diff --git a/neuronxcc-2.14.213.0+013d129b/0_REGISTRY/0.0.24.dev0/inference/bloom/hf-internal-testing/tiny-random-BloomForCausalLM/502c174d959be18e9532.json b/neuronxcc-2.14.213.0+013d129b/0_REGISTRY/0.0.24.dev0/inference/bloom/hf-internal-testing/tiny-random-BloomForCausalLM/502c174d959be18e9532.json
new file mode 100644
index 0000000000000000000000000000000000000000..14b1677ac695214209351982dda7497b926402a8
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/0_REGISTRY/0.0.24.dev0/inference/bloom/hf-internal-testing/tiny-random-BloomForCausalLM/502c174d959be18e9532.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": "2.14.213.0+013d129b", "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-2.14.213.0+013d129b/0_REGISTRY/0.0.24.dev0/inference/bloom/hf-internal-testing/tiny-random-BloomForCausalLM/de8fa96c7bfc636c1862.json b/neuronxcc-2.14.213.0+013d129b/0_REGISTRY/0.0.24.dev0/inference/bloom/hf-internal-testing/tiny-random-BloomForCausalLM/de8fa96c7bfc636c1862.json
new file mode 100644
index 0000000000000000000000000000000000000000..45b8738682bac1b47d56eeacda7f40d33dcc3ff1
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/0_REGISTRY/0.0.24.dev0/inference/bloom/hf-internal-testing/tiny-random-BloomForCausalLM/de8fa96c7bfc636c1862.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": "2.14.213.0+013d129b", "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-2.14.213.0+013d129b/0_REGISTRY/0.0.24.dev0/inference/bloom/hf-internal-testing/tiny-random-BloomForCausalLM/f8993574c5ad6dd46126.json b/neuronxcc-2.14.213.0+013d129b/0_REGISTRY/0.0.24.dev0/inference/bloom/hf-internal-testing/tiny-random-BloomForCausalLM/f8993574c5ad6dd46126.json
new file mode 100644
index 0000000000000000000000000000000000000000..a66ab8215df505d60cbaec34ac269da71aa61d83
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/0_REGISTRY/0.0.24.dev0/inference/bloom/hf-internal-testing/tiny-random-BloomForCausalLM/f8993574c5ad6dd46126.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": "2.14.213.0+013d129b", "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-2.14.213.0+013d129b/0_REGISTRY/0.0.24.dev0/inference/gpt2/gpt2/2f32aee440dc0531b512.json b/neuronxcc-2.14.213.0+013d129b/0_REGISTRY/0.0.24.dev0/inference/gpt2/gpt2/2f32aee440dc0531b512.json
new file mode 100644
index 0000000000000000000000000000000000000000..bcd13e210b4885582156fbcc89a8dcb0033428d3
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/0_REGISTRY/0.0.24.dev0/inference/gpt2/gpt2/2f32aee440dc0531b512.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": "2.14.213.0+013d129b", "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-2.14.213.0+013d129b/0_REGISTRY/0.0.24.dev0/inference/gpt2/hf-internal-testing/tiny-random-gpt2/27b58df3bfc4df73d037.json b/neuronxcc-2.14.213.0+013d129b/0_REGISTRY/0.0.24.dev0/inference/gpt2/hf-internal-testing/tiny-random-gpt2/27b58df3bfc4df73d037.json
new file mode 100644
index 0000000000000000000000000000000000000000..807c19b7e8eb7aa863a7fa9cb57cc246fa4d5c51
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/0_REGISTRY/0.0.24.dev0/inference/gpt2/hf-internal-testing/tiny-random-gpt2/27b58df3bfc4df73d037.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": "2.14.213.0+013d129b", "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-2.14.213.0+013d129b/0_REGISTRY/0.0.24.dev0/inference/gpt2/hf-internal-testing/tiny-random-gpt2/89abea0c2d8d0238662e.json b/neuronxcc-2.14.213.0+013d129b/0_REGISTRY/0.0.24.dev0/inference/gpt2/hf-internal-testing/tiny-random-gpt2/89abea0c2d8d0238662e.json
new file mode 100644
index 0000000000000000000000000000000000000000..6e0a644027f3f51441879ba4b38809a33a846d21
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/0_REGISTRY/0.0.24.dev0/inference/gpt2/hf-internal-testing/tiny-random-gpt2/89abea0c2d8d0238662e.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": "2.14.213.0+013d129b", "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-2.14.213.0+013d129b/0_REGISTRY/0.0.24.dev0/inference/gpt2/hf-internal-testing/tiny-random-gpt2/acae9d99b368f7e213f8.json b/neuronxcc-2.14.213.0+013d129b/0_REGISTRY/0.0.24.dev0/inference/gpt2/hf-internal-testing/tiny-random-gpt2/acae9d99b368f7e213f8.json
new file mode 100644
index 0000000000000000000000000000000000000000..6706bc62d72569d38c38bfbf04ffe75664bd21fa
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/0_REGISTRY/0.0.24.dev0/inference/gpt2/hf-internal-testing/tiny-random-gpt2/acae9d99b368f7e213f8.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": "2.14.213.0+013d129b", "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-2.14.213.0+013d129b/0_REGISTRY/0.0.24.dev0/inference/llama/dacorvo/tiny-random-llama/7d494ae543777c6ceced.json b/neuronxcc-2.14.213.0+013d129b/0_REGISTRY/0.0.24.dev0/inference/llama/dacorvo/tiny-random-llama/7d494ae543777c6ceced.json
new file mode 100644
index 0000000000000000000000000000000000000000..07b77eedae7a26693bab8bc5cc4fe683c3521e6b
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/0_REGISTRY/0.0.24.dev0/inference/llama/dacorvo/tiny-random-llama/7d494ae543777c6ceced.json
@@ -0,0 +1 @@
+{"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 1, "eos_token_id": 2, "hidden_act": "silu", "hidden_size": 128, "initializer_range": 0.02, "intermediate_size": 256, "max_position_embeddings": 512, "mlp_bias": false, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 2, "checkpoint_id": "dacorvo/tiny-random-llama", "checkpoint_revision": "7fdafd2fe6a2d31c6abb72ae60db606d8bb23196", "compiler_type": "neuronx-cc", "compiler_version": "2.14.213.0+013d129b", "num_cores": 2, "sequence_length": 100, "task": "text-generation"}, "num_attention_heads": 1, "num_hidden_layers": 1, "num_key_value_heads": 1, "pretraining_tp": 1, "rms_norm_eps": 1e-06, "rope_scaling": null, "rope_theta": 10000.0, "tie_word_embeddings": false, "torch_dtype": "float32", "use_cache": true, "vocab_size": 32000}
\ No newline at end of file
diff --git a/neuronxcc-2.14.213.0+013d129b/0_REGISTRY/0.0.24.dev0/inference/llama/dacorvo/tiny-random-llama/ae94dd17cae6a47e6149.json b/neuronxcc-2.14.213.0+013d129b/0_REGISTRY/0.0.24.dev0/inference/llama/dacorvo/tiny-random-llama/ae94dd17cae6a47e6149.json
new file mode 100644
index 0000000000000000000000000000000000000000..1f8180e51c9748850da967013f713657c93a0616
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/0_REGISTRY/0.0.24.dev0/inference/llama/dacorvo/tiny-random-llama/ae94dd17cae6a47e6149.json
@@ -0,0 +1 @@
+{"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 1, "eos_token_id": 2, "hidden_act": "silu", "hidden_size": 128, "initializer_range": 0.02, "intermediate_size": 256, "max_position_embeddings": 512, "mlp_bias": false, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 1, "checkpoint_id": "dacorvo/tiny-random-llama", "checkpoint_revision": "7fdafd2fe6a2d31c6abb72ae60db606d8bb23196", "compiler_type": "neuronx-cc", "compiler_version": "2.14.213.0+013d129b", "num_cores": 2, "sequence_length": 100, "task": "text-generation"}, "num_attention_heads": 1, "num_hidden_layers": 1, "num_key_value_heads": 1, "pretraining_tp": 1, "rms_norm_eps": 1e-06, "rope_scaling": null, "rope_theta": 10000.0, "tie_word_embeddings": false, "torch_dtype": "float32", "use_cache": true, "vocab_size": 32000}
\ No newline at end of file
diff --git a/neuronxcc-2.14.213.0+013d129b/0_REGISTRY/0.0.24.dev0/inference/llama/dacorvo/tiny-random-llama/e1b2ea90dc6ea1de8bfd.json b/neuronxcc-2.14.213.0+013d129b/0_REGISTRY/0.0.24.dev0/inference/llama/dacorvo/tiny-random-llama/e1b2ea90dc6ea1de8bfd.json
new file mode 100644
index 0000000000000000000000000000000000000000..24566751112563f8010b1e16b37b26e5d0822945
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/0_REGISTRY/0.0.24.dev0/inference/llama/dacorvo/tiny-random-llama/e1b2ea90dc6ea1de8bfd.json
@@ -0,0 +1 @@
+{"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 1, "eos_token_id": 2, "hidden_act": "silu", "hidden_size": 128, "initializer_range": 0.02, "intermediate_size": 256, "max_position_embeddings": 512, "mlp_bias": false, "model_type": "llama", "neuron": {"auto_cast_type": "fp32", "batch_size": 1, "checkpoint_id": "dacorvo/tiny-random-llama", "checkpoint_revision": "7fdafd2fe6a2d31c6abb72ae60db606d8bb23196", "compiler_type": "neuronx-cc", "compiler_version": "2.14.213.0+013d129b", "num_cores": 2, "sequence_length": 100, "task": "text-generation"}, "num_attention_heads": 1, "num_hidden_layers": 1, "num_key_value_heads": 1, "pretraining_tp": 1, "rms_norm_eps": 1e-06, "rope_scaling": null, "rope_theta": 10000.0, "tie_word_embeddings": false, "torch_dtype": "float32", "use_cache": true, "vocab_size": 32000}
\ No newline at end of file
diff --git a/neuronxcc-2.14.213.0+013d129b/0_REGISTRY/0.0.24.dev0/inference/mistral/dacorvo/tiny-random-MistralForCausalLM/030abc51a5f2f76b79e7.json b/neuronxcc-2.14.213.0+013d129b/0_REGISTRY/0.0.24.dev0/inference/mistral/dacorvo/tiny-random-MistralForCausalLM/030abc51a5f2f76b79e7.json
new file mode 100644
index 0000000000000000000000000000000000000000..9f56417c7913f68b2cdd0640947572e1f2b99dc3
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/0_REGISTRY/0.0.24.dev0/inference/mistral/dacorvo/tiny-random-MistralForCausalLM/030abc51a5f2f76b79e7.json
@@ -0,0 +1 @@
+{"architectures": ["MistralForCausalLM"], "attention_dropout": 0.0, "attention_probs_dropout_prob": 0.1, "bos_token_id": 1, "eos_token_id": 2, "hidden_act": "gelu", "hidden_dropout_prob": 0.1, "hidden_size": 32, "initializer_range": 0.02, "intermediate_size": 37, "is_decoder": true, "max_position_embeddings": 512, "model_type": "mistral", "neuron": {"auto_cast_type": "fp32", "batch_size": 1, "checkpoint_id": "dacorvo/tiny-random-MistralForCausalLM", "checkpoint_revision": "81d453e3c8985649e9ee3d4c9378461029d1c73a", "compiler_type": "neuronx-cc", "compiler_version": "2.14.213.0+013d129b", "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-2.14.213.0+013d129b/0_REGISTRY/0.0.24.dev0/inference/mistral/dacorvo/tiny-random-MistralForCausalLM/a8de305007c583a8655b.json b/neuronxcc-2.14.213.0+013d129b/0_REGISTRY/0.0.24.dev0/inference/mistral/dacorvo/tiny-random-MistralForCausalLM/a8de305007c583a8655b.json
new file mode 100644
index 0000000000000000000000000000000000000000..d05d0480e3b1057ada1a520cab39463ddfd10302
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/0_REGISTRY/0.0.24.dev0/inference/mistral/dacorvo/tiny-random-MistralForCausalLM/a8de305007c583a8655b.json
@@ -0,0 +1 @@
+{"architectures": ["MistralForCausalLM"], "attention_dropout": 0.0, "attention_probs_dropout_prob": 0.1, "bos_token_id": 1, "eos_token_id": 2, "hidden_act": "gelu", "hidden_dropout_prob": 0.1, "hidden_size": 32, "initializer_range": 0.02, "intermediate_size": 37, "is_decoder": true, "max_position_embeddings": 512, "model_type": "mistral", "neuron": {"auto_cast_type": "fp16", "batch_size": 2, "checkpoint_id": "dacorvo/tiny-random-MistralForCausalLM", "checkpoint_revision": "81d453e3c8985649e9ee3d4c9378461029d1c73a", "compiler_type": "neuronx-cc", "compiler_version": "2.14.213.0+013d129b", "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-2.14.213.0+013d129b/0_REGISTRY/0.0.24.dev0/inference/mistral/dacorvo/tiny-random-MistralForCausalLM/b5387a4261fa49a90350.json b/neuronxcc-2.14.213.0+013d129b/0_REGISTRY/0.0.24.dev0/inference/mistral/dacorvo/tiny-random-MistralForCausalLM/b5387a4261fa49a90350.json
new file mode 100644
index 0000000000000000000000000000000000000000..1380fa23431b6cb272938d2276732b0a09888d7e
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/0_REGISTRY/0.0.24.dev0/inference/mistral/dacorvo/tiny-random-MistralForCausalLM/b5387a4261fa49a90350.json
@@ -0,0 +1 @@
+{"architectures": ["MistralForCausalLM"], "attention_dropout": 0.0, "attention_probs_dropout_prob": 0.1, "bos_token_id": 1, "eos_token_id": 2, "hidden_act": "gelu", "hidden_dropout_prob": 0.1, "hidden_size": 32, "initializer_range": 0.02, "intermediate_size": 37, "is_decoder": true, "max_position_embeddings": 512, "model_type": "mistral", "neuron": {"auto_cast_type": "fp16", "batch_size": 1, "checkpoint_id": "dacorvo/tiny-random-MistralForCausalLM", "checkpoint_revision": "81d453e3c8985649e9ee3d4c9378461029d1c73a", "compiler_type": "neuronx-cc", "compiler_version": "2.14.213.0+013d129b", "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-2.14.213.0+013d129b/0_REGISTRY/0.0.24.dev0/inference/mixtral/dacorvo/Mixtral-tiny/0d04f4d4b987d19e2758.json b/neuronxcc-2.14.213.0+013d129b/0_REGISTRY/0.0.24.dev0/inference/mixtral/dacorvo/Mixtral-tiny/0d04f4d4b987d19e2758.json
new file mode 100644
index 0000000000000000000000000000000000000000..2f032b7acfd394ab98c5169520bc9e2e943c1e7a
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/0_REGISTRY/0.0.24.dev0/inference/mixtral/dacorvo/Mixtral-tiny/0d04f4d4b987d19e2758.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": "2.14.213.0+013d129b", "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-2.14.213.0+013d129b/0_REGISTRY/0.0.24.dev0/inference/mixtral/dacorvo/Mixtral-tiny/81569bedad5c44215faa.json b/neuronxcc-2.14.213.0+013d129b/0_REGISTRY/0.0.24.dev0/inference/mixtral/dacorvo/Mixtral-tiny/81569bedad5c44215faa.json
new file mode 100644
index 0000000000000000000000000000000000000000..f55915c26788c48b56881df8452019e63549d2bf
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/0_REGISTRY/0.0.24.dev0/inference/mixtral/dacorvo/Mixtral-tiny/81569bedad5c44215faa.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": "2.14.213.0+013d129b", "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-2.14.213.0+013d129b/0_REGISTRY/0.0.24.dev0/inference/mixtral/dacorvo/Mixtral-tiny/8739d715a7c9acaf4c0b.json b/neuronxcc-2.14.213.0+013d129b/0_REGISTRY/0.0.24.dev0/inference/mixtral/dacorvo/Mixtral-tiny/8739d715a7c9acaf4c0b.json
new file mode 100644
index 0000000000000000000000000000000000000000..f638e5cc49117e6be28848a8ce9c63d43fb86a5a
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/0_REGISTRY/0.0.24.dev0/inference/mixtral/dacorvo/Mixtral-tiny/8739d715a7c9acaf4c0b.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": "2.14.213.0+013d129b", "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-2.14.213.0+013d129b/0_REGISTRY/0.0.24.dev0/inference/opt/hf-internal-testing/tiny-random-OPTForCausalLM/6823d660d178eb94eca0.json b/neuronxcc-2.14.213.0+013d129b/0_REGISTRY/0.0.24.dev0/inference/opt/hf-internal-testing/tiny-random-OPTForCausalLM/6823d660d178eb94eca0.json
new file mode 100644
index 0000000000000000000000000000000000000000..cd42105a9ee06a71c5f65612dda9ea24e9a53233
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/0_REGISTRY/0.0.24.dev0/inference/opt/hf-internal-testing/tiny-random-OPTForCausalLM/6823d660d178eb94eca0.json
@@ -0,0 +1 @@
+{"_remove_final_layer_norm": false, "activation_function": "relu", "architectures": ["OPTForCausalLM"], "attention_dropout": 0.1, "bos_token_id": 2, "do_layer_norm_before": true, "dropout": 0.1, "embed_dim": 16, "enable_bias": true, "eos_token_id": 2, "ffn_dim": 4, "hidden_size": 16, "init_std": 0.02, "is_decoder": true, "layer_norm_elementwise_affine": true, "layerdrop": 0.0, "max_position_embeddings": 100, "model_type": "opt", "neuron": {"auto_cast_type": "fp16", "batch_size": 2, "checkpoint_id": "hf-internal-testing/tiny-random-OPTForCausalLM", "checkpoint_revision": "190d1f4fc0011d2eaeaa05282e0fbd2445e4b11f", "compiler_type": "neuronx-cc", "compiler_version": "2.14.213.0+013d129b", "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-2.14.213.0+013d129b/0_REGISTRY/0.0.24.dev0/inference/opt/hf-internal-testing/tiny-random-OPTForCausalLM/bb6453044adc53f42f22.json b/neuronxcc-2.14.213.0+013d129b/0_REGISTRY/0.0.24.dev0/inference/opt/hf-internal-testing/tiny-random-OPTForCausalLM/bb6453044adc53f42f22.json
new file mode 100644
index 0000000000000000000000000000000000000000..c137802831305388f5878bcbbc9360344ed9c3eb
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/0_REGISTRY/0.0.24.dev0/inference/opt/hf-internal-testing/tiny-random-OPTForCausalLM/bb6453044adc53f42f22.json
@@ -0,0 +1 @@
+{"_remove_final_layer_norm": false, "activation_function": "relu", "architectures": ["OPTForCausalLM"], "attention_dropout": 0.1, "bos_token_id": 2, "do_layer_norm_before": true, "dropout": 0.1, "embed_dim": 16, "enable_bias": true, "eos_token_id": 2, "ffn_dim": 4, "hidden_size": 16, "init_std": 0.02, "is_decoder": true, "layer_norm_elementwise_affine": true, "layerdrop": 0.0, "max_position_embeddings": 100, "model_type": "opt", "neuron": {"auto_cast_type": "fp32", "batch_size": 1, "checkpoint_id": "hf-internal-testing/tiny-random-OPTForCausalLM", "checkpoint_revision": "190d1f4fc0011d2eaeaa05282e0fbd2445e4b11f", "compiler_type": "neuronx-cc", "compiler_version": "2.14.213.0+013d129b", "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-2.14.213.0+013d129b/0_REGISTRY/0.0.24.dev0/inference/opt/hf-internal-testing/tiny-random-OPTForCausalLM/e48297c242b96882b93b.json b/neuronxcc-2.14.213.0+013d129b/0_REGISTRY/0.0.24.dev0/inference/opt/hf-internal-testing/tiny-random-OPTForCausalLM/e48297c242b96882b93b.json
new file mode 100644
index 0000000000000000000000000000000000000000..4085831bbd0f03c199a96bbe4a131b05c8cde60a
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/0_REGISTRY/0.0.24.dev0/inference/opt/hf-internal-testing/tiny-random-OPTForCausalLM/e48297c242b96882b93b.json
@@ -0,0 +1 @@
+{"_remove_final_layer_norm": false, "activation_function": "relu", "architectures": ["OPTForCausalLM"], "attention_dropout": 0.1, "bos_token_id": 2, "do_layer_norm_before": true, "dropout": 0.1, "embed_dim": 16, "enable_bias": true, "eos_token_id": 2, "ffn_dim": 4, "hidden_size": 16, "init_std": 0.02, "is_decoder": true, "layer_norm_elementwise_affine": true, "layerdrop": 0.0, "max_position_embeddings": 100, "model_type": "opt", "neuron": {"auto_cast_type": "fp16", "batch_size": 1, "checkpoint_id": "hf-internal-testing/tiny-random-OPTForCausalLM", "checkpoint_revision": "190d1f4fc0011d2eaeaa05282e0fbd2445e4b11f", "compiler_type": "neuronx-cc", "compiler_version": "2.14.213.0+013d129b", "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-2.14.213.0+013d129b/MODULE_0b9060298e5901893068+2c2d707e/compile_flags.txt b/neuronxcc-2.14.213.0+013d129b/MODULE_0b9060298e5901893068+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_0b9060298e5901893068+2c2d707e/compile_flags.txt
@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none
\ No newline at end of file
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_0b9060298e5901893068+2c2d707e/model.hlo_module.pb b/neuronxcc-2.14.213.0+013d129b/MODULE_0b9060298e5901893068+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..9a48667d7e67d27acc47aced5a2bebf9dd197d64
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_0b9060298e5901893068+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c7d78190172a5596e4c18b735e6f3b730c56c05c9c05685feb3bb3da6de729c7
+size 64090
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_0b9060298e5901893068+2c2d707e/model.neff b/neuronxcc-2.14.213.0+013d129b/MODULE_0b9060298e5901893068+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..b40b45b00c228951acb60a9e91508f57d247b69f
Binary files /dev/null and b/neuronxcc-2.14.213.0+013d129b/MODULE_0b9060298e5901893068+2c2d707e/model.neff differ
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_1ddfe7bb03e6b5141bca+2c2d707e/compile_flags.txt b/neuronxcc-2.14.213.0+013d129b/MODULE_1ddfe7bb03e6b5141bca+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_1ddfe7bb03e6b5141bca+2c2d707e/compile_flags.txt
@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none
\ No newline at end of file
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_1ddfe7bb03e6b5141bca+2c2d707e/model.hlo_module.pb b/neuronxcc-2.14.213.0+013d129b/MODULE_1ddfe7bb03e6b5141bca+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..b9cdb1f0045ed11fd87fae80ba6ee9f6b26314da
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_1ddfe7bb03e6b5141bca+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0beb3bc81714d1f70e77131f347a87578e245d93457b5aa875e18761c434e459
+size 1900
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_1ddfe7bb03e6b5141bca+2c2d707e/model.neff b/neuronxcc-2.14.213.0+013d129b/MODULE_1ddfe7bb03e6b5141bca+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..f71f5ca76c4cc617dfeda26ed668bc0c5d306df8
Binary files /dev/null and b/neuronxcc-2.14.213.0+013d129b/MODULE_1ddfe7bb03e6b5141bca+2c2d707e/model.neff differ
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_1f3d02b2530b83c4abf7+2c2d707e/compile_flags.txt b/neuronxcc-2.14.213.0+013d129b/MODULE_1f3d02b2530b83c4abf7+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_1f3d02b2530b83c4abf7+2c2d707e/compile_flags.txt
@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none
\ No newline at end of file
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_1f3d02b2530b83c4abf7+2c2d707e/model.hlo_module.pb b/neuronxcc-2.14.213.0+013d129b/MODULE_1f3d02b2530b83c4abf7+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..73388db93515b2d61ed100e6221869e02df0ff66
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_1f3d02b2530b83c4abf7+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c016cb9348980e5ceadfeb94168eab02c144bb5693c61751c2d3367d09ef5b5f
+size 57433
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_1f3d02b2530b83c4abf7+2c2d707e/model.neff b/neuronxcc-2.14.213.0+013d129b/MODULE_1f3d02b2530b83c4abf7+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..5b288d876a3398db95457b311ee3350e04f3cb1e
Binary files /dev/null and b/neuronxcc-2.14.213.0+013d129b/MODULE_1f3d02b2530b83c4abf7+2c2d707e/model.neff differ
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_23ced3c113e97afdb062+2c2d707e/compile_flags.txt b/neuronxcc-2.14.213.0+013d129b/MODULE_23ced3c113e97afdb062+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_23ced3c113e97afdb062+2c2d707e/compile_flags.txt
@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none
\ No newline at end of file
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_23ced3c113e97afdb062+2c2d707e/model.hlo_module.pb b/neuronxcc-2.14.213.0+013d129b/MODULE_23ced3c113e97afdb062+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..011a26c008928c9bab374987e5d96931b7d0fc46
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_23ced3c113e97afdb062+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:76ee8fcd669c0344010019e40d232b2beb49ae91140f62839d3464f9d6c4bd3a
+size 60871
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_23ced3c113e97afdb062+2c2d707e/model.neff b/neuronxcc-2.14.213.0+013d129b/MODULE_23ced3c113e97afdb062+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..2e81b009ce2b4ea1256d5a64cbacf4c145bbfc38
Binary files /dev/null and b/neuronxcc-2.14.213.0+013d129b/MODULE_23ced3c113e97afdb062+2c2d707e/model.neff differ
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_257d339d60d0fce6e129+2c2d707e/compile_flags.txt b/neuronxcc-2.14.213.0+013d129b/MODULE_257d339d60d0fce6e129+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_257d339d60d0fce6e129+2c2d707e/compile_flags.txt
@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none
\ No newline at end of file
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_257d339d60d0fce6e129+2c2d707e/model.hlo_module.pb b/neuronxcc-2.14.213.0+013d129b/MODULE_257d339d60d0fce6e129+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..ceb6eda844e17e952ff6a301aea3adc8ab4abdb3
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_257d339d60d0fce6e129+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:15567e4d8acb5d4d60c377ea8defce5048d45c48f339a6185772a293a8afa793
+size 49801
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_257d339d60d0fce6e129+2c2d707e/model.neff b/neuronxcc-2.14.213.0+013d129b/MODULE_257d339d60d0fce6e129+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..ce54fd02526c1e98375038aeb968504618471aae
Binary files /dev/null and b/neuronxcc-2.14.213.0+013d129b/MODULE_257d339d60d0fce6e129+2c2d707e/model.neff differ
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_2aa937204f63179c0db3+2c2d707e/compile_flags.txt b/neuronxcc-2.14.213.0+013d129b/MODULE_2aa937204f63179c0db3+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_2aa937204f63179c0db3+2c2d707e/compile_flags.txt
@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none
\ No newline at end of file
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_2aa937204f63179c0db3+2c2d707e/model.hlo_module.pb b/neuronxcc-2.14.213.0+013d129b/MODULE_2aa937204f63179c0db3+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..6c9dfa48abaf3f3738951588d04b87cf1b87e4b0
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_2aa937204f63179c0db3+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d86873653e3f68563af1f244d97fd91d90568d1dfc5ede3cdceb914bb08460a7
+size 57064
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_2aa937204f63179c0db3+2c2d707e/model.neff b/neuronxcc-2.14.213.0+013d129b/MODULE_2aa937204f63179c0db3+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..94d6abee7d1adb399940f9b8e509f5913c3a7a98
Binary files /dev/null and b/neuronxcc-2.14.213.0+013d129b/MODULE_2aa937204f63179c0db3+2c2d707e/model.neff differ
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_34480d97464f3bf86d90+2c2d707e/compile_flags.txt b/neuronxcc-2.14.213.0+013d129b/MODULE_34480d97464f3bf86d90+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_34480d97464f3bf86d90+2c2d707e/compile_flags.txt
@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none
\ No newline at end of file
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_34480d97464f3bf86d90+2c2d707e/model.hlo_module.pb b/neuronxcc-2.14.213.0+013d129b/MODULE_34480d97464f3bf86d90+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..fc62b327caf46463e40ebd4c2e32a5ce862bc713
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_34480d97464f3bf86d90+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:634c3398b71dcfaabba007cabe8d9a72a6020285edef8ab2bd1cde5422a1bf5f
+size 46625
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_34480d97464f3bf86d90+2c2d707e/model.neff b/neuronxcc-2.14.213.0+013d129b/MODULE_34480d97464f3bf86d90+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..30bd13b6de1044b74927f740782a4cd4c0b81211
Binary files /dev/null and b/neuronxcc-2.14.213.0+013d129b/MODULE_34480d97464f3bf86d90+2c2d707e/model.neff differ
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_46948ddb997e0e72bed0+2c2d707e/compile_flags.txt b/neuronxcc-2.14.213.0+013d129b/MODULE_46948ddb997e0e72bed0+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_46948ddb997e0e72bed0+2c2d707e/compile_flags.txt
@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none
\ No newline at end of file
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_46948ddb997e0e72bed0+2c2d707e/model.hlo_module.pb b/neuronxcc-2.14.213.0+013d129b/MODULE_46948ddb997e0e72bed0+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..933e5b9078476eb0ca1240a1c380b284c3baf527
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_46948ddb997e0e72bed0+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c6b93438bd3b5a59ce97c1ae510ec134652a81d958b6ca06f2122467a6569b1f
+size 1900
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_46948ddb997e0e72bed0+2c2d707e/model.neff b/neuronxcc-2.14.213.0+013d129b/MODULE_46948ddb997e0e72bed0+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..1800b86cb3af0911b33b4fc80aeb02e32e93bfd8
Binary files /dev/null and b/neuronxcc-2.14.213.0+013d129b/MODULE_46948ddb997e0e72bed0+2c2d707e/model.neff differ
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_4a486888f48de6666282+2c2d707e/compile_flags.txt b/neuronxcc-2.14.213.0+013d129b/MODULE_4a486888f48de6666282+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_4a486888f48de6666282+2c2d707e/compile_flags.txt
@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none
\ No newline at end of file
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_4a486888f48de6666282+2c2d707e/model.hlo_module.pb b/neuronxcc-2.14.213.0+013d129b/MODULE_4a486888f48de6666282+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..163e31246e15436044ee43ea753a73da774508cf
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_4a486888f48de6666282+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:401229f357b8f66524080f98dd7eb6a847f069c7ac6908aa712968abc981b181
+size 11805
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_4a486888f48de6666282+2c2d707e/model.neff b/neuronxcc-2.14.213.0+013d129b/MODULE_4a486888f48de6666282+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..9abf18eaf3754403e27530a1b199217bf1a63ebb
Binary files /dev/null and b/neuronxcc-2.14.213.0+013d129b/MODULE_4a486888f48de6666282+2c2d707e/model.neff differ
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_4fd66239b44449a4f2f7+2c2d707e/compile_flags.txt b/neuronxcc-2.14.213.0+013d129b/MODULE_4fd66239b44449a4f2f7+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_4fd66239b44449a4f2f7+2c2d707e/compile_flags.txt
@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none
\ No newline at end of file
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_4fd66239b44449a4f2f7+2c2d707e/model.hlo_module.pb b/neuronxcc-2.14.213.0+013d129b/MODULE_4fd66239b44449a4f2f7+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..fc26b477e2396f0901d32bd26a217e5b5a478ac7
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_4fd66239b44449a4f2f7+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:befe42fedd8b9425909c5facb895e1105a46e1910c668329289cc3d4a17ee184
+size 11430
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_4fd66239b44449a4f2f7+2c2d707e/model.neff b/neuronxcc-2.14.213.0+013d129b/MODULE_4fd66239b44449a4f2f7+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..c8bfa5615dfb0f5140b2bbb05a8e958cb6e990da
Binary files /dev/null and b/neuronxcc-2.14.213.0+013d129b/MODULE_4fd66239b44449a4f2f7+2c2d707e/model.neff differ
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_667f1d540de4ddfad7c8+2c2d707e/compile_flags.txt b/neuronxcc-2.14.213.0+013d129b/MODULE_667f1d540de4ddfad7c8+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_667f1d540de4ddfad7c8+2c2d707e/compile_flags.txt
@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none
\ No newline at end of file
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_667f1d540de4ddfad7c8+2c2d707e/model.hlo_module.pb b/neuronxcc-2.14.213.0+013d129b/MODULE_667f1d540de4ddfad7c8+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..7215233c65ee8ef2dd00b18e162c5f8882d03d9f
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_667f1d540de4ddfad7c8+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b792380611b31e01aad73c5237fc73e36813c0a5fdb22b8d084a1f95a0689638
+size 56823
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_667f1d540de4ddfad7c8+2c2d707e/model.neff b/neuronxcc-2.14.213.0+013d129b/MODULE_667f1d540de4ddfad7c8+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..5614b027a012da35a51dc9142d3a57afea7ff84e
Binary files /dev/null and b/neuronxcc-2.14.213.0+013d129b/MODULE_667f1d540de4ddfad7c8+2c2d707e/model.neff differ
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_762891d6ec9d87fe667d+2c2d707e/compile_flags.txt b/neuronxcc-2.14.213.0+013d129b/MODULE_762891d6ec9d87fe667d+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_762891d6ec9d87fe667d+2c2d707e/compile_flags.txt
@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none
\ No newline at end of file
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_762891d6ec9d87fe667d+2c2d707e/model.hlo_module.pb b/neuronxcc-2.14.213.0+013d129b/MODULE_762891d6ec9d87fe667d+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..53da9a99784d38de534922ec4357c658d33769dc
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_762891d6ec9d87fe667d+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d94b968c5c8dff6a4638aa68e36fd1fa431e592d5e4372717b7aeb0e7a11b7c8
+size 49571
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_762891d6ec9d87fe667d+2c2d707e/model.neff b/neuronxcc-2.14.213.0+013d129b/MODULE_762891d6ec9d87fe667d+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..c33465d2bc32dba112575c8329cd396a4bb20c51
Binary files /dev/null and b/neuronxcc-2.14.213.0+013d129b/MODULE_762891d6ec9d87fe667d+2c2d707e/model.neff differ
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_85c78a8ee4342175b5f7+2c2d707e/compile_flags.txt b/neuronxcc-2.14.213.0+013d129b/MODULE_85c78a8ee4342175b5f7+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_85c78a8ee4342175b5f7+2c2d707e/compile_flags.txt
@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none
\ No newline at end of file
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_85c78a8ee4342175b5f7+2c2d707e/model.hlo_module.pb b/neuronxcc-2.14.213.0+013d129b/MODULE_85c78a8ee4342175b5f7+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..0867080f1a9bcc2d9fd479b09227899318459c93
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_85c78a8ee4342175b5f7+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4ad04ceffc006a3115f299fecf8b5793dc4627b8b4cfba00c6fd9542da6c18dd
+size 1900
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_85c78a8ee4342175b5f7+2c2d707e/model.neff b/neuronxcc-2.14.213.0+013d129b/MODULE_85c78a8ee4342175b5f7+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..41c38e621b6c2d102568ebb9a405aca6111a328f
Binary files /dev/null and b/neuronxcc-2.14.213.0+013d129b/MODULE_85c78a8ee4342175b5f7+2c2d707e/model.neff differ
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_873305dba4b4ddafd5a3+2c2d707e/compile_flags.txt b/neuronxcc-2.14.213.0+013d129b/MODULE_873305dba4b4ddafd5a3+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_873305dba4b4ddafd5a3+2c2d707e/compile_flags.txt
@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none
\ No newline at end of file
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_873305dba4b4ddafd5a3+2c2d707e/model.hlo_module.pb b/neuronxcc-2.14.213.0+013d129b/MODULE_873305dba4b4ddafd5a3+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..abf0f871151d8b5cb7524d3a61c7dfdca5d6008c
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_873305dba4b4ddafd5a3+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9113508a9d7f74d5c4e1c1cd0684454022dacc678dc5cd146c6828a3387c66c6
+size 137320
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_873305dba4b4ddafd5a3+2c2d707e/model.neff b/neuronxcc-2.14.213.0+013d129b/MODULE_873305dba4b4ddafd5a3+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..6397f697b961acf5439778eef32ae5a222b30868
Binary files /dev/null and b/neuronxcc-2.14.213.0+013d129b/MODULE_873305dba4b4ddafd5a3+2c2d707e/model.neff differ
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_893f186ba6918788aa7b+2c2d707e/compile_flags.txt b/neuronxcc-2.14.213.0+013d129b/MODULE_893f186ba6918788aa7b+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_893f186ba6918788aa7b+2c2d707e/compile_flags.txt
@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none
\ No newline at end of file
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_893f186ba6918788aa7b+2c2d707e/model.hlo_module.pb b/neuronxcc-2.14.213.0+013d129b/MODULE_893f186ba6918788aa7b+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..367845051dfb489471a3cc85754e5f1109833ede
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_893f186ba6918788aa7b+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:eb188c38126e065b6109e14200f32c00a1594d0532f13ff3c31cba503186efe6
+size 11219
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_893f186ba6918788aa7b+2c2d707e/model.neff b/neuronxcc-2.14.213.0+013d129b/MODULE_893f186ba6918788aa7b+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..0a922dae47ee4be49f3e3731fc752cc44f093c39
Binary files /dev/null and b/neuronxcc-2.14.213.0+013d129b/MODULE_893f186ba6918788aa7b+2c2d707e/model.neff differ
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_8f4979cf58247a14d465+2c2d707e/compile_flags.txt b/neuronxcc-2.14.213.0+013d129b/MODULE_8f4979cf58247a14d465+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_8f4979cf58247a14d465+2c2d707e/compile_flags.txt
@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none
\ No newline at end of file
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_8f4979cf58247a14d465+2c2d707e/model.hlo_module.pb b/neuronxcc-2.14.213.0+013d129b/MODULE_8f4979cf58247a14d465+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..04d80b88d7098cc356079287d91ada0411294b1e
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_8f4979cf58247a14d465+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c94f2b4246b601ce896fc815a1a78a55863638173d2af5e9384715c45c79c1b2
+size 45678
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_8f4979cf58247a14d465+2c2d707e/model.neff b/neuronxcc-2.14.213.0+013d129b/MODULE_8f4979cf58247a14d465+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..8a2a12447692676a16e0931a09b2b026860350e4
Binary files /dev/null and b/neuronxcc-2.14.213.0+013d129b/MODULE_8f4979cf58247a14d465+2c2d707e/model.neff differ
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_8fcbfec50e9a92cc7b98+2c2d707e/compile_flags.txt b/neuronxcc-2.14.213.0+013d129b/MODULE_8fcbfec50e9a92cc7b98+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_8fcbfec50e9a92cc7b98+2c2d707e/compile_flags.txt
@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none
\ No newline at end of file
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_8fcbfec50e9a92cc7b98+2c2d707e/model.hlo_module.pb b/neuronxcc-2.14.213.0+013d129b/MODULE_8fcbfec50e9a92cc7b98+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..d76ae6cc7e278cc6018651595376a8f35f3bc269
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_8fcbfec50e9a92cc7b98+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1d076ad5b375f54c4c7d9a01d50a6b2c401df57cac718472b932b578877559ce
+size 44219
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_8fcbfec50e9a92cc7b98+2c2d707e/model.neff b/neuronxcc-2.14.213.0+013d129b/MODULE_8fcbfec50e9a92cc7b98+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..dc4dcaa5e96560fd7bbf8be7391cdfb42472c9ea
Binary files /dev/null and b/neuronxcc-2.14.213.0+013d129b/MODULE_8fcbfec50e9a92cc7b98+2c2d707e/model.neff differ
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_94b2507e085493db16e1+2c2d707e/compile_flags.txt b/neuronxcc-2.14.213.0+013d129b/MODULE_94b2507e085493db16e1+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_94b2507e085493db16e1+2c2d707e/compile_flags.txt
@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none
\ No newline at end of file
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_94b2507e085493db16e1+2c2d707e/model.hlo_module.pb b/neuronxcc-2.14.213.0+013d129b/MODULE_94b2507e085493db16e1+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..0fb07ba0cb1d0c8307281351f5786147870b4a81
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_94b2507e085493db16e1+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0f6b361321152cb3163a46ac1fcfa724b1720a8b0f655ee6e367d44e5e2de788
+size 10051
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_94b2507e085493db16e1+2c2d707e/model.neff b/neuronxcc-2.14.213.0+013d129b/MODULE_94b2507e085493db16e1+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..a062a080f21935f6502d2b21fa41f95605ae852b
Binary files /dev/null and b/neuronxcc-2.14.213.0+013d129b/MODULE_94b2507e085493db16e1+2c2d707e/model.neff differ
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_a059efcff1f3cab6b3fe+2c2d707e/compile_flags.txt b/neuronxcc-2.14.213.0+013d129b/MODULE_a059efcff1f3cab6b3fe+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_a059efcff1f3cab6b3fe+2c2d707e/compile_flags.txt
@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none
\ No newline at end of file
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_a059efcff1f3cab6b3fe+2c2d707e/model.hlo_module.pb b/neuronxcc-2.14.213.0+013d129b/MODULE_a059efcff1f3cab6b3fe+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..c1e8911a200e98cc4146fbb7ff7ccf619d50316a
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_a059efcff1f3cab6b3fe+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:eb7e8a9531a5e7cf9c0509cd1ad4c39ea22f372768f62deace35d9ec784ac8e9
+size 21546
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_a059efcff1f3cab6b3fe+2c2d707e/model.neff b/neuronxcc-2.14.213.0+013d129b/MODULE_a059efcff1f3cab6b3fe+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..c14afcdcacf44fa9e75ee0bacc96f19c120a7d26
Binary files /dev/null and b/neuronxcc-2.14.213.0+013d129b/MODULE_a059efcff1f3cab6b3fe+2c2d707e/model.neff differ
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_a3ee40f3a8ab254a2daf+2c2d707e/compile_flags.txt b/neuronxcc-2.14.213.0+013d129b/MODULE_a3ee40f3a8ab254a2daf+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_a3ee40f3a8ab254a2daf+2c2d707e/compile_flags.txt
@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none
\ No newline at end of file
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_a3ee40f3a8ab254a2daf+2c2d707e/model.hlo_module.pb b/neuronxcc-2.14.213.0+013d129b/MODULE_a3ee40f3a8ab254a2daf+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..f79c3632624bdc40390cf237389a45fab7845212
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_a3ee40f3a8ab254a2daf+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4b5ae56337d8de58f46726edcea7204f9a095438215fd0474dcfa78e9bd3f705
+size 137320
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_a3ee40f3a8ab254a2daf+2c2d707e/model.neff b/neuronxcc-2.14.213.0+013d129b/MODULE_a3ee40f3a8ab254a2daf+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..a6d3e64757a10b51835534728a52533400c7a76b
Binary files /dev/null and b/neuronxcc-2.14.213.0+013d129b/MODULE_a3ee40f3a8ab254a2daf+2c2d707e/model.neff differ
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_a607a0e4832ed43b0421+2c2d707e/compile_flags.txt b/neuronxcc-2.14.213.0+013d129b/MODULE_a607a0e4832ed43b0421+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_a607a0e4832ed43b0421+2c2d707e/compile_flags.txt
@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none
\ No newline at end of file
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_a607a0e4832ed43b0421+2c2d707e/model.hlo_module.pb b/neuronxcc-2.14.213.0+013d129b/MODULE_a607a0e4832ed43b0421+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..9178bdf0dc925cdd33aa0d3bb38e691bda67380b
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_a607a0e4832ed43b0421+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c04a01222601c1720ec6c3111fd35fa7c565d4e2a9f3029eab92955f2a3cb01d
+size 56984
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_a607a0e4832ed43b0421+2c2d707e/model.neff b/neuronxcc-2.14.213.0+013d129b/MODULE_a607a0e4832ed43b0421+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..7dfd73a32098fa94a81e54a07cba41c3e4b1ffd1
Binary files /dev/null and b/neuronxcc-2.14.213.0+013d129b/MODULE_a607a0e4832ed43b0421+2c2d707e/model.neff differ
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_a6c43b72dffb6d6e9f80+2c2d707e/compile_flags.txt b/neuronxcc-2.14.213.0+013d129b/MODULE_a6c43b72dffb6d6e9f80+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_a6c43b72dffb6d6e9f80+2c2d707e/compile_flags.txt
@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none
\ No newline at end of file
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_a6c43b72dffb6d6e9f80+2c2d707e/model.hlo_module.pb b/neuronxcc-2.14.213.0+013d129b/MODULE_a6c43b72dffb6d6e9f80+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..a35fb53dc32c78fa8b9fb2ab447f4ea4df09b33b
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_a6c43b72dffb6d6e9f80+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3d5f691e3c29b3fe42f7af31687ebb9cd76d83e40db015194b9bf34b2345c857
+size 10089
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_a6c43b72dffb6d6e9f80+2c2d707e/model.neff b/neuronxcc-2.14.213.0+013d129b/MODULE_a6c43b72dffb6d6e9f80+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..a256224f0cc66c8159676a50732736154b95f987
Binary files /dev/null and b/neuronxcc-2.14.213.0+013d129b/MODULE_a6c43b72dffb6d6e9f80+2c2d707e/model.neff differ
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_a77cd506706133f93c7b+2c2d707e/compile_flags.txt b/neuronxcc-2.14.213.0+013d129b/MODULE_a77cd506706133f93c7b+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_a77cd506706133f93c7b+2c2d707e/compile_flags.txt
@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none
\ No newline at end of file
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_a77cd506706133f93c7b+2c2d707e/model.hlo_module.pb b/neuronxcc-2.14.213.0+013d129b/MODULE_a77cd506706133f93c7b+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..37228d09d13a664881f3da6ac978ab6589732b50
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_a77cd506706133f93c7b+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6138fe20b06d8e8555b5e431932c8b52221251fc4defcdec25d77301f1363a0d
+size 46300
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_a77cd506706133f93c7b+2c2d707e/model.neff b/neuronxcc-2.14.213.0+013d129b/MODULE_a77cd506706133f93c7b+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..8bd074fd0e59e7bfa7c30467f2decd7eb17d6ab3
Binary files /dev/null and b/neuronxcc-2.14.213.0+013d129b/MODULE_a77cd506706133f93c7b+2c2d707e/model.neff differ
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_ab36e350fba6fe7154f8+2c2d707e/compile_flags.txt b/neuronxcc-2.14.213.0+013d129b/MODULE_ab36e350fba6fe7154f8+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_ab36e350fba6fe7154f8+2c2d707e/compile_flags.txt
@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none
\ No newline at end of file
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_ab36e350fba6fe7154f8+2c2d707e/model.hlo_module.pb b/neuronxcc-2.14.213.0+013d129b/MODULE_ab36e350fba6fe7154f8+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..5dcaae8fe82c1d19232ba0741cc002730320de48
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_ab36e350fba6fe7154f8+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9bcb6888543e28b850db513ef38618b1c44e913e98b3294e75f5c7945d589ea6
+size 137320
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_ab36e350fba6fe7154f8+2c2d707e/model.neff b/neuronxcc-2.14.213.0+013d129b/MODULE_ab36e350fba6fe7154f8+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..75063a2dd7ad385425651f588b29c95f83744193
Binary files /dev/null and b/neuronxcc-2.14.213.0+013d129b/MODULE_ab36e350fba6fe7154f8+2c2d707e/model.neff differ
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_ae887d3f328332ccc603+2c2d707e/compile_flags.txt b/neuronxcc-2.14.213.0+013d129b/MODULE_ae887d3f328332ccc603+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_ae887d3f328332ccc603+2c2d707e/compile_flags.txt
@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none
\ No newline at end of file
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_ae887d3f328332ccc603+2c2d707e/model.hlo_module.pb b/neuronxcc-2.14.213.0+013d129b/MODULE_ae887d3f328332ccc603+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..5db918da3dc8f246a182985e1cbe7c0e65f351fb
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_ae887d3f328332ccc603+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c37df6693dcfc652696f56dccedeb984dd0e52b27ae40fa8fe6e39b2dd9c0617
+size 23402
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_ae887d3f328332ccc603+2c2d707e/model.neff b/neuronxcc-2.14.213.0+013d129b/MODULE_ae887d3f328332ccc603+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..495bd687c255ffab919bef1b4e4afc1adf6c90ef
Binary files /dev/null and b/neuronxcc-2.14.213.0+013d129b/MODULE_ae887d3f328332ccc603+2c2d707e/model.neff differ
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_afac27ba717c0798c826+2c2d707e/compile_flags.txt b/neuronxcc-2.14.213.0+013d129b/MODULE_afac27ba717c0798c826+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_afac27ba717c0798c826+2c2d707e/compile_flags.txt
@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none
\ No newline at end of file
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_afac27ba717c0798c826+2c2d707e/model.hlo_module.pb b/neuronxcc-2.14.213.0+013d129b/MODULE_afac27ba717c0798c826+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..388d536e0a1e068d4ccc4eead42f4640109cf3ae
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_afac27ba717c0798c826+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a68490357cec88895f6c4dbb708ccf8a328a1b8e59b1b262b34b9301f735ab89
+size 59641
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_afac27ba717c0798c826+2c2d707e/model.neff b/neuronxcc-2.14.213.0+013d129b/MODULE_afac27ba717c0798c826+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..17c7e0e24bbc10833406240f2e8ebbde5faf5108
Binary files /dev/null and b/neuronxcc-2.14.213.0+013d129b/MODULE_afac27ba717c0798c826+2c2d707e/model.neff differ
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_b2f9e0eb00860609779e+2c2d707e/compile_flags.txt b/neuronxcc-2.14.213.0+013d129b/MODULE_b2f9e0eb00860609779e+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_b2f9e0eb00860609779e+2c2d707e/compile_flags.txt
@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none
\ No newline at end of file
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_b2f9e0eb00860609779e+2c2d707e/model.hlo_module.pb b/neuronxcc-2.14.213.0+013d129b/MODULE_b2f9e0eb00860609779e+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..37ec052b640381160175751de7efb21ef1a6152f
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_b2f9e0eb00860609779e+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:26608d57a07bac99faa67e11d540583125cd6237557ddcea73a813b7d71b4ab7
+size 10159
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_b2f9e0eb00860609779e+2c2d707e/model.neff b/neuronxcc-2.14.213.0+013d129b/MODULE_b2f9e0eb00860609779e+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..9f89692bebb31eaf080ee4e8d6ea3fada5644c9a
Binary files /dev/null and b/neuronxcc-2.14.213.0+013d129b/MODULE_b2f9e0eb00860609779e+2c2d707e/model.neff differ
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_b6f6f37b2fd0358f490f+2c2d707e/compile_flags.txt b/neuronxcc-2.14.213.0+013d129b/MODULE_b6f6f37b2fd0358f490f+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_b6f6f37b2fd0358f490f+2c2d707e/compile_flags.txt
@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none
\ No newline at end of file
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_b6f6f37b2fd0358f490f+2c2d707e/model.hlo_module.pb b/neuronxcc-2.14.213.0+013d129b/MODULE_b6f6f37b2fd0358f490f+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..5c88875119154324fae322d19571b790b04f6af2
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_b6f6f37b2fd0358f490f+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e92d2b9f8def0eed41e9f9c048fa5566e88159976da01ffc0236d23044ce4b44
+size 56444
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_b6f6f37b2fd0358f490f+2c2d707e/model.neff b/neuronxcc-2.14.213.0+013d129b/MODULE_b6f6f37b2fd0358f490f+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..ae740d940ecc2de4705c469ae8f06fd89cf7a290
Binary files /dev/null and b/neuronxcc-2.14.213.0+013d129b/MODULE_b6f6f37b2fd0358f490f+2c2d707e/model.neff differ
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_c672fc907b762e3c3650+2c2d707e/compile_flags.txt b/neuronxcc-2.14.213.0+013d129b/MODULE_c672fc907b762e3c3650+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_c672fc907b762e3c3650+2c2d707e/compile_flags.txt
@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none
\ No newline at end of file
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_c672fc907b762e3c3650+2c2d707e/model.hlo_module.pb b/neuronxcc-2.14.213.0+013d129b/MODULE_c672fc907b762e3c3650+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..a3e4a2f1549ea7300474f77c36ec68a672d13dfa
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_c672fc907b762e3c3650+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:391145f49fa12d1076e2c6dd473eb1cb154b780006bca0e35139e67365507d8f
+size 43653
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_c672fc907b762e3c3650+2c2d707e/model.neff b/neuronxcc-2.14.213.0+013d129b/MODULE_c672fc907b762e3c3650+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..8cf69d58e133839320f386f4d523fae799cf21f4
Binary files /dev/null and b/neuronxcc-2.14.213.0+013d129b/MODULE_c672fc907b762e3c3650+2c2d707e/model.neff differ
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_d36dd359e1a3701af2e8+2c2d707e/compile_flags.txt b/neuronxcc-2.14.213.0+013d129b/MODULE_d36dd359e1a3701af2e8+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_d36dd359e1a3701af2e8+2c2d707e/compile_flags.txt
@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none
\ No newline at end of file
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_d36dd359e1a3701af2e8+2c2d707e/model.hlo_module.pb b/neuronxcc-2.14.213.0+013d129b/MODULE_d36dd359e1a3701af2e8+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..20e5291e19a1031524eb537d6a8f3c0e8d42d351
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_d36dd359e1a3701af2e8+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:01762334646745ea64672aebc0cb19c8446134cd684508f20bc3055226ffb7cb
+size 9872
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_d36dd359e1a3701af2e8+2c2d707e/model.neff b/neuronxcc-2.14.213.0+013d129b/MODULE_d36dd359e1a3701af2e8+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..ace28c599c559d1352ca43fadd855fb5998e4e93
Binary files /dev/null and b/neuronxcc-2.14.213.0+013d129b/MODULE_d36dd359e1a3701af2e8+2c2d707e/model.neff differ
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_d4e387899da5e1d0cffb+2c2d707e/compile_flags.txt b/neuronxcc-2.14.213.0+013d129b/MODULE_d4e387899da5e1d0cffb+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_d4e387899da5e1d0cffb+2c2d707e/compile_flags.txt
@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none
\ No newline at end of file
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_d4e387899da5e1d0cffb+2c2d707e/model.hlo_module.pb b/neuronxcc-2.14.213.0+013d129b/MODULE_d4e387899da5e1d0cffb+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..cf5e345f3e7563bc3f1eb559a0748c12d36a1159
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_d4e387899da5e1d0cffb+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a5cc47c8333aa0d5981410b7c6850034df283c6fb6066e13e9d04a2c63120867
+size 57353
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_d4e387899da5e1d0cffb+2c2d707e/model.neff b/neuronxcc-2.14.213.0+013d129b/MODULE_d4e387899da5e1d0cffb+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..1c6ad27a1b6dbb8db871c8750821f5dc2e15075f
Binary files /dev/null and b/neuronxcc-2.14.213.0+013d129b/MODULE_d4e387899da5e1d0cffb+2c2d707e/model.neff differ
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_ddfd7c58057c6453d736+2c2d707e/compile_flags.txt b/neuronxcc-2.14.213.0+013d129b/MODULE_ddfd7c58057c6453d736+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_ddfd7c58057c6453d736+2c2d707e/compile_flags.txt
@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none
\ No newline at end of file
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_ddfd7c58057c6453d736+2c2d707e/model.hlo_module.pb b/neuronxcc-2.14.213.0+013d129b/MODULE_ddfd7c58057c6453d736+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..efdbffc2d406e9be5bb42f8c0ac598241c725271
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_ddfd7c58057c6453d736+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1acbce7e67e72e84aa6e812146904823019f8191a304fdc1d7fd6ea194596252
+size 10130
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_ddfd7c58057c6453d736+2c2d707e/model.neff b/neuronxcc-2.14.213.0+013d129b/MODULE_ddfd7c58057c6453d736+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..166af4ffd887758a4c2994b314b8ab5c7986a163
Binary files /dev/null and b/neuronxcc-2.14.213.0+013d129b/MODULE_ddfd7c58057c6453d736+2c2d707e/model.neff differ
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_dfd016cccf28e4d1a656+2c2d707e/compile_flags.txt b/neuronxcc-2.14.213.0+013d129b/MODULE_dfd016cccf28e4d1a656+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_dfd016cccf28e4d1a656+2c2d707e/compile_flags.txt
@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none
\ No newline at end of file
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_dfd016cccf28e4d1a656+2c2d707e/model.hlo_module.pb b/neuronxcc-2.14.213.0+013d129b/MODULE_dfd016cccf28e4d1a656+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..579a8d99430160ee7c5a3a5268e3b7c846d6eb1e
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_dfd016cccf28e4d1a656+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9965ef44e48e47cd63beff6f3a42b1d3a8174d2caec1b0eeacb268a9e1a033dd
+size 24034
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_dfd016cccf28e4d1a656+2c2d707e/model.neff b/neuronxcc-2.14.213.0+013d129b/MODULE_dfd016cccf28e4d1a656+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..32d30d5b5b2d507edd5793a2a9fb47d4f1b134d4
Binary files /dev/null and b/neuronxcc-2.14.213.0+013d129b/MODULE_dfd016cccf28e4d1a656+2c2d707e/model.neff differ
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_e594326003aca64d3c8e+2c2d707e/compile_flags.txt b/neuronxcc-2.14.213.0+013d129b/MODULE_e594326003aca64d3c8e+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_e594326003aca64d3c8e+2c2d707e/compile_flags.txt
@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none
\ No newline at end of file
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_e594326003aca64d3c8e+2c2d707e/model.hlo_module.pb b/neuronxcc-2.14.213.0+013d129b/MODULE_e594326003aca64d3c8e+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..300225fed6c6c24ce969e2728b34c13ffa82e3b8
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_e594326003aca64d3c8e+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8bf18cff49ee7726efd5ed1eae93985e61d593c3051bec1d6181c8fafb366aa5
+size 49826
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_e594326003aca64d3c8e+2c2d707e/model.neff b/neuronxcc-2.14.213.0+013d129b/MODULE_e594326003aca64d3c8e+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..ee026b3cb30adcd870e50c9da8ca043f4cc3e5d6
Binary files /dev/null and b/neuronxcc-2.14.213.0+013d129b/MODULE_e594326003aca64d3c8e+2c2d707e/model.neff differ
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_e77f571bdc588dc2613f+2c2d707e/compile_flags.txt b/neuronxcc-2.14.213.0+013d129b/MODULE_e77f571bdc588dc2613f+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_e77f571bdc588dc2613f+2c2d707e/compile_flags.txt
@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none
\ No newline at end of file
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_e77f571bdc588dc2613f+2c2d707e/model.hlo_module.pb b/neuronxcc-2.14.213.0+013d129b/MODULE_e77f571bdc588dc2613f+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..c115ecb19ffec7846562ce04b7c47477de08575d
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_e77f571bdc588dc2613f+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:33f0179a416403db0f60617af296ce30b6fd60b4adb9c5334671d01c2a2d2382
+size 22132
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_e77f571bdc588dc2613f+2c2d707e/model.neff b/neuronxcc-2.14.213.0+013d129b/MODULE_e77f571bdc588dc2613f+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..21af802e6a6968ef5d7bba4834fea1a1a5270abe
Binary files /dev/null and b/neuronxcc-2.14.213.0+013d129b/MODULE_e77f571bdc588dc2613f+2c2d707e/model.neff differ
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_f16230777a0107771eef+2c2d707e/compile_flags.txt b/neuronxcc-2.14.213.0+013d129b/MODULE_f16230777a0107771eef+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_f16230777a0107771eef+2c2d707e/compile_flags.txt
@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none
\ No newline at end of file
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_f16230777a0107771eef+2c2d707e/model.hlo_module.pb b/neuronxcc-2.14.213.0+013d129b/MODULE_f16230777a0107771eef+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..721956871657a5ae57ce923cd94b141b8d331e48
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_f16230777a0107771eef+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1c67694c75986472bd9f6692f6beddf0181577fdc9476ee5171dd9ed6e919cb2
+size 24383
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_f16230777a0107771eef+2c2d707e/model.neff b/neuronxcc-2.14.213.0+013d129b/MODULE_f16230777a0107771eef+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..b22956325c380d7b78f1ec4b4e1eee2415619da6
Binary files /dev/null and b/neuronxcc-2.14.213.0+013d129b/MODULE_f16230777a0107771eef+2c2d707e/model.neff differ
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_f2fc770203bd6f435b81+2c2d707e/compile_flags.txt b/neuronxcc-2.14.213.0+013d129b/MODULE_f2fc770203bd6f435b81+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_f2fc770203bd6f435b81+2c2d707e/compile_flags.txt
@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none
\ No newline at end of file
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_f2fc770203bd6f435b81+2c2d707e/model.hlo_module.pb b/neuronxcc-2.14.213.0+013d129b/MODULE_f2fc770203bd6f435b81+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..9a685842614e1299cdaa30d10bd2d60b75fe804e
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_f2fc770203bd6f435b81+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:818c5f9d4243ecdd4cc2a4d5879de3ca132144254461f627a97c7fd619fbe22c
+size 41821
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_f2fc770203bd6f435b81+2c2d707e/model.neff b/neuronxcc-2.14.213.0+013d129b/MODULE_f2fc770203bd6f435b81+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..cb9ba763629d030a9b5d9e3f52d9ed96ff7f5a61
Binary files /dev/null and b/neuronxcc-2.14.213.0+013d129b/MODULE_f2fc770203bd6f435b81+2c2d707e/model.neff differ
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_f394bb9283bee0635fa7+2c2d707e/compile_flags.txt b/neuronxcc-2.14.213.0+013d129b/MODULE_f394bb9283bee0635fa7+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_f394bb9283bee0635fa7+2c2d707e/compile_flags.txt
@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none
\ No newline at end of file
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_f394bb9283bee0635fa7+2c2d707e/model.hlo_module.pb b/neuronxcc-2.14.213.0+013d129b/MODULE_f394bb9283bee0635fa7+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..d3d86b1614b0b511181c441b2c62611e933b823b
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_f394bb9283bee0635fa7+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6c6a8ab0d0ac9e1202ffbccc7f93099b699114271c71a75815a3184619055b99
+size 12237
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_f394bb9283bee0635fa7+2c2d707e/model.neff b/neuronxcc-2.14.213.0+013d129b/MODULE_f394bb9283bee0635fa7+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..e8aa6ba1735d69a9f0df0df9eb23fdc19635c6cb
Binary files /dev/null and b/neuronxcc-2.14.213.0+013d129b/MODULE_f394bb9283bee0635fa7+2c2d707e/model.neff differ
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_fa766b7e9c712d21d3d8+2c2d707e/compile_flags.txt b/neuronxcc-2.14.213.0+013d129b/MODULE_fa766b7e9c712d21d3d8+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_fa766b7e9c712d21d3d8+2c2d707e/compile_flags.txt
@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none
\ No newline at end of file
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_fa766b7e9c712d21d3d8+2c2d707e/model.hlo_module.pb b/neuronxcc-2.14.213.0+013d129b/MODULE_fa766b7e9c712d21d3d8+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..3ec9fe122ab1d8c39ebb56627aa332aea1c290e6
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_fa766b7e9c712d21d3d8+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5431241ec261247dd83b5980adad0acefb4f18a36130bdd5070c0545abb423d8
+size 60791
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_fa766b7e9c712d21d3d8+2c2d707e/model.neff b/neuronxcc-2.14.213.0+013d129b/MODULE_fa766b7e9c712d21d3d8+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..d6c1715bcf2cef7e851c31bc9470a3f2c2f539ed
Binary files /dev/null and b/neuronxcc-2.14.213.0+013d129b/MODULE_fa766b7e9c712d21d3d8+2c2d707e/model.neff differ
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_fe94638cf5800fe61eca+2c2d707e/compile_flags.txt b/neuronxcc-2.14.213.0+013d129b/MODULE_fe94638cf5800fe61eca+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_fe94638cf5800fe61eca+2c2d707e/compile_flags.txt
@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none
\ No newline at end of file
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_fe94638cf5800fe61eca+2c2d707e/model.hlo_module.pb b/neuronxcc-2.14.213.0+013d129b/MODULE_fe94638cf5800fe61eca+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..d4c0669ee723abdb28a050fa1de1d70cd5ac9774
--- /dev/null
+++ b/neuronxcc-2.14.213.0+013d129b/MODULE_fe94638cf5800fe61eca+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0cd407f616887e7e057beda08319d867f56a3c2f1d69c020291a455a471f7564
+size 137320
diff --git a/neuronxcc-2.14.213.0+013d129b/MODULE_fe94638cf5800fe61eca+2c2d707e/model.neff b/neuronxcc-2.14.213.0+013d129b/MODULE_fe94638cf5800fe61eca+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..442dd953495636c6f3bf80ab15da8bf299c25590
Binary files /dev/null and b/neuronxcc-2.14.213.0+013d129b/MODULE_fe94638cf5800fe61eca+2c2d707e/model.neff differ