diff --git a/.gitattributes b/.gitattributes
index 783f25fa4387612ce563637e53eb5be0e38e48df..47ef8678568a5aee12b4a7fec52d230275d674a6 100644
--- a/.gitattributes
+++ b/.gitattributes
@@ -16604,3 +16604,15 @@ neuronxcc-2.21.33363.0+82129205/MODULE_9e68ef7eab52f091b941+300990b0/wrapped_nef
 neuronxcc-2.21.33363.0+82129205/MODULE_a155b32467ad91efdb56+a02c3a36/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.21.33363.0+82129205/MODULE_a155b32467ad91efdb56+a02c3a36/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.21.33363.0+82129205/MODULE_e45e2c512692918b8ac7+24129607/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.33363.0+82129205/MODULE_196e8f2481cd5ccf08a0+8a07fa10/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.33363.0+82129205/MODULE_196e8f2481cd5ccf08a0+8a07fa10/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.33363.0+82129205/MODULE_3c1208c8dbdb83a6bda4+cdfec012/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.33363.0+82129205/MODULE_3c1208c8dbdb83a6bda4+cdfec012/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.33363.0+82129205/MODULE_8b05ce6ede44eb411643+c476990f/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.33363.0+82129205/MODULE_8b05ce6ede44eb411643+c476990f/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.33363.0+82129205/MODULE_c49a2268db57a615eb6d+6fabf34d/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.33363.0+82129205/MODULE_c49a2268db57a615eb6d+6fabf34d/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.33363.0+82129205/MODULE_de4db98e666fc07883ed+57ca39a9/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.33363.0+82129205/MODULE_de4db98e666fc07883ed+57ca39a9/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.33363.0+82129205/MODULE_e22b5a814ee72d89608b+ea37e1a6/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.33363.0+82129205/MODULE_e22b5a814ee72d89608b+ea37e1a6/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
diff --git a/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.6.dev0/gemma3_text/unsloth/gemma-3-270m-it/f632bb119dd8a1b24f5e.json b/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.6.dev0/gemma3_text/unsloth/gemma-3-270m-it/f632bb119dd8a1b24f5e.json
new file mode 100644
index 0000000000000000000000000000000000000000..31f90e6e0ca66a4bbff827d339f27d7f69bb2755
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.6.dev0/gemma3_text/unsloth/gemma-3-270m-it/f632bb119dd8a1b24f5e.json
@@ -0,0 +1,81 @@
+{
+  "_entry_class": "SingleModelCacheEntry",
+  "_model_id": "unsloth/gemma-3-270m-it",
+  "_sliding_window_pattern": 6,
+  "_task": "text-generation",
+  "architectures": [
+    "Gemma3ForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "attn_logit_softcapping": null,
+  "dtype": "bfloat16",
+  "final_logit_softcapping": null,
+  "head_dim": 256,
+  "hidden_activation": "gelu_pytorch_tanh",
+  "hidden_size": 640,
+  "initializer_range": 0.02,
+  "intermediate_size": 2048,
+  "layer_types": [
+    "sliding_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "full_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "full_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "full_attention"
+  ],
+  "max_position_embeddings": 32768,
+  "model_type": "gemma3_text",
+  "neuron": {
+    "_serialized_key": "NxDNeuronConfig",
+    "batch_size": 1,
+    "capacity_factor": null,
+    "checkpoint_id": "unsloth/gemma-3-270m-it",
+    "checkpoint_revision": "23cf460f6bb16954176b3ddcc8d4f250501458a9",
+    "continuous_batching": false,
+    "ep_degree": 1,
+    "fused_qkv": true,
+    "glu_mlp": true,
+    "local_ranks_size": 2,
+    "max_batch_size": 1,
+    "max_context_length": 1024,
+    "max_topk": 256,
+    "n_active_tokens": 1024,
+    "neuronxcc_version": "2.21.33363.0+82129205",
+    "on_device_sampling": true,
+    "optimum_neuron_version": "0.4.6.dev0",
+    "output_logits": false,
+    "pp_degree": 1,
+    "sequence_length": 1024,
+    "speculation_length": 0,
+    "start_rank_id": 0,
+    "target": "trn1",
+    "torch_dtype": "bfloat16",
+    "tp_degree": 2
+  },
+  "num_attention_heads": 4,
+  "num_hidden_layers": 18,
+  "num_key_value_heads": 1,
+  "query_pre_attn_scalar": 256,
+  "rms_norm_eps": 1e-06,
+  "rope_local_base_freq": 10000.0,
+  "rope_scaling": null,
+  "rope_theta": 1000000.0,
+  "sliding_window": 512,
+  "unsloth_fixed": true,
+  "use_bidirectional_attention": false,
+  "use_cache": true,
+  "vocab_size": 262144
+}
\ No newline at end of file
diff --git a/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.6.dev0/llama/unsloth/Llama-3.2-1B-Instruct/b906aa10209b714ed9c0.json b/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.6.dev0/llama/unsloth/Llama-3.2-1B-Instruct/b906aa10209b714ed9c0.json
new file mode 100644
index 0000000000000000000000000000000000000000..bd8827415272f32353129c09693db442919ac8d1
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.6.dev0/llama/unsloth/Llama-3.2-1B-Instruct/b906aa10209b714ed9c0.json
@@ -0,0 +1,63 @@
+{
+  "_entry_class": "SingleModelCacheEntry",
+  "_model_id": "unsloth/Llama-3.2-1B-Instruct",
+  "_task": "text-generation",
+  "architectures": [
+    "LlamaForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "dtype": "bfloat16",
+  "head_dim": 64,
+  "hidden_act": "silu",
+  "hidden_size": 2048,
+  "initializer_range": 0.02,
+  "intermediate_size": 8192,
+  "max_position_embeddings": 131072,
+  "mlp_bias": false,
+  "model_type": "llama",
+  "neuron": {
+    "_serialized_key": "NxDNeuronConfig",
+    "batch_size": 1,
+    "capacity_factor": null,
+    "checkpoint_id": "unsloth/Llama-3.2-1B-Instruct",
+    "checkpoint_revision": null,
+    "continuous_batching": false,
+    "ep_degree": 1,
+    "fused_qkv": false,
+    "glu_mlp": true,
+    "local_ranks_size": 2,
+    "max_batch_size": 1,
+    "max_context_length": 4096,
+    "max_topk": 256,
+    "n_active_tokens": 4096,
+    "neuronxcc_version": "2.21.33363.0+82129205",
+    "on_device_sampling": false,
+    "optimum_neuron_version": "0.4.6.dev0",
+    "output_logits": false,
+    "pp_degree": 1,
+    "sequence_length": 4096,
+    "speculation_length": 5,
+    "start_rank_id": 0,
+    "target": "trn1",
+    "torch_dtype": "bfloat16",
+    "tp_degree": 2
+  },
+  "num_attention_heads": 32,
+  "num_hidden_layers": 16,
+  "num_key_value_heads": 8,
+  "pretraining_tp": 1,
+  "rms_norm_eps": 1e-05,
+  "rope_scaling": {
+    "factor": 32.0,
+    "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": true,
+  "unsloth_fixed": true,
+  "use_cache": true,
+  "vocab_size": 128256
+}
\ No newline at end of file
diff --git a/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.6.dev0/qwen2/Qwen/Qwen2.5-0.5B/e31330a79f96b75c3b92.json b/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.6.dev0/qwen2/Qwen/Qwen2.5-0.5B/e31330a79f96b75c3b92.json
new file mode 100644
index 0000000000000000000000000000000000000000..024a35ae67c4682840d46710e0b4abcdc1f093ed
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.6.dev0/qwen2/Qwen/Qwen2.5-0.5B/e31330a79f96b75c3b92.json
@@ -0,0 +1,82 @@
+{
+  "_entry_class": "SingleModelCacheEntry",
+  "_model_id": "Qwen/Qwen2.5-0.5B",
+  "_task": "text-generation",
+  "architectures": [
+    "Qwen2ForCausalLM"
+  ],
+  "attention_dropout": 0.0,
+  "dtype": "bfloat16",
+  "hidden_act": "silu",
+  "hidden_size": 896,
+  "initializer_range": 0.02,
+  "intermediate_size": 4864,
+  "layer_types": [
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention"
+  ],
+  "max_position_embeddings": 32768,
+  "max_window_layers": 24,
+  "model_type": "qwen2",
+  "neuron": {
+    "_serialized_key": "NxDNeuronConfig",
+    "batch_size": 2,
+    "capacity_factor": null,
+    "checkpoint_id": "Qwen/Qwen2.5-0.5B",
+    "checkpoint_revision": "060db6499f32faf8b98477b0a26969ef7d8b9987",
+    "continuous_batching": true,
+    "ep_degree": 1,
+    "fused_qkv": false,
+    "glu_mlp": true,
+    "local_ranks_size": 2,
+    "max_batch_size": 2,
+    "max_context_length": 1024,
+    "max_topk": 256,
+    "n_active_tokens": 1024,
+    "neuronxcc_version": "2.21.33363.0+82129205",
+    "on_device_sampling": false,
+    "optimum_neuron_version": "0.4.6.dev0",
+    "output_logits": false,
+    "pp_degree": 1,
+    "sequence_length": 1024,
+    "speculation_length": 0,
+    "start_rank_id": 0,
+    "target": "trn1",
+    "torch_dtype": "bfloat16",
+    "tp_degree": 2
+  },
+  "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": null,
+  "tie_word_embeddings": true,
+  "use_cache": true,
+  "use_mrope": false,
+  "use_sliding_window": false,
+  "vocab_size": 151936
+}
\ No newline at end of file
diff --git a/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.6.dev0/smollm3/HuggingFaceTB/SmolLM3-3B/ca162bc98de3faa6a1cd.json b/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.6.dev0/smollm3/HuggingFaceTB/SmolLM3-3B/ca162bc98de3faa6a1cd.json
new file mode 100644
index 0000000000000000000000000000000000000000..0c7f51b20d88a0a54a4af9a516b6770ccc5f6702
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.6.dev0/smollm3/HuggingFaceTB/SmolLM3-3B/ca162bc98de3faa6a1cd.json
@@ -0,0 +1,134 @@
+{
+  "_entry_class": "SingleModelCacheEntry",
+  "_model_id": "HuggingFaceTB/SmolLM3-3B",
+  "_task": "text-generation",
+  "architectures": [
+    "SmolLM3ForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "dtype": "bfloat16",
+  "hidden_act": "silu",
+  "hidden_size": 2048,
+  "initializer_range": 0.02,
+  "intermediate_size": 11008,
+  "layer_types": [
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention"
+  ],
+  "max_position_embeddings": 65536,
+  "max_window_layers": 28,
+  "mlp_bias": false,
+  "model_type": "smollm3",
+  "neuron": {
+    "_serialized_key": "NxDNeuronConfig",
+    "batch_size": 1,
+    "capacity_factor": null,
+    "checkpoint_id": "HuggingFaceTB/SmolLM3-3B",
+    "checkpoint_revision": "a07cc9a04f16550a088caea529712d1d335b0ac1",
+    "continuous_batching": false,
+    "ep_degree": 1,
+    "fused_qkv": true,
+    "glu_mlp": true,
+    "local_ranks_size": 2,
+    "max_batch_size": 1,
+    "max_context_length": 1024,
+    "max_topk": 256,
+    "n_active_tokens": 1024,
+    "neuronxcc_version": "2.21.33363.0+82129205",
+    "on_device_sampling": true,
+    "optimum_neuron_version": "0.4.6.dev0",
+    "output_logits": false,
+    "pp_degree": 1,
+    "sequence_length": 1024,
+    "speculation_length": 0,
+    "start_rank_id": 0,
+    "target": "trn1",
+    "torch_dtype": "bfloat16",
+    "tp_degree": 2
+  },
+  "no_rope_layer_interval": 4,
+  "no_rope_layers": [
+    1,
+    1,
+    1,
+    0,
+    1,
+    1,
+    1,
+    0,
+    1,
+    1,
+    1,
+    0,
+    1,
+    1,
+    1,
+    0,
+    1,
+    1,
+    1,
+    0,
+    1,
+    1,
+    1,
+    0,
+    1,
+    1,
+    1,
+    0,
+    1,
+    1,
+    1,
+    0,
+    1,
+    1,
+    1,
+    0
+  ],
+  "num_attention_heads": 16,
+  "num_hidden_layers": 36,
+  "num_key_value_heads": 4,
+  "pretraining_tp": 2,
+  "rms_norm_eps": 1e-06,
+  "rope_scaling": null,
+  "rope_theta": 5000000.0,
+  "sliding_window": null,
+  "use_cache": false,
+  "use_sliding_window": false,
+  "vocab_size": 128256
+}
\ No newline at end of file
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_10217061096959125489+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_10217061096959125489+e30acd3a/model.neff
index 3e34c2980ade69a1dcbdf8e51d89aaaf03d6c963..dbacf490d23ba23f4402ac29d5341e1a064e5efb 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_10217061096959125489+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_10217061096959125489+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_10244305442015770634+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_10244305442015770634+e30acd3a/model.neff
index 94e503153a287916fd28fe9aef32c4c4151dbaf3..e0a87e5c2616fb98397e08f4f9f6ee6bbe13e0d1 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_10244305442015770634+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_10244305442015770634+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_10606948783918825529+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_10606948783918825529+e30acd3a/model.neff
index 4afd85112aafdca0bad8c3279dae09aaf5dfb428..c50ebaefc1393de1dba2d6a1d9cf40e7d1e64d2f 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_10606948783918825529+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_10606948783918825529+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_10645643398657092095+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_10645643398657092095+e30acd3a/model.neff
index 38a73f6cfb3877e6bb78bde6d26152b2dd55569f..09f1f8d97b248f81f5ff82322e0b9664c35c377c 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_10645643398657092095+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_10645643398657092095+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_10661660426924300837+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_10661660426924300837+e30acd3a/model.neff
index 524ac606e31e910b462136c03fb3bc1dfcbc2858..e8d8bfa2561ac4042df13636830a21c360db80c1 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_10661660426924300837+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_10661660426924300837+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_10746122569655005679+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_10746122569655005679+e30acd3a/model.neff
index d1d315a125b585e67d1ad1d7b9ee1b6f15d1f168..f83a2d99c14ca92038903f493250dac6d71ab117 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_10746122569655005679+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_10746122569655005679+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_11086318750207148626+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_11086318750207148626+e30acd3a/model.neff
index 8ee8108e85e0aa04e983e233156a02dd149c1aaa..0c34bf3058af50ee7d21453a1fb2bcff8a86cf3a 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_11086318750207148626+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_11086318750207148626+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_11909310600244571805+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_11909310600244571805+e30acd3a/model.neff
index 5e76821f3bc5879affa5ba117c789b715cde9f23..d7b20b175bde4fb59645eaf2c501f0bcb2ff37a9 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_11909310600244571805+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_11909310600244571805+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_11933818254123612383+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_11933818254123612383+e30acd3a/model.neff
index 8f33f6e4cff05f2588f38cad798fb91c9314d8d0..0aa3c19178cfccada7d611bb946f5c6f2dccbd5f 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_11933818254123612383+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_11933818254123612383+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_1210392327607194823+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_1210392327607194823+e30acd3a/model.neff
index b8fa2dd295472d599a0b17f753684a37768f9e31..6336c87808f8623c560e2949e78415b51587bb13 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_1210392327607194823+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_1210392327607194823+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_12487216553200321032+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_12487216553200321032+e30acd3a/model.neff
index 7bf545041fe4645f55df110cb23585368a05ad66..9e5f504883d30291f1e49f1bc6a7c4b01a21f20d 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_12487216553200321032+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_12487216553200321032+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_13085549342645515693+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_13085549342645515693+e30acd3a/model.neff
index 3420bfb9c6605094925028360840fc69afb37ed9..b66378bc3945bf9fe8eb76dae4cdc54bc8a5fccf 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_13085549342645515693+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_13085549342645515693+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_13102010590082783346+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_13102010590082783346+e30acd3a/model.neff
index afda08ee0c8111aa761538199f2355dc2222a35b..bbe5c4fbb5e02d501dd529d07eed2276d114af63 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_13102010590082783346+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_13102010590082783346+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_13475273426270778455+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_13475273426270778455+e30acd3a/model.neff
index 6877b4e004a8fd429dacbf0c4c195f66184965a3..01de3bc345b22105cbe102f899485a7f70725709 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_13475273426270778455+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_13475273426270778455+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_13728813963059599796+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_13728813963059599796+e30acd3a/model.neff
index 92aba1535f88f3aa36e315e0d34f4babd83edf58..ff839943df4215b1d1b37710e0b89373cc511954 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_13728813963059599796+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_13728813963059599796+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_13783872956654054643+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_13783872956654054643+e30acd3a/model.neff
index c695d10ea694597a5f836a20bb33199270f7e2e3..58bab538b9dcafe995a838a764f9fecfca256340 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_13783872956654054643+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_13783872956654054643+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_14097776130612710282+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_14097776130612710282+e30acd3a/model.neff
index 4eaceba6f6ed01c901ff1d1ca927883ac16c1bde..a69b520e5faf0af1977db5710457dbaa87f3c544 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_14097776130612710282+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_14097776130612710282+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_14155105943057125178+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_14155105943057125178+e30acd3a/model.neff
index 4331845a6c13fe24cc72177cbc445395aedecf01..7f715cd04ecda1e0e7970fdf21f3177afe29a005 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_14155105943057125178+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_14155105943057125178+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_14173682194645188821+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_14173682194645188821+e30acd3a/model.neff
index c60c4f4de7ea200bcc007bd81dbcd58ef14ce813..14a7adc0e2a8d63247e55e36981efa6f18510745 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_14173682194645188821+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_14173682194645188821+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_14274833313744358348+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_14274833313744358348+e30acd3a/model.neff
index 3c16c750c25551743ed51d8fee31b570e1e853a7..36f46acba695b2fe4dcb9d3ab1a8a54a9da98488 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_14274833313744358348+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_14274833313744358348+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_14469716438804935215+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_14469716438804935215+e30acd3a/model.neff
index f3c334b13638ef45c83d2dc6e295ffed31dd1d85..2da86fde3d166f5371e644013f28b293d90d33c6 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_14469716438804935215+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_14469716438804935215+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_14739672283771660808+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_14739672283771660808+e30acd3a/model.neff
index f531bb624e28f3dfae8a4ae0ffe5ee5b30c77d23..138aaabef6f8037216f9058225a66719fcd1e7c9 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_14739672283771660808+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_14739672283771660808+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_15442663025941492357+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_15442663025941492357+e30acd3a/model.neff
index 6b05605e29cb12b7d4b8998a468ff4554018fba6..eec7c1da5439f63e26cdb4f991504f095fb2833a 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_15442663025941492357+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_15442663025941492357+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_15557971880365771457+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_15557971880365771457+e30acd3a/model.neff
index 3023dc4e990d00a11ca7e339d82d186be9c90c10..24c5c8420cdd5812b450fe2c26ea49c1f98b4b6f 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_15557971880365771457+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_15557971880365771457+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_15564425602929126510+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_15564425602929126510+e30acd3a/model.neff
index c1fba6e674a392724f9131a7347f4744d99ddcb2..da42086895b2b3a831bff0b6eb8d86d1d19ea5f4 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_15564425602929126510+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_15564425602929126510+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_15589838460896944293+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_15589838460896944293+e30acd3a/model.neff
index 8b0ae39312f828f284f6ec71e05204a0c3dfaedc..3e70a2ab2790a7c77eac1bc4167153922e2b0af7 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_15589838460896944293+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_15589838460896944293+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_15654572849171857535+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_15654572849171857535+e30acd3a/model.neff
index 04666f926997f4653009ea7f63578d6a0eb9e60c..0d32f2ff846713495cec24a999cb55ec3885da38 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_15654572849171857535+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_15654572849171857535+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_15767383571209512795+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_15767383571209512795+e30acd3a/model.neff
index cfeadce60e8e54713c4a559b98eb8fd46c801ab1..1a8ef29da60b8260dd8d58f080a27b2b8ab16649 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_15767383571209512795+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_15767383571209512795+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_15888836045088309511+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_15888836045088309511+e30acd3a/model.neff
index bc4b35b433a56f9d4ac6d39bfc5ff38528300d88..0f33d8260ecf8da422b397060b8393cd5acefa9a 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_15888836045088309511+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_15888836045088309511+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_15929451261464042997+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_15929451261464042997+e30acd3a/model.neff
index 447ecfa1b02993d988bf86b8023b0f83ff8e401a..5524fd843ead89ed214cc7ce4b06239281491520 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_15929451261464042997+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_15929451261464042997+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_16016505958416521648+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_16016505958416521648+e30acd3a/model.neff
index 497ef42aac8d0eababc9bebcf69aa0df74fcecac..46df60341ce8ee2ceb7873e792e418a70bbb24f2 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_16016505958416521648+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_16016505958416521648+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_16053163479112702088+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_16053163479112702088+e30acd3a/model.neff
index 4aa612e25df7928495fa30f732e08cb19f45fc62..e5c64c47acbdbf6977b2b4d8939479bf5243443b 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_16053163479112702088+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_16053163479112702088+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_16204387852795926216+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_16204387852795926216+e30acd3a/model.neff
index ac7809b46bc53d7175f07072a976b394c803fd41..4b745af904a54b53dfb663dd7f434d27a0a62408 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_16204387852795926216+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_16204387852795926216+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_1656719109221189948+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_1656719109221189948+e30acd3a/model.neff
index 38e1d2275f79df310f00ad7ec40d1e22f8446a10..0a7cde0dc7896e992f28cf62479d211bb29b9fd4 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_1656719109221189948+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_1656719109221189948+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_16599571375348449904+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_16599571375348449904+e30acd3a/model.neff
index beceb51ca517e679535a11799d475cd9227b1c6a..57e8ba36358ac1d3c54925d814b2d34a2eccd88e 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_16599571375348449904+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_16599571375348449904+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_1663757798483801648+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_1663757798483801648+e30acd3a/model.neff
index 4346f4b419928b8b00d6fa493a399aeb300abf4d..8c272ca0906bc1990c7c9578c9c9a1a893701e33 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_1663757798483801648+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_1663757798483801648+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_16738296820980389103+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_16738296820980389103+e30acd3a/model.neff
index 335873944d950adefc661f5ab799297533b25ffd..355e4d8e5f138603a85e949cdb990b0e300f1a07 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_16738296820980389103+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_16738296820980389103+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_1696825468766062114+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_1696825468766062114+e30acd3a/model.neff
index 284277094d772b94567460091de68dc048d8006c..e75ad95845598d5dabc5ac15d4d08481c56ad0e0 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_1696825468766062114+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_1696825468766062114+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_17476209562158013765+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_17476209562158013765+e30acd3a/model.neff
index 16454301701bcde1e8cc715e4bf84698b69012ea..6cfe5cb895668cb0a1dbb4293b769ec805d08187 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_17476209562158013765+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_17476209562158013765+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_17781596253725927902+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_17781596253725927902+e30acd3a/model.neff
index 36c6edd3d83841bcc1e0d54b9fcff9621aa4ca32..e71ae558642444d12d37ca9f714edcc0985f10a8 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_17781596253725927902+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_17781596253725927902+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_1795740353831178306+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_1795740353831178306+e30acd3a/model.neff
index 014069929c2228b68b56387b85554909ef1902ec..082483fe46140cfa01d140b9ffa6defb198d7704 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_1795740353831178306+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_1795740353831178306+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_1800832390737682969+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_1800832390737682969+e30acd3a/model.neff
index 63e72692a003b9ba35e2551994053b673badd852..4060e7d3998ba2d12c05819c117ee48167deb95c 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_1800832390737682969+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_1800832390737682969+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_1805481651134498710+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_1805481651134498710+e30acd3a/model.neff
index 6965229004e0570152d561ab363f8a7f33c15bbe..f57bbfe315c849eb4eb531f850373aa9a4216dbd 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_1805481651134498710+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_1805481651134498710+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_1805728312057401221+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_1805728312057401221+e30acd3a/model.neff
index 77243f36bf1684adce7f7f6c30b41189277003f9..0518b15f04f8337266662098c28799cdc9c9e063 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_1805728312057401221+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_1805728312057401221+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_18070749384531238695+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_18070749384531238695+e30acd3a/model.neff
index e8775b3de5ccbd9cb0d4d9b1a20e274788c1465c..f1142e0b1c2df3abd9fbc57fb5772e036b7e6e3f 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_18070749384531238695+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_18070749384531238695+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_18134517905922687964+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_18134517905922687964+e30acd3a/model.neff
index 4be93b54b187e68e3b7df7f60d4ed156c1c66554..e0c46c78d712997347aeef069e912524e56d8e8a 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_18134517905922687964+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_18134517905922687964+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_18207475553554060390+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_18207475553554060390+e30acd3a/model.neff
index 52fc9d82a6b0450b6528f5f6193ef1652a5cf8f1..033548be6fb7307825dd71ca8d78487bd07829b8 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_18207475553554060390+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_18207475553554060390+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_196e8f2481cd5ccf08a0+8a07fa10/compile_flags.json b/neuronxcc-2.21.33363.0+82129205/MODULE_196e8f2481cd5ccf08a0+8a07fa10/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..ee7f5b985b00cbff6c16724858bc6260258dbb8a
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_196e8f2481cd5ccf08a0+8a07fa10/compile_flags.json
@@ -0,0 +1 @@
+["--target=trn1", "--enable-saturate-infinity", "--auto-cast=none", "--model-type=transformer", "-O1", "--logfile=/tmp/nxdi_test_d6f3b9ab-010f-461c-818a-798fe0472842/compiler_workdir/DecoderLayerWrapper/_tp0_bk0/log-neuron-cc.txt", "--enable-internal-neff-wrapper"]
\ No newline at end of file
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_196e8f2481cd5ccf08a0+8a07fa10/model.done b/neuronxcc-2.21.33363.0+82129205/MODULE_196e8f2481cd5ccf08a0+8a07fa10/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_196e8f2481cd5ccf08a0+8a07fa10/model.hlo_module.pb b/neuronxcc-2.21.33363.0+82129205/MODULE_196e8f2481cd5ccf08a0+8a07fa10/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..be5670c6258a72b3253c262c04b0a3c084a1810d
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_196e8f2481cd5ccf08a0+8a07fa10/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3dc99fd8fa13fb76a08ae04c8275958cd87878cfbf86f63ae613d7efcf775bfc
+size 29412
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_196e8f2481cd5ccf08a0+8a07fa10/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_196e8f2481cd5ccf08a0+8a07fa10/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..f746d00ed0c201935305e032f435a66e3c71dd76
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_196e8f2481cd5ccf08a0+8a07fa10/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a17238f1332fc01ebe2959cb313369d26825336bcb498e5bb54df61a1329e228
+size 328704
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_196e8f2481cd5ccf08a0+8a07fa10/wrapped_neff.hlo b/neuronxcc-2.21.33363.0+82129205/MODULE_196e8f2481cd5ccf08a0+8a07fa10/wrapped_neff.hlo
new file mode 100644
index 0000000000000000000000000000000000000000..bb9b708a972e6039f61b88ee6523421d4573969c
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_196e8f2481cd5ccf08a0+8a07fa10/wrapped_neff.hlo
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:43f1052e739b2ae8746f5fd851831c92edc53fe5443d44401b078bf704ee81c8
+size 334452
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_2157613318347839507+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_2157613318347839507+e30acd3a/model.neff
index e0a8c6c2117864c56bd0bba9165d85b68f969313..600d473db4032e83d9d6102c37c110c6388b44cd 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_2157613318347839507+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_2157613318347839507+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_2203200257461345827+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_2203200257461345827+e30acd3a/model.neff
index 3a76d39a6278a193c3d712e05682b6a01a8e1392..0f3dd83a171ea01a41f1df13f4d886907f0bbfaf 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_2203200257461345827+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_2203200257461345827+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_2410426589418994260+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_2410426589418994260+e30acd3a/model.neff
index a54a8dd6871783d1ebac2c4d7a9616b292b2206c..3b9fa23910cda2ca23a61a0d0f115698df2830df 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_2410426589418994260+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_2410426589418994260+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_2426414314187505427+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_2426414314187505427+e30acd3a/model.neff
index 65a510ca4889e820d6d030892636957b0a9587b8..ede25efb715cd33d292222169d6c1ba781af3ca8 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_2426414314187505427+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_2426414314187505427+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_2576759111807165188+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_2576759111807165188+e30acd3a/model.neff
index e98316277cb98efc3a722242025eb358ea4c16a0..03702793e85c8202441e99d953c9518bce2c739d 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_2576759111807165188+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_2576759111807165188+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_2650447136112456251+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_2650447136112456251+e30acd3a/model.neff
index 47954bf2f12b5aee90a2c822c3b444c2fe06909a..bafb859e677b6d8e81af0efe7169e5ac9935308b 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_2650447136112456251+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_2650447136112456251+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_2682953264353234433+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_2682953264353234433+e30acd3a/model.neff
index b2bed8d8905566df58161d4b6e69dd38e8e91c85..e76c12c8f336bed5537a8e195f04154ba47df961 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_2682953264353234433+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_2682953264353234433+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_2699280712748688265+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_2699280712748688265+e30acd3a/model.neff
index 1728d386a0b989c0d6b989420eefe8e6a1fc01d9..b679e629ecb405c0f41db310782d6d769320c80b 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_2699280712748688265+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_2699280712748688265+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_2811495285170804454+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_2811495285170804454+e30acd3a/model.neff
index 33980202cb922eb8192bc3c1565dcc7c1f555b6f..9851df7c2b120b12b92d4215af9eacd0e0edcbc0 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_2811495285170804454+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_2811495285170804454+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_2ce3d3065088729c7d7f+24129607/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_2ce3d3065088729c7d7f+24129607/model.neff
index aead2b3226d482bda7c015ad4d479b20d6fb5559..960928a7f6edfd19d171be3f3c40c32e9bea392f 100644
--- a/neuronxcc-2.21.33363.0+82129205/MODULE_2ce3d3065088729c7d7f+24129607/model.neff
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_2ce3d3065088729c7d7f+24129607/model.neff
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:da3045f0f7af6c2235fc2428acc84add9712c1569dc34e19b016a9e2f1a980ea
+oid sha256:3bf93955aefc8032fef422430de11249c4b59e5fa6e8711123283bb35b6b0d7f
 size 3769344
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_302950992539679003+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_302950992539679003+e30acd3a/model.neff
index 96f49bca6dccb1aa20c7c38b9b46e57fe7152c74..1f3ed754219f313e25d2af4e12fe81ce3967ab66 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_302950992539679003+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_302950992539679003+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_3c1208c8dbdb83a6bda4+cdfec012/compile_flags.json b/neuronxcc-2.21.33363.0+82129205/MODULE_3c1208c8dbdb83a6bda4+cdfec012/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..8cc734dbfbd3b6fea3d5dbbe4813556dd711aae5
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_3c1208c8dbdb83a6bda4+cdfec012/compile_flags.json
@@ -0,0 +1 @@
+["--target=trn1", "--enable-saturate-infinity", "--auto-cast=none", "--model-type=transformer", "-O1", "--logfile=/tmp/nxdi_test_a048987d-35ef-4711-ad7b-f6a46d6bdb1c/compiler_workdir/DecoderLayerWrapper/_tp0_bk0/log-neuron-cc.txt", "--enable-internal-neff-wrapper"]
\ No newline at end of file
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_3c1208c8dbdb83a6bda4+cdfec012/model.done b/neuronxcc-2.21.33363.0+82129205/MODULE_3c1208c8dbdb83a6bda4+cdfec012/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_3c1208c8dbdb83a6bda4+cdfec012/model.hlo_module.pb b/neuronxcc-2.21.33363.0+82129205/MODULE_3c1208c8dbdb83a6bda4+cdfec012/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..07bf62460bc7813e4a95f208598e2fa4ae551d41
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_3c1208c8dbdb83a6bda4+cdfec012/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:76dc525043c8d0fc6b46382b17659e6a2f2e981788c635021a8a61aa832d2866
+size 14480
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_3c1208c8dbdb83a6bda4+cdfec012/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_3c1208c8dbdb83a6bda4+cdfec012/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..a4536fb41c459c9bda64b59fddc30ce972e7ca52
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_3c1208c8dbdb83a6bda4+cdfec012/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5d6c2a671da55eb5ab5462b8438bcc5bb269578dc2599c503d6054b72e5f8d68
+size 267264
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_3c1208c8dbdb83a6bda4+cdfec012/wrapped_neff.hlo b/neuronxcc-2.21.33363.0+82129205/MODULE_3c1208c8dbdb83a6bda4+cdfec012/wrapped_neff.hlo
new file mode 100644
index 0000000000000000000000000000000000000000..69da42ba02caa247e4e4fbf996bc297848337ee4
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_3c1208c8dbdb83a6bda4+cdfec012/wrapped_neff.hlo
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:585b737af5e66fdcbd03e3adfd6e521a83d64b7ffdd63fc372e6729b8c630c95
+size 272962
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_4013984518400530928+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_4013984518400530928+e30acd3a/model.neff
index 51c2ea245cd9f79c0fe705c6f4d5b09d635c655c..890a025501d70c411d9614a6735e1582c5850ab0 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_4013984518400530928+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_4013984518400530928+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_4052264821257342969+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_4052264821257342969+e30acd3a/model.neff
index 4b17c026f321b138997918acae111f001ce0cb2c..6d0ce9d166860e65250070ab617c1c791fb43ab2 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_4052264821257342969+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_4052264821257342969+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_4166695227062860792+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_4166695227062860792+e30acd3a/model.neff
index 5261ccf43b518357ad89932c9076edad4f67a12e..7791d262cc82036e11fee3b763a02dbfad7292f8 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_4166695227062860792+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_4166695227062860792+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_4232557591140652245+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_4232557591140652245+e30acd3a/model.neff
index cd96b0905c4cff7fd81d1dd4ead8b5586d3a18c5..6f014a9fb543e454c0e18f5b2922b376ad9c9d81 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_4232557591140652245+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_4232557591140652245+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_4606491920565381523+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_4606491920565381523+e30acd3a/model.neff
index 8a0d3c5cb4b2ccec8da55057d1b408b01b8cf0ee..9c788815296177d0e2c4ef7844904f02908c0c30 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_4606491920565381523+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_4606491920565381523+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_4608333438381222049+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_4608333438381222049+e30acd3a/model.neff
index 7050d73a0f8ace8a3fc70595f4d9b4d165aea28f..67b448eaa620a0783433f2d4a62b7397ee131f94 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_4608333438381222049+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_4608333438381222049+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_4683016291147007110+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_4683016291147007110+e30acd3a/model.neff
index aaacc1c7d43f2bed931f5f435e11c043dbb77cb0..9b90a00b22a48498895abffd583c028792a525e7 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_4683016291147007110+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_4683016291147007110+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_4816422433712790690+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_4816422433712790690+e30acd3a/model.neff
index 9944a440631105219131dfe4c9e4c3fcc1b8dcbe..211d2997439608df71a6324f62fb4b6151820aee 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_4816422433712790690+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_4816422433712790690+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_4842952601010372434+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_4842952601010372434+e30acd3a/model.neff
index 7fd21fcc81dfdeae164ff3a2b21fb6ffc8514b3b..1c644d723e3fecce7fd699db579e38fbd5a77245 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_4842952601010372434+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_4842952601010372434+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_4939721357859779936+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_4939721357859779936+e30acd3a/model.neff
index e11224ec542fc6a357e804a3cc0f99c8de6ce41c..51e93cbbaac25be1d0df2d0129be2853e703ed0f 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_4939721357859779936+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_4939721357859779936+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_4b17cd74ff5b9c533464+010126f4/compile_flags.json b/neuronxcc-2.21.33363.0+82129205/MODULE_4b17cd74ff5b9c533464+010126f4/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..b7c6353a35aa9c54a8fbaaa97c2af3f35c67df6e
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_4b17cd74ff5b9c533464+010126f4/compile_flags.json
@@ -0,0 +1 @@
+["--target=trn1", "--enable-saturate-infinity", "--auto-cast=none", "--model-type=transformer", "-O1", "--logfile=/tmp/nxdi_test_c1eda606-54e7-4da6-b7fb-71e3c39738f5/compiler_workdir/Llama3RotaryEmbedding/_tp0_bk0/log-neuron-cc.txt", "--enable-internal-neff-wrapper"]
\ No newline at end of file
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_4b17cd74ff5b9c533464+010126f4/model.done b/neuronxcc-2.21.33363.0+82129205/MODULE_4b17cd74ff5b9c533464+010126f4/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_4b17cd74ff5b9c533464+010126f4/model.hlo_module.pb b/neuronxcc-2.21.33363.0+82129205/MODULE_4b17cd74ff5b9c533464+010126f4/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..f1e2473c54cffac949ecf31d3b6ec636c290545b
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_4b17cd74ff5b9c533464+010126f4/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f4a29d4cf27aec9af90e53324eff9ee69e033679482f4016a9dd3e5416c7c68a
+size 1584
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_4b17cd74ff5b9c533464+010126f4/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_4b17cd74ff5b9c533464+010126f4/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..7ecc9d45c0631cd0690c9cce884bdca9cf8659f6
Binary files /dev/null and b/neuronxcc-2.21.33363.0+82129205/MODULE_4b17cd74ff5b9c533464+010126f4/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_5397223456911199516+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_5397223456911199516+e30acd3a/model.neff
index 93c0980779695d339c0888ed12917bd7bc1d912a..7750b73a0fd895db1d24b24e68eb3cea0b285e72 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_5397223456911199516+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_5397223456911199516+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_5724544066416780383+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_5724544066416780383+e30acd3a/model.neff
index 502fed883bb7e0cc853af1724f20c7b973f3df56..33412c09d2bf1e1c10af054df4e329f80453c981 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_5724544066416780383+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_5724544066416780383+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_5765916484866650909+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_5765916484866650909+e30acd3a/model.neff
index 3b8b479b40f7076ec8d71689ee6689e130b3f2ed..e29295f8e4034552179460e4a4ef89d5c31d6be9 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_5765916484866650909+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_5765916484866650909+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_6368409506294683105+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_6368409506294683105+e30acd3a/model.neff
index a6b42ce07a0758b1480f5b33bb20e4c334b40380..76cc00646492024f60371b74667e98e8cd1df95b 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_6368409506294683105+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_6368409506294683105+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_6375780938374445148+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_6375780938374445148+e30acd3a/model.neff
index 7936be0b32383f09638a0834a34b44340158feda..66d38d93aa9fa356cbe81986784e784a33b5183d 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_6375780938374445148+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_6375780938374445148+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_6417297157123190451+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_6417297157123190451+e30acd3a/model.neff
index da3e3422e0bb327195ef1a998e23e67a1b4cd6ca..4d9469ded7cb471f6870867e6ea830d692c26b86 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_6417297157123190451+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_6417297157123190451+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_665145310648895012+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_665145310648895012+e30acd3a/model.neff
index 39fff34bc8d71b38bfeff8d9346bea1e19d7a87e..ee316c5f618d78d2983b53690d52c9f5e4d33a24 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_665145310648895012+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_665145310648895012+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_6711765444274762085+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_6711765444274762085+e30acd3a/model.neff
index 51e7dd1a7aeeb6497c26194890d5f245049c3b04..951427f01844968523868732b91a135f10a85ed7 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_6711765444274762085+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_6711765444274762085+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_7049343467335223052+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_7049343467335223052+e30acd3a/model.neff
index a93ae187d6afa5d1d82f2922bea29a68a052bd68..e9bf1cd34f1f24683c72ee9670438cfc08aab968 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_7049343467335223052+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_7049343467335223052+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_7877866862268499659+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_7877866862268499659+e30acd3a/model.neff
index d3e444ce58ba9972dc3431955f36a41eda1882d8..1623267b6ce6c60df973dfe1de0a78ce7e5dacc1 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_7877866862268499659+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_7877866862268499659+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_793722132614349680+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_793722132614349680+e30acd3a/model.neff
index 0d6729c29003825dc883a5ad28975f453f58cfd8..3f8aa18cf592ff983fb7ccfd087f65d4b0adfeb9 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_793722132614349680+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_793722132614349680+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_8142158350842512240+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_8142158350842512240+e30acd3a/model.neff
index 40ff123d4db3543e1eee73268b3ece50ad3534b9..5dd7e8f45ff709519e284e6724aa1b3c7a4a1168 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_8142158350842512240+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_8142158350842512240+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_8146958338478347620+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_8146958338478347620+e30acd3a/model.neff
index c65dee50813279b7dbe666563a76f9750ba569ce..ec1f6029c7d565834b3afb50796c0d5a241ee12a 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_8146958338478347620+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_8146958338478347620+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_8245165830758578911+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_8245165830758578911+e30acd3a/model.neff
index e167aa87c7fd64041cca9a70ac63fbe87f32a90c..eedad878aa049049e8854b893582e3e536d4ce7e 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_8245165830758578911+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_8245165830758578911+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_8679096817551502409+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_8679096817551502409+e30acd3a/model.neff
index 512fe55750892ccde9127f57c9e00c67a4f5092e..33b1ae60d5da9f9a073f7b3dfcab4d58df22fdb1 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_8679096817551502409+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_8679096817551502409+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_8882592820567938515+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_8882592820567938515+e30acd3a/model.neff
index 23d5d7e6154801644cca994278618d27c3951fd2..5e5412b327542dcec5635d40e51b8b7de313ee99 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_8882592820567938515+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_8882592820567938515+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_8b05ce6ede44eb411643+c476990f/compile_flags.json b/neuronxcc-2.21.33363.0+82129205/MODULE_8b05ce6ede44eb411643+c476990f/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..460823fb641ae9bff76a4e61cd7115bd5ab27688
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_8b05ce6ede44eb411643+c476990f/compile_flags.json
@@ -0,0 +1 @@
+["--target=trn1", "--enable-saturate-infinity", "--auto-cast=none", "--model-type=transformer", "-O1", "--logfile=/tmp/nxdi_test_b826a256-7dc1-4bc3-b813-c1c729ca32be/compiler_workdir/AttentionModelWrapper/_tp0_bk0/log-neuron-cc.txt", "--enable-internal-neff-wrapper"]
\ No newline at end of file
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_8b05ce6ede44eb411643+c476990f/model.done b/neuronxcc-2.21.33363.0+82129205/MODULE_8b05ce6ede44eb411643+c476990f/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_8b05ce6ede44eb411643+c476990f/model.hlo_module.pb b/neuronxcc-2.21.33363.0+82129205/MODULE_8b05ce6ede44eb411643+c476990f/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..8496781848c02ecd308b5b2c5a3c342b702425e7
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_8b05ce6ede44eb411643+c476990f/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:51e79a5cc325f9a317d158d931d75ec73ae1cfb16a4d83343a2593edd95d0044
+size 8979
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_8b05ce6ede44eb411643+c476990f/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_8b05ce6ede44eb411643+c476990f/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..730e13132d47a12079e7340b84833d855f5820b0
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_8b05ce6ede44eb411643+c476990f/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:30eadc9e457938098996cc75d85cb9be5c8e77f4d4bcffd7f18446d290bf7aa8
+size 246784
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_8b05ce6ede44eb411643+c476990f/wrapped_neff.hlo b/neuronxcc-2.21.33363.0+82129205/MODULE_8b05ce6ede44eb411643+c476990f/wrapped_neff.hlo
new file mode 100644
index 0000000000000000000000000000000000000000..9b8affcbe8714dc805b34e01440e814e45be8c77
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_8b05ce6ede44eb411643+c476990f/wrapped_neff.hlo
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2643311acd46ce414465312f39d2ef532646aa19da33525981c73d4aaf3dff64
+size 249608
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_8d5c7639490b29a58f59+24129607/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_8d5c7639490b29a58f59+24129607/model.neff
index 6d1a4e87872478e4413248eb929d1e999c32196b..9f9816c42ef6d1d1476078bc6bdafbda0bdf3b86 100644
--- a/neuronxcc-2.21.33363.0+82129205/MODULE_8d5c7639490b29a58f59+24129607/model.neff
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_8d5c7639490b29a58f59+24129607/model.neff
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ec3b3a30a58e872543f04aec56ad92e14636838909cc7af96bac1c3dbfb159b7
+oid sha256:91b57ad4edc4ae73246db5b3e94a279b058571ab31a6538177c2912805818613
 size 42322944
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_9168843780052024308+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_9168843780052024308+e30acd3a/model.neff
index fc9d1287a6cab646ee15a6e2921108e6d02955c2..6971a9fe7c6042048cd780361b920091da7c0b12 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_9168843780052024308+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_9168843780052024308+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_9527558113976496538+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_9527558113976496538+e30acd3a/model.neff
index 3dbbfe0002a7498b16cebe5589303a99a865b77e..c0a7f8337923fd66b79ba138dd5b12d24716de38 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_9527558113976496538+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_9527558113976496538+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_9533302140457976101+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_9533302140457976101+e30acd3a/model.neff
index 612249c69427750828b77e59094ed9b33cedd3c8..da3eb4480735237bfef1dcc0e371b201a3735c99 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_9533302140457976101+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_9533302140457976101+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_9770544877851564228+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_9770544877851564228+e30acd3a/model.neff
index 09659879aff5213ecd778370b340b450bfd39f25..bc33655a9748410045e3bab323a27dc4e0691b48 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_9770544877851564228+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_9770544877851564228+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_c49a2268db57a615eb6d+6fabf34d/compile_flags.json b/neuronxcc-2.21.33363.0+82129205/MODULE_c49a2268db57a615eb6d+6fabf34d/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..79c179745e8f7ac30f64ce42a96b188000b63c19
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_c49a2268db57a615eb6d+6fabf34d/compile_flags.json
@@ -0,0 +1 @@
+["--target=trn1", "--enable-saturate-infinity", "--auto-cast=none", "--model-type=transformer", "-O1", "--logfile=/tmp/nxdi_test_1e77ae6a-e32d-4ef2-ab2f-818db0fa550c/compiler_workdir/NeuronRMSNorm/_tp0_bk0/log-neuron-cc.txt", "--enable-internal-neff-wrapper"]
\ No newline at end of file
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_c49a2268db57a615eb6d+6fabf34d/model.done b/neuronxcc-2.21.33363.0+82129205/MODULE_c49a2268db57a615eb6d+6fabf34d/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_c49a2268db57a615eb6d+6fabf34d/model.hlo_module.pb b/neuronxcc-2.21.33363.0+82129205/MODULE_c49a2268db57a615eb6d+6fabf34d/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..6a2752c25c9f7e1092e2b55c010694e168610ce3
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_c49a2268db57a615eb6d+6fabf34d/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1c774c6626fc91684e924b843c1fe807dadbbac6d8a8e19443663b43c13eac35
+size 1165
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_c49a2268db57a615eb6d+6fabf34d/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_c49a2268db57a615eb6d+6fabf34d/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..117ef3eae340e847885dd085f3c9448645f463f8
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_c49a2268db57a615eb6d+6fabf34d/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:03089fc97d2a44018a3fbd59503fd1385544d41e7b723a34cbfb452a8afadd83
+size 103424
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_c49a2268db57a615eb6d+6fabf34d/wrapped_neff.hlo b/neuronxcc-2.21.33363.0+82129205/MODULE_c49a2268db57a615eb6d+6fabf34d/wrapped_neff.hlo
new file mode 100644
index 0000000000000000000000000000000000000000..f1d5eaec675b158f7efcb62cb2153a4338105fef
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_c49a2268db57a615eb6d+6fabf34d/wrapped_neff.hlo
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e168843725d1098bda575488262d11df35198a102926075097f0e72bcf089794
+size 104320
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_de4db98e666fc07883ed+57ca39a9/compile_flags.json b/neuronxcc-2.21.33363.0+82129205/MODULE_de4db98e666fc07883ed+57ca39a9/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..cace416bea747457567f75595f02c6c788d14ca4
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_de4db98e666fc07883ed+57ca39a9/compile_flags.json
@@ -0,0 +1 @@
+["--target=trn1", "--enable-saturate-infinity", "--auto-cast=none", "--model-type=transformer", "-O1", "--logfile=/tmp/nxdi_test_a1024c95-4824-4d39-9861-90ff01abe0a8/compiler_workdir/NeuronLlamaMLP/_tp0_bk0/log-neuron-cc.txt", "--enable-internal-neff-wrapper"]
\ No newline at end of file
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_de4db98e666fc07883ed+57ca39a9/model.done b/neuronxcc-2.21.33363.0+82129205/MODULE_de4db98e666fc07883ed+57ca39a9/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_de4db98e666fc07883ed+57ca39a9/model.hlo_module.pb b/neuronxcc-2.21.33363.0+82129205/MODULE_de4db98e666fc07883ed+57ca39a9/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..e6084ce6d845ff91b01c0cbfd758d24019f38ef9
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_de4db98e666fc07883ed+57ca39a9/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a99355de7043422bc03a8df795afecf586999ec744407bc4fbffe63f002ba2df
+size 1931
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_de4db98e666fc07883ed+57ca39a9/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_de4db98e666fc07883ed+57ca39a9/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..53a13ad044c42fa3fdaea8a7eb0dca67991079b8
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_de4db98e666fc07883ed+57ca39a9/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5297e8bcd422f08b2e0af1489094bce2f207484dab7451821fab6beec7a6a924
+size 134144
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_de4db98e666fc07883ed+57ca39a9/wrapped_neff.hlo b/neuronxcc-2.21.33363.0+82129205/MODULE_de4db98e666fc07883ed+57ca39a9/wrapped_neff.hlo
new file mode 100644
index 0000000000000000000000000000000000000000..1c3c36b17848f1c01a0706b8ef059775b3e8640f
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_de4db98e666fc07883ed+57ca39a9/wrapped_neff.hlo
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5c442cb3c8af766a2963c862133e56866205aae958401c938851645bd70fc111
+size 136222
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_e22b5a814ee72d89608b+ea37e1a6/compile_flags.json b/neuronxcc-2.21.33363.0+82129205/MODULE_e22b5a814ee72d89608b+ea37e1a6/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..2e01732701efcc4a9af20e98873c3e3508e67bf4
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_e22b5a814ee72d89608b+ea37e1a6/compile_flags.json
@@ -0,0 +1 @@
+["--target=trn1", "--enable-saturate-infinity", "--auto-cast=none", "--model-type=transformer", "-O1", "--logfile=/tmp/nxdi_test_0b6ada24-7772-422d-89eb-fa025b21f005/compiler_workdir/AttentionModelWrapper/_tp0_bk0/log-neuron-cc.txt", "--enable-internal-neff-wrapper"]
\ No newline at end of file
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_e22b5a814ee72d89608b+ea37e1a6/model.done b/neuronxcc-2.21.33363.0+82129205/MODULE_e22b5a814ee72d89608b+ea37e1a6/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_e22b5a814ee72d89608b+ea37e1a6/model.hlo_module.pb b/neuronxcc-2.21.33363.0+82129205/MODULE_e22b5a814ee72d89608b+ea37e1a6/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..a11cc062167102b0f4faa9fdf16388d72d9e0015
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_e22b5a814ee72d89608b+ea37e1a6/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4590b82eb167fd46963abdbb717cc18cb9511961a3393d100eeacdbdc45b0243
+size 11280
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_e22b5a814ee72d89608b+ea37e1a6/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_e22b5a814ee72d89608b+ea37e1a6/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..baf7d7aa7d9627440dc497338e7ad5027985639e
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_e22b5a814ee72d89608b+ea37e1a6/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2890df215c2cfc41bfce07d5a44e04b906ddc04c930efc1159c4a43a1f1f6696
+size 1444864
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_e22b5a814ee72d89608b+ea37e1a6/wrapped_neff.hlo b/neuronxcc-2.21.33363.0+82129205/MODULE_e22b5a814ee72d89608b+ea37e1a6/wrapped_neff.hlo
new file mode 100644
index 0000000000000000000000000000000000000000..45f1944df278b5fb6cb9741c7360f69f70cb3c64
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_e22b5a814ee72d89608b+ea37e1a6/wrapped_neff.hlo
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9db018e5c4a0957630b8ecca51b2bf4e2ae8e988a596acbf2a42a58d9dc3c394
+size 1447845
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_f19ecf2e2bcc3a4b1bd9+a02c3a36/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_f19ecf2e2bcc3a4b1bd9+a02c3a36/model.neff
index 2e304f5f34c4473ccaf6e0d3745f4d1f6ee5e8e9..3c306c10c871c48496d696082a2943964c10939b 100644
--- a/neuronxcc-2.21.33363.0+82129205/MODULE_f19ecf2e2bcc3a4b1bd9+a02c3a36/model.neff
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_f19ecf2e2bcc3a4b1bd9+a02c3a36/model.neff
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6fee6df0d8e815c51f82b7158798b7add293f37bb80ac24f1c373fbcbbed9710
+oid sha256:c24f8a8a17f98924056937c429dc942fb7eb20b256091a121d75c2b7cee3595d
 size 1547264
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_f19ecf2e2bcc3a4b1bd9+a02c3a36/wrapped_neff.hlo b/neuronxcc-2.21.33363.0+82129205/MODULE_f19ecf2e2bcc3a4b1bd9+a02c3a36/wrapped_neff.hlo
index 88a0d5d3f2831cf021957ecb3454157426d77484..669b603d077941939591b47ebf14c49d9f6e6447 100644
--- a/neuronxcc-2.21.33363.0+82129205/MODULE_f19ecf2e2bcc3a4b1bd9+a02c3a36/wrapped_neff.hlo
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_f19ecf2e2bcc3a4b1bd9+a02c3a36/wrapped_neff.hlo
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3179cb717e4aabf2107e6f14781201172d17fb10c114b830d04a7472bf9b009e
+oid sha256:0e558717ccc8f47d226617fc449fae05f0f0f2209f618dca4b18f3f732b195a8
 size 1703455
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_fde2b1111ad10bd7504e+a02c3a36/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_fde2b1111ad10bd7504e+a02c3a36/model.neff
index 2ee7685225d01109cabca2c77360e3dd24b7722b..fe4fbf0a45ca6640d54ee04f69e32f6fd1589354 100644
--- a/neuronxcc-2.21.33363.0+82129205/MODULE_fde2b1111ad10bd7504e+a02c3a36/model.neff
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_fde2b1111ad10bd7504e+a02c3a36/model.neff
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f6b2df4e44f711cd0e1f04d8a201cecb4eef47cea703205520508d24777b3bb6
+oid sha256:9e487c1cbe54818225ff482fc2fe4dba83692a4d96b2af8e1009003612806ccb
 size 2100224
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_fde2b1111ad10bd7504e+a02c3a36/wrapped_neff.hlo b/neuronxcc-2.21.33363.0+82129205/MODULE_fde2b1111ad10bd7504e+a02c3a36/wrapped_neff.hlo
index ba9bf7cd4202c74d051ffd415c643db6b20bf348..08b52e85d5cea4bf3aca9b95a523ff9c723dca9f 100644
--- a/neuronxcc-2.21.33363.0+82129205/MODULE_fde2b1111ad10bd7504e+a02c3a36/wrapped_neff.hlo
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_fde2b1111ad10bd7504e+a02c3a36/wrapped_neff.hlo
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3c71e9d7a2ed1c070836de52d8db409df98b800fc373e8694162ad067655ae0f
+oid sha256:6ffb2ce4faa67cb457150d722f4263068943930cab8e706c3c63f972868c63fd
 size 2174311