diff --git a/.gitattributes b/.gitattributes
index 8db7f8ef22d04ee2f398caf1280d87e1d4b831c5..57d3df4ea6f3716048527fccb78c1d55cbfccaca 100644
--- a/.gitattributes
+++ b/.gitattributes
@@ -14679,3 +14679,15 @@ neuronxcc-2.21.33363.0+82129205/MODULE_e1f1656f0920f8eaa2cf+d1033e07/wrapped_nef
 neuronxcc-2.21.33363.0+82129205/MODULE_113098831bb4bba15e6c+24129607/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.21.33363.0+82129205/MODULE_71106d0443c3197ff0a7+a02c3a36/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.21.33363.0+82129205/MODULE_71106d0443c3197ff0a7+a02c3a36/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.33363.0+82129205/MODULE_471b99120f603ce294bf+7057e9b1/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.33363.0+82129205/MODULE_471b99120f603ce294bf+7057e9b1/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.33363.0+82129205/MODULE_50f9204751de3f6cb66a+5dff2d72/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.33363.0+82129205/MODULE_50f9204751de3f6cb66a+5dff2d72/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.33363.0+82129205/MODULE_6021c0ffec11dc78e457+cbaca00b/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.33363.0+82129205/MODULE_6021c0ffec11dc78e457+cbaca00b/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.33363.0+82129205/MODULE_a13cc5665e7d154b7429+695bec87/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.33363.0+82129205/MODULE_a13cc5665e7d154b7429+695bec87/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.33363.0+82129205/MODULE_a60ee6952450ce742f72+f00715ce/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.33363.0+82129205/MODULE_a60ee6952450ce742f72+f00715ce/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.33363.0+82129205/MODULE_e0dc711f77a602f2be1f+25971132/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.33363.0+82129205/MODULE_e0dc711f77a602f2be1f+25971132/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
diff --git a/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.4.dev1/llama/unsloth/Llama-3.2-1B-Instruct/c7bdc28395608bf889f0.json b/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.4.dev1/llama/unsloth/Llama-3.2-1B-Instruct/c7bdc28395608bf889f0.json
new file mode 100644
index 0000000000000000000000000000000000000000..ef0c9a67bbd5387d464408421a4a789127845b6a
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.4.dev1/llama/unsloth/Llama-3.2-1B-Instruct/c7bdc28395608bf889f0.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": 4,
+    "capacity_factor": null,
+    "checkpoint_id": "unsloth/Llama-3.2-1B-Instruct",
+    "checkpoint_revision": "5a8abab4a5d6f164389b1079fb721cfab8d7126c",
+    "continuous_batching": true,
+    "ep_degree": 1,
+    "fused_qkv": true,
+    "glu_mlp": true,
+    "local_ranks_size": 2,
+    "max_batch_size": 4,
+    "max_context_length": 4096,
+    "max_topk": 256,
+    "n_active_tokens": 4096,
+    "neuronxcc_version": "2.21.33363.0+82129205",
+    "on_device_sampling": true,
+    "optimum_neuron_version": "0.4.4.dev1",
+    "output_logits": false,
+    "pp_degree": 1,
+    "sequence_length": 4096,
+    "speculation_length": 0,
+    "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/MODULE_10217061096959125489+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_10217061096959125489+e30acd3a/model.neff
index 28e711fb7d5cecaa069fe028aaf088dbd65ca358..fb4d3b4651830ae2d5b62100ba54c2f3f2ccbbd0 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 96538d052cc568c00a78302f169100f7faaca2e2..05be6a630403a216a1cf52becdbe467769c1e493 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 e8c7f399f906931ee2ad0de76e3eb0a7db4073d2..2b24c268eb636c27295f7bee20ed5df978f43751 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 fc48d78198aedb5bedd24e60b3d21a71f07b072e..ded9b2b982f7986eb9c29cd97207544d3f829a7c 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 f7edd77139251f274899a146040c8795782a3338..e271eb26492f2cdc4b3bc17c4483fb1b78612646 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 775e9a55d970e235b9c17b8713e908c87e81ece5..abfec0dc0c5c2c775465ca076b3d02dc109bd712 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 a36451ed18f0275d19af82f9563882204c14673e..c862b6527ef9fd3b5b1234107550870d30e50d68 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 f3d8893a7402e92a7a7e8bffee6ca8eebdbacb58..8d7b3c9e634569e420a61cb966302e7661e81e21 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 abd1bc237fe0ac534c2e7f248f2bdaa8d015f379..3e43bef24e75237200b5fe03ccc119e6369e39aa 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 a04645dc724d3b8330533dc8f5b2f6c154bb2ea7..a2fd5ff82f4103fd414dd178ca70ef4ec37cf0fe 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 23aee8b9a7016c3cef8b060bd3b6518db39d1e79..d809c72a786eb7016113cdd994dea25b89325485 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 f1586c3b2ef08e9ef497fe6ffcc8c26a78dc71c7..4453ddbb12343da87edc9300d755a258c5201bca 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 c541e9a7684294d68c7add3bbc8d6ecee4989808..b43b3bf249d8068096c11fc60984df384240b8f0 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 f9a1ae4206031e6cd4b772b0f41f2c00656ef218..c7a085584afbf07bdad4721e5d6ec163a3ae8b70 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 0ea74b1fbe8637dbf329f46cf03a86b64c3ce60b..b29f779386b6692c4e65dfc22b41cc387e31018c 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 01f54b0af5f3b33cc55e51788ba4fec1da837f84..14e1c54a7f20a00472f1beeaad89397a8275e9e7 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 323644a679f5a9418aca14f5e45a7f3602c76eeb..6aa7d0d04abf82f23395da65a84a7834b9236b9f 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 b39c0d4ab8c8e2d0ba52c1211260a65debafdd86..b5856997d0a100854527df66f37aa5e25eb2c633 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 2fa9381e3a790229d6561dfe85332c85e7371f83..8495465b6008a40d4a7440c01d99fc8754b6cd6f 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 03fa5d0b102d941f9bdde0b40668ddfb8276be41..432e00a14ce9b1fb79643e29d70aa075d4c4c752 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 d5482120d1491e06875177583b482d0f5a2d5d3f..f0a5503ac94c5286e5c591993fd8dfa62f45b63f 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 7cdd0f17f53683b9f09829aa150bc7d15dfc84bf..a8a827f51b1c0939fd29bf9d77c44abe1748737b 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 c856a15da61a869feb5609fa9831925b7a9527c9..acdda90b98b846915b3dfddb49c982c84ae9a814 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 89d6222da8c37f7158d0d69da0797626dcb20e08..ac53f291a16c0b3374d4d896287e25b794d1a293 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 9d43351ad577eaa86cd5c4b0ed80480d545669d6..158077f52b74fd502ed9d4e5e494fa8b14ff19b0 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 135b2666b46aa5556d2867691e753c27eeb7321d..e78465dff0222da01a32f7206e49bb7426d573e6 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 a8f37f18f2815d0f69c8d26b1cd446f5e6595332..eae3cfdafd30db913da70c50cf619d0cf11b3cd5 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 1c355be45ce2b8df3cdc653b07cf2a7746dc0255..650f710e2eefaf1ac2a9ed1da41ad879fa120b46 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 3b98d24b2a7833f85af7f883d6085ef0957d2f2f..180f318d216853f0ee3037d4236ca74ec9fb7585 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 4081f033646f8aa5890703b74cf3fdb34164b010..250e634f706f35de4a888ef9f4603e19a1f9b7f6 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 a901c468c757c863b10496a1bd433c881af24a21..a6f9f1996136b250b9b77695c8f7e0d854238dfb 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 6aaab36f17171885443bd3a26f608a6965575733..5b2794ed887e412f32480ce07847e79b4ba71ba7 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 08e8c28adc13ea6fa01cb8a0f209d3638a2165bc..2dc6e9a8c585edab23e38c2508bcee7cc1286873 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 bcebe0764cd13bebd8760feb7c7a45341a0d6348..6f21cfe1ae60f82d2a5ae8a65399e966caa5da31 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 d4b3210a579dc28157ebaaf92e8be7705aa47ea9..6490297f97bb2c438f9ae97f6772b0a249324396 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 cd71e06090cacc7ec3a0c16528a647e4be86cfc3..eae8cb39a6a674b70cc9faefac7b7ce23f54f61d 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 aaab617f45cc38cdd7091e13ffc2c164aca027bb..d66f0291c5d48a8a493d7764fd70d709f05583a5 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 d2a823147e0d42047031f33fbe48c57b3b6875b9..de10751a9210b3ea5507c352256b2bd24e7f4a9c 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 f15a6950d6091e56e226c959e12a75d565bc64ef..68020a019dbac8d4859fd5688c76d7c316b26193 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 a4e29ec934f6d5e6c5914bbc845a95afe7076070..53b87d55f7936b1cbfb58f742a2c370d45cbe3cd 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 16d0a782066595fddc097052b923b0445881d305..b2afab1062dafbb483f78490e7d49228ea3156f4 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 694113d4d292f1974ba24d96f4ff9411c970799f..19dfbb7257ea77e23256753876fb6bee93293741 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 c909401184a4fca5cbb14641bd049be99eb703ca..fa0fc5e50e943a776f201690f2d13d0febfca605 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 8030319c46a02c9ab3b1f76cbeeb4b63ae6e73ae..50d569ac63173b736fd95952318c32e25c191474 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 61c06c28676450e5ad3d712e8eac8e4c843680db..76a58d6b32f53eb1a668175437d64a9b0c462a6b 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 b066ff48de850ef9e1448099f2845558dceed3dd..5a08313959ed634274770dd5811c1367cfd649b9 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 2464411630af54ede059d4a93709151a0567744c..d2cf3cacccd6dcff5474c77e703629f1c9e1efbb 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_2157613318347839507+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_2157613318347839507+e30acd3a/model.neff
index f9ee866bf69eaff8790008acc0571fb2f74262da..243f51e964297e94ac23cb06789d09a4ef353e13 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 bb959d947c6e11b943bb200437ea3abdb2427d91..5dad5fc59a24480ff1ca184d5ff773c87d4bae23 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 f0af402338775f681587e66c3bf99525d983eaa9..7d14fd99a8dfe2fe3aac0b77b2a5d1565d80d1a8 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 fe9e711c380b7530237659d481afcd8e0261de25..6db66274a0f53594890862e9c87d988fb80d6990 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 64f44d02e77de00c3e4611a0720998c4759643ac..a33d6455f282e86e328af0286533155f4ee1db18 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 cb1cc2450a227e05eb3722e453dc709ff6ed4937..deff4b489c33460ff576c96d22bbeb44e95fa3f7 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 a170e55a0c00dcf2d18739c73507fa304e07354e..bc147e967271122c10a6c12ef104e5977bd0f6c9 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 87118d7804511de2e9c5690df74c7974b9251bfb..3aba5d24f06d4652c8b685aedd9db16fd5c55c0b 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 41ac9a3f7913164acbbe4572bb2196ffa09a77e7..3ed60ffdd521764d22cf871b528ffa1263d496a0 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_302950992539679003+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_302950992539679003+e30acd3a/model.neff
index 3c20acf11ae3587ed9fbff9bd75fd9a6a2ee0edf..e7e7970eea0775562e1e2d1751a43f12a069b471 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_4013984518400530928+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_4013984518400530928+e30acd3a/model.neff
index c37964e2bd7679bbc3217f8d2f163db239b72afc..3ff5274ddbe8115c9ea7867d9fe4bfd87734d1b7 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 8a8cae4124e60768790e119528965fa3f661cb1a..d41d224010694d9946413779c16454e40cc49694 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 b3a6af05407f86fb707af4d401962c08e27e5cc8..3521dc6d7bbfdd977e16d7bad3e6636b4b7986f9 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 46986d5248ff85c6fbeecddc6f4b97bd5bcafc66..fad6de3c976ae07863c8ae97321406d7852afc99 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 09316f8e9bc9b4ba7a850396e73042e2d2a13d62..5b00a1d5cfc1dd045d37f7ac46477928f6977685 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 341d9d5d536ce5c8c9407a9025720c2b9fb83228..e401ee510f54eab905cd7931c82c326d6a69eb7c 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 a488892415f07dfb4e3ba88fa55c3e84b2d7315a..9d8d689363f7771abb13d1d9341d80dba7ce7fd3 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_471b99120f603ce294bf+7057e9b1/compile_flags.json b/neuronxcc-2.21.33363.0+82129205/MODULE_471b99120f603ce294bf+7057e9b1/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..8beb98af049105efc76dcc55313f8c2d19fc94bc
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_471b99120f603ce294bf+7057e9b1/compile_flags.json
@@ -0,0 +1 @@
+["--target=trn1", "--enable-saturate-infinity", "--auto-cast=none", "--model-type=transformer", "-O1", "--logfile=/tmp/nxdi_test_159809f9-af5b-43fa-b1af-07a9ce3a3e92/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_471b99120f603ce294bf+7057e9b1/model.done b/neuronxcc-2.21.33363.0+82129205/MODULE_471b99120f603ce294bf+7057e9b1/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_471b99120f603ce294bf+7057e9b1/model.hlo_module.pb b/neuronxcc-2.21.33363.0+82129205/MODULE_471b99120f603ce294bf+7057e9b1/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..be5670c6258a72b3253c262c04b0a3c084a1810d
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_471b99120f603ce294bf+7057e9b1/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_471b99120f603ce294bf+7057e9b1/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_471b99120f603ce294bf+7057e9b1/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..c6e315c45d7db35eec104fd92730dd3e841e866e
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_471b99120f603ce294bf+7057e9b1/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f4009d88a4050f0350159eae9063ba99615a46a6ae63770ae53b5ea090357127
+size 328704
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_471b99120f603ce294bf+7057e9b1/wrapped_neff.hlo b/neuronxcc-2.21.33363.0+82129205/MODULE_471b99120f603ce294bf+7057e9b1/wrapped_neff.hlo
new file mode 100644
index 0000000000000000000000000000000000000000..72598aca5571ce0aec4c8ef3351b482d13f89798
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_471b99120f603ce294bf+7057e9b1/wrapped_neff.hlo
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:941a1ae8af3a77dd2ed2fb87e8761d835cb2e0c07d1f465fbd2895a242a7722d
+size 334452
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 3bc896d2c04fbf9244ac9d81163199ab994bd14c..a87e5c515987c9f1f3ac9b4eb5e81bea83f2ff4e 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 e3760818526a16d2b478e4496484be044d1d5975..1727954a830caaca8bc0fd07b66755e006a113f5 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 fa942d78e6f1b0c28e1b7f2bff4ffd68d73b1503..bce77287f64ba1bb286b505ca877cad76d63b2b5 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_50f9204751de3f6cb66a+5dff2d72/compile_flags.json b/neuronxcc-2.21.33363.0+82129205/MODULE_50f9204751de3f6cb66a+5dff2d72/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..01052a5a4a9a6a71f10fdac45e89fbea563b7738
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_50f9204751de3f6cb66a+5dff2d72/compile_flags.json
@@ -0,0 +1 @@
+["--target=trn1", "--enable-saturate-infinity", "--auto-cast=none", "--model-type=transformer", "-O1", "--logfile=/tmp/nxdi_test_e71a9f78-8c49-4638-90e7-7121f95aa9bd/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_50f9204751de3f6cb66a+5dff2d72/model.done b/neuronxcc-2.21.33363.0+82129205/MODULE_50f9204751de3f6cb66a+5dff2d72/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_50f9204751de3f6cb66a+5dff2d72/model.hlo_module.pb b/neuronxcc-2.21.33363.0+82129205/MODULE_50f9204751de3f6cb66a+5dff2d72/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..e6084ce6d845ff91b01c0cbfd758d24019f38ef9
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_50f9204751de3f6cb66a+5dff2d72/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_50f9204751de3f6cb66a+5dff2d72/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_50f9204751de3f6cb66a+5dff2d72/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..5f5bcade00dfc4ccb93d543b95edf1347d01fa45
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_50f9204751de3f6cb66a+5dff2d72/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4bf314c031b3935094b6e0105c52f1153e11f3e30942e1aa70adb77ebf18c453
+size 134144
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_50f9204751de3f6cb66a+5dff2d72/wrapped_neff.hlo b/neuronxcc-2.21.33363.0+82129205/MODULE_50f9204751de3f6cb66a+5dff2d72/wrapped_neff.hlo
new file mode 100644
index 0000000000000000000000000000000000000000..93e6304631c01f8913d4944732fc755037cffaca
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_50f9204751de3f6cb66a+5dff2d72/wrapped_neff.hlo
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cbbf9a4111b5845a284ff3be60278141a51b6b4c20c08a8f3a7160b05fe0b025
+size 136222
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 24ec73923764414f31f6176556a490fc8c5c865f..70de752684c96a92cc4138368202ebabf4b0babd 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 eccc5ca3cd4a96d99eaf8fb3d06a6dcc7912116c..ffae6ecd3593d503c7aa3671992ac940e7159c3a 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 d2a35e2977f625e831d13f43db69183ea0df0393..49f6a29b1452b8694296cd3ff4ad9ae4c957308c 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_6021c0ffec11dc78e457+cbaca00b/compile_flags.json b/neuronxcc-2.21.33363.0+82129205/MODULE_6021c0ffec11dc78e457+cbaca00b/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..68043f8c826bbd893412a7a4080ba13516153a15
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_6021c0ffec11dc78e457+cbaca00b/compile_flags.json
@@ -0,0 +1 @@
+["--target=trn1", "--enable-saturate-infinity", "--auto-cast=none", "--model-type=transformer", "-O1", "--logfile=/tmp/nxdi_test_f931d687-6d7c-4d18-bc18-4cf24c32056a/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_6021c0ffec11dc78e457+cbaca00b/model.done b/neuronxcc-2.21.33363.0+82129205/MODULE_6021c0ffec11dc78e457+cbaca00b/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_6021c0ffec11dc78e457+cbaca00b/model.hlo_module.pb b/neuronxcc-2.21.33363.0+82129205/MODULE_6021c0ffec11dc78e457+cbaca00b/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..8496781848c02ecd308b5b2c5a3c342b702425e7
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_6021c0ffec11dc78e457+cbaca00b/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_6021c0ffec11dc78e457+cbaca00b/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_6021c0ffec11dc78e457+cbaca00b/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..a22f964d33358097f0ea1efaed19ee60be1cf506
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_6021c0ffec11dc78e457+cbaca00b/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5df86eaa884f3df0c15300ef036ae3dfce1b421bb455ab8d3c8360a5082ccef8
+size 246784
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_6021c0ffec11dc78e457+cbaca00b/wrapped_neff.hlo b/neuronxcc-2.21.33363.0+82129205/MODULE_6021c0ffec11dc78e457+cbaca00b/wrapped_neff.hlo
new file mode 100644
index 0000000000000000000000000000000000000000..ab94bada8afab818ea67c102f1f266a60b61629f
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_6021c0ffec11dc78e457+cbaca00b/wrapped_neff.hlo
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bf48eefa0ce4b159f6b7a37ce856c0b8ca3b39714840c2e05cb18811749621f9
+size 249608
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 968e7e35709d7af1a23204314cdb4a604eb0fafb..3a713a87e23e272c1c096b08035ae44ff532fab4 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 ab7cf883f4db7207afd820712064fc02922cc1a8..772c2f6f82fd23454b11cb1d53b8da41469533a9 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 1ad1f88f7226dfb9a49515125b6f3e1f1132c18f..9e0582a7c8b90623db0b32818d3a94f9fe1f9cb8 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 ac91f9d5f8c75407fb239734ad56e8fa3abfe45c..3b4cc421ee86dcee89d78e55c8dbb6947855d9cd 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 7759b3accb239b3624e50e517d9f082169c1d8d1..2fffcc899e42dcb35802a02c22ddb628de29c50c 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 b091ebebd573ab8f77c463cb57302eb42780ef46..49510ef141df67e3f038b87ec121e6e8579353e7 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 54bed5fbe893c56c1c163e100e4d9e016217da74..d93b9127936ebc11902cd7ca33a5ffb740428945 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 76f78c36750c81f188d47700ff9703b7a5442b60..4fdace68f831dd002f978ff82c1bccdc6ad4e36b 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 935d0f619548338ef08558ab287e21defa721821..fcde4e08bdf8f90e6fb681361eda2ffba74b9d6c 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 69dbde92d435954edc009a15116792dae31649c3..da8cabf61a17d3d701ca4f7ec9bb0c1bd95e4470 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 ebf782e9e9b205fa54943340a1358fe1f186036c..30fd5e2f7791295d458fae3ddd7d0e0d09b895e9 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 313ae7affe764331c23555e63dff43816c9f4a79..b13e74db6b3bec7b994e53d9c6309d7078cac8f5 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 af654cef1081034b223352930d6c1d4059d70d0b..3e9008b48501b9c6bd8c80481d7a5ff86d8ec901 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_9168843780052024308+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_9168843780052024308+e30acd3a/model.neff
index 4240fecb374129dfe0d8833a3dcd1b9b15d52cbc..8bc741aad68b61c10064ad06c314252a883c8dad 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 f14b398360560b48107ffcb485a9d1e1ebef2bb4..93e9c1e1d53d70e02978598cad7250f01bb9a3b6 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 8d01f0362fda4b40c81e6bd1102444692e651c1c..99939d1b3d9fd9b356eaf7bde9ea6fd6844c72ce 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 6b6f60eaeff7b5e3a3bd5290937072974cffbd82..7891c56743c331887cb92cf419b40b1175694473 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_a13cc5665e7d154b7429+695bec87/compile_flags.json b/neuronxcc-2.21.33363.0+82129205/MODULE_a13cc5665e7d154b7429+695bec87/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..7005cef356cfe5e14f4844d33ae4cff7122d15c2
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_a13cc5665e7d154b7429+695bec87/compile_flags.json
@@ -0,0 +1 @@
+["--target=trn1", "--enable-saturate-infinity", "--auto-cast=none", "--model-type=transformer", "-O1", "--logfile=/tmp/nxdi_test_d03955e1-200a-4942-9836-95a18cab45b5/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_a13cc5665e7d154b7429+695bec87/model.done b/neuronxcc-2.21.33363.0+82129205/MODULE_a13cc5665e7d154b7429+695bec87/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_a13cc5665e7d154b7429+695bec87/model.hlo_module.pb b/neuronxcc-2.21.33363.0+82129205/MODULE_a13cc5665e7d154b7429+695bec87/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..6a2752c25c9f7e1092e2b55c010694e168610ce3
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_a13cc5665e7d154b7429+695bec87/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_a13cc5665e7d154b7429+695bec87/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_a13cc5665e7d154b7429+695bec87/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..14a06cc1987099e36806736c6d543a5f082036d9
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_a13cc5665e7d154b7429+695bec87/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7e63d3fbb9595b3fa4cb085bf9ffcd5e0cc19c821c61826c04fe2db91472bc22
+size 103424
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_a13cc5665e7d154b7429+695bec87/wrapped_neff.hlo b/neuronxcc-2.21.33363.0+82129205/MODULE_a13cc5665e7d154b7429+695bec87/wrapped_neff.hlo
new file mode 100644
index 0000000000000000000000000000000000000000..aca8358aa92146cd05458e5ec9feb7a015e623c7
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_a13cc5665e7d154b7429+695bec87/wrapped_neff.hlo
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a8d6565429f6f758b416dc39538ceff0ef8e9c6e846362923f8b6b27293ca13c
+size 104320
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_a60ee6952450ce742f72+f00715ce/compile_flags.json b/neuronxcc-2.21.33363.0+82129205/MODULE_a60ee6952450ce742f72+f00715ce/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..f282c912d46019cde56608c1d77f20e79692d05e
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_a60ee6952450ce742f72+f00715ce/compile_flags.json
@@ -0,0 +1 @@
+["--target=trn1", "--enable-saturate-infinity", "--auto-cast=none", "--model-type=transformer", "-O1", "--logfile=/tmp/nxdi_test_79334c47-cfd5-4d71-96e6-96ebc5d4656f/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_a60ee6952450ce742f72+f00715ce/model.done b/neuronxcc-2.21.33363.0+82129205/MODULE_a60ee6952450ce742f72+f00715ce/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_a60ee6952450ce742f72+f00715ce/model.hlo_module.pb b/neuronxcc-2.21.33363.0+82129205/MODULE_a60ee6952450ce742f72+f00715ce/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..07bf62460bc7813e4a95f208598e2fa4ae551d41
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_a60ee6952450ce742f72+f00715ce/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_a60ee6952450ce742f72+f00715ce/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_a60ee6952450ce742f72+f00715ce/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..5440976b0e07b8cc84eb0e3aa993e9261192558d
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_a60ee6952450ce742f72+f00715ce/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6ab77c165e6bd93659829b57fa15665a7073439ed0a5e6ee3e6f628b385e4cf3
+size 267264
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_a60ee6952450ce742f72+f00715ce/wrapped_neff.hlo b/neuronxcc-2.21.33363.0+82129205/MODULE_a60ee6952450ce742f72+f00715ce/wrapped_neff.hlo
new file mode 100644
index 0000000000000000000000000000000000000000..925e7dd833ba36cac5819fa04ad4721ee9b2736f
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_a60ee6952450ce742f72+f00715ce/wrapped_neff.hlo
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:edb3014a1a3406b61514c09c3685d41be776b2bbdadb464c3a2b865d3f3ce19f
+size 272962
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_e0dc711f77a602f2be1f+25971132/compile_flags.json b/neuronxcc-2.21.33363.0+82129205/MODULE_e0dc711f77a602f2be1f+25971132/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..94f8171328ab44bb44f25ec6173fc944e7af11e5
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_e0dc711f77a602f2be1f+25971132/compile_flags.json
@@ -0,0 +1 @@
+["--target=trn1", "--enable-saturate-infinity", "--auto-cast=none", "--model-type=transformer", "-O1", "--logfile=/tmp/nxdi_test_f0e96e7f-9d68-4032-9666-dc77f08fb1ea/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_e0dc711f77a602f2be1f+25971132/model.done b/neuronxcc-2.21.33363.0+82129205/MODULE_e0dc711f77a602f2be1f+25971132/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_e0dc711f77a602f2be1f+25971132/model.hlo_module.pb b/neuronxcc-2.21.33363.0+82129205/MODULE_e0dc711f77a602f2be1f+25971132/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..a11cc062167102b0f4faa9fdf16388d72d9e0015
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_e0dc711f77a602f2be1f+25971132/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_e0dc711f77a602f2be1f+25971132/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_e0dc711f77a602f2be1f+25971132/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..d2132b6bf69fb6ee98d6fafcce75ea77377120d7
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_e0dc711f77a602f2be1f+25971132/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b05078b98a6b4e75abf2245eea204e15795571d0adee9afe0aa1b5414db4eecf
+size 1444864
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_e0dc711f77a602f2be1f+25971132/wrapped_neff.hlo b/neuronxcc-2.21.33363.0+82129205/MODULE_e0dc711f77a602f2be1f+25971132/wrapped_neff.hlo
new file mode 100644
index 0000000000000000000000000000000000000000..59843a1f3d5fb7f88c341d5d90f1f23bd996fd4d
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_e0dc711f77a602f2be1f+25971132/wrapped_neff.hlo
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:993ef0093066453a0dcc29caedf5423867360b52b3ec981ba60a12d9a7ac9a29
+size 1447845
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_e2b1af2a61862b81d7e4+e15b50a0/compile_flags.json b/neuronxcc-2.21.33363.0+82129205/MODULE_e2b1af2a61862b81d7e4+e15b50a0/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..a6cb8600d5669579343406239126c9003fa0642d
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_e2b1af2a61862b81d7e4+e15b50a0/compile_flags.json
@@ -0,0 +1 @@
+["--target=trn1", "--enable-saturate-infinity", "--auto-cast=none", "--model-type=transformer", "-O1", "--logfile=/tmp/nxdi_test_d6051fc0-df39-4891-8288-efb851b83b76/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_e2b1af2a61862b81d7e4+e15b50a0/model.done b/neuronxcc-2.21.33363.0+82129205/MODULE_e2b1af2a61862b81d7e4+e15b50a0/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_e2b1af2a61862b81d7e4+e15b50a0/model.hlo_module.pb b/neuronxcc-2.21.33363.0+82129205/MODULE_e2b1af2a61862b81d7e4+e15b50a0/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..f1e2473c54cffac949ecf31d3b6ec636c290545b
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_e2b1af2a61862b81d7e4+e15b50a0/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_e2b1af2a61862b81d7e4+e15b50a0/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_e2b1af2a61862b81d7e4+e15b50a0/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..52926f37fa9a69573778a60778b34857c87d2794
Binary files /dev/null and b/neuronxcc-2.21.33363.0+82129205/MODULE_e2b1af2a61862b81d7e4+e15b50a0/model.neff differ