diff --git a/.gitattributes b/.gitattributes
index ba09e9a94216105513dc76e66f149854c7973225..44c997c7b5caf5a9ea8b8ea39e475fd559ba68b3 100644
--- a/.gitattributes
+++ b/.gitattributes
@@ -2137,3 +2137,15 @@ neuronxcc-2.16.372.0+4a9b2326/MODULE_fadc62f786e9e6e1bf75+613edded/model.neff fi
 neuronxcc-2.16.372.0+4a9b2326/MODULE_fce469267b2ad1b5d80e+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.16.372.0+4a9b2326/MODULE_8ea2de719456a80196f9+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.16.372.0+4a9b2326/MODULE_c064389074584e41f54f+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_01c445025d99d71be2b3+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_15eb37e3fb556fd3decb+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_2ae37ee2c7255cce8028+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_498320a96768c9eab266+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_53d5c248a142eced45e6+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_74a533aa34a2ab0fc7aa+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_857e6d9a3a28c7fe8c0c+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_892a0bb27ce39228be75+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_9b4b8c3000ad341d2ddc+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_a361a81ac38223fcadef+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_d3e5e4b762fd9115b68b+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_f5b6904839bc118bcd54+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
diff --git a/neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.2.0.dev1/inference/llama/llamafactory/tiny-random-Llama-3/51231e30eb944c5bc1f8.json b/neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.2.0.dev1/inference/llama/llamafactory/tiny-random-Llama-3/51231e30eb944c5bc1f8.json
new file mode 100644
index 0000000000000000000000000000000000000000..864fc653dd7308bf25c2d711be401c4770458e6f
--- /dev/null
+++ b/neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.2.0.dev1/inference/llama/llamafactory/tiny-random-Llama-3/51231e30eb944c5bc1f8.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": 131072, "mlp_bias": false, "model_type": "llama", "neuron": {"auto_cast_type": "bf16", "batch_size": 2, "checkpoint_id": "llamafactory/tiny-random-Llama-3", "checkpoint_revision": "bf2a2e3bf199ad2ee96f02a3c00246c608db22a8", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 1, "sequence_length": 128, "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-2.16.372.0+4a9b2326/0_REGISTRY/0.2.0.dev1/inference/llama/llamafactory/tiny-random-Llama-3/588e5102c25f59898011.json b/neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.2.0.dev1/inference/llama/llamafactory/tiny-random-Llama-3/588e5102c25f59898011.json
new file mode 100644
index 0000000000000000000000000000000000000000..90c35333bf7710ede2f9e35aaa6aa3945f525ef6
--- /dev/null
+++ b/neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.2.0.dev1/inference/llama/llamafactory/tiny-random-Llama-3/588e5102c25f59898011.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": 131072, "mlp_bias": false, "model_type": "llama", "neuron": {"auto_cast_type": "fp32", "batch_size": 1, "checkpoint_id": "llamafactory/tiny-random-Llama-3", "checkpoint_revision": "bf2a2e3bf199ad2ee96f02a3c00246c608db22a8", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 2, "sequence_length": 131072, "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-2.16.372.0+4a9b2326/0_REGISTRY/0.2.0.dev1/inference/llama/llamafactory/tiny-random-Llama-3/5c109283b389f09f634b.json b/neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.2.0.dev1/inference/llama/llamafactory/tiny-random-Llama-3/5c109283b389f09f634b.json
new file mode 100644
index 0000000000000000000000000000000000000000..9061a325c5507c14ce36fa6b3296ef522a8277fd
--- /dev/null
+++ b/neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.2.0.dev1/inference/llama/llamafactory/tiny-random-Llama-3/5c109283b389f09f634b.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": 131072, "mlp_bias": false, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 1, "checkpoint_id": "llamafactory/tiny-random-Llama-3", "checkpoint_revision": "bf2a2e3bf199ad2ee96f02a3c00246c608db22a8", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 2, "sequence_length": 512, "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-2.16.372.0+4a9b2326/0_REGISTRY/0.2.0.dev1/inference/llama/llamafactory/tiny-random-Llama-3/648252251e4a8bcdcbed.json b/neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.2.0.dev1/inference/llama/llamafactory/tiny-random-Llama-3/648252251e4a8bcdcbed.json
new file mode 100644
index 0000000000000000000000000000000000000000..5a4e02c58435c44fb358e71ba3688607f7548967
--- /dev/null
+++ b/neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.2.0.dev1/inference/llama/llamafactory/tiny-random-Llama-3/648252251e4a8bcdcbed.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": 131072, "mlp_bias": false, "model_type": "llama", "neuron": {"auto_cast_type": "bf16", "batch_size": 2, "checkpoint_id": "llamafactory/tiny-random-Llama-3", "checkpoint_revision": "bf2a2e3bf199ad2ee96f02a3c00246c608db22a8", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 2, "sequence_length": 128, "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-2.16.372.0+4a9b2326/0_REGISTRY/0.2.0.dev1/inference/llama/llamafactory/tiny-random-Llama-3/e76fd07fe3037f2e366b.json b/neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.2.0.dev1/inference/llama/llamafactory/tiny-random-Llama-3/e76fd07fe3037f2e366b.json
new file mode 100644
index 0000000000000000000000000000000000000000..558249aade1e462ab64f9a660568cd9d21a5a658
--- /dev/null
+++ b/neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.2.0.dev1/inference/llama/llamafactory/tiny-random-Llama-3/e76fd07fe3037f2e366b.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": 131072, "mlp_bias": false, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 1, "checkpoint_id": "llamafactory/tiny-random-Llama-3", "checkpoint_revision": "bf2a2e3bf199ad2ee96f02a3c00246c608db22a8", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 1, "sequence_length": 512, "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-2.16.372.0+4a9b2326/0_REGISTRY/0.2.0.dev1/inference/qwen2/Qwen/Qwen2.5-0.5B/e28ef40ae22096954a46.json b/neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.2.0.dev1/inference/qwen2/Qwen/Qwen2.5-0.5B/e28ef40ae22096954a46.json
new file mode 100644
index 0000000000000000000000000000000000000000..77a110ebfce3adb4af9aefda9c5491d6ae664c1c
--- /dev/null
+++ b/neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.2.0.dev1/inference/qwen2/Qwen/Qwen2.5-0.5B/e28ef40ae22096954a46.json
@@ -0,0 +1 @@
+{"architectures": ["Qwen2ForCausalLM"], "attention_dropout": 0.0, "bos_token_id": 151643, "eos_token_id": 151643, "hidden_act": "silu", "hidden_size": 896, "initializer_range": 0.02, "intermediate_size": 4864, "max_position_embeddings": 32768, "max_window_layers": 24, "model_type": "qwen2", "neuron": {"auto_cast_type": "fp32", "batch_size": 1, "checkpoint_id": "Qwen/Qwen2.5-0.5B", "checkpoint_revision": "060db6499f32faf8b98477b0a26969ef7d8b9987", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 24, "sequence_length": 128, "task": "text-generation"}, "num_attention_heads": 14, "num_hidden_layers": 24, "num_key_value_heads": 2, "rms_norm_eps": 1e-06, "rope_scaling": null, "rope_theta": 1000000.0, "sliding_window": 32768, "tie_word_embeddings": true, "torch_dtype": "bfloat16", "use_cache": true, "use_mrope": false, "use_sliding_window": false, "vocab_size": 151936}
\ No newline at end of file
diff --git a/neuronxcc-2.16.372.0+4a9b2326/MODULE_01c445025d99d71be2b3+613edded/compile_flags.json b/neuronxcc-2.16.372.0+4a9b2326/MODULE_01c445025d99d71be2b3+613edded/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..0c655e4bcad17c00d4d9ec479b31fc9e637d4287
--- /dev/null
+++ b/neuronxcc-2.16.372.0+4a9b2326/MODULE_01c445025d99d71be2b3+613edded/compile_flags.json
@@ -0,0 +1 @@
+["--target=trn1", "--model-type=transformer", "--auto-cast=none", "--execute-repetition=1"]
\ No newline at end of file
diff --git a/neuronxcc-2.16.372.0+4a9b2326/MODULE_01c445025d99d71be2b3+613edded/model.done b/neuronxcc-2.16.372.0+4a9b2326/MODULE_01c445025d99d71be2b3+613edded/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.16.372.0+4a9b2326/MODULE_01c445025d99d71be2b3+613edded/model.hlo_module.pb b/neuronxcc-2.16.372.0+4a9b2326/MODULE_01c445025d99d71be2b3+613edded/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..20e24226176413defa38074f1430e2e09f406964
--- /dev/null
+++ b/neuronxcc-2.16.372.0+4a9b2326/MODULE_01c445025d99d71be2b3+613edded/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:25ea5e996df5161b77f2848d7a7b352abd10853e44eb9a1ea521a369a8bd1978
+size 16017
diff --git a/neuronxcc-2.16.372.0+4a9b2326/MODULE_01c445025d99d71be2b3+613edded/model.neff b/neuronxcc-2.16.372.0+4a9b2326/MODULE_01c445025d99d71be2b3+613edded/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..f2a5537185a2ef8889f1ce6b226df6aa79b27de0
--- /dev/null
+++ b/neuronxcc-2.16.372.0+4a9b2326/MODULE_01c445025d99d71be2b3+613edded/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e7032e5c57479ee3cc37e828d55e2c502e3f73a187fb95ac169463b8b10b5129
+size 164864
diff --git a/neuronxcc-2.16.372.0+4a9b2326/MODULE_15eb37e3fb556fd3decb+613edded/compile_flags.json b/neuronxcc-2.16.372.0+4a9b2326/MODULE_15eb37e3fb556fd3decb+613edded/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..0c655e4bcad17c00d4d9ec479b31fc9e637d4287
--- /dev/null
+++ b/neuronxcc-2.16.372.0+4a9b2326/MODULE_15eb37e3fb556fd3decb+613edded/compile_flags.json
@@ -0,0 +1 @@
+["--target=trn1", "--model-type=transformer", "--auto-cast=none", "--execute-repetition=1"]
\ No newline at end of file
diff --git a/neuronxcc-2.16.372.0+4a9b2326/MODULE_15eb37e3fb556fd3decb+613edded/model.done b/neuronxcc-2.16.372.0+4a9b2326/MODULE_15eb37e3fb556fd3decb+613edded/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.16.372.0+4a9b2326/MODULE_15eb37e3fb556fd3decb+613edded/model.hlo_module.pb b/neuronxcc-2.16.372.0+4a9b2326/MODULE_15eb37e3fb556fd3decb+613edded/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..247ab58a5f601c34212c726132f2a92e7365f3e0
--- /dev/null
+++ b/neuronxcc-2.16.372.0+4a9b2326/MODULE_15eb37e3fb556fd3decb+613edded/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:72fa7b18ee25d91aa7213eae5fa35a6365a890ba7c42cdf5dcc6dca6941bd4c7
+size 22249
diff --git a/neuronxcc-2.16.372.0+4a9b2326/MODULE_15eb37e3fb556fd3decb+613edded/model.neff b/neuronxcc-2.16.372.0+4a9b2326/MODULE_15eb37e3fb556fd3decb+613edded/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..853e31b2b310a0716ee26c27c1b3406fd3b0b6f5
--- /dev/null
+++ b/neuronxcc-2.16.372.0+4a9b2326/MODULE_15eb37e3fb556fd3decb+613edded/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:95e23e7ef5c4fe6850767e4f53f3c9b732fa8cba098b90bcec750b96c33ac0cc
+size 144384
diff --git a/neuronxcc-2.16.372.0+4a9b2326/MODULE_2ae37ee2c7255cce8028+613edded/compile_flags.json b/neuronxcc-2.16.372.0+4a9b2326/MODULE_2ae37ee2c7255cce8028+613edded/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..0c655e4bcad17c00d4d9ec479b31fc9e637d4287
--- /dev/null
+++ b/neuronxcc-2.16.372.0+4a9b2326/MODULE_2ae37ee2c7255cce8028+613edded/compile_flags.json
@@ -0,0 +1 @@
+["--target=trn1", "--model-type=transformer", "--auto-cast=none", "--execute-repetition=1"]
\ No newline at end of file
diff --git a/neuronxcc-2.16.372.0+4a9b2326/MODULE_2ae37ee2c7255cce8028+613edded/model.done b/neuronxcc-2.16.372.0+4a9b2326/MODULE_2ae37ee2c7255cce8028+613edded/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.16.372.0+4a9b2326/MODULE_2ae37ee2c7255cce8028+613edded/model.hlo_module.pb b/neuronxcc-2.16.372.0+4a9b2326/MODULE_2ae37ee2c7255cce8028+613edded/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..2809e92c6b169bba548d1ac9d7835631ddedea72
--- /dev/null
+++ b/neuronxcc-2.16.372.0+4a9b2326/MODULE_2ae37ee2c7255cce8028+613edded/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c7f2ade06e6bd7f532c121ba34190998ff35dd497f016a58850b862ba532e025
+size 253034
diff --git a/neuronxcc-2.16.372.0+4a9b2326/MODULE_2ae37ee2c7255cce8028+613edded/model.neff b/neuronxcc-2.16.372.0+4a9b2326/MODULE_2ae37ee2c7255cce8028+613edded/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..86d724ea3235ba6a24b499925caddf7e7572f368
--- /dev/null
+++ b/neuronxcc-2.16.372.0+4a9b2326/MODULE_2ae37ee2c7255cce8028+613edded/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:079a299d91b20604217523d226ec3c9494b5b539c78c76b48ca4aa5ff81e50cb
+size 461824
diff --git a/neuronxcc-2.16.372.0+4a9b2326/MODULE_498320a96768c9eab266+613edded/compile_flags.json b/neuronxcc-2.16.372.0+4a9b2326/MODULE_498320a96768c9eab266+613edded/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..0c655e4bcad17c00d4d9ec479b31fc9e637d4287
--- /dev/null
+++ b/neuronxcc-2.16.372.0+4a9b2326/MODULE_498320a96768c9eab266+613edded/compile_flags.json
@@ -0,0 +1 @@
+["--target=trn1", "--model-type=transformer", "--auto-cast=none", "--execute-repetition=1"]
\ No newline at end of file
diff --git a/neuronxcc-2.16.372.0+4a9b2326/MODULE_498320a96768c9eab266+613edded/model.done b/neuronxcc-2.16.372.0+4a9b2326/MODULE_498320a96768c9eab266+613edded/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.16.372.0+4a9b2326/MODULE_498320a96768c9eab266+613edded/model.hlo_module.pb b/neuronxcc-2.16.372.0+4a9b2326/MODULE_498320a96768c9eab266+613edded/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..f46c947428afa87fc57a40144868185fca216164
--- /dev/null
+++ b/neuronxcc-2.16.372.0+4a9b2326/MODULE_498320a96768c9eab266+613edded/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8f1d7c2eec6e1858da932b982108c03d01f06ad91af02e22ebe3e0217290640e
+size 25797
diff --git a/neuronxcc-2.16.372.0+4a9b2326/MODULE_498320a96768c9eab266+613edded/model.neff b/neuronxcc-2.16.372.0+4a9b2326/MODULE_498320a96768c9eab266+613edded/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..ee6d9a4bbbf89757e46d796bcfe47e0bc5e4b8d1
--- /dev/null
+++ b/neuronxcc-2.16.372.0+4a9b2326/MODULE_498320a96768c9eab266+613edded/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:56c70a5f40c8bd35deee8753b029d7aec0e7df8bbc11feafc2a3b25391870410
+size 211907584
diff --git a/neuronxcc-2.16.372.0+4a9b2326/MODULE_53d5c248a142eced45e6+613edded/compile_flags.json b/neuronxcc-2.16.372.0+4a9b2326/MODULE_53d5c248a142eced45e6+613edded/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..0c655e4bcad17c00d4d9ec479b31fc9e637d4287
--- /dev/null
+++ b/neuronxcc-2.16.372.0+4a9b2326/MODULE_53d5c248a142eced45e6+613edded/compile_flags.json
@@ -0,0 +1 @@
+["--target=trn1", "--model-type=transformer", "--auto-cast=none", "--execute-repetition=1"]
\ No newline at end of file
diff --git a/neuronxcc-2.16.372.0+4a9b2326/MODULE_53d5c248a142eced45e6+613edded/model.done b/neuronxcc-2.16.372.0+4a9b2326/MODULE_53d5c248a142eced45e6+613edded/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.16.372.0+4a9b2326/MODULE_53d5c248a142eced45e6+613edded/model.hlo_module.pb b/neuronxcc-2.16.372.0+4a9b2326/MODULE_53d5c248a142eced45e6+613edded/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..94e303c042bfee5d0b1885a478da925af195a751
--- /dev/null
+++ b/neuronxcc-2.16.372.0+4a9b2326/MODULE_53d5c248a142eced45e6+613edded/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e71d28835e0bac2749ae20c3544805424c7085646099c71c09926731a18541a9
+size 16957
diff --git a/neuronxcc-2.16.372.0+4a9b2326/MODULE_53d5c248a142eced45e6+613edded/model.neff b/neuronxcc-2.16.372.0+4a9b2326/MODULE_53d5c248a142eced45e6+613edded/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..12eee2ffff43063ea0d94a51de88dabff405494c
--- /dev/null
+++ b/neuronxcc-2.16.372.0+4a9b2326/MODULE_53d5c248a142eced45e6+613edded/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:19415ca65c0508316343f385f5e2659e76000f503ccd32d040c759ae2d72be7c
+size 154624
diff --git a/neuronxcc-2.16.372.0+4a9b2326/MODULE_74a533aa34a2ab0fc7aa+613edded/compile_flags.json b/neuronxcc-2.16.372.0+4a9b2326/MODULE_74a533aa34a2ab0fc7aa+613edded/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..0c655e4bcad17c00d4d9ec479b31fc9e637d4287
--- /dev/null
+++ b/neuronxcc-2.16.372.0+4a9b2326/MODULE_74a533aa34a2ab0fc7aa+613edded/compile_flags.json
@@ -0,0 +1 @@
+["--target=trn1", "--model-type=transformer", "--auto-cast=none", "--execute-repetition=1"]
\ No newline at end of file
diff --git a/neuronxcc-2.16.372.0+4a9b2326/MODULE_74a533aa34a2ab0fc7aa+613edded/model.done b/neuronxcc-2.16.372.0+4a9b2326/MODULE_74a533aa34a2ab0fc7aa+613edded/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.16.372.0+4a9b2326/MODULE_74a533aa34a2ab0fc7aa+613edded/model.hlo_module.pb b/neuronxcc-2.16.372.0+4a9b2326/MODULE_74a533aa34a2ab0fc7aa+613edded/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..22b164a32c1595602622450de11b42323b33174a
--- /dev/null
+++ b/neuronxcc-2.16.372.0+4a9b2326/MODULE_74a533aa34a2ab0fc7aa+613edded/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8c9a5c78e6a03342265a2e7fb6b0115ae7273464dcc9a91a3b6fc3a02f5a1a42
+size 21313
diff --git a/neuronxcc-2.16.372.0+4a9b2326/MODULE_74a533aa34a2ab0fc7aa+613edded/model.neff b/neuronxcc-2.16.372.0+4a9b2326/MODULE_74a533aa34a2ab0fc7aa+613edded/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..6e0ead8fde17f036784ba47c9f0a920df8c63c9d
--- /dev/null
+++ b/neuronxcc-2.16.372.0+4a9b2326/MODULE_74a533aa34a2ab0fc7aa+613edded/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:68611abe11ae6fd76851c150ccb6f53c191d49f98cc0c1ec7da8077318be63b2
+size 164864
diff --git a/neuronxcc-2.16.372.0+4a9b2326/MODULE_857e6d9a3a28c7fe8c0c+613edded/compile_flags.json b/neuronxcc-2.16.372.0+4a9b2326/MODULE_857e6d9a3a28c7fe8c0c+613edded/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..0c655e4bcad17c00d4d9ec479b31fc9e637d4287
--- /dev/null
+++ b/neuronxcc-2.16.372.0+4a9b2326/MODULE_857e6d9a3a28c7fe8c0c+613edded/compile_flags.json
@@ -0,0 +1 @@
+["--target=trn1", "--model-type=transformer", "--auto-cast=none", "--execute-repetition=1"]
\ No newline at end of file
diff --git a/neuronxcc-2.16.372.0+4a9b2326/MODULE_857e6d9a3a28c7fe8c0c+613edded/model.done b/neuronxcc-2.16.372.0+4a9b2326/MODULE_857e6d9a3a28c7fe8c0c+613edded/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.16.372.0+4a9b2326/MODULE_857e6d9a3a28c7fe8c0c+613edded/model.hlo_module.pb b/neuronxcc-2.16.372.0+4a9b2326/MODULE_857e6d9a3a28c7fe8c0c+613edded/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..6ad3e73f2bc846fd41127d11d1987b8ed5fa2579
--- /dev/null
+++ b/neuronxcc-2.16.372.0+4a9b2326/MODULE_857e6d9a3a28c7fe8c0c+613edded/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:10522391bb6a18d7514c5e4bbd5263485db968acc7d1919b77f31ce89d5b9bdb
+size 211801
diff --git a/neuronxcc-2.16.372.0+4a9b2326/MODULE_857e6d9a3a28c7fe8c0c+613edded/model.neff b/neuronxcc-2.16.372.0+4a9b2326/MODULE_857e6d9a3a28c7fe8c0c+613edded/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..3f00acabc065ae7730ff4c42a259bd8071ea2d8e
--- /dev/null
+++ b/neuronxcc-2.16.372.0+4a9b2326/MODULE_857e6d9a3a28c7fe8c0c+613edded/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ab5353b0f6920e0f9751a1fae3e62d423c2302af910caffab35d0c0dd75ad940
+size 482304
diff --git a/neuronxcc-2.16.372.0+4a9b2326/MODULE_892a0bb27ce39228be75+613edded/compile_flags.json b/neuronxcc-2.16.372.0+4a9b2326/MODULE_892a0bb27ce39228be75+613edded/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..0c655e4bcad17c00d4d9ec479b31fc9e637d4287
--- /dev/null
+++ b/neuronxcc-2.16.372.0+4a9b2326/MODULE_892a0bb27ce39228be75+613edded/compile_flags.json
@@ -0,0 +1 @@
+["--target=trn1", "--model-type=transformer", "--auto-cast=none", "--execute-repetition=1"]
\ No newline at end of file
diff --git a/neuronxcc-2.16.372.0+4a9b2326/MODULE_892a0bb27ce39228be75+613edded/model.done b/neuronxcc-2.16.372.0+4a9b2326/MODULE_892a0bb27ce39228be75+613edded/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.16.372.0+4a9b2326/MODULE_892a0bb27ce39228be75+613edded/model.hlo_module.pb b/neuronxcc-2.16.372.0+4a9b2326/MODULE_892a0bb27ce39228be75+613edded/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..30aebc1c4da20feb21eb95e882b3cf9b2110fc9a
--- /dev/null
+++ b/neuronxcc-2.16.372.0+4a9b2326/MODULE_892a0bb27ce39228be75+613edded/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:51bbc25d2825e7c7f0b6598c9c83f441b70cbe384b1b927cc6aa410b26925f7b
+size 19946
diff --git a/neuronxcc-2.16.372.0+4a9b2326/MODULE_892a0bb27ce39228be75+613edded/model.neff b/neuronxcc-2.16.372.0+4a9b2326/MODULE_892a0bb27ce39228be75+613edded/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..f758d6047abab1d7c5cf144fca87b2571b904602
--- /dev/null
+++ b/neuronxcc-2.16.372.0+4a9b2326/MODULE_892a0bb27ce39228be75+613edded/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e912bfa34bce8ffe47a9d72c901e5c1e2934476dab1bc3ea38fdcac275c6daf4
+size 1137664
diff --git a/neuronxcc-2.16.372.0+4a9b2326/MODULE_9b4b8c3000ad341d2ddc+613edded/compile_flags.json b/neuronxcc-2.16.372.0+4a9b2326/MODULE_9b4b8c3000ad341d2ddc+613edded/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..0c655e4bcad17c00d4d9ec479b31fc9e637d4287
--- /dev/null
+++ b/neuronxcc-2.16.372.0+4a9b2326/MODULE_9b4b8c3000ad341d2ddc+613edded/compile_flags.json
@@ -0,0 +1 @@
+["--target=trn1", "--model-type=transformer", "--auto-cast=none", "--execute-repetition=1"]
\ No newline at end of file
diff --git a/neuronxcc-2.16.372.0+4a9b2326/MODULE_9b4b8c3000ad341d2ddc+613edded/model.done b/neuronxcc-2.16.372.0+4a9b2326/MODULE_9b4b8c3000ad341d2ddc+613edded/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.16.372.0+4a9b2326/MODULE_9b4b8c3000ad341d2ddc+613edded/model.hlo_module.pb b/neuronxcc-2.16.372.0+4a9b2326/MODULE_9b4b8c3000ad341d2ddc+613edded/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..ca95641d4e43d0dc0dfb3fc24ca82f13bfeae74f
--- /dev/null
+++ b/neuronxcc-2.16.372.0+4a9b2326/MODULE_9b4b8c3000ad341d2ddc+613edded/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:02541b4f1a227208a8fca19e8df0b1bb10a326d7b099dc9e1e0994ca0bd2dc97
+size 18642
diff --git a/neuronxcc-2.16.372.0+4a9b2326/MODULE_9b4b8c3000ad341d2ddc+613edded/model.neff b/neuronxcc-2.16.372.0+4a9b2326/MODULE_9b4b8c3000ad341d2ddc+613edded/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..03174f1a98422d9c3449abf59b43ef0719b7caa6
--- /dev/null
+++ b/neuronxcc-2.16.372.0+4a9b2326/MODULE_9b4b8c3000ad341d2ddc+613edded/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6e54c0fc769fa2b8a8d66af1ed639710141c0a78e0b816629385300658506839
+size 164864
diff --git a/neuronxcc-2.16.372.0+4a9b2326/MODULE_a361a81ac38223fcadef+613edded/compile_flags.json b/neuronxcc-2.16.372.0+4a9b2326/MODULE_a361a81ac38223fcadef+613edded/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..0c655e4bcad17c00d4d9ec479b31fc9e637d4287
--- /dev/null
+++ b/neuronxcc-2.16.372.0+4a9b2326/MODULE_a361a81ac38223fcadef+613edded/compile_flags.json
@@ -0,0 +1 @@
+["--target=trn1", "--model-type=transformer", "--auto-cast=none", "--execute-repetition=1"]
\ No newline at end of file
diff --git a/neuronxcc-2.16.372.0+4a9b2326/MODULE_a361a81ac38223fcadef+613edded/model.done b/neuronxcc-2.16.372.0+4a9b2326/MODULE_a361a81ac38223fcadef+613edded/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.16.372.0+4a9b2326/MODULE_a361a81ac38223fcadef+613edded/model.hlo_module.pb b/neuronxcc-2.16.372.0+4a9b2326/MODULE_a361a81ac38223fcadef+613edded/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..ff7c2409cd59041e332cf85bdb84ede4999b11bc
--- /dev/null
+++ b/neuronxcc-2.16.372.0+4a9b2326/MODULE_a361a81ac38223fcadef+613edded/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:62e8063648a4e054790d23b22ded8e4071c955149bab368a27365a1b0119c960
+size 19770
diff --git a/neuronxcc-2.16.372.0+4a9b2326/MODULE_a361a81ac38223fcadef+613edded/model.neff b/neuronxcc-2.16.372.0+4a9b2326/MODULE_a361a81ac38223fcadef+613edded/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..34ae4ce5eca62aa39c7b63bc741ed110d5f21160
--- /dev/null
+++ b/neuronxcc-2.16.372.0+4a9b2326/MODULE_a361a81ac38223fcadef+613edded/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8cae4926be6ce0a5742289e00679047c631c7c9ddd866c245f1f469212f8cfa7
+size 164864
diff --git a/neuronxcc-2.16.372.0+4a9b2326/MODULE_d3e5e4b762fd9115b68b+613edded/compile_flags.json b/neuronxcc-2.16.372.0+4a9b2326/MODULE_d3e5e4b762fd9115b68b+613edded/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..0c655e4bcad17c00d4d9ec479b31fc9e637d4287
--- /dev/null
+++ b/neuronxcc-2.16.372.0+4a9b2326/MODULE_d3e5e4b762fd9115b68b+613edded/compile_flags.json
@@ -0,0 +1 @@
+["--target=trn1", "--model-type=transformer", "--auto-cast=none", "--execute-repetition=1"]
\ No newline at end of file
diff --git a/neuronxcc-2.16.372.0+4a9b2326/MODULE_d3e5e4b762fd9115b68b+613edded/model.done b/neuronxcc-2.16.372.0+4a9b2326/MODULE_d3e5e4b762fd9115b68b+613edded/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.16.372.0+4a9b2326/MODULE_d3e5e4b762fd9115b68b+613edded/model.hlo_module.pb b/neuronxcc-2.16.372.0+4a9b2326/MODULE_d3e5e4b762fd9115b68b+613edded/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..72ccdcbf42033d8e5e614ed10839757fb60e11ae
--- /dev/null
+++ b/neuronxcc-2.16.372.0+4a9b2326/MODULE_d3e5e4b762fd9115b68b+613edded/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:23013ccec68df254df596a1fd8e25827188451a402482f46e8cd8c01de19f1dd
+size 19582
diff --git a/neuronxcc-2.16.372.0+4a9b2326/MODULE_d3e5e4b762fd9115b68b+613edded/model.neff b/neuronxcc-2.16.372.0+4a9b2326/MODULE_d3e5e4b762fd9115b68b+613edded/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..efd934fc66284c58d3f5e8d402f5acc3c945dfc6
--- /dev/null
+++ b/neuronxcc-2.16.372.0+4a9b2326/MODULE_d3e5e4b762fd9115b68b+613edded/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a4b1bfb6f295eb7ff7456a9f50c16988c5692a9f85b2009cbae3db7c2ec5848b
+size 154624
diff --git a/neuronxcc-2.16.372.0+4a9b2326/MODULE_f5b6904839bc118bcd54+613edded/compile_flags.json b/neuronxcc-2.16.372.0+4a9b2326/MODULE_f5b6904839bc118bcd54+613edded/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..0c655e4bcad17c00d4d9ec479b31fc9e637d4287
--- /dev/null
+++ b/neuronxcc-2.16.372.0+4a9b2326/MODULE_f5b6904839bc118bcd54+613edded/compile_flags.json
@@ -0,0 +1 @@
+["--target=trn1", "--model-type=transformer", "--auto-cast=none", "--execute-repetition=1"]
\ No newline at end of file
diff --git a/neuronxcc-2.16.372.0+4a9b2326/MODULE_f5b6904839bc118bcd54+613edded/model.done b/neuronxcc-2.16.372.0+4a9b2326/MODULE_f5b6904839bc118bcd54+613edded/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.16.372.0+4a9b2326/MODULE_f5b6904839bc118bcd54+613edded/model.hlo_module.pb b/neuronxcc-2.16.372.0+4a9b2326/MODULE_f5b6904839bc118bcd54+613edded/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..bb3dbf80636e1d359031768747c4098f0fcd0647
--- /dev/null
+++ b/neuronxcc-2.16.372.0+4a9b2326/MODULE_f5b6904839bc118bcd54+613edded/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c0213b4ed4aec983d88fee5e5a306218855ab9155d35caa84e767e5367993056
+size 20706
diff --git a/neuronxcc-2.16.372.0+4a9b2326/MODULE_f5b6904839bc118bcd54+613edded/model.neff b/neuronxcc-2.16.372.0+4a9b2326/MODULE_f5b6904839bc118bcd54+613edded/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..82f90e78e98b6183512474d9aed96bd47828cd3c
--- /dev/null
+++ b/neuronxcc-2.16.372.0+4a9b2326/MODULE_f5b6904839bc118bcd54+613edded/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:905cbd244bc79bb8576eb25218a6a1eeb8e5d9fab1c5ae1c0c0178912167de72
+size 154624