043/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6bffcb2823f8201e760f5723b6cd6c5a63aed40089257262070efcf665a54cc3
+size 327080
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_c7dd940fbcfda75b0522+39f12043/model.neff b/neuronxcc-2.15.143.0+e39249ad/MODULE_c7dd940fbcfda75b0522+39f12043/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..1c45e6ecc86098d023e3577d4671e382b56fbb77
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/MODULE_c7dd940fbcfda75b0522+39f12043/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:743a90711a47a78d549ae8eb6af77671e0b61888ffa045807c1a6142385d12dd
+size 7281664
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_c9e2909fdfcfd10ac3b0+39f12043/compile_flags.json b/neuronxcc-2.15.143.0+e39249ad/MODULE_c9e2909fdfcfd10ac3b0+39f12043/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..146361c3d3be9576d184ec08a57699cb27736489
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/MODULE_c9e2909fdfcfd10ac3b0+39f12043/compile_flags.json
@@ -0,0 +1 @@
+["--model-type=transformer", "--auto-cast=none", "--execute-repetition=1"]
\ No newline at end of file
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_c9e2909fdfcfd10ac3b0+39f12043/model.done b/neuronxcc-2.15.143.0+e39249ad/MODULE_c9e2909fdfcfd10ac3b0+39f12043/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_c9e2909fdfcfd10ac3b0+39f12043/model.hlo_module.pb b/neuronxcc-2.15.143.0+e39249ad/MODULE_c9e2909fdfcfd10ac3b0+39f12043/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..e6d0943174096a2f77ec512e9bb694e9c23a1135
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/MODULE_c9e2909fdfcfd10ac3b0+39f12043/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b83412ec92554d74d4b38d218a4506c213126b7077da3b97fe0ac909e489e814
+size 52462
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_c9e2909fdfcfd10ac3b0+39f12043/model.neff b/neuronxcc-2.15.143.0+e39249ad/MODULE_c9e2909fdfcfd10ac3b0+39f12043/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..ef7086d93fb33474b6d64752b7311c80287038fe
Binary files /dev/null and b/neuronxcc-2.15.143.0+e39249ad/MODULE_c9e2909fdfcfd10ac3b0+39f12043/model.neff differ
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_d2d015220cf99f09e551+39f12043/compile_flags.json b/neuronxcc-2.15.143.0+e39249ad/MODULE_d2d015220cf99f09e551+39f12043/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..146361c3d3be9576d184ec08a57699cb27736489
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/MODULE_d2d015220cf99f09e551+39f12043/compile_flags.json
@@ -0,0 +1 @@
+["--model-type=transformer", "--auto-cast=none", "--execute-repetition=1"]
\ No newline at end of file
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_d2d015220cf99f09e551+39f12043/model.done b/neuronxcc-2.15.143.0+e39249ad/MODULE_d2d015220cf99f09e551+39f12043/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_d2d015220cf99f09e551+39f12043/model.hlo_module.pb b/neuronxcc-2.15.143.0+e39249ad/MODULE_d2d015220cf99f09e551+39f12043/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..69555acd84437767909aca21afb590dd241645ad
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/MODULE_d2d015220cf99f09e551+39f12043/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:890b699709f557b1bff9e48827b260cda17574e909549f1ceba0bcdf1f802fe7
+size 52462
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_d2d015220cf99f09e551+39f12043/model.neff b/neuronxcc-2.15.143.0+e39249ad/MODULE_d2d015220cf99f09e551+39f12043/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..c79f053b2774761063a1a6f2004d426f0ebfe979
Binary files /dev/null and b/neuronxcc-2.15.143.0+e39249ad/MODULE_d2d015220cf99f09e551+39f12043/model.neff differ
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_e1bacfe9616dab23b889+39f12043/compile_flags.json b/neuronxcc-2.15.143.0+e39249ad/MODULE_e1bacfe9616dab23b889+39f12043/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..146361c3d3be9576d184ec08a57699cb27736489
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/MODULE_e1bacfe9616dab23b889+39f12043/compile_flags.json
@@ -0,0 +1 @@
+["--model-type=transformer", "--auto-cast=none", "--execute-repetition=1"]
\ No newline at end of file
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_e1bacfe9616dab23b889+39f12043/model.done b/neuronxcc-2.15.143.0+e39249ad/MODULE_e1bacfe9616dab23b889+39f12043/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_e1bacfe9616dab23b889+39f12043/model.hlo_module.pb b/neuronxcc-2.15.143.0+e39249ad/MODULE_e1bacfe9616dab23b889+39f12043/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..c0bdc49c66b390ea6737b74757d1e5fa92fe2212
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/MODULE_e1bacfe9616dab23b889+39f12043/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:928585b8e5f71dd4a28c288f89edbe330b39b0bbc6e1987f8362deb35d7208d1
+size 420857
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_e1bacfe9616dab23b889+39f12043/model.neff b/neuronxcc-2.15.143.0+e39249ad/MODULE_e1bacfe9616dab23b889+39f12043/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..67ab20f3464dac031bddda344f62634cdb8e53ee
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/MODULE_e1bacfe9616dab23b889+39f12043/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:856a9358cb893a54137bbbab1e9e74ca0b3ba32637588c79a3db47ebc18b3aea
+size 2796544