diff --git a/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.28.dev0/inference/llama/llamafactory/tiny-random-Llama-3/2d90eff859bef42c3a80.json b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.28.dev0/inference/llama/llamafactory/tiny-random-Llama-3/2d90eff859bef42c3a80.json
new file mode 100644
index 0000000000000000000000000000000000000000..30626ab8b97cea6ed795edea35a4da9f2a69e2d8
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.28.dev0/inference/llama/llamafactory/tiny-random-Llama-3/2d90eff859bef42c3a80.json
@@ -0,0 +1 @@
+{"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 128000, "eos_token_id": [128001, 128008, 128009], "head_dim": 4, "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": "b6475c3a7bd11f8d5056144b1c13d3d4ceb10026", "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": {"factor": 8.0, "high_freq_factor": 4.0, "low_freq_factor": 1.0, "original_max_position_embeddings": 8192, "rope_type": "llama3"}, "rope_theta": 500000.0, "tie_word_embeddings": false, "torch_dtype": "float16", "use_cache": true, "vocab_size": 128256}
\ No newline at end of file
diff --git a/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.28.dev0/inference/llama/llamafactory/tiny-random-Llama-3/64e733e0df11827bb701.json b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.28.dev0/inference/llama/llamafactory/tiny-random-Llama-3/64e733e0df11827bb701.json
new file mode 100644
index 0000000000000000000000000000000000000000..36fe4dcac87bdf64beef4405d4a4a44d477afcab
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.28.dev0/inference/llama/llamafactory/tiny-random-Llama-3/64e733e0df11827bb701.json
@@ -0,0 +1 @@
+{"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 128000, "eos_token_id": [128001, 128008, 128009], "head_dim": 4, "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": "b6475c3a7bd11f8d5056144b1c13d3d4ceb10026", "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": {"factor": 8.0, "high_freq_factor": 4.0, "low_freq_factor": 1.0, "original_max_position_embeddings": 8192, "rope_type": "llama3"}, "rope_theta": 500000.0, "tie_word_embeddings": false, "torch_dtype": "float16", "use_cache": true, "vocab_size": 128256}
\ No newline at end of file
diff --git a/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.28.dev0/inference/llama/llamafactory/tiny-random-Llama-3/8e2a857a01d8d2193ba3.json b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.28.dev0/inference/llama/llamafactory/tiny-random-Llama-3/8e2a857a01d8d2193ba3.json
new file mode 100644
index 0000000000000000000000000000000000000000..b3bfe77a007bd4e6cc2ee2283cb3ee3893ba0d11
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.28.dev0/inference/llama/llamafactory/tiny-random-Llama-3/8e2a857a01d8d2193ba3.json
@@ -0,0 +1 @@
+{"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 128000, "eos_token_id": [128001, 128008, 128009], "head_dim": 4, "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": "b6475c3a7bd11f8d5056144b1c13d3d4ceb10026", "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": {"factor": 8.0, "high_freq_factor": 4.0, "low_freq_factor": 1.0, "original_max_position_embeddings": 8192, "rope_type": "llama3"}, "rope_theta": 500000.0, "tie_word_embeddings": false, "torch_dtype": "float16", "use_cache": true, "vocab_size": 128256}
\ No newline at end of file
diff --git a/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.28.dev0/inference/mixtral/dacorvo/Mixtral-tiny/50328eaa8ef14cde8b99.json b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.28.dev0/inference/mixtral/dacorvo/Mixtral-tiny/50328eaa8ef14cde8b99.json
new file mode 100644
index 0000000000000000000000000000000000000000..26df9670615a78020443b5697aade2cfb4274c3b
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.28.dev0/inference/mixtral/dacorvo/Mixtral-tiny/50328eaa8ef14cde8b99.json
@@ -0,0 +1 @@
+{"architectures": ["MixtralForCausalLM"], "attention_dropout": 0.0, "bos_token_id": 1, "eos_token_id": 2, "head_dim": 32, "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/0.0.28.dev0/inference/mixtral/dacorvo/Mixtral-tiny/9e5cb5eaf4e7d583d446.json b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.28.dev0/inference/mixtral/dacorvo/Mixtral-tiny/9e5cb5eaf4e7d583d446.json
new file mode 100644
index 0000000000000000000000000000000000000000..63651ede70443350c0bb654b635a2d69687d39e3
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.28.dev0/inference/mixtral/dacorvo/Mixtral-tiny/9e5cb5eaf4e7d583d446.json
@@ -0,0 +1 @@
+{"architectures": ["MixtralForCausalLM"], "attention_dropout": 0.0, "bos_token_id": 1, "eos_token_id": 2, "head_dim": 32, "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/0.0.28.dev0/inference/mixtral/dacorvo/Mixtral-tiny/fb61ef026f730d03d8a9.json b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.28.dev0/inference/mixtral/dacorvo/Mixtral-tiny/fb61ef026f730d03d8a9.json
new file mode 100644
index 0000000000000000000000000000000000000000..a5c798372cb8908bde5a53e715ccee7271f51e97
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.28.dev0/inference/mixtral/dacorvo/Mixtral-tiny/fb61ef026f730d03d8a9.json
@@ -0,0 +1 @@
+{"architectures": ["MixtralForCausalLM"], "attention_dropout": 0.0, "bos_token_id": 1, "eos_token_id": 2, "head_dim": 32, "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/0.0.28.dev1/inference/bloom/hf-internal-testing/tiny-random-BloomForCausalLM/4538babd01182da34f41.json b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.28.dev1/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/0.0.28.dev1/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/0.0.28.dev1/inference/bloom/hf-internal-testing/tiny-random-BloomForCausalLM/b40a794eff5ea470980f.json b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.28.dev1/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/0.0.28.dev1/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/0.0.28.dev1/inference/bloom/hf-internal-testing/tiny-random-BloomForCausalLM/cdbdbf737272e98ae974.json b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.28.dev1/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/0.0.28.dev1/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/0.0.28.dev1/inference/gpt2/gpt2/9125a78a53c4a0cd3507.json b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.28.dev1/inference/gpt2/gpt2/9125a78a53c4a0cd3507.json
new file mode 100644
index 0000000000000000000000000000000000000000..2ea36cfeed1f69b520bf5f2ac5bbc3ea2f02739d
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.28.dev1/inference/gpt2/gpt2/9125a78a53c4a0cd3507.json
@@ -0,0 +1 @@
+{"activation_function": "gelu_new", "architectures": ["GPT2LMHeadModel"], "attn_pdrop": 0.1, "bos_token_id": 50256, "embd_pdrop": 0.1, "eos_token_id": 50256, "initializer_range": 0.02, "layer_norm_epsilon": 1e-05, "model_type": "gpt2", "n_ctx": 1024, "n_embd": 768, "n_head": 12, "n_inner": null, "n_layer": 12, "n_positions": 1024, "neuron": {"auto_cast_type": "fp32", "batch_size": 1, "checkpoint_id": "gpt2", "checkpoint_revision": "607a30d783dfa663caf39e06633721c8d4cfcd7e", "compiler_type": "neuronx-cc", "compiler_version": "2.15.143.0+e39249ad", "num_cores": 2, "sequence_length": 128, "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/0.0.28.dev1/inference/gpt2/gpt2/aadac442fccfdd730bb5.json b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.28.dev1/inference/gpt2/gpt2/aadac442fccfdd730bb5.json
new file mode 100644
index 0000000000000000000000000000000000000000..60376d7ece85795e4beeb4a28074f2021203851e
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.28.dev1/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/0.0.28.dev1/inference/gpt2/hf-internal-testing/tiny-random-gpt2/1f8668afaa87da7a4bc8.json b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.28.dev1/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/0.0.28.dev1/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/0.0.28.dev1/inference/gpt2/hf-internal-testing/tiny-random-gpt2/9055305e649c4ec32010.json b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.28.dev1/inference/gpt2/hf-internal-testing/tiny-random-gpt2/9055305e649c4ec32010.json
new file mode 100644
index 0000000000000000000000000000000000000000..2f8ae5129f36063c1cc195b60ccd46c35a0be795
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.28.dev1/inference/gpt2/hf-internal-testing/tiny-random-gpt2/9055305e649c4ec32010.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": 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/0.0.28.dev1/inference/gpt2/hf-internal-testing/tiny-random-gpt2/af0ef68e4ec57e23cab5.json b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.28.dev1/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/0.0.28.dev1/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/0.0.28.dev1/inference/gpt2/hf-internal-testing/tiny-random-gpt2/e0c5534b610fbfb22b6e.json b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.28.dev1/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/0.0.28.dev1/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/0.0.28.dev1/inference/llama/llamafactory/tiny-random-Llama-3/2d90eff859bef42c3a80.json b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.28.dev1/inference/llama/llamafactory/tiny-random-Llama-3/2d90eff859bef42c3a80.json
new file mode 100644
index 0000000000000000000000000000000000000000..30626ab8b97cea6ed795edea35a4da9f2a69e2d8
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.28.dev1/inference/llama/llamafactory/tiny-random-Llama-3/2d90eff859bef42c3a80.json
@@ -0,0 +1 @@
+{"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 128000, "eos_token_id": [128001, 128008, 128009], "head_dim": 4, "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": "b6475c3a7bd11f8d5056144b1c13d3d4ceb10026", "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": {"factor": 8.0, "high_freq_factor": 4.0, "low_freq_factor": 1.0, "original_max_position_embeddings": 8192, "rope_type": "llama3"}, "rope_theta": 500000.0, "tie_word_embeddings": false, "torch_dtype": "float16", "use_cache": true, "vocab_size": 128256}
\ No newline at end of file
diff --git a/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.28.dev1/inference/llama/llamafactory/tiny-random-Llama-3/64e733e0df11827bb701.json b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.28.dev1/inference/llama/llamafactory/tiny-random-Llama-3/64e733e0df11827bb701.json
new file mode 100644
index 0000000000000000000000000000000000000000..36fe4dcac87bdf64beef4405d4a4a44d477afcab
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.28.dev1/inference/llama/llamafactory/tiny-random-Llama-3/64e733e0df11827bb701.json
@@ -0,0 +1 @@
+{"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 128000, "eos_token_id": [128001, 128008, 128009], "head_dim": 4, "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": "b6475c3a7bd11f8d5056144b1c13d3d4ceb10026", "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": {"factor": 8.0, "high_freq_factor": 4.0, "low_freq_factor": 1.0, "original_max_position_embeddings": 8192, "rope_type": "llama3"}, "rope_theta": 500000.0, "tie_word_embeddings": false, "torch_dtype": "float16", "use_cache": true, "vocab_size": 128256}
\ No newline at end of file
diff --git a/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.28.dev1/inference/llama/llamafactory/tiny-random-Llama-3/8e2a857a01d8d2193ba3.json b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.28.dev1/inference/llama/llamafactory/tiny-random-Llama-3/8e2a857a01d8d2193ba3.json
new file mode 100644
index 0000000000000000000000000000000000000000..b3bfe77a007bd4e6cc2ee2283cb3ee3893ba0d11
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.28.dev1/inference/llama/llamafactory/tiny-random-Llama-3/8e2a857a01d8d2193ba3.json
@@ -0,0 +1 @@
+{"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 128000, "eos_token_id": [128001, 128008, 128009], "head_dim": 4, "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": "b6475c3a7bd11f8d5056144b1c13d3d4ceb10026", "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": {"factor": 8.0, "high_freq_factor": 4.0, "low_freq_factor": 1.0, "original_max_position_embeddings": 8192, "rope_type": "llama3"}, "rope_theta": 500000.0, "tie_word_embeddings": false, "torch_dtype": "float16", "use_cache": true, "vocab_size": 128256}
\ No newline at end of file
diff --git a/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.28.dev1/inference/mistral/dacorvo/tiny-random-MistralForCausalLM/7d2df9c19098ee676fa8.json b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.28.dev1/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/0.0.28.dev1/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/0.0.28.dev1/inference/mistral/dacorvo/tiny-random-MistralForCausalLM/ac0287246ef4575f8b78.json b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.28.dev1/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/0.0.28.dev1/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/0.0.28.dev1/inference/mistral/dacorvo/tiny-random-MistralForCausalLM/af0f1b79454574c744eb.json b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.28.dev1/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/0.0.28.dev1/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/0.0.28.dev1/inference/mixtral/dacorvo/Mixtral-tiny/50328eaa8ef14cde8b99.json b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.28.dev1/inference/mixtral/dacorvo/Mixtral-tiny/50328eaa8ef14cde8b99.json
new file mode 100644
index 0000000000000000000000000000000000000000..26df9670615a78020443b5697aade2cfb4274c3b
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.28.dev1/inference/mixtral/dacorvo/Mixtral-tiny/50328eaa8ef14cde8b99.json
@@ -0,0 +1 @@
+{"architectures": ["MixtralForCausalLM"], "attention_dropout": 0.0, "bos_token_id": 1, "eos_token_id": 2, "head_dim": 32, "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/0.0.28.dev1/inference/mixtral/dacorvo/Mixtral-tiny/9e5cb5eaf4e7d583d446.json b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.28.dev1/inference/mixtral/dacorvo/Mixtral-tiny/9e5cb5eaf4e7d583d446.json
new file mode 100644
index 0000000000000000000000000000000000000000..63651ede70443350c0bb654b635a2d69687d39e3
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.28.dev1/inference/mixtral/dacorvo/Mixtral-tiny/9e5cb5eaf4e7d583d446.json
@@ -0,0 +1 @@
+{"architectures": ["MixtralForCausalLM"], "attention_dropout": 0.0, "bos_token_id": 1, "eos_token_id": 2, "head_dim": 32, "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/0.0.28.dev1/inference/mixtral/dacorvo/Mixtral-tiny/fb61ef026f730d03d8a9.json b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.28.dev1/inference/mixtral/dacorvo/Mixtral-tiny/fb61ef026f730d03d8a9.json
new file mode 100644
index 0000000000000000000000000000000000000000..a5c798372cb8908bde5a53e715ccee7271f51e97
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.28.dev1/inference/mixtral/dacorvo/Mixtral-tiny/fb61ef026f730d03d8a9.json
@@ -0,0 +1 @@
+{"architectures": ["MixtralForCausalLM"], "attention_dropout": 0.0, "bos_token_id": 1, "eos_token_id": 2, "head_dim": 32, "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/0.0.28.dev1/inference/opt/hf-internal-testing/tiny-random-OPTForCausalLM/93d7843b7abe79348924.json b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.28.dev1/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/0.0.28.dev1/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/0.0.28.dev1/inference/opt/hf-internal-testing/tiny-random-OPTForCausalLM/b392ca0eec3368da72ec.json b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.28.dev1/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/0.0.28.dev1/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/0.0.28.dev1/inference/opt/hf-internal-testing/tiny-random-OPTForCausalLM/f829b04255e2c75cd7c8.json b/neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.28.dev1/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/0.0.28.dev1/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/MODULE_088e597b461b7797e98a+39f12043/compile_flags.json b/neuronxcc-2.15.143.0+e39249ad/MODULE_088e597b461b7797e98a+39f12043/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..146361c3d3be9576d184ec08a57699cb27736489
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/MODULE_088e597b461b7797e98a+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_088e597b461b7797e98a+39f12043/model.done b/neuronxcc-2.15.143.0+e39249ad/MODULE_088e597b461b7797e98a+39f12043/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_088e597b461b7797e98a+39f12043/model.hlo_module.pb b/neuronxcc-2.15.143.0+e39249ad/MODULE_088e597b461b7797e98a+39f12043/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..28bafbcd573a7529b5e30ebe12e33d94783ee55f
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/MODULE_088e597b461b7797e98a+39f12043/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8185f5ae3f7c385d6a477b89670134081ddf8a9e6c491067a3c541d3cd7e4116
+size 19815
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_088e597b461b7797e98a+39f12043/model.neff b/neuronxcc-2.15.143.0+e39249ad/MODULE_088e597b461b7797e98a+39f12043/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..73b1f436733aa5a1c77e0cf904b5438085105009
Binary files /dev/null and b/neuronxcc-2.15.143.0+e39249ad/MODULE_088e597b461b7797e98a+39f12043/model.neff differ
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_388f4cccc44b60415fd8+39f12043/compile_flags.json b/neuronxcc-2.15.143.0+e39249ad/MODULE_388f4cccc44b60415fd8+39f12043/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..146361c3d3be9576d184ec08a57699cb27736489
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/MODULE_388f4cccc44b60415fd8+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_388f4cccc44b60415fd8+39f12043/model.done b/neuronxcc-2.15.143.0+e39249ad/MODULE_388f4cccc44b60415fd8+39f12043/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_388f4cccc44b60415fd8+39f12043/model.hlo_module.pb b/neuronxcc-2.15.143.0+e39249ad/MODULE_388f4cccc44b60415fd8+39f12043/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..203c3fb14d7e2998f35fab152ded637727a658c9
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/MODULE_388f4cccc44b60415fd8+39f12043/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7d87abf84bb903e2033d6193ef7927943a0cb3838ae96310d2ef33294710adf7
+size 16706
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_388f4cccc44b60415fd8+39f12043/model.neff b/neuronxcc-2.15.143.0+e39249ad/MODULE_388f4cccc44b60415fd8+39f12043/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..e56f0c7a7db4871ad34aa152dcd7001b02d15803
Binary files /dev/null and b/neuronxcc-2.15.143.0+e39249ad/MODULE_388f4cccc44b60415fd8+39f12043/model.neff differ
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_42ca10d6c86a3705a9b1+39f12043/compile_flags.json b/neuronxcc-2.15.143.0+e39249ad/MODULE_42ca10d6c86a3705a9b1+39f12043/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..146361c3d3be9576d184ec08a57699cb27736489
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/MODULE_42ca10d6c86a3705a9b1+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_42ca10d6c86a3705a9b1+39f12043/model.done b/neuronxcc-2.15.143.0+e39249ad/MODULE_42ca10d6c86a3705a9b1+39f12043/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_42ca10d6c86a3705a9b1+39f12043/model.hlo_module.pb b/neuronxcc-2.15.143.0+e39249ad/MODULE_42ca10d6c86a3705a9b1+39f12043/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..41af7ea401f9d7a6ecf1d8b02da06f345c7b404c
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/MODULE_42ca10d6c86a3705a9b1+39f12043/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ef7f8031707a83cb7f74fb92e8ccf231663c4ff90ac86617c39dda40275e2bf0
+size 16350
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_42ca10d6c86a3705a9b1+39f12043/model.neff b/neuronxcc-2.15.143.0+e39249ad/MODULE_42ca10d6c86a3705a9b1+39f12043/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..9f798a7e548d95284c654db1891e52185ba86fde
Binary files /dev/null and b/neuronxcc-2.15.143.0+e39249ad/MODULE_42ca10d6c86a3705a9b1+39f12043/model.neff differ
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_90d99a26fb5f52ef45ea+39f12043/compile_flags.json b/neuronxcc-2.15.143.0+e39249ad/MODULE_90d99a26fb5f52ef45ea+39f12043/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..146361c3d3be9576d184ec08a57699cb27736489
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/MODULE_90d99a26fb5f52ef45ea+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_90d99a26fb5f52ef45ea+39f12043/model.done b/neuronxcc-2.15.143.0+e39249ad/MODULE_90d99a26fb5f52ef45ea+39f12043/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_90d99a26fb5f52ef45ea+39f12043/model.hlo_module.pb b/neuronxcc-2.15.143.0+e39249ad/MODULE_90d99a26fb5f52ef45ea+39f12043/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..2dd77bd920bde029bff3361b7e5a18bd64436ae7
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/MODULE_90d99a26fb5f52ef45ea+39f12043/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c9d414397a377eaecbc5659cc7ffbb9be6c8c1f6b0ab971d3c1eff5b4f358034
+size 20632
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_90d99a26fb5f52ef45ea+39f12043/model.neff b/neuronxcc-2.15.143.0+e39249ad/MODULE_90d99a26fb5f52ef45ea+39f12043/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..f9a7b50f6158c43a5c321805a2c0354f3d5d91e4
Binary files /dev/null and b/neuronxcc-2.15.143.0+e39249ad/MODULE_90d99a26fb5f52ef45ea+39f12043/model.neff differ
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_b4c874058a6bc07fb9cc+39f12043/compile_flags.json b/neuronxcc-2.15.143.0+e39249ad/MODULE_b4c874058a6bc07fb9cc+39f12043/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..146361c3d3be9576d184ec08a57699cb27736489
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/MODULE_b4c874058a6bc07fb9cc+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_b4c874058a6bc07fb9cc+39f12043/model.done b/neuronxcc-2.15.143.0+e39249ad/MODULE_b4c874058a6bc07fb9cc+39f12043/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_b4c874058a6bc07fb9cc+39f12043/model.hlo_module.pb b/neuronxcc-2.15.143.0+e39249ad/MODULE_b4c874058a6bc07fb9cc+39f12043/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..816c8e1068133f6c248403174df8c27d1b3f567d
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/MODULE_b4c874058a6bc07fb9cc+39f12043/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:580bda056e115966ef731bd7e39aa9dc34c4b0bd62b0226f8e820d7b8c31caff
+size 22182
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_b4c874058a6bc07fb9cc+39f12043/model.neff b/neuronxcc-2.15.143.0+e39249ad/MODULE_b4c874058a6bc07fb9cc+39f12043/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..929b874d6942115a4a16050ea46788f4dfcae00f
Binary files /dev/null and b/neuronxcc-2.15.143.0+e39249ad/MODULE_b4c874058a6bc07fb9cc+39f12043/model.neff differ
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_dd5ec373a4cca12fef42+39f12043/compile_flags.json b/neuronxcc-2.15.143.0+e39249ad/MODULE_dd5ec373a4cca12fef42+39f12043/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..146361c3d3be9576d184ec08a57699cb27736489
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/MODULE_dd5ec373a4cca12fef42+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_dd5ec373a4cca12fef42+39f12043/model.done b/neuronxcc-2.15.143.0+e39249ad/MODULE_dd5ec373a4cca12fef42+39f12043/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_dd5ec373a4cca12fef42+39f12043/model.hlo_module.pb b/neuronxcc-2.15.143.0+e39249ad/MODULE_dd5ec373a4cca12fef42+39f12043/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..9418667ef875fe54b8f656845634850884637aee
--- /dev/null
+++ b/neuronxcc-2.15.143.0+e39249ad/MODULE_dd5ec373a4cca12fef42+39f12043/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:801fa7ea731610acf3198d1e5d029129bd2e97b0650aba8d4635bffe245f726a
+size 19309
diff --git a/neuronxcc-2.15.143.0+e39249ad/MODULE_dd5ec373a4cca12fef42+39f12043/model.neff b/neuronxcc-2.15.143.0+e39249ad/MODULE_dd5ec373a4cca12fef42+39f12043/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..6e3b07931f8ee8341b6c05c8c02a269e10d79234
Binary files /dev/null and b/neuronxcc-2.15.143.0+e39249ad/MODULE_dd5ec373a4cca12fef42+39f12043/model.neff differ