diff --git a/.gitattributes b/.gitattributes
index 6852c2fc66a95e610a9202eb49c55fb00fea3651..17c66a89242ba39c395b833a53abd2b08c99a7e7 100644
--- a/.gitattributes
+++ b/.gitattributes
@@ -14802,3 +14802,15 @@ neuronxcc-2.21.33363.0+82129205/MODULE_ec8190569be823d3a7e0+78204f11/model.neff
 neuronxcc-2.21.33363.0+82129205/MODULE_ec8190569be823d3a7e0+78204f11/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.21.33363.0+82129205/MODULE_fed7336848aa3eedef35+3da1c170/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.21.33363.0+82129205/MODULE_fed7336848aa3eedef35+3da1c170/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.33363.0+82129205/MODULE_291400d642b5ad75c433+ab70272a/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.33363.0+82129205/MODULE_291400d642b5ad75c433+ab70272a/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.33363.0+82129205/MODULE_2c57dc97d3eca428336b+b02d6f1f/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.33363.0+82129205/MODULE_2c57dc97d3eca428336b+b02d6f1f/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.33363.0+82129205/MODULE_8beea6718ed2fea04975+490b9be1/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.33363.0+82129205/MODULE_8beea6718ed2fea04975+490b9be1/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.33363.0+82129205/MODULE_d9c41545be3dd4360c60+fc60ae29/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.33363.0+82129205/MODULE_d9c41545be3dd4360c60+fc60ae29/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.33363.0+82129205/MODULE_e2a55b2ffbb844debb55+f8e432bd/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.33363.0+82129205/MODULE_e2a55b2ffbb844debb55+f8e432bd/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.33363.0+82129205/MODULE_e9f45339826914e7685d+573f850f/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.33363.0+82129205/MODULE_e9f45339826914e7685d+573f850f/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/e000ebe2224a5d805b7a.json b/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.4.dev1/llama/unsloth/Llama-3.2-1B-Instruct/e000ebe2224a5d805b7a.json
new file mode 100644
index 0000000000000000000000000000000000000000..d33e0e446abc0abf477501f55e36e1a2e5e5bea8
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.4.dev1/llama/unsloth/Llama-3.2-1B-Instruct/e000ebe2224a5d805b7a.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": "5a8abab4a5d6f164389b1079fb721cfab8d7126c",
+    "continuous_batching": false,
+    "ep_degree": 1,
+    "fused_qkv": true,
+    "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": 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_04d083e7ddd421914edd+a02c3a36/model.hlo_module.pb b/neuronxcc-2.21.33363.0+82129205/MODULE_04d083e7ddd421914edd+a02c3a36/model.hlo_module.pb
index 0e09de6a37cf6cddcee906bf7dc92a8fd68cc432..02d5635e470b37da95faf60ddd821a40309a57cb 100644
--- a/neuronxcc-2.21.33363.0+82129205/MODULE_04d083e7ddd421914edd+a02c3a36/model.hlo_module.pb
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_04d083e7ddd421914edd+a02c3a36/model.hlo_module.pb
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ca7cc6d8e758f625d6f0c885073edb0b271b36fd6298ee026e8192a6764df484
+oid sha256:ba35978ae34a2d849c0f0b731c412c2ae336c89307133c39f2913ef746911eaf
 size 426072
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_04d083e7ddd421914edd+a02c3a36/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_04d083e7ddd421914edd+a02c3a36/model.neff
index ee1fde88ee18063fba83eb4148ca448a684dbbb2..42500a25ea4c84b4e929fa0371b274050992b38b 100644
--- a/neuronxcc-2.21.33363.0+82129205/MODULE_04d083e7ddd421914edd+a02c3a36/model.neff
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_04d083e7ddd421914edd+a02c3a36/model.neff
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:762757cc5c4594f1bb9da7f513be76d1e60d1afc33e95d07135b0161c4d886ad
+oid sha256:b2f2e1fd51f5a6cc33ca4f98bf96f931433e6bf7fcc58f11194f6cd583681432
 size 2100224
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_04d083e7ddd421914edd+a02c3a36/wrapped_neff.hlo b/neuronxcc-2.21.33363.0+82129205/MODULE_04d083e7ddd421914edd+a02c3a36/wrapped_neff.hlo
index 551c1615d262ee3f91cfb7e91f38eda966cf90e2..c30cb28a9b79256f3c217e730c37318e90097235 100644
--- a/neuronxcc-2.21.33363.0+82129205/MODULE_04d083e7ddd421914edd+a02c3a36/wrapped_neff.hlo
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_04d083e7ddd421914edd+a02c3a36/wrapped_neff.hlo
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:60ea1cef0048c2c53649e2bb82a3ed6fc5949255b01379d34dc30037683711e9
+oid sha256:c1bf1adbd4dd98e2a5bc158dd7a3844a3ea3e7fd33b9182c1bd50ae1af57c6cc
 size 2174311
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 6fc9b243f7ee317965247faff8f6007fc4ad300c..1078db196ac58f46f60b8017f3431069abe05965 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 09b91781af48a861e31743589e322a2a983e373a..e46b0c2c968614f50c9525572690242067beb214 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 a4d27b722800cb268d8db04fb62a5aa4c81a71eb..aef7158559bce9afe0e3cf39fddaca526290fbb7 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 8f053f862545b56db4dfb10a2cfa5e2510900c78..161d6bf8d6cf9b0f96be16b14bf9fe8b7a21025e 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 d2b496c8a4c9b1ff4330c51fb530920df4b1b5a7..a1a635cfc8a9ec3e3b47dc185ddf0ba303bd547e 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 1ac1636f6fed9f3825371d7650fc37c3578de102..57f5e34f96c5b746ec5eb00f251bbeb9b2c379f7 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 05ac88e0a469ff89150137387085ce3339d92aa4..73fcc86149bdf98291dc1d49178c8cb1b1e3cb63 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 b8ede32a9ce40a09813bdae40e83cbb2ad15ddf3..ee9cc2989351a709be43158e320203c19ccb6613 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 3ccdbb9e115cb46436e232f34066026b84614364..a5613acde3d11679b9fa0a41f28ef7db1df20b56 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 175134d23e770f9876f40a16decbfe83e2053bf0..ca57c363ea8f65915e5f8e9090d1c1061ed4d08b 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 7a99f78b8da3bc906b326b16cfd3ae05cfe356cf..96801ab6f9269f478a725f46b7d523a9cb1abbc9 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 e291307878ecc8194c874c7ea8ca4ceff8c57988..f0291eb1682d35cab0b3883f6eb59077bdc1ea95 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 3773e7e4ccb503e0607670273d3861ab1a42cea5..827a4454dda697a16c40d324a847478c17235f5a 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 aab53569bb2740b395a7271fc6c4b8f005c0c51f..90cf8aa4f20bdbb307e8b53effea5894837f3e0c 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 3a04e8484b3bbe5f3068ea7e0655f7b2a06ee02f..da6ae8e9161b15626c2e85d2dbb5aee9fd1f5799 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 22aae2f696bf9e80c8bf30bb2c11fa6728df6321..76be0ac0ffa59fb203a39e10ae00546397209d8e 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 d3c3efba2ff18c0939394fb5f22788b0bc80dca3..6947380fb198e325bb204e6a45eb6c1f715e006f 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 ad07ad32ec8de33008dad35109738c899580873b..161e1c1526eab00bd1f95939b436bae24e17261e 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 2d364592ca6f469a2e8e9986f8ba52fe61af363b..9461c2745122493248b3ae611ac4171217e85082 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 0ce167f276a4e80dc0e03cb9aafd2b88008e6d0f..427a3aa2edeacdd0d972d93aaa7e311b83bfa4b5 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 e92a94046c6aa5a11b151c6646509574ecaf1caf..fbbd29394d3c2dfc8aff4b8dcace20e3e8ddcfdb 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 15f70a48f7a6dc8f8d3fdd1989812ab1a6081748..9ed8b41a7ad6bf8215c176e7a711f1074a4b23c7 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 7fcdcb5b05f95bc2befdb593d606f55edb6d686b..7f90c2838f4e10ae35e0ce3ebb0b2df0a1d5bf16 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 c40ae8729d3a4a84cc5c3b8b62686e79967e39aa..56acaaeae6207a1100550e096a276408023cafbf 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 0cb53f111b795f9f326e4b410d2a13e5fd785e8e..8521c3661b173efc0609ea29ece252e06dbe8a3b 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 264f175ea60e34bc462d1e89fb17e5017716318b..3459a5a5ab866e8e21968c9ef686ff51761bc12e 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 0337809c13fd92e36e7d41da425dc68316142455..a1ae3a272e6764449cd2433d4aed16073e1751f8 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 8b64aee9cd072a8ce6aced7beb9aab4315ed0ca7..9885d626bf76fc20b3b8e794ae4344dd045274fc 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 8173dd980f342158f9dac0bfab54a1037eae0d7a..9a6d18ef6d3bc13d77ff38153ac835cf1fbbfb2e 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 bbd41979e7c9ca3afe2ecb6e3f1542ed07701e04..c5a0ec9ebb27fe483240ad1e1abb4a5b7346b856 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 3ca44fe213ab78eb8058d8ca127e0be6244ae984..bd13c281792e61ade7ff856fad075017e0d1ba84 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 808f5ded2438e3b56524a79a1edba7ab3a22ba86..2b882d02be56c27c6862ba2017e0bf7eca34780b 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 e7d1258e88407ebe49a057851af24d0ce1570ecf..f9fee26657aeb02bd09fce9dd735eae7fdb15308 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 21269e6cfb5f9e8eb23062c6f2f5e4fee911995b..674f844bdc24f7b63ee897613d94d11067fc215d 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 47ac5a826cb57dbf3824de0755a3f65e3b195fef..d1142be161affa65f2b52aca8b9939f555d5ce21 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 729b66a8a8d9aabdf65ebc03f6893f169a76a60f..36dc5913c25e0b6fb0e59111988aa6c5e726b126 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 9d7e4493f649bcfff18d8399b938f4d3696d3d66..413c5d91f7de7571a505dc8fc024a3683423f69b 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 d86d500920050499edd5456d3e8f225c222bebd9..cb8571e90b0a0e6f341e8ab8449eae9ee03d251b 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 c0b659c2e4dc5f2fe381355a8e3be884fb80d4a1..024552464c271686905a82e5845dac02b8304c1a 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 8def4cba9e0bd8727c9a8b32cdff482c94660142..a59ac28800fa4c8885a28e2e775375701be1b950 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 a3d19d9b81b99381f84f67e78625efd03f577856..e417d287fb4e77eaa8e90fd41111b43bd22a2a07 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 fac73d1e6a57723f68d319c906edafa7688ee361..061978e02907e7c1afe05f46c89f7164e409f7ac 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 260e78bd39beb595374a538c2ba4258011828bfc..079aa2a020de5aad2e62b534870fc6482c806683 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 ac41543d74039559ed83e53b4f60319a1a99227a..5ebc3eaeaf3b6a03e40ae9890fab8d6660e2bf80 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 4b2c17bd6cac8f9d1f1f51fa500ccf5ccdd2e6fd..5bdd68b21b0e5755ef80e0368f8fa160d7afbdd5 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 52adfe790fce1f326e9ae24e176cf72a25332866..8bf314c2d37547c37f2c7da9290bee3cb92adaad 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 25194be0ce0a3074b872d718e2732e2763746ae9..495b23a5f7a4b9204c82fd477298758f4d119d52 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_1df28a40cf4145bdfcae+24129607/model.hlo_module.pb b/neuronxcc-2.21.33363.0+82129205/MODULE_1df28a40cf4145bdfcae+24129607/model.hlo_module.pb
index 4d2d52b719ee95069817408f27c1b116e509195b..784a3ee3f203e74d81da6e0636587eae686e78dc 100644
--- a/neuronxcc-2.21.33363.0+82129205/MODULE_1df28a40cf4145bdfcae+24129607/model.hlo_module.pb
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_1df28a40cf4145bdfcae+24129607/model.hlo_module.pb
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c8512bf4e6aeb6df80fb45dc7894b2d89f3f2b4b170d276d4c3314ada0ddeabe
+oid sha256:186dab3857ec413ea82efe060aa6ff3019614f5934e8fc61ad4d338750d2288d
 size 593552
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_1df28a40cf4145bdfcae+24129607/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_1df28a40cf4145bdfcae+24129607/model.neff
index 40f4b4ddaaa70c2238730eef5c86cd9ce01f606d..93a722b3a3138154db6bf031ae5f043ff6124bf5 100644
--- a/neuronxcc-2.21.33363.0+82129205/MODULE_1df28a40cf4145bdfcae+24129607/model.neff
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_1df28a40cf4145bdfcae+24129607/model.neff
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b08f8d2b97c97f7110a50b3404f5e3719ab8b1c47928621acf663f3c0a5d07b2
+oid sha256:e47179548ff270cb7c5dbf254d7b67ea6fa1dcd0e57b3be76432ba270eb2c76d
 size 3769344
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 105b276331dd933eedd9822c420aa84fc2d6372e..560f6ba7e258c88bcedf6d205689011b111d987c 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 8e541c1a570b184ab604d0ee5b8bdafe07f183c4..d18749e631a45823fd3bc085b0befe5afc2cb374 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 1f1380185f9186f8ac623d3445b553aa522d1f1e..39fba3865d31395efd9cd2854aa4c813c39866eb 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 584433f8bfe4ddaec9ba384714210de58a0a4d49..59d707b7648e742e3df3dced43ee2c777604a108 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 44da94d15c68ecb176624daaa37c07cd05b5d1da..9d6d02cb03ab0b3dd2ad9eeef73e35eb24a77ec6 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 832796a6e406c2cf4b4e2c58dfd067d6e94a4fab..2cb75e9cfac8c64bca5437eccec9e4e8687f775c 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 ae6c2586d942a10f94ea2235448df68397d0ba2a..3e79948b2607fc7bb17983be882e6d06dd65086a 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 cc1398a66b95cce7e90b3ab6f31e472c21e48572..25a2517d583094f51837d599656f390f4f668bac 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_27143ff5b0e3602becba+24129607/model.hlo_module.pb b/neuronxcc-2.21.33363.0+82129205/MODULE_27143ff5b0e3602becba+24129607/model.hlo_module.pb
index c60900ba381154eb791edc399fd642d82e2d7a8b..551c5e8fbfd3c3fc6eb28e28eafc28a666525df1 100644
--- a/neuronxcc-2.21.33363.0+82129205/MODULE_27143ff5b0e3602becba+24129607/model.hlo_module.pb
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_27143ff5b0e3602becba+24129607/model.hlo_module.pb
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f4da7912903ac15c84f4ae7dd2bb2dffde034a0754b5746a0fbd328ea034f552
+oid sha256:533f33597e64d3998f8c1e1e3c9c75ab4c0b8a8eda6faf3a88bf1049a0048bce
 size 474402
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_27143ff5b0e3602becba+24129607/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_27143ff5b0e3602becba+24129607/model.neff
index c9618c98b7fec4e8d09648963ef6b82b5d05a7d1..2cd59e40c1ef4397bf449b042e647e4f5134128f 100644
--- a/neuronxcc-2.21.33363.0+82129205/MODULE_27143ff5b0e3602becba+24129607/model.neff
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_27143ff5b0e3602becba+24129607/model.neff
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d947016590971cae215cdccdc63138c7a2576148fc5073d4ea6eb6f5fd78dd10
+oid sha256:8173430a6c94b1b6db86668758d37a5f9e57b28772e64a13eb46d630a4abe609
 size 42322944
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 dee77903bf80398ddc52be36c94e3a675372f35d..50bc35a37e6f7b60076bea24b13ffbf186036ac3 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_291400d642b5ad75c433+ab70272a/compile_flags.json b/neuronxcc-2.21.33363.0+82129205/MODULE_291400d642b5ad75c433+ab70272a/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..59515781faff5178e25ba9c564ea204336a42d86
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_291400d642b5ad75c433+ab70272a/compile_flags.json
@@ -0,0 +1 @@
+["--target=trn1", "--enable-saturate-infinity", "--auto-cast=none", "--model-type=transformer", "-O1", "--logfile=/tmp/nxdi_test_0b50f051-4d60-4984-b3f9-17e95bd54730/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_291400d642b5ad75c433+ab70272a/model.done b/neuronxcc-2.21.33363.0+82129205/MODULE_291400d642b5ad75c433+ab70272a/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_291400d642b5ad75c433+ab70272a/model.hlo_module.pb b/neuronxcc-2.21.33363.0+82129205/MODULE_291400d642b5ad75c433+ab70272a/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..8496781848c02ecd308b5b2c5a3c342b702425e7
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_291400d642b5ad75c433+ab70272a/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_291400d642b5ad75c433+ab70272a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_291400d642b5ad75c433+ab70272a/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..2877d5db20606ec1562b484e645181cc42c8c3f8
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_291400d642b5ad75c433+ab70272a/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0e022c7ae4a8a43ea5e7889a1dab92b8b7815c71b243dafa06d36b5dc96875cd
+size 246784
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_291400d642b5ad75c433+ab70272a/wrapped_neff.hlo b/neuronxcc-2.21.33363.0+82129205/MODULE_291400d642b5ad75c433+ab70272a/wrapped_neff.hlo
new file mode 100644
index 0000000000000000000000000000000000000000..148954f34251eae88188db7e2dec355ed7b4a0ea
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_291400d642b5ad75c433+ab70272a/wrapped_neff.hlo
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8bfffc5a5b5146a15ad9f5615b18918592c50b6eb8bceb58d097d1292ea82ec2
+size 249608
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_2c57dc97d3eca428336b+b02d6f1f/compile_flags.json b/neuronxcc-2.21.33363.0+82129205/MODULE_2c57dc97d3eca428336b+b02d6f1f/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..63b891e0ed4360e2de21b41eadd59714a953cb72
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_2c57dc97d3eca428336b+b02d6f1f/compile_flags.json
@@ -0,0 +1 @@
+["--target=trn1", "--enable-saturate-infinity", "--auto-cast=none", "--model-type=transformer", "-O1", "--logfile=/tmp/nxdi_test_6555f51c-4d32-40c4-bb09-8e2ea941571b/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_2c57dc97d3eca428336b+b02d6f1f/model.done b/neuronxcc-2.21.33363.0+82129205/MODULE_2c57dc97d3eca428336b+b02d6f1f/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_2c57dc97d3eca428336b+b02d6f1f/model.hlo_module.pb b/neuronxcc-2.21.33363.0+82129205/MODULE_2c57dc97d3eca428336b+b02d6f1f/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..a11cc062167102b0f4faa9fdf16388d72d9e0015
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_2c57dc97d3eca428336b+b02d6f1f/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_2c57dc97d3eca428336b+b02d6f1f/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_2c57dc97d3eca428336b+b02d6f1f/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..2dd8505b2b1e905aa1f2a2f46ca4cd9231b9cf40
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_2c57dc97d3eca428336b+b02d6f1f/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:73b55fd5ec128941bae204eaae6848f9428fe4309f7898c199fcc921b3761d00
+size 1444864
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_2c57dc97d3eca428336b+b02d6f1f/wrapped_neff.hlo b/neuronxcc-2.21.33363.0+82129205/MODULE_2c57dc97d3eca428336b+b02d6f1f/wrapped_neff.hlo
new file mode 100644
index 0000000000000000000000000000000000000000..84886bea1d1eaeda916ee1a53c106744c1962239
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_2c57dc97d3eca428336b+b02d6f1f/wrapped_neff.hlo
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a7f6922405c6059c23c9ca39929da06734556a6afddc3c880da347a2bd7c224e
+size 1447845
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 ae8d1529449a5556f610fe1015a1880c36312bf8..b63698803c16be4846c65b9946d8cd2e0b3b32b6 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 5166f1728ad28658b7c03c426b85effee553b0ec..34065bfbe801270bd75ead5cfea912e62bf6a490 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 c63d9114429b4fb83fa766010a33dfd3a6e8e5d6..a47d9c3d42c7784f328aa679ef565136fea00f1b 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 1c67a914de65704afc76a3ed42699561be1c5bce..f1850fbff72bfed284e65615ae0ece1a3f42cc68 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 a3acb171f298f850e474f9cffc919b1271255150..ea69c01f4d68a5970bcad40fcf72a8ca3b300be2 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 d73529fae541d8a70b96d52c15f34ad5d71f8f6d..1325f27124c0925dd2baa8565b68373c11c84445 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 33552c08011b9d2ff0c6048c1d317f9d1e690b7c..0786ffcb81db648dacd47cb59164bcfa0b61ebf8 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 7f0e841e511cc6130eeefa0e8365cba51d1dbcae..d8bd0ccc4cb6a015880bf525f3653bb759b1a40b 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 3a737dde1df398ec82d5be5ece2386f9d766cc0d..69b54f9816a3a8be6037443b27ec41c61e0b6d07 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 736c3eb2f65b49c6fe137e34f9e04a595b2402b6..1fe87f9f2f705ac581c8ad64dc67fefe5ce04755 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 cafa75630c4d229d0e4453134487a381188f97f2..abace410929203c8fcd9a9dc53f33bccfe5c011a 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_5397223456911199516+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_5397223456911199516+e30acd3a/model.neff
index e20844dadf2d97b20818539e8c0b58d2b0497348..ed7fb97f4ffd08a454a3318e687da53478fdb00b 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 6ef75c76042e00dd60534fa5ca604e7bf457f7f5..7383be9f6bb09b283e2fe7d8d50791737f2d8c2a 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 c3155a902b666cbe5016a9c2fe402d9360be8a4d..cf7ade24726d1a95d8f464aca33c85aa9c969cae 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_63438d1b3946efd1c46a+c15df11e/compile_flags.json b/neuronxcc-2.21.33363.0+82129205/MODULE_63438d1b3946efd1c46a+c15df11e/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..5d8fa62a0623d5c522408a5d136bd166dc55eaa1
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_63438d1b3946efd1c46a+c15df11e/compile_flags.json
@@ -0,0 +1 @@
+["--target=trn1", "--enable-saturate-infinity", "--auto-cast=none", "--model-type=transformer", "-O1", "--logfile=/tmp/nxdi_test_c120441f-1554-44b7-a045-bec8026c4962/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_63438d1b3946efd1c46a+c15df11e/model.done b/neuronxcc-2.21.33363.0+82129205/MODULE_63438d1b3946efd1c46a+c15df11e/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_63438d1b3946efd1c46a+c15df11e/model.hlo_module.pb b/neuronxcc-2.21.33363.0+82129205/MODULE_63438d1b3946efd1c46a+c15df11e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..f1e2473c54cffac949ecf31d3b6ec636c290545b
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_63438d1b3946efd1c46a+c15df11e/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_63438d1b3946efd1c46a+c15df11e/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_63438d1b3946efd1c46a+c15df11e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..173cead72063551b1ee65839f2d76aea97829a2e
Binary files /dev/null and b/neuronxcc-2.21.33363.0+82129205/MODULE_63438d1b3946efd1c46a+c15df11e/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 f1b54c12bfa350744143b491281f579092256c6f..a7f0e04bb10382c38282b82a92d13054edb45f5e 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 888de92d4a2bb0ace05f6ba66c9d37a1ab780151..fd5c0202fedf63294412fb9f753a355f222fa0c5 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 c375c9cbf575b6e37d2693b1468ea9a9a013d2a9..02d64ab1cbc2d1d3f57364918e530a417187c7a8 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 80423c9369eb398b0ac765aeaffafaa9c4269bb4..8f6cb51f0b6f59c630462f03c17358935f3a1b3a 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 db8409bc23c06cbe2d7ddcfe8560cf9b39ac83aa..17e8d6c7003aa1bb413ad64848aa4a943703042e 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 cf0148ff77986772aec35a2442cb32255ecb242d..afe276f427813ca72a7fa4abf4e9954e704e2308 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 dc0e394226968fa31c56461d3426e904c4820fc8..e0675518d0e29d1b02c295deb0f2ab5432f6c235 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 f5557b4268e59d15e130b5ac3a648d60416ba021..919b30612a73bb7882140bf3520b38fea61630d2 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 f17d337daade487261ec151fbc1e62981c5d0e0a..c8fc9534d50bcd91a6a35c8b07404eec6fee3362 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 34fbb6fc6198877b872d21e9ec302e86c07ed64e..978e42945151d1ac38a0a55dbd8fde0acc117f4d 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 090beec320ea457d249de50f9a70f9030364f3c5..4814b1b4e26350771d02cb58b677f94c07cc27ae 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 650249609f67cf68383596604c5054c2cc7431b5..ef8612a9b99d4bc0c0de606d4bf68dbdc6d9f16c 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 ad1812fd814b4d8a86f9fbba97512d3a031ec708..6b25ff337de6f0f7ba8f8b1df1bca24b1e024778 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_8beea6718ed2fea04975+490b9be1/compile_flags.json b/neuronxcc-2.21.33363.0+82129205/MODULE_8beea6718ed2fea04975+490b9be1/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..535d8e9c708e7bfc65473caafe207f3c3ade68fe
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_8beea6718ed2fea04975+490b9be1/compile_flags.json
@@ -0,0 +1 @@
+["--target=trn1", "--enable-saturate-infinity", "--auto-cast=none", "--model-type=transformer", "-O1", "--logfile=/tmp/nxdi_test_3e64ee13-5fb1-4c22-9ad3-09d6983bc390/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_8beea6718ed2fea04975+490b9be1/model.done b/neuronxcc-2.21.33363.0+82129205/MODULE_8beea6718ed2fea04975+490b9be1/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_8beea6718ed2fea04975+490b9be1/model.hlo_module.pb b/neuronxcc-2.21.33363.0+82129205/MODULE_8beea6718ed2fea04975+490b9be1/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..6a2752c25c9f7e1092e2b55c010694e168610ce3
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_8beea6718ed2fea04975+490b9be1/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_8beea6718ed2fea04975+490b9be1/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_8beea6718ed2fea04975+490b9be1/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..c3127366adeb1e20940e9d7a762dc5c18ece40db
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_8beea6718ed2fea04975+490b9be1/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8d779cd3be22d6fc0b9fa7ac723c115906ce6b202f4eb8542a834b2f7db810b8
+size 103424
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_8beea6718ed2fea04975+490b9be1/wrapped_neff.hlo b/neuronxcc-2.21.33363.0+82129205/MODULE_8beea6718ed2fea04975+490b9be1/wrapped_neff.hlo
new file mode 100644
index 0000000000000000000000000000000000000000..4ca146b3e47fac8da6414e7cc06db1035c7e0bc5
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_8beea6718ed2fea04975+490b9be1/wrapped_neff.hlo
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a55be484714fd74d3c0b90dda9f3601e55514ca9671113f5322718800e3cba56
+size 104320
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 81ade31991dc1e92294d57cd7d19a59ce951b1a1..36e0d05a2d387feb3c4be3c2283f2d4b28ccb394 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 35704c3bed8def02ceaff64d5c8c4ed874668145..65dd5d26240002b4af00c27331c68a3a9dc9d1ae 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 838192b2d461384d60c793c401a51125541c041a..bc452c37427818775658cf16e5a84b61ffb9ceaf 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 70ba9590c6b64fac18a222f029c81815d0b4924e..e8274ff8a9fc6b3de6e05d92189bc926393b533e 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_ba6231346e52f5ca827b+a02c3a36/model.hlo_module.pb b/neuronxcc-2.21.33363.0+82129205/MODULE_ba6231346e52f5ca827b+a02c3a36/model.hlo_module.pb
index 7a1d77a94199152750a0d9ef9008e4af87353eb8..9eb026e722a4ee122a16fd9be4a9de66e2d1e292 100644
--- a/neuronxcc-2.21.33363.0+82129205/MODULE_ba6231346e52f5ca827b+a02c3a36/model.hlo_module.pb
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_ba6231346e52f5ca827b+a02c3a36/model.hlo_module.pb
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4e8b912560da43fc635c4044cfb7d1f65300cdc0afe421d9c6d0b386f4838634
+oid sha256:1844c82063e3d5290e8111f7e2aeabe387926990dff36fad5d51c6230dede8fa
 size 590808
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_ba6231346e52f5ca827b+a02c3a36/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_ba6231346e52f5ca827b+a02c3a36/model.neff
index 8418cc1c009477a727dc349bf20408b487a15b5b..56dd67a20259383872a764ec2ad160c98a7204a7 100644
--- a/neuronxcc-2.21.33363.0+82129205/MODULE_ba6231346e52f5ca827b+a02c3a36/model.neff
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_ba6231346e52f5ca827b+a02c3a36/model.neff
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:022c5bfaf034dff82d9958b91d3a444a48494d496b589f25c87f7cb6148193e7
+oid sha256:0625de1836ed48174dadb2f553476a9ae6ea7a92ac66c8407021220bd91cb1d2
 size 1547264
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_ba6231346e52f5ca827b+a02c3a36/wrapped_neff.hlo b/neuronxcc-2.21.33363.0+82129205/MODULE_ba6231346e52f5ca827b+a02c3a36/wrapped_neff.hlo
index c25c102ea5fc2356b3e7f166a6b23bc1cf006eb2..20e553d45f82355512338b2e2005072f0cf1991e 100644
--- a/neuronxcc-2.21.33363.0+82129205/MODULE_ba6231346e52f5ca827b+a02c3a36/wrapped_neff.hlo
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_ba6231346e52f5ca827b+a02c3a36/wrapped_neff.hlo
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0cb1d4ae05fbb5c716576213608a211006302872feef27fdd9c1dc42fde598d7
+oid sha256:304ec5997c34b2ba237f0774feb55999d09bb3d4ec8eeeb4137461534a2ad0da
 size 1703455
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_d9c41545be3dd4360c60+fc60ae29/compile_flags.json b/neuronxcc-2.21.33363.0+82129205/MODULE_d9c41545be3dd4360c60+fc60ae29/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..799d83ea7321938b1c45d705d50b6e22cd4877f6
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_d9c41545be3dd4360c60+fc60ae29/compile_flags.json
@@ -0,0 +1 @@
+["--target=trn1", "--enable-saturate-infinity", "--auto-cast=none", "--model-type=transformer", "-O1", "--logfile=/tmp/nxdi_test_a63904be-30a9-4d86-b3b5-9b5c319e4952/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_d9c41545be3dd4360c60+fc60ae29/model.done b/neuronxcc-2.21.33363.0+82129205/MODULE_d9c41545be3dd4360c60+fc60ae29/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_d9c41545be3dd4360c60+fc60ae29/model.hlo_module.pb b/neuronxcc-2.21.33363.0+82129205/MODULE_d9c41545be3dd4360c60+fc60ae29/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..be5670c6258a72b3253c262c04b0a3c084a1810d
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_d9c41545be3dd4360c60+fc60ae29/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_d9c41545be3dd4360c60+fc60ae29/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_d9c41545be3dd4360c60+fc60ae29/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..3bd51f75202d06797344e82ece4ece4debe21eda
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_d9c41545be3dd4360c60+fc60ae29/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1811e275cf968902507086bb9aa98b081c039cf06db8f311b4d1ec11c0c8e212
+size 328704
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_d9c41545be3dd4360c60+fc60ae29/wrapped_neff.hlo b/neuronxcc-2.21.33363.0+82129205/MODULE_d9c41545be3dd4360c60+fc60ae29/wrapped_neff.hlo
new file mode 100644
index 0000000000000000000000000000000000000000..d42ae1ada3f438320b9ec53ec5a38eabd9ab0c65
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_d9c41545be3dd4360c60+fc60ae29/wrapped_neff.hlo
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:79f31fe01e34a786e1117be1b8667a3546f99868dd267301281798c7644bcd04
+size 334452
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_e2a55b2ffbb844debb55+f8e432bd/compile_flags.json b/neuronxcc-2.21.33363.0+82129205/MODULE_e2a55b2ffbb844debb55+f8e432bd/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..186e8c4c11e8c63957bc2e1e1a4f584f8be5a8c8
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_e2a55b2ffbb844debb55+f8e432bd/compile_flags.json
@@ -0,0 +1 @@
+["--target=trn1", "--enable-saturate-infinity", "--auto-cast=none", "--model-type=transformer", "-O1", "--logfile=/tmp/nxdi_test_9150e3bd-397d-4380-b6ff-99a69c4256b5/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_e2a55b2ffbb844debb55+f8e432bd/model.done b/neuronxcc-2.21.33363.0+82129205/MODULE_e2a55b2ffbb844debb55+f8e432bd/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_e2a55b2ffbb844debb55+f8e432bd/model.hlo_module.pb b/neuronxcc-2.21.33363.0+82129205/MODULE_e2a55b2ffbb844debb55+f8e432bd/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..07bf62460bc7813e4a95f208598e2fa4ae551d41
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_e2a55b2ffbb844debb55+f8e432bd/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_e2a55b2ffbb844debb55+f8e432bd/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_e2a55b2ffbb844debb55+f8e432bd/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..f220afb58b139ebcf7e303da84cd05e87a58d017
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_e2a55b2ffbb844debb55+f8e432bd/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:daf764cbb2cd783abaf3d6104e3901c4c6f16e9f6f5a8efdf50c1ee47717502b
+size 267264
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_e2a55b2ffbb844debb55+f8e432bd/wrapped_neff.hlo b/neuronxcc-2.21.33363.0+82129205/MODULE_e2a55b2ffbb844debb55+f8e432bd/wrapped_neff.hlo
new file mode 100644
index 0000000000000000000000000000000000000000..411da124e5614ddf3fa0dd6a98bbcb75f7b83203
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_e2a55b2ffbb844debb55+f8e432bd/wrapped_neff.hlo
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:85334f38650281aee0dbaa543d2b0d383adf442e3890542e3e0d1480ea3d5946
+size 272962
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_e9f45339826914e7685d+573f850f/compile_flags.json b/neuronxcc-2.21.33363.0+82129205/MODULE_e9f45339826914e7685d+573f850f/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..7264a3dd14c578c8af5d55897fc49e1821aed9ab
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_e9f45339826914e7685d+573f850f/compile_flags.json
@@ -0,0 +1 @@
+["--target=trn1", "--enable-saturate-infinity", "--auto-cast=none", "--model-type=transformer", "-O1", "--logfile=/tmp/nxdi_test_e16b1fd7-a5da-4d8c-8f92-6ef2d2e932a0/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_e9f45339826914e7685d+573f850f/model.done b/neuronxcc-2.21.33363.0+82129205/MODULE_e9f45339826914e7685d+573f850f/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_e9f45339826914e7685d+573f850f/model.hlo_module.pb b/neuronxcc-2.21.33363.0+82129205/MODULE_e9f45339826914e7685d+573f850f/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..e6084ce6d845ff91b01c0cbfd758d24019f38ef9
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_e9f45339826914e7685d+573f850f/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_e9f45339826914e7685d+573f850f/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_e9f45339826914e7685d+573f850f/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..df761539acbbceca9484caa0f0c8c53566d85f54
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_e9f45339826914e7685d+573f850f/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bd58faaf6d0b5a35ef740d2438f75d97691dece68dab364000b1c462d153281b
+size 134144
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_e9f45339826914e7685d+573f850f/wrapped_neff.hlo b/neuronxcc-2.21.33363.0+82129205/MODULE_e9f45339826914e7685d+573f850f/wrapped_neff.hlo
new file mode 100644
index 0000000000000000000000000000000000000000..3e4e05ce2c34646f0f53bc633881054133573cbc
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_e9f45339826914e7685d+573f850f/wrapped_neff.hlo
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9bb2fea85eb0585a042bddaaccc087bab17bb7d0363a6705a049fffccc436f35
+size 136222