diff --git a/.gitattributes b/.gitattributes
index 618eb3a57ea7bf850bb31e13a14cdf11b1843a6e..7b95b29dc64b15048efa90956e486ee5247b69d6 100644
--- a/.gitattributes
+++ b/.gitattributes
@@ -16687,3 +16687,15 @@ neuronxcc-2.21.33363.0+82129205/MODULE_99a4c8ce2f07c82d25d4+76eb1f45/model.neff
 neuronxcc-2.21.33363.0+82129205/MODULE_99a4c8ce2f07c82d25d4+76eb1f45/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.21.33363.0+82129205/MODULE_9cd1f2706a9a7386b4a1+edf8250f/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.21.33363.0+82129205/MODULE_9cd1f2706a9a7386b4a1+edf8250f/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.33363.0+82129205/MODULE_053e06ffe169b4c659fc+1f529cf9/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.33363.0+82129205/MODULE_053e06ffe169b4c659fc+1f529cf9/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.33363.0+82129205/MODULE_104a571a8528da5086cf+396e89c0/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.33363.0+82129205/MODULE_104a571a8528da5086cf+396e89c0/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.33363.0+82129205/MODULE_40fbbca6d04d4522a405+00862aac/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.33363.0+82129205/MODULE_40fbbca6d04d4522a405+00862aac/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.33363.0+82129205/MODULE_6823a7a9d96c090ebb77+4dcc0058/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.33363.0+82129205/MODULE_6823a7a9d96c090ebb77+4dcc0058/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.33363.0+82129205/MODULE_a8e0ba8e05a954f8347b+a5aa275c/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.33363.0+82129205/MODULE_a8e0ba8e05a954f8347b+a5aa275c/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.33363.0+82129205/MODULE_b81bbded724bb14afb4a+299b35f0/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.33363.0+82129205/MODULE_b81bbded724bb14afb4a+299b35f0/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
diff --git a/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.6.dev1/llama/unsloth/Llama-3.2-1B-Instruct/da6950729a5bb39e8b09.json b/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.6.dev1/llama/unsloth/Llama-3.2-1B-Instruct/da6950729a5bb39e8b09.json
new file mode 100644
index 0000000000000000000000000000000000000000..9394dacd4355f7db8d23bc51193db70dea7885d2
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.6.dev1/llama/unsloth/Llama-3.2-1B-Instruct/da6950729a5bb39e8b09.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": "0.4.6.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_053e06ffe169b4c659fc+1f529cf9/compile_flags.json b/neuronxcc-2.21.33363.0+82129205/MODULE_053e06ffe169b4c659fc+1f529cf9/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..53d3993c84cee8d2a0cf1ce0f66ba0312b988a54
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_053e06ffe169b4c659fc+1f529cf9/compile_flags.json
@@ -0,0 +1 @@
+["--target=trn1", "--enable-saturate-infinity", "--auto-cast=none", "--model-type=transformer", "-O1", "--logfile=/tmp/nxdi_test_51c9690e-c821-4720-b336-3bb21ce6ab6a/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_053e06ffe169b4c659fc+1f529cf9/model.done b/neuronxcc-2.21.33363.0+82129205/MODULE_053e06ffe169b4c659fc+1f529cf9/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_053e06ffe169b4c659fc+1f529cf9/model.hlo_module.pb b/neuronxcc-2.21.33363.0+82129205/MODULE_053e06ffe169b4c659fc+1f529cf9/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..b7f9f80de01bab3afae30f106439a7600609730b
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_053e06ffe169b4c659fc+1f529cf9/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3a4f4cdcd60e648ee1c6f483fd7d775daeec9d14a980ac23a3446e8df4ef38a3
+size 8979
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_053e06ffe169b4c659fc+1f529cf9/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_053e06ffe169b4c659fc+1f529cf9/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..92cb35c2be61da624a4666b6029172a8faaa77c9
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_053e06ffe169b4c659fc+1f529cf9/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1bc4cb159855e795ab36b8223e0dc30fd38fd7341ec1fdaa1b6e9c4416edc509
+size 246784
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_053e06ffe169b4c659fc+1f529cf9/wrapped_neff.hlo b/neuronxcc-2.21.33363.0+82129205/MODULE_053e06ffe169b4c659fc+1f529cf9/wrapped_neff.hlo
new file mode 100644
index 0000000000000000000000000000000000000000..3109ec5d67eb139b1dc226e58cdcb7be509ad4e3
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_053e06ffe169b4c659fc+1f529cf9/wrapped_neff.hlo
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d9f62569c6b6e053701cb34a63e84fc72e9fa575fa8c88003ea1400fc44c0f27
+size 249608
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 435c2abdd651364499b1f0ac62751c9350712d79..825718996a7729bbd821ac4978acb712a241e3ab 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 bde72ffaa2aa79846edaa5d93eb2f252e3972335..a8606cef4ecf1e11bde21fd15e59610ad4a47f83 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_104a571a8528da5086cf+396e89c0/compile_flags.json b/neuronxcc-2.21.33363.0+82129205/MODULE_104a571a8528da5086cf+396e89c0/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..0718d03a2e82fefebdff02eab79b09df1256c8c0
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_104a571a8528da5086cf+396e89c0/compile_flags.json
@@ -0,0 +1 @@
+["--target=trn1", "--enable-saturate-infinity", "--auto-cast=none", "--model-type=transformer", "-O1", "--logfile=/tmp/nxdi_test_de173fb5-5b60-47d5-a98d-5caccd41d0e0/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_104a571a8528da5086cf+396e89c0/model.done b/neuronxcc-2.21.33363.0+82129205/MODULE_104a571a8528da5086cf+396e89c0/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_104a571a8528da5086cf+396e89c0/model.hlo_module.pb b/neuronxcc-2.21.33363.0+82129205/MODULE_104a571a8528da5086cf+396e89c0/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..618358c2b96cde9d0fdd8a5363f4e3756584720d
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_104a571a8528da5086cf+396e89c0/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:42a756345a0e47699f8c540a039458d52f4072f93c3d432124a626815c5383b0
+size 14480
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_104a571a8528da5086cf+396e89c0/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_104a571a8528da5086cf+396e89c0/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..92de1f9da5234b5ce3d00d676056c96b8241f1c3
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_104a571a8528da5086cf+396e89c0/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7db7de816196dd9a2b1a05dcca2b36b4b255bd47471d461abb7b0a9732f309c4
+size 267264
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_104a571a8528da5086cf+396e89c0/wrapped_neff.hlo b/neuronxcc-2.21.33363.0+82129205/MODULE_104a571a8528da5086cf+396e89c0/wrapped_neff.hlo
new file mode 100644
index 0000000000000000000000000000000000000000..92b2e4d4d4612f052796dcffc42caa59897e5450
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_104a571a8528da5086cf+396e89c0/wrapped_neff.hlo
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:73b03521c97b7ff382596b28a58ae98a356f8d7ff447fd8189f224f6559b67e5
+size 272962
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 637b8accb2eeda5f6cb831487c8b0caea01467b3..dc993c479ba2e8eba852d67ec4921eecf2866327 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 2b7a4c287ebdb1656bfb7982d57981d0341d2575..8b0ed41d93842fb649a771f51cc3c67a50527add 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 94051db7208bf785d762b655108b74cd460abe74..5eeabe500332635f3de3b9a219afa23cbf345c8c 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 311c8ea2e19310e784b642eca461407e62d59c16..655c7df498a0eefe66e15357b3ff908c9970962c 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 9d37ef2d75514ffba611661c207b2e1eab5caa4a..dfc620176ff019c1743b5f2e7609f6281d673d8d 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 1acc57c3bee517306af3a2abdc45e0dd517fee06..77a913fc6af039c549045000394facddd423edd2 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 36e8233e52394b154fc69ca9ccd224e2afca1fe6..febd76225c7aed9f27a371387f682b56272aed23 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 3224d48163d743419823f06cf466285301238c9a..d4a3ee581bbf32ecbeaad6bd02576cd642901d7a 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 5a4f44d2b5f326b5d926a197f8ae748a949b5214..d3d4d5d6d3eeb7fbeff95f96565f070abb3cf475 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 c4833af468e77f1f559bc16a688887c01947d443..7c4d6d7d4dbcf65ab1bc7b6a5e8b1753dd9f71ef 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 f8c64737ff4d2c2d5c8498fbd53b74edf7bfadb0..bb21a30124bf2b6e2a420394e9acd193e720e383 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 a4aa658eeb8c2a78ab14c2ff1f81f99936d97645..14b8890877311c150270cfc660cdd68167e6ed86 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 962f243914a38d52790567f1e66c8eecc721cf67..c301e14ad2ca998cfb8b2dee9a5204b926ec12ba 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 7a0ed1d194e2440b728f85cfa0557248d5cba403..0ce737b31871905ff51341b98b0c15ad9915a77e 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 2bcc3a174ca6a5224cd0b87d2920c3f4cefd6f5e..00f597960b6520225b1b5700cd01b1b220819b4f 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 05c1e52235ab7443ab409ec9169ae0a67e6296ca..0b09bedd8655394baf2df2eb3224a0db93996a5d 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 29efb6f5b26cfcb9a8b2de24701df44966b62af3..faf81f22d0a607a35e35a655f448d08d0a99f477 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 40b2801575fb52827895d30ad3ef06e8a300899b..8e179b820ff158382253ada7846dbf3a09832f61 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 63641399c78893e9eef87e54e0b48c6b8af707ac..9783f73ca61cd64e76591a01487616e7f7f39435 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 7ff49e54b0ead1ab29eef887178269c401ad80b3..e5afcaf440bbd9e9e317f95f0c77e5c2deb7c86e 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 6fb7ea45af29ec97c08f7d652cf2f51ba57e72a8..19401ae5037a69bb26ad0560e9880950f812a7db 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 164524129c422bde0eef8df7f4b01f4c995748c2..90a90f5a20c75fde7eaf08370f6589aaad169bcb 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 754b4cfc257299803a760a2dd1749c0a4e0584bc..5050507c22afed7e641b2ca1aab3f3e8293fb043 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 ceb869334a53b3fabad158f7c273aa9f68fb2bf4..df0aea2de78b69bb39926ef966f754e7109b3ca5 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 28f1350d29f3dca2345a49605a8b2c014f4218ae..f8b195f534933983bb9126f78d65f31053fb3287 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 0de7ad99af42302f194956d0b6aeaa508972f197..a8272a8985c95a56f455ccf64776ed3f69324c38 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 4ca930b1861afb43646b7cde930742a2f69b6f09..4a79766a563798c53b1043a8c99c861c62c9f42c 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 0f6d7f7d458a415b8763dd785caeba448a847dc6..f8d8679292217de71fb8e743b63bd62b91445e77 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 fa23a80bc46721fd1914a725c185d4acec199226..5c3284f425688eb15a40bb03137e7317b495080e 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 47fbf97336fd751dfba553364f2f38aca61d61c9..c6c03386f27cdbc0c34a4d99d851211c6e3cb35b 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 6763a132fa1eec6635bae8c4a64937c18adb6a5a..29442f1d86d515a26b5e4a2a37d7ea7334187853 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 cfb01d83a442e0d7d002e492dca044cd87602f1f..52e198ed647f516462b33c7490b81d45b98bb98e 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 f230c24d6866c3858250ca318f135c63078b468d..18302b6ef279113bfb9beb4c920a35192d09d00f 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 bba0860034032c6c17790a205e3fb3800a09b239..a99bb1b22d0585c87ea3fd0290b51131e5070f02 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 cc4debad2a50d6ab5f7a1e3fa20f540c896c5de5..8bd3ed758f9175d2e7386f4f6f492bf319c460a1 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 1e90367a3e0aac53aa277109fc5977ca94d462e5..718b80e0c715865b723818041e5469de6c78f8d5 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 1c410478789dd90f75e4d479250d0194ce588427..fecb8387df20f7344a970375e68d18741dc3e665 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 4dbb02688ef9ae3ae2bc3eb7b6486f4c54bc3d28..7640b7e55db9882284deb3ba2ff1d101bbfd7ae9 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 3b40fd9bef1a1e504e18b6477961a755f9531b58..ef025fea120ea674f640202c14f25e9daf7aa661 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 8ef409e722d647a00cea116d9bdfc8c67fc795ed..301e52bd81e5211c8e12e9e38329433dc3242734 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 026fb1e3dd805e496b11bc29455df08f173b9ec0..5608fd418b1e67ff917846bcae9ebd6fc4d7d3b1 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 2d297c98ec43c00b3161f8d105320470ac0463d1..be91fa26932c76e729f5e6240c9ce5ec282202d4 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 b570d3c0319a32f03c576c845f735ae3cd04b983..3ddffa7f395bfbfffe9b7f26ab6fe128575e4f3b 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 a463c0011a8ece70b1eac21776bf8f19855889ab..1fefb3958aa04fcd6c795ba2a4a5dde1510ae5f4 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 0a20a239827b3f8d5d43590b20e10ce8cccf8a68..c750d6730d68e7dec17ba7ea898f853e707b1a9f 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_2132d7650a8d2d4fdb70+c80c3385/compile_flags.json b/neuronxcc-2.21.33363.0+82129205/MODULE_2132d7650a8d2d4fdb70+c80c3385/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..d600bb7a7ed005f746a538d193be8425a1812cb9
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_2132d7650a8d2d4fdb70+c80c3385/compile_flags.json
@@ -0,0 +1 @@
+["--target=trn1", "--enable-saturate-infinity", "--auto-cast=none", "--model-type=transformer", "-O1", "--logfile=/tmp/nxdi_test_bf38c405-6d3c-4256-860f-918a8fcef636/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_2132d7650a8d2d4fdb70+c80c3385/model.done b/neuronxcc-2.21.33363.0+82129205/MODULE_2132d7650a8d2d4fdb70+c80c3385/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_2132d7650a8d2d4fdb70+c80c3385/model.hlo_module.pb b/neuronxcc-2.21.33363.0+82129205/MODULE_2132d7650a8d2d4fdb70+c80c3385/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..f1e2473c54cffac949ecf31d3b6ec636c290545b
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_2132d7650a8d2d4fdb70+c80c3385/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_2132d7650a8d2d4fdb70+c80c3385/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_2132d7650a8d2d4fdb70+c80c3385/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..f81f96e71efed26f43482682cf62d449f6a35592
Binary files /dev/null and b/neuronxcc-2.21.33363.0+82129205/MODULE_2132d7650a8d2d4fdb70+c80c3385/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 e6497d8d5582abf6ca693933e63563ff4d5ed6a8..6d747f702578abd0c86faf364fd7aae985b75a58 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 309ce8c85e109b8d9e48b57ce6b4726c2313f95f..059d1f56012551cd227f0a6b3f2db12880ab72d5 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 8ffe66d286445bac1e54c8cf41e641e975f80ab8..c61876140be1db595e9269ee4dd48fc71e6741a3 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 7dab360da833f650a69d299ee87f72ce22eac321..944fc5b0ee99fb8494ff40bff11c2ae57577d5a5 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 dd70bd622e3c9c25642a5f3d25657ff29535657d..8775a02bee0b92d7bf2cf852a8c931190490ae45 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 7ef66754f77c43e7ab116ca92d23a07f00d1b256..7a76658f39b15320e7fbf2bcfe9710f2873f4d24 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 5f6764beb108725e0cbc9d6a89e3edf1f585db7c..c7bc4734d8281fac5d0b17faf5403c5c2184f966 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 1553c4008b17197b895c5a27d6785f16210ed8ab..af7071a0d83cfb70378fdd4ab64194161c5a5536 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 760252b18942ef703b3e5c97f523510f4aed3ef5..79e55d75b5413395d65e984c6f7e35895c5323ce 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_2811495285170804454+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_2811495285170804454+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_2ce3d3065088729c7d7f+24129607/model.hlo_module.pb b/neuronxcc-2.21.33363.0+82129205/MODULE_2ce3d3065088729c7d7f+24129607/model.hlo_module.pb
index 5b31a795fa051fe6382001027af513aab21cff12..1b8352e1ee9da452fd713c1a756c86482df2737e 100644
--- a/neuronxcc-2.21.33363.0+82129205/MODULE_2ce3d3065088729c7d7f+24129607/model.hlo_module.pb
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_2ce3d3065088729c7d7f+24129607/model.hlo_module.pb
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9a587d5ddb67a42533df39b1b05311d022b9cb500122dfc7b382e9f459a2b148
+oid sha256:e635e3fd90e62f2f22ffdf53f78d7bdd4fa50499755a35b844ef028afd5c49ab
 size 593552
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_2ce3d3065088729c7d7f+24129607/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_2ce3d3065088729c7d7f+24129607/model.neff
index 19d57ab036df5a355c0b32336d85a582ca5ff866..4c835106a85cb95f49deaf1931e2168c3e8e72e1 100644
--- a/neuronxcc-2.21.33363.0+82129205/MODULE_2ce3d3065088729c7d7f+24129607/model.neff
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_2ce3d3065088729c7d7f+24129607/model.neff
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a9a1689ed7d2b0223c8cb4efa62045beb080793c0aece10a790120c2720e031a
+oid sha256:2d2ecefc8ed2770f153acd71390244b54bda1ee5865d1b29705b1b1e8c67e471
 size 3769344
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_302950992539679003+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_302950992539679003+e30acd3a/model.neff
index 3be737032057c462d0b982aecdc05addec8ef013..2d962c9ac1967b53965c850071f9fe9005faac90 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 e5e8557c70eb56e9fb34eb0bf26d44221d8d89de..889ff47931ea3eb54475e077bd14ac4801f56d85 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 7c82691617d7c5fa30942e34fbddec88e112ad36..cd7d33786c3f3d8371a6df600cf344b876b1ae2f 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_40fbbca6d04d4522a405+00862aac/compile_flags.json b/neuronxcc-2.21.33363.0+82129205/MODULE_40fbbca6d04d4522a405+00862aac/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..b1bd41b512113b0c1df31911e7957dba125fa641
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_40fbbca6d04d4522a405+00862aac/compile_flags.json
@@ -0,0 +1 @@
+["--target=trn1", "--enable-saturate-infinity", "--auto-cast=none", "--model-type=transformer", "-O1", "--logfile=/tmp/nxdi_test_29a03076-720a-475c-bb73-16dd94bc8e46/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_40fbbca6d04d4522a405+00862aac/model.done b/neuronxcc-2.21.33363.0+82129205/MODULE_40fbbca6d04d4522a405+00862aac/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_40fbbca6d04d4522a405+00862aac/model.hlo_module.pb b/neuronxcc-2.21.33363.0+82129205/MODULE_40fbbca6d04d4522a405+00862aac/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..6a2752c25c9f7e1092e2b55c010694e168610ce3
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_40fbbca6d04d4522a405+00862aac/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_40fbbca6d04d4522a405+00862aac/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_40fbbca6d04d4522a405+00862aac/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..f5551667de5b8f942d6b272db62e4cb0edd0f575
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_40fbbca6d04d4522a405+00862aac/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:451af314773a8ca0b9859f591053d70a21c3c02853fee5c5e896854bbc39db82
+size 103424
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_40fbbca6d04d4522a405+00862aac/wrapped_neff.hlo b/neuronxcc-2.21.33363.0+82129205/MODULE_40fbbca6d04d4522a405+00862aac/wrapped_neff.hlo
new file mode 100644
index 0000000000000000000000000000000000000000..deaeaf9a9e099daf706f76e60cf39af7a1a85e66
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_40fbbca6d04d4522a405+00862aac/wrapped_neff.hlo
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c8d3f4a816b22bed72f563e3d82c84ecadc56eda8a3fe98328938a0b900c252a
+size 104320
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 abaa4018e5c85fd073d456d8ebc85c486e914e00..6dafd667c879807e2cf6aeb5066c174f99d9393e 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 d0f3f98f3ecbc0bbaf7c24a79f501ac46a5f6eeb..be060b96a512165362fbadc53d5b35844681565b 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 85146c5536f79b29558ae47c0e556acb8e160658..e6f1f78d55f28b7fb3aa5c11346b15901927d11f 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 6612f2f2d5e4211efda76ba1bd85948f6f5b600f..fe9ac2e6a56a2c7729bc626fac24f95aec30c4ea 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 e397e677b38621e4c4847b8ddcc19df5bda1a4a5..5b2d5620a4e6abc960d1e3123dbf430e1459010f 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 1dcaeeb9c80a58560d4d8dcc00880f104e091a86..23fc07456c16b151494f1baddd3b102db899de92 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 17d5eefff9e308359854f2d29fb64ef42205bafe..b959bdaa4737a6f82626fd9db436671a9fc03d77 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 04955733352a3926ebfd33d41f3bf9fe15fc2d70..b1088e4f8dd5c035952501497798be4f69f4ec11 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 7fde6cff05ee74e0fdf85e814d301a8d5ba16241..ae5ebdcec64ac037537823cc6689c00cfd2ff6f5 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 508930a68cb468573b210d3daf83e55b489cf8bf..256504c4a16627419f30ac189ab927cf09666417 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 23811d085c782a0f434b18f4af8e81860338cbbe..b421ed90f76bf316af1b9af21405ff347437f721 100644
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_5765916484866650909+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_5765916484866650909+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_6368409506294683105+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_6368409506294683105+e30acd3a/model.neff
index 5f3d71376dddf4c0d9223d029e32a15f5b45261a..e4c2de99367d40927501238f5c734b05669e5e02 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 b945bedf0b95a3c105e4c2f38554ec79810312fc..d6a72bba1fcd1c111557ac5296748ec99f8f74a2 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 04fa848502b723e268e14c32f7e5c94d090a22c6..f91f85fce31559653db0e9c293c4e9f20ac42982 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 840088dfe9afb15b81d0e8c53b71c3a83c436518..c90715bb823d824ca962467996ebfbf41be4bae0 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 e54f39b0b72dc0ea9fb135e44242a00e935bf67a..0b9b626509a3173430c6ffaf4a0bdcda026c46a4 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_6823a7a9d96c090ebb77+4dcc0058/compile_flags.json b/neuronxcc-2.21.33363.0+82129205/MODULE_6823a7a9d96c090ebb77+4dcc0058/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..dfd7c454dbfd328f336f6c1ca37ab9cc208aff87
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_6823a7a9d96c090ebb77+4dcc0058/compile_flags.json
@@ -0,0 +1 @@
+["--target=trn1", "--enable-saturate-infinity", "--auto-cast=none", "--model-type=transformer", "-O1", "--logfile=/tmp/nxdi_test_86ad91e2-c59a-4081-872b-619a7126d67c/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_6823a7a9d96c090ebb77+4dcc0058/model.done b/neuronxcc-2.21.33363.0+82129205/MODULE_6823a7a9d96c090ebb77+4dcc0058/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_6823a7a9d96c090ebb77+4dcc0058/model.hlo_module.pb b/neuronxcc-2.21.33363.0+82129205/MODULE_6823a7a9d96c090ebb77+4dcc0058/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..4bb45842ea0894b69b62a879499055cb6a5cb9ae
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_6823a7a9d96c090ebb77+4dcc0058/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:be2fc82d01eae877238b23e1fa62d9bd75d648e5ffdf58b8cf01e4f672c3e27e
+size 11280
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_6823a7a9d96c090ebb77+4dcc0058/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_6823a7a9d96c090ebb77+4dcc0058/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..f936d29614f5e08323ffd1a2e5dc23909621d530
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_6823a7a9d96c090ebb77+4dcc0058/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7dcb374d8c0cecc9a7ffbdc66d9db7186f9e536ba86cdf42f6055e3c7f4b2975
+size 1444864
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_6823a7a9d96c090ebb77+4dcc0058/wrapped_neff.hlo b/neuronxcc-2.21.33363.0+82129205/MODULE_6823a7a9d96c090ebb77+4dcc0058/wrapped_neff.hlo
new file mode 100644
index 0000000000000000000000000000000000000000..2bf218e14ac4050dc50b7f247bd65937990327df
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_6823a7a9d96c090ebb77+4dcc0058/wrapped_neff.hlo
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3934f3b5cbe744c6d6c94c266d4ece763d21e6b8989bacacd2a3271edfb9177e
+size 1447845
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 872e340fd0fc40da6e9d7d163003107cad7d1f7e..a29ff43eadf9716ce39ea68d6728c0f8096e8e05 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 7b1376525cc384529b32c647e86acf03681373cb..e28ba3d8d0f6f49c3ce88d65008043d16d7536f9 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 7cc9d35f53b0c1067c0c21b6b5f9865bed6a3bcb..0f8c130ca3b5ee8f3366ba6179cb35299f449795 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 de31f3165d9264d681c52b5f5ce10651aa56fa33..95b2d284a76dcdc59b903d9d12a330e993475b96 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 330bf4f60ffe2f50ecba95399305154cb85c071a..df3b471fb0d0249aef0cfbaf534ac8ba7aeb1e44 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 e754db87565601ad17a045d878b7b1c4dfdaa6bb..fa527a501809e483bdb3cd2eb00188481e0ea792 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 8f125c2be65638a2692a23d1a4da1cef0c9eb82b..bd8c3bb88ee5b5aab163dbc04a5c3610db467cc2 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 0a3160c8a0c5b8647e0a564ae2f8700c694b347d..8ea6713900adb01d1d3265719a1fc8a64d11538e 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 21423c8824e63fca810dbf7ba5c44158f43defd5..6e958451c8b323ad8c50c1bf22ac29e8158aaa93 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 e31837372b6afd5677c207a71b996b3360978735..a41129081a0562e3a821abeb3b91b3185d82875f 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 9dfa11caf2c6754b995e0990b038f5f5570f8cd4..f6e84dcc6098fbe31c3b86fa4161e84068112257 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 0823d6c1877f724d047ae7ea776c5d6bb41b8e07..c336ecde1ed1f9bffc64336349d675b8e69bdb32 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_a8e0ba8e05a954f8347b+a5aa275c/compile_flags.json b/neuronxcc-2.21.33363.0+82129205/MODULE_a8e0ba8e05a954f8347b+a5aa275c/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..0005d004a7b3880177587bbc5c6d9519431505a5
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_a8e0ba8e05a954f8347b+a5aa275c/compile_flags.json
@@ -0,0 +1 @@
+["--target=trn1", "--enable-saturate-infinity", "--auto-cast=none", "--model-type=transformer", "-O1", "--logfile=/tmp/nxdi_test_de882b8d-b195-4bc1-9769-672bd5aa1afd/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_a8e0ba8e05a954f8347b+a5aa275c/model.done b/neuronxcc-2.21.33363.0+82129205/MODULE_a8e0ba8e05a954f8347b+a5aa275c/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_a8e0ba8e05a954f8347b+a5aa275c/model.hlo_module.pb b/neuronxcc-2.21.33363.0+82129205/MODULE_a8e0ba8e05a954f8347b+a5aa275c/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..6da2cba66863fc471486cc2a5da22b16c00d6480
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_a8e0ba8e05a954f8347b+a5aa275c/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:993f99d2d09d34bd152af4c7f96a1e6e1d8788e1cd4aefb845c601d2f4d5fcfb
+size 1931
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_a8e0ba8e05a954f8347b+a5aa275c/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_a8e0ba8e05a954f8347b+a5aa275c/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..7a8b5ee68791daed9bfd8188af44aa0dee320c99
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_a8e0ba8e05a954f8347b+a5aa275c/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ada7c9a7bd10c7383ea998b240b95c0d343d4533e17443cfb899aa2a49c6a423
+size 134144
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_a8e0ba8e05a954f8347b+a5aa275c/wrapped_neff.hlo b/neuronxcc-2.21.33363.0+82129205/MODULE_a8e0ba8e05a954f8347b+a5aa275c/wrapped_neff.hlo
new file mode 100644
index 0000000000000000000000000000000000000000..73a73a0de41289b6ca34aed07a8a9748ed23d5ec
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_a8e0ba8e05a954f8347b+a5aa275c/wrapped_neff.hlo
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:75c8bd3754aa0227f6ad95d522bbb8cef485720ed0beda4c6db73164ef101f6e
+size 136222
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_b81bbded724bb14afb4a+299b35f0/compile_flags.json b/neuronxcc-2.21.33363.0+82129205/MODULE_b81bbded724bb14afb4a+299b35f0/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..1433b66b7563173493e3a81e699ded1f8c5c5684
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_b81bbded724bb14afb4a+299b35f0/compile_flags.json
@@ -0,0 +1 @@
+["--target=trn1", "--enable-saturate-infinity", "--auto-cast=none", "--model-type=transformer", "-O1", "--logfile=/tmp/nxdi_test_ba7b71f7-6343-4409-abf9-81c1313ad456/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_b81bbded724bb14afb4a+299b35f0/model.done b/neuronxcc-2.21.33363.0+82129205/MODULE_b81bbded724bb14afb4a+299b35f0/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_b81bbded724bb14afb4a+299b35f0/model.hlo_module.pb b/neuronxcc-2.21.33363.0+82129205/MODULE_b81bbded724bb14afb4a+299b35f0/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..0e51cb39a846e4f58cfe3d5ec50ea34e7babca50
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_b81bbded724bb14afb4a+299b35f0/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c233e7013daa344cb368079ec790fbf10f597bc7f8708f8c91ff6684cdc1cd2e
+size 29412
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_b81bbded724bb14afb4a+299b35f0/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_b81bbded724bb14afb4a+299b35f0/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..5eb5e111e1ae328907091e3b73284877c586c4f2
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_b81bbded724bb14afb4a+299b35f0/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1f91af4adbba18c638d0c26d46ad0e1a234c9518105024f93d1dc616e78f7b4a
+size 328704
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_b81bbded724bb14afb4a+299b35f0/wrapped_neff.hlo b/neuronxcc-2.21.33363.0+82129205/MODULE_b81bbded724bb14afb4a+299b35f0/wrapped_neff.hlo
new file mode 100644
index 0000000000000000000000000000000000000000..c43406194ff591d2e19d2d4bb44f7823048a7767
--- /dev/null
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_b81bbded724bb14afb4a+299b35f0/wrapped_neff.hlo
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9baf1f88ca954fec3c238ccaabaf2aca1ef08c7dae378a86be5a16c17e9249d6
+size 334452
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_f19ecf2e2bcc3a4b1bd9+a02c3a36/model.hlo_module.pb b/neuronxcc-2.21.33363.0+82129205/MODULE_f19ecf2e2bcc3a4b1bd9+a02c3a36/model.hlo_module.pb
index db90fa54202e67f9c2b63a9f960524c316db768c..087adad3be38716000a37f13a671422143669eca 100644
--- a/neuronxcc-2.21.33363.0+82129205/MODULE_f19ecf2e2bcc3a4b1bd9+a02c3a36/model.hlo_module.pb
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_f19ecf2e2bcc3a4b1bd9+a02c3a36/model.hlo_module.pb
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8ed2e09754ffce7b5957680d04d07b9b18166a7809582694d1e73f3af4b21434
+oid sha256:18a82b3beb86fc5ad8db17a90281a3d8d1e499f2379ca89b897d6d0517a715c4
 size 590808
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_f19ecf2e2bcc3a4b1bd9+a02c3a36/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_f19ecf2e2bcc3a4b1bd9+a02c3a36/model.neff
index 1c4d36523ab361879edddbeb9386f0255d2ef751..8838db0bf7abf44cf5a1cdcdf0bf1e7318ed9635 100644
--- a/neuronxcc-2.21.33363.0+82129205/MODULE_f19ecf2e2bcc3a4b1bd9+a02c3a36/model.neff
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_f19ecf2e2bcc3a4b1bd9+a02c3a36/model.neff
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8a4d1820de61a7d69c692dda846a2beefab202121ece6bf3b2f3d3f57b6615ef
+oid sha256:ac0a2d934d862240565860cc6900461f35bc7b15f7ddd942e609544c31ccb79a
 size 1547264
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_f19ecf2e2bcc3a4b1bd9+a02c3a36/wrapped_neff.hlo b/neuronxcc-2.21.33363.0+82129205/MODULE_f19ecf2e2bcc3a4b1bd9+a02c3a36/wrapped_neff.hlo
index 6bd39e951a18efffa4c7afec9d14b8dc12b2564c..8d10dbbd497448a055e817c3a90af3f70db04a01 100644
--- a/neuronxcc-2.21.33363.0+82129205/MODULE_f19ecf2e2bcc3a4b1bd9+a02c3a36/wrapped_neff.hlo
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_f19ecf2e2bcc3a4b1bd9+a02c3a36/wrapped_neff.hlo
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:09068bc76e1d7f12bfc2a2a0a894345a7b43e7ba0364c7d8ca52c83125a85fce
+oid sha256:d34ff2994ccd2e4398318481f851d110a5f77534cc397f743c28bd4f2381b907
 size 1703455
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_fde2b1111ad10bd7504e+a02c3a36/model.hlo_module.pb b/neuronxcc-2.21.33363.0+82129205/MODULE_fde2b1111ad10bd7504e+a02c3a36/model.hlo_module.pb
index 1f4ba21c9f53777d279e086808276ab93c3b0615..47dc969631220fdb40e5fe10f7bfae7a5454d613 100644
--- a/neuronxcc-2.21.33363.0+82129205/MODULE_fde2b1111ad10bd7504e+a02c3a36/model.hlo_module.pb
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_fde2b1111ad10bd7504e+a02c3a36/model.hlo_module.pb
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6661ed9cdcff742b06a6c5ab009981de926925468498a831dec14ba213178f65
+oid sha256:018a62e9a6d7b944baff314113a3d60b4bee472c3df99216573823afea25c4f2
 size 426072
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_fde2b1111ad10bd7504e+a02c3a36/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_fde2b1111ad10bd7504e+a02c3a36/model.neff
index 2d9f1255e7f21cac515745f0bf3e5d2619eb74a6..91d1676bd61d9a919e54cae306e29b7e029a1a23 100644
--- a/neuronxcc-2.21.33363.0+82129205/MODULE_fde2b1111ad10bd7504e+a02c3a36/model.neff
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_fde2b1111ad10bd7504e+a02c3a36/model.neff
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:78a45a412b6e0562fe26aed1091353d940bb875ab9302dc24eb0c24d6e52029c
+oid sha256:21d92962c69cd74fe90afa7ad69519546e4b8f9005bbf62f45b3079ebc309329
 size 2100224
diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_fde2b1111ad10bd7504e+a02c3a36/wrapped_neff.hlo b/neuronxcc-2.21.33363.0+82129205/MODULE_fde2b1111ad10bd7504e+a02c3a36/wrapped_neff.hlo
index d245795d2f67279f3fbe1eafc884ef5e4daa8773..0fa957b30018f80503ba6f53250614d2df173674 100644
--- a/neuronxcc-2.21.33363.0+82129205/MODULE_fde2b1111ad10bd7504e+a02c3a36/wrapped_neff.hlo
+++ b/neuronxcc-2.21.33363.0+82129205/MODULE_fde2b1111ad10bd7504e+a02c3a36/wrapped_neff.hlo
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9a1dee73cc6b9bb4a4a4dc8dc072f92e7f134800fa3f17b1317b073dfb40bd24
+oid sha256:7010ceb9c9761c037b9054de44457b35a1612d73d1af9eaa94e5216e06e32d0d
 size 2174311