diff --git a/.gitattributes b/.gitattributes
index 167226225ede27d075a3d9bc391be549e1719920..61c5f2fd6c1f9cfbace3814a93cbcb266328351f 100644
--- a/.gitattributes
+++ b/.gitattributes
@@ -13084,3 +13084,15 @@ neuronxcc-2.21.18209.0+043b1bf7/MODULE_c7d8772c9f964f249b02+23655fdd/model.neff
 neuronxcc-2.21.18209.0+043b1bf7/MODULE_c7d8772c9f964f249b02+23655fdd/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.21.18209.0+043b1bf7/MODULE_f29c0c4d0640b6ea7a78+468a8606/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.21.18209.0+043b1bf7/MODULE_f29c0c4d0640b6ea7a78+468a8606/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.18209.0+043b1bf7/MODULE_0625536f1892e1f4e662+6a341b1b/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.18209.0+043b1bf7/MODULE_0625536f1892e1f4e662+6a341b1b/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.18209.0+043b1bf7/MODULE_335a05732a8d8bc68ae5+82442274/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.18209.0+043b1bf7/MODULE_335a05732a8d8bc68ae5+82442274/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.18209.0+043b1bf7/MODULE_774c8effbfb7a0624377+ab016a3f/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.18209.0+043b1bf7/MODULE_774c8effbfb7a0624377+ab016a3f/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.18209.0+043b1bf7/MODULE_8c1e274b62e5a0e83b8f+9c04ca4d/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.18209.0+043b1bf7/MODULE_8c1e274b62e5a0e83b8f+9c04ca4d/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.18209.0+043b1bf7/MODULE_8f4495da47e020c01d90+e67d5dd5/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.18209.0+043b1bf7/MODULE_8f4495da47e020c01d90+e67d5dd5/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.18209.0+043b1bf7/MODULE_e6929db0dbb95c6feea7+43861c3b/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.18209.0+043b1bf7/MODULE_e6929db0dbb95c6feea7+43861c3b/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/0_REGISTRY/0.4.3.dev0/llama/unsloth/Llama-3.2-1B-Instruct/ebe02ff9a8c339b67979.json b/neuronxcc-2.21.18209.0+043b1bf7/0_REGISTRY/0.4.3.dev0/llama/unsloth/Llama-3.2-1B-Instruct/ebe02ff9a8c339b67979.json
new file mode 100644
index 0000000000000000000000000000000000000000..180c3bb64e196835f955c9c486c2d290dd270710
--- /dev/null
+++ b/neuronxcc-2.21.18209.0+043b1bf7/0_REGISTRY/0.4.3.dev0/llama/unsloth/Llama-3.2-1B-Instruct/ebe02ff9a8c339b67979.json
@@ -0,0 +1,63 @@
+{
+  "_entry_class": "SingleModelCacheEntry",
+  "_model_id": "unsloth/Llama-3.2-1B-Instruct",
+  "_task": "text-generation",
+  "architectures": [
+    "LlamaForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "dtype": "bfloat16",
+  "head_dim": 64,
+  "hidden_act": "silu",
+  "hidden_size": 2048,
+  "initializer_range": 0.02,
+  "intermediate_size": 8192,
+  "max_position_embeddings": 131072,
+  "mlp_bias": false,
+  "model_type": "llama",
+  "neuron": {
+    "_serialized_key": "NxDNeuronConfig",
+    "batch_size": 4,
+    "capacity_factor": null,
+    "checkpoint_id": "unsloth/Llama-3.2-1B-Instruct",
+    "checkpoint_revision": "5a8abab4a5d6f164389b1079fb721cfab8d7126c",
+    "continuous_batching": true,
+    "ep_degree": 1,
+    "fused_qkv": true,
+    "glu_mlp": true,
+    "local_ranks_size": 2,
+    "max_batch_size": 4,
+    "max_context_length": 4096,
+    "max_topk": 256,
+    "n_active_tokens": 4096,
+    "neuronxcc_version": "2.21.18209.0+043b1bf7",
+    "on_device_sampling": true,
+    "optimum_neuron_version": "0.4.3.dev0",
+    "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-2.21.18209.0+043b1bf7/MODULE_0625536f1892e1f4e662+6a341b1b/compile_flags.json b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_0625536f1892e1f4e662+6a341b1b/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..7c8746fa5e0acbe647bf9c55b311ab0ebe88e849
--- /dev/null
+++ b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_0625536f1892e1f4e662+6a341b1b/compile_flags.json
@@ -0,0 +1 @@
+["--target=trn1", "--enable-saturate-infinity", "--auto-cast=none", "--model-type=transformer", "-O1", "--logfile=/tmp/nxdi_test_283a264f-e386-4f6f-958c-6400fa4a475e/compiler_workdir/NeuronLlamaMLP/_tp0_bk0/log-neuron-cc.txt", "--enable-internal-neff-wrapper"]
\ No newline at end of file
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_0625536f1892e1f4e662+6a341b1b/model.done b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_0625536f1892e1f4e662+6a341b1b/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_0625536f1892e1f4e662+6a341b1b/model.hlo_module.pb b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_0625536f1892e1f4e662+6a341b1b/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..e6084ce6d845ff91b01c0cbfd758d24019f38ef9
--- /dev/null
+++ b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_0625536f1892e1f4e662+6a341b1b/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-2.21.18209.0+043b1bf7/MODULE_0625536f1892e1f4e662+6a341b1b/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_0625536f1892e1f4e662+6a341b1b/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..cf2e9171e07f2e615df2f6ce395f0d3d59a4209c
--- /dev/null
+++ b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_0625536f1892e1f4e662+6a341b1b/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a8d0b88510e33cb637f36fb4f93d9ae7943ae71771a11b1915ce027f9f75c45b
+size 134144
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_0625536f1892e1f4e662+6a341b1b/wrapped_neff.hlo b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_0625536f1892e1f4e662+6a341b1b/wrapped_neff.hlo
new file mode 100644
index 0000000000000000000000000000000000000000..93f4de7454520523c447d3c1d37778ff3a2ad295
--- /dev/null
+++ b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_0625536f1892e1f4e662+6a341b1b/wrapped_neff.hlo
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8ff61a57bba2ad6e07b18e1c228614706d635da0d20843b253e4b7bdf96caf34
+size 136222
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_10217061096959125489+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_10217061096959125489+e30acd3a/model.neff
index 83d8858e5099cceac4eceaf0a61a35945f63b7be..2d561d78113b98b8dbcb8405bcebecd7e609559c 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_10217061096959125489+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_10217061096959125489+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_10244305442015770634+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_10244305442015770634+e30acd3a/model.neff
index 2636cffd0f45d1fd0ccea3c7edd75c219e45d358..18865e16d7062d3ac2685029a6c55011ae20a791 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_10244305442015770634+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_10244305442015770634+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_10606948783918825529+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_10606948783918825529+e30acd3a/model.neff
index ff92d5a2d3b02541fedfc2576564328e33219109..4426a2af15c5a968e5f50a63592c1762f756a91c 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_10606948783918825529+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_10606948783918825529+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_10645643398657092095+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_10645643398657092095+e30acd3a/model.neff
index 4ccadb7dcd1cff2b5c4909d566a0d8c96aad223b..f1b4558b7288b4afb325cf52a3be8948bae323e8 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_10645643398657092095+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_10645643398657092095+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_10661660426924300837+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_10661660426924300837+e30acd3a/model.neff
index 4581a9fa69a3b59944a6331c896230a5318874f6..6f4349fae31d6d83dbbdd79e906c5962ab852688 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_10661660426924300837+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_10661660426924300837+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_10746122569655005679+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_10746122569655005679+e30acd3a/model.neff
index 8654021b677b8bcbef3c10a40e53a4b6e8ca6223..e47825f474c2d39a1554c47056258904f2a479a8 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_10746122569655005679+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_10746122569655005679+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_11086318750207148626+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_11086318750207148626+e30acd3a/model.neff
index 9e8e3d57e44cbc8564a9f0005ac140c2f40316af..a2ff9f5ee6a6942ecd46a7e26f437eb867cc8ce5 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_11086318750207148626+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_11086318750207148626+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_11909310600244571805+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_11909310600244571805+e30acd3a/model.neff
index 3dff1aad0465de83c9f56053db0b448c0ead3ba0..d83037e2605f9f89394205cb23b8f1bfd2514e3d 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_11909310600244571805+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_11909310600244571805+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_11933818254123612383+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_11933818254123612383+e30acd3a/model.neff
index ca37698e0dbbdf46cb1c3f7067e2c0b4175fd3c5..390206dddb8e6785f901d2802a81a49e5cdc489a 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_11933818254123612383+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_11933818254123612383+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_1210392327607194823+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_1210392327607194823+e30acd3a/model.neff
index 2052efdd5fb25f0e040026f1e597e21e57593206..5ddf7e6ebf0016eee4667d754f221d25f4502956 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_1210392327607194823+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_1210392327607194823+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_12487216553200321032+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_12487216553200321032+e30acd3a/model.neff
index 05b8e4a55197d15a663d7bcacb44a9e7d9d93068..dfef5d3937916394c7b5a472c9c8bcf690167552 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_12487216553200321032+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_12487216553200321032+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_13085549342645515693+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_13085549342645515693+e30acd3a/model.neff
index efcf619fc75c48d81ffcd82f9aae5ef663e46825..673668ffa119c21a02e9e4841bf676472dde0c58 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_13085549342645515693+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_13085549342645515693+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_13102010590082783346+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_13102010590082783346+e30acd3a/model.neff
index da6057a04a965171dfffdbe370b25d13b5805e84..b28b4c1b425f45aa7fa35622664cb867da9641c1 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_13102010590082783346+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_13102010590082783346+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_13475273426270778455+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_13475273426270778455+e30acd3a/model.neff
index 9661013dbdb4627bfd64ecad41f682d679e96a2c..2385b5869ec05a286ffda714fd80a1d8bcdcfe96 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_13475273426270778455+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_13475273426270778455+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_13728813963059599796+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_13728813963059599796+e30acd3a/model.neff
index 01bdcebe735efed5ee6cb1d612e83a2a5d638c20..1a64233326db82e6d1fef736734929c948f53f4b 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_13728813963059599796+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_13728813963059599796+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_13783872956654054643+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_13783872956654054643+e30acd3a/model.neff
index aa4efeb83ad8fe2c6527cbe6e5d75acd34228f7c..24ed3314d978a3984e0837fcb1cd97e371e8b3d5 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_13783872956654054643+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_13783872956654054643+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_14097776130612710282+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_14097776130612710282+e30acd3a/model.neff
index 7f795c5197ab46ade40153bc6389a3b540fe9a38..70ecae0d932fc016c66d8786eb1d0eb02a792c7c 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_14097776130612710282+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_14097776130612710282+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_14155105943057125178+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_14155105943057125178+e30acd3a/model.neff
index 8e2a2196c3498d7ea848711d0ffee16491b3f0d3..7168efad6d5da07ed8dc61eaedcdf194afd8ed42 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_14155105943057125178+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_14155105943057125178+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_14173682194645188821+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_14173682194645188821+e30acd3a/model.neff
index 22abf3b9bc5866e3e7649f1152f1d63408c88fd3..4bf3f71469d912a3c9126218b14491c8b2e6baa3 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_14173682194645188821+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_14173682194645188821+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_14274833313744358348+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_14274833313744358348+e30acd3a/model.neff
index c746c7b7e1047ad037da760d6fa0fa24dd4aa100..2a2766811a9de61292bfb5a6dbb8c439c146bdac 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_14274833313744358348+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_14274833313744358348+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_14469716438804935215+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_14469716438804935215+e30acd3a/model.neff
index 57d0ccc5b8897f613b68411dfe5e1c41d757c93a..54f20729a0234b2937c1787171c085c6b2862a9c 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_14469716438804935215+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_14469716438804935215+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_14739672283771660808+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_14739672283771660808+e30acd3a/model.neff
index b107a35470a62a9d4a39a5d87aead187a9b89044..09a91911ccc3fe4bdc0671352428827398a11bcf 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_14739672283771660808+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_14739672283771660808+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_15442663025941492357+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_15442663025941492357+e30acd3a/model.neff
index 7f3b2eb11f6a84c784db28ec130dad3dd7c13d10..db3ee19582e3be9b14821b3f17ad9764552fb4ff 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_15442663025941492357+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_15442663025941492357+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_15557971880365771457+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_15557971880365771457+e30acd3a/model.neff
index 1e316bfa85cbff39dfda1aaac539d605fa491960..9ddb706daf35be5e16840f212644ee86ff3ed97b 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_15557971880365771457+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_15557971880365771457+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_15564425602929126510+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_15564425602929126510+e30acd3a/model.neff
index 4fc9a47c4f6f04183b87489dc165ef0e1622bef5..8980e921f1940fa6ccd8e31ff56ae04668f675f3 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_15564425602929126510+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_15564425602929126510+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_15589838460896944293+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_15589838460896944293+e30acd3a/model.neff
index 4792ea6d3c84bf4603d04781449806f5f1c91868..f943ac1985b7306d08c21ed3d7fb06306b511375 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_15589838460896944293+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_15589838460896944293+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_15654572849171857535+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_15654572849171857535+e30acd3a/model.neff
index 4eb31b44be2c31090bc85f834295478295a326e7..8fdeab0188ab7c0e89faee36dbb006eda566dc39 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_15654572849171857535+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_15654572849171857535+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_15767383571209512795+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_15767383571209512795+e30acd3a/model.neff
index a549edd2754900e04672e61ad299cc7636bfa390..0b12d05f889bb3222ca5bb7c2163077327b9a71e 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_15767383571209512795+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_15767383571209512795+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_15888836045088309511+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_15888836045088309511+e30acd3a/model.neff
index 64893f26514cdb0661137b5bda7a41cc7e38495b..4def4398264671e567c80a8ccef2a15e5c35850e 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_15888836045088309511+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_15888836045088309511+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_15929451261464042997+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_15929451261464042997+e30acd3a/model.neff
index 51c08786967dcac6c19df4b8fb15d03056f64684..11add02068ea23a70f4b39c8f3790cfaf6571bff 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_15929451261464042997+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_15929451261464042997+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_16016505958416521648+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_16016505958416521648+e30acd3a/model.neff
index 70b2804fdd96a9d260ec2f1b31fc9ad1caac7f4b..58c1a05d8bcfbd628e2c903cb9cac86db64390a3 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_16016505958416521648+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_16016505958416521648+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_16053163479112702088+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_16053163479112702088+e30acd3a/model.neff
index cf0c60f2d4488c2d7bc3c3866d42f53c29c4861a..6603f188ad6311aa95d78e43f9a4f3525b0ea82f 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_16053163479112702088+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_16053163479112702088+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_16204387852795926216+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_16204387852795926216+e30acd3a/model.neff
index 1903023bbb205ce324e5a16ee5729e7e0f576ec2..a3b7e5657fbd5f9907e9d311a86633582c9d9a43 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_16204387852795926216+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_16204387852795926216+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_1656719109221189948+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_1656719109221189948+e30acd3a/model.neff
index 2eef4927923304073a1ceed91f75bb0cf50ed3d2..6560f0ceb4f1e9b601ef6ad0adaec91fe1b95fb9 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_1656719109221189948+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_1656719109221189948+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_16599571375348449904+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_16599571375348449904+e30acd3a/model.neff
index e96fb89a3eaf51ed84d33e2be1471ba0baf0e082..9ed19a4c27ad8386a52efa3b67c4772c56382e58 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_16599571375348449904+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_16599571375348449904+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_1663757798483801648+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_1663757798483801648+e30acd3a/model.neff
index 2cda3befe4e1eb47d7cf7d30567f4ab247a9a527..d20d9ddde32bd77839f2c5f61c5ecac1ffe9f2bf 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_1663757798483801648+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_1663757798483801648+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_16738296820980389103+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_16738296820980389103+e30acd3a/model.neff
index 8bc8880bbf869abeaea1f3f9f331baf96c7359b8..7fc64900d48c4e1b200d7d5a7a80688cc91bcd2e 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_16738296820980389103+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_16738296820980389103+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_1696825468766062114+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_1696825468766062114+e30acd3a/model.neff
index c7876eca8491d8507b5a966fd068f9f93593bbb3..76da716a533b29bd2b7f90a45bb3e008d6dad294 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_1696825468766062114+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_1696825468766062114+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_17476209562158013765+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_17476209562158013765+e30acd3a/model.neff
index 9e29721c21d3c406910970f8bcbf3776555e82c0..14a5936bbb2ac35a2b2327f3e99ad6edc21a610c 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_17476209562158013765+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_17476209562158013765+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_17781596253725927902+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_17781596253725927902+e30acd3a/model.neff
index b93fc47c16a37ba02f256735755eece218b1af49..1f2fb04126196cc126948482fb70191f19d4a89b 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_17781596253725927902+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_17781596253725927902+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_1795740353831178306+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_1795740353831178306+e30acd3a/model.neff
index d392e3e330765e878a7deb9a3f5f81c9b40ac6f0..6c1665c6376e2f9c5eded54541b2a96940cbe3d8 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_1795740353831178306+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_1795740353831178306+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_1800832390737682969+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_1800832390737682969+e30acd3a/model.neff
index b0745079be4596bfbad7cfcc5f8cad1d92a9060b..25180004173707121a4de5c4baf2fc8e85b06ecb 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_1800832390737682969+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_1800832390737682969+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_1805481651134498710+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_1805481651134498710+e30acd3a/model.neff
index d14f90d74aa43ce27f3a564206e9a245b8661f93..0e50ead69939665875fc10fa8aa308e0dc537399 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_1805481651134498710+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_1805481651134498710+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_1805728312057401221+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_1805728312057401221+e30acd3a/model.neff
index 6624c55106244cdaa581f595e73389da74cfefd9..f98b6b33c0419befd38842d67a4ae731ba1d613a 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_1805728312057401221+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_1805728312057401221+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_18070749384531238695+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_18070749384531238695+e30acd3a/model.neff
index 168fbb3bec9567bcf7494d4f47e83eea05c1a83d..825dcea3c5b42aeeb576e1e5ffbdeb836094a770 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_18070749384531238695+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_18070749384531238695+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_18134517905922687964+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_18134517905922687964+e30acd3a/model.neff
index eba46d44d75239702a969f27cfeae2576f9d09fd..a43679ab6fd487daac63ae1ecbbef4379c930ee7 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_18134517905922687964+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_18134517905922687964+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_18207475553554060390+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_18207475553554060390+e30acd3a/model.neff
index e38d5d70d24f6ccfc3b64ef9f176b00e10f17f2b..664373558697af8dce0c41920021dada1046ca47 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_18207475553554060390+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_18207475553554060390+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_2157613318347839507+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_2157613318347839507+e30acd3a/model.neff
index 05431489bedebf98a14226f3fba7f29c7154fbd0..d09a9557dd0b7a650c09a7822de6f3d0d397f3b6 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_2157613318347839507+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_2157613318347839507+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_2203200257461345827+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_2203200257461345827+e30acd3a/model.neff
index 04f5018bf16de0aa05f95a91f3b50d3df1835bf5..d9ca1e83222368a42e0cdd1d79fcb74f6b26a684 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_2203200257461345827+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_2203200257461345827+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_2410426589418994260+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_2410426589418994260+e30acd3a/model.neff
index 71b105a7e2ab1e364d988b98b97171e5cf274b05..9f1365eb35655e09b9ffdd3ad68910b475d120d9 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_2410426589418994260+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_2410426589418994260+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_2426414314187505427+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_2426414314187505427+e30acd3a/model.neff
index 9de4bd19b2f07fc1a4007d68ad8b3af03c8fef69..c21e667e90250141a522ac9afbe657c7d7f6fea9 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_2426414314187505427+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_2426414314187505427+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_2576759111807165188+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_2576759111807165188+e30acd3a/model.neff
index 5e7716f4a4abd1bcd35d18f28ef208c091885952..4b8a9f1448419e7abe3b756ae536619e6999d442 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_2576759111807165188+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_2576759111807165188+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_2650447136112456251+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_2650447136112456251+e30acd3a/model.neff
index e245f4ae05cb2764042a7a162ae254611d01d756..7eae9d5d09122f6f368fb946409d59d75b41443f 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_2650447136112456251+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_2650447136112456251+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_2682953264353234433+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_2682953264353234433+e30acd3a/model.neff
index ff415063d882e63afab0c888fd01a977a981876b..22cb851cd2a0c55a2276d90821a1e054addae290 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_2682953264353234433+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_2682953264353234433+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_2699280712748688265+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_2699280712748688265+e30acd3a/model.neff
index 03834f5c3e914c7cec66667b7b8da8c1ff8c6ad6..4a354fa9ac9352204c9c38b5425aaef69ae137af 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_2699280712748688265+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_2699280712748688265+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_2811495285170804454+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_2811495285170804454+e30acd3a/model.neff
index cc2ddae810fe87389003478f98df3d0d35145335..178db0cd91a0869f88d5c0b8e9ea77f200d2a6e2 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_2811495285170804454+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_2811495285170804454+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_302950992539679003+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_302950992539679003+e30acd3a/model.neff
index d3bd1565ff954756947c94e88a771b2de678807a..68c2e8cac735ac0b615065390c7d34d9a697b49e 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_302950992539679003+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_302950992539679003+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_335a05732a8d8bc68ae5+82442274/compile_flags.json b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_335a05732a8d8bc68ae5+82442274/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..3812e5a87c49680b4152874c2927afbb15ee084a
--- /dev/null
+++ b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_335a05732a8d8bc68ae5+82442274/compile_flags.json
@@ -0,0 +1 @@
+["--target=trn1", "--enable-saturate-infinity", "--auto-cast=none", "--model-type=transformer", "-O1", "--logfile=/tmp/nxdi_test_210e9cee-a739-400c-b736-1113fbfef615/compiler_workdir/AttentionModelWrapper/_tp0_bk0/log-neuron-cc.txt", "--enable-internal-neff-wrapper"]
\ No newline at end of file
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_335a05732a8d8bc68ae5+82442274/model.done b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_335a05732a8d8bc68ae5+82442274/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_335a05732a8d8bc68ae5+82442274/model.hlo_module.pb b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_335a05732a8d8bc68ae5+82442274/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..8496781848c02ecd308b5b2c5a3c342b702425e7
--- /dev/null
+++ b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_335a05732a8d8bc68ae5+82442274/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-2.21.18209.0+043b1bf7/MODULE_335a05732a8d8bc68ae5+82442274/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_335a05732a8d8bc68ae5+82442274/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..98047be9de5ef0efcb49f3120d7ce327732480df
--- /dev/null
+++ b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_335a05732a8d8bc68ae5+82442274/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4eefb6d64e80bd6a3dee6d50b18a020ac4c3e6f998d623e9271d0b5ff283c5c8
+size 246784
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_335a05732a8d8bc68ae5+82442274/wrapped_neff.hlo b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_335a05732a8d8bc68ae5+82442274/wrapped_neff.hlo
new file mode 100644
index 0000000000000000000000000000000000000000..380e807e12552496bfd2f7bb5d6af9db6bddb795
--- /dev/null
+++ b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_335a05732a8d8bc68ae5+82442274/wrapped_neff.hlo
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d46fb59367838a282334efb966f64bda16b09061a17a2f0d32927202e3f4bead
+size 249608
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_4013984518400530928+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_4013984518400530928+e30acd3a/model.neff
index 13796ced2c5886433358c24304d5f8c3cc8f45fb..52b401e2cd7884fe714fe75055e68191b8d15a6a 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_4013984518400530928+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_4013984518400530928+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_4052264821257342969+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_4052264821257342969+e30acd3a/model.neff
index 4ddbb5f85feb644114be7acff97b08de7a1d25ef..f2c1e87e484c9409580fec45cf8642c3306500e1 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_4052264821257342969+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_4052264821257342969+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_4166695227062860792+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_4166695227062860792+e30acd3a/model.neff
index 3951b6506310698bdd32c8c081761fdade607fc2..10fc11175f2bfd1bd4f605287b1ca8505e678523 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_4166695227062860792+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_4166695227062860792+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_4232557591140652245+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_4232557591140652245+e30acd3a/model.neff
index 8f549a2c7e395be3a73ee59761582f3da92fa345..c3952de7a978f1ba107328c5fa08359ce2ae2c10 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_4232557591140652245+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_4232557591140652245+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_4606491920565381523+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_4606491920565381523+e30acd3a/model.neff
index b333a30b9faedd601a6d26d8deee2d409031ad3b..144ed4b6380ecb8330efc018d0d6740ff1994ed1 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_4606491920565381523+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_4606491920565381523+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_4608333438381222049+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_4608333438381222049+e30acd3a/model.neff
index b29100b8ba1aea2e42bba8e0a0e0063a235859bd..0c9435bd5bad9a86b9a29fe8c88283acb0f3b818 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_4608333438381222049+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_4608333438381222049+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_4683016291147007110+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_4683016291147007110+e30acd3a/model.neff
index ee27fe5f6b8d0edde96d861943b06f491f536748..1334107222e70c548dfbe239d809f84ba3aa8a3d 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_4683016291147007110+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_4683016291147007110+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_4816422433712790690+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_4816422433712790690+e30acd3a/model.neff
index d84488c6155d8fe3c004ef4d62913281fa7a3f47..05957ba561d9ccf49a6fffc7e19cb428cfe1bd34 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_4816422433712790690+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_4816422433712790690+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_4842952601010372434+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_4842952601010372434+e30acd3a/model.neff
index cf306e66523471b1a44ecf05861c3e750a799847..3fba86c92e7d06afc3276a1e5ccb1da880f07209 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_4842952601010372434+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_4842952601010372434+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_4939721357859779936+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_4939721357859779936+e30acd3a/model.neff
index 5203278d2fb993502be4064c7d6b097fd1ee131b..18566b0be999a8da92476e03e3aae7254a391c5c 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_4939721357859779936+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_4939721357859779936+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_5397223456911199516+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_5397223456911199516+e30acd3a/model.neff
index 828fa8e84036dbc50f0aa6fc4c2a2bd07591294f..052376af4469f322ddbc9e9b1bf1f78bb8d0d65a 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_5397223456911199516+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_5397223456911199516+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_5724544066416780383+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_5724544066416780383+e30acd3a/model.neff
index 06a66cd2fe751308257f34044cee8f9f48ca4d44..d97216a181657b36bd581601dd41127384ecfd01 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_5724544066416780383+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_5724544066416780383+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_5765916484866650909+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_5765916484866650909+e30acd3a/model.neff
index 620cf265332f6407df79a8a081ea9a12616f9a5d..c1e1a99a09110d19c31ff52e541170bdfddea7a4 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_5765916484866650909+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_5765916484866650909+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_6368409506294683105+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_6368409506294683105+e30acd3a/model.neff
index dca6896a3614de09d3d7b41d43e1cf481fe58f45..edd6eda09559724c0fc57dc5df2af6a9661c6e4c 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_6368409506294683105+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_6368409506294683105+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_6375780938374445148+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_6375780938374445148+e30acd3a/model.neff
index 868db6281fdab66f3cabe7e2fe9bd0361b024502..a3a96f28997bbb0e04f4b43cdc7274681c5b7cd4 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_6375780938374445148+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_6375780938374445148+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_6417297157123190451+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_6417297157123190451+e30acd3a/model.neff
index 142ab181a58e53640ba3ee3317aefe0d1e8dac2f..34ec6d640bee4b209c264ad1752491049e028cfa 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_6417297157123190451+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_6417297157123190451+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_665145310648895012+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_665145310648895012+e30acd3a/model.neff
index 2d4ec2fd0b91ecca26872a719bc001690e81349f..9788e30d9196b9f1b8a6b33ef8d4ab0b160ac34a 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_665145310648895012+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_665145310648895012+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_6711765444274762085+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_6711765444274762085+e30acd3a/model.neff
index 52b46a8a8a8d5a5402381950f292dec5ed4c1de2..572efe390d3de938b520f544d861e099d7659011 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_6711765444274762085+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_6711765444274762085+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_7049343467335223052+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_7049343467335223052+e30acd3a/model.neff
index c45bb31b22e858cdd34414853bb6f0b1a7ef111a..6e2e67e3fabc275887003c454a2f30d0fee3b2af 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_7049343467335223052+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_7049343467335223052+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_774c8effbfb7a0624377+ab016a3f/compile_flags.json b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_774c8effbfb7a0624377+ab016a3f/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..3354daabb34ad45cc1fb9478d35c8c142a0aaa55
--- /dev/null
+++ b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_774c8effbfb7a0624377+ab016a3f/compile_flags.json
@@ -0,0 +1 @@
+["--target=trn1", "--enable-saturate-infinity", "--auto-cast=none", "--model-type=transformer", "-O1", "--logfile=/tmp/nxdi_test_235d9abb-084c-477b-86a5-31ae2353a30c/compiler_workdir/DecoderLayerWrapper/_tp0_bk0/log-neuron-cc.txt", "--enable-internal-neff-wrapper"]
\ No newline at end of file
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_774c8effbfb7a0624377+ab016a3f/model.done b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_774c8effbfb7a0624377+ab016a3f/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_774c8effbfb7a0624377+ab016a3f/model.hlo_module.pb b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_774c8effbfb7a0624377+ab016a3f/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..be5670c6258a72b3253c262c04b0a3c084a1810d
--- /dev/null
+++ b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_774c8effbfb7a0624377+ab016a3f/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-2.21.18209.0+043b1bf7/MODULE_774c8effbfb7a0624377+ab016a3f/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_774c8effbfb7a0624377+ab016a3f/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..f9d272194266dc993bf00292c3828597290b29d3
--- /dev/null
+++ b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_774c8effbfb7a0624377+ab016a3f/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3473e11b5b8a867aa1855f20606dceba8059f2173201195d1ff8fcbbddb4008a
+size 328704
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_774c8effbfb7a0624377+ab016a3f/wrapped_neff.hlo b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_774c8effbfb7a0624377+ab016a3f/wrapped_neff.hlo
new file mode 100644
index 0000000000000000000000000000000000000000..f5242dc641798cc6411cb8b39289a7c9aa2df640
--- /dev/null
+++ b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_774c8effbfb7a0624377+ab016a3f/wrapped_neff.hlo
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cc49abde8db3d0a18dba2b4d8d0e204970cc8888a92fce56bb9890fcecc65b07
+size 334452
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_7877866862268499659+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_7877866862268499659+e30acd3a/model.neff
index 2dfbba41be4343d7ac7ec4f6c78e16850195ed3f..7d0691acd43dc1f62392d36087b5480f67e86828 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_7877866862268499659+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_7877866862268499659+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_793722132614349680+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_793722132614349680+e30acd3a/model.neff
index 1cb26eb588ef6fe775b269a32d3ff36d384597ae..297751a519b124a1da16d5ff2cb52779d57d7d59 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_793722132614349680+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_793722132614349680+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_8142158350842512240+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_8142158350842512240+e30acd3a/model.neff
index fab7fbba9aa237e351286d8b012149ab5f821165..60c9db77d85f808859d11b195b4fd89ed5574d9c 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_8142158350842512240+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_8142158350842512240+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_8146958338478347620+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_8146958338478347620+e30acd3a/model.neff
index 40d10ff1020f48099475ecef5ce6be240a0dab60..9fb0b89577bd550ddbea26e21e966fc316cb3d82 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_8146958338478347620+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_8146958338478347620+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_8245165830758578911+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_8245165830758578911+e30acd3a/model.neff
index 35f228c30507ca3c980feed86d1ac8449a2b6b93..c2aa8dd2e96b8febae8313a66bf5e3abb2aba216 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_8245165830758578911+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_8245165830758578911+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_8679096817551502409+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_8679096817551502409+e30acd3a/model.neff
index 3e73f67eebd44538a5938668945cefbdf2d07d55..1a546fd288006087f4e4377949a92c03870b5db6 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_8679096817551502409+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_8679096817551502409+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_8882592820567938515+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_8882592820567938515+e30acd3a/model.neff
index 8fe5a611cfa6e19ae6b9525c936652e97fb99912..7e9b3484894a13bfb91dfd980956ee70bfd99237 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_8882592820567938515+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_8882592820567938515+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_8c1e274b62e5a0e83b8f+9c04ca4d/compile_flags.json b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_8c1e274b62e5a0e83b8f+9c04ca4d/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..a6153ba34ccfd90d2492f98baf496360d87fdbc4
--- /dev/null
+++ b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_8c1e274b62e5a0e83b8f+9c04ca4d/compile_flags.json
@@ -0,0 +1 @@
+["--target=trn1", "--enable-saturate-infinity", "--auto-cast=none", "--model-type=transformer", "-O1", "--logfile=/tmp/nxdi_test_5d7e7eb6-25f3-4036-8355-61ae790596d7/compiler_workdir/DecoderLayerWrapper/_tp0_bk0/log-neuron-cc.txt", "--enable-internal-neff-wrapper"]
\ No newline at end of file
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_8c1e274b62e5a0e83b8f+9c04ca4d/model.done b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_8c1e274b62e5a0e83b8f+9c04ca4d/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_8c1e274b62e5a0e83b8f+9c04ca4d/model.hlo_module.pb b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_8c1e274b62e5a0e83b8f+9c04ca4d/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..07bf62460bc7813e4a95f208598e2fa4ae551d41
--- /dev/null
+++ b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_8c1e274b62e5a0e83b8f+9c04ca4d/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-2.21.18209.0+043b1bf7/MODULE_8c1e274b62e5a0e83b8f+9c04ca4d/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_8c1e274b62e5a0e83b8f+9c04ca4d/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..01084cb280e3c8ba0808f387478d7fac614e8856
--- /dev/null
+++ b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_8c1e274b62e5a0e83b8f+9c04ca4d/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5f9d7c98ea8a9d81bd842e9c8a37fddc1f757ae32d809d1c224b9eea46c3df93
+size 267264
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_8c1e274b62e5a0e83b8f+9c04ca4d/wrapped_neff.hlo b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_8c1e274b62e5a0e83b8f+9c04ca4d/wrapped_neff.hlo
new file mode 100644
index 0000000000000000000000000000000000000000..70e222b3d6f491acdcf69980ecf98d020304c57f
--- /dev/null
+++ b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_8c1e274b62e5a0e83b8f+9c04ca4d/wrapped_neff.hlo
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:988affcc90eea98394598ea674394daaa2369c6577d879c8bdad600a6b8436ea
+size 272962
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_8f4495da47e020c01d90+e67d5dd5/compile_flags.json b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_8f4495da47e020c01d90+e67d5dd5/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..0da3e053aad9382749cce37a514d478e9e77dd41
--- /dev/null
+++ b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_8f4495da47e020c01d90+e67d5dd5/compile_flags.json
@@ -0,0 +1 @@
+["--target=trn1", "--enable-saturate-infinity", "--auto-cast=none", "--model-type=transformer", "-O1", "--logfile=/tmp/nxdi_test_6fdbdcbc-2431-480c-8b78-e7e7008ed744/compiler_workdir/AttentionModelWrapper/_tp0_bk0/log-neuron-cc.txt", "--enable-internal-neff-wrapper"]
\ No newline at end of file
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_8f4495da47e020c01d90+e67d5dd5/model.done b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_8f4495da47e020c01d90+e67d5dd5/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_8f4495da47e020c01d90+e67d5dd5/model.hlo_module.pb b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_8f4495da47e020c01d90+e67d5dd5/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..a11cc062167102b0f4faa9fdf16388d72d9e0015
--- /dev/null
+++ b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_8f4495da47e020c01d90+e67d5dd5/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-2.21.18209.0+043b1bf7/MODULE_8f4495da47e020c01d90+e67d5dd5/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_8f4495da47e020c01d90+e67d5dd5/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..b55c9cd33233d445e072b40c7d959d749090c47b
--- /dev/null
+++ b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_8f4495da47e020c01d90+e67d5dd5/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c62317b9517376d8f82340aff529356c18f08810cf257eceb26c5d69eb71ddc3
+size 1444864
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_8f4495da47e020c01d90+e67d5dd5/wrapped_neff.hlo b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_8f4495da47e020c01d90+e67d5dd5/wrapped_neff.hlo
new file mode 100644
index 0000000000000000000000000000000000000000..76d0ad9c8161485cb979d3bc262e16a17c43d9d3
--- /dev/null
+++ b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_8f4495da47e020c01d90+e67d5dd5/wrapped_neff.hlo
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:31bbd61664ccd41c7b5dca5c5db26d1f29d71bc6a0acfd6c1c0804661b8e6db4
+size 1447845
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_9168843780052024308+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_9168843780052024308+e30acd3a/model.neff
index 7043b484abe388a58194f437d9ad72fd0e31d183..db69bae13e895e5d2f2e431ed282c91b541b4bb4 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_9168843780052024308+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_9168843780052024308+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_9527558113976496538+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_9527558113976496538+e30acd3a/model.neff
index cdc88f263b4e8ef1fb3d5255284db11106dc171b..1547c318ad88d8c8489c4a3829dff16d4fb8c485 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_9527558113976496538+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_9527558113976496538+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_9533302140457976101+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_9533302140457976101+e30acd3a/model.neff
index 714f6fe99cda57de8157298e90343c3d607ec4d4..3f980f0e61e989b47e24155ecaed09e0c09ada1e 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_9533302140457976101+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_9533302140457976101+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_9770544877851564228+e30acd3a/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_9770544877851564228+e30acd3a/model.neff
index 33d2485511b2597c39b00f677374824e6ffd6bc4..a5d30aa24437eb4858bf3b6142763a4c9cd78f7f 100644
Binary files a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_9770544877851564228+e30acd3a/model.neff and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_9770544877851564228+e30acd3a/model.neff differ
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_e6929db0dbb95c6feea7+43861c3b/compile_flags.json b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_e6929db0dbb95c6feea7+43861c3b/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..7228753e152d60353d2a5e38db4677f682442d50
--- /dev/null
+++ b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_e6929db0dbb95c6feea7+43861c3b/compile_flags.json
@@ -0,0 +1 @@
+["--target=trn1", "--enable-saturate-infinity", "--auto-cast=none", "--model-type=transformer", "-O1", "--logfile=/tmp/nxdi_test_01a40801-205d-4fba-a140-7937a02a3660/compiler_workdir/NeuronRMSNorm/_tp0_bk0/log-neuron-cc.txt", "--enable-internal-neff-wrapper"]
\ No newline at end of file
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_e6929db0dbb95c6feea7+43861c3b/model.done b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_e6929db0dbb95c6feea7+43861c3b/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_e6929db0dbb95c6feea7+43861c3b/model.hlo_module.pb b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_e6929db0dbb95c6feea7+43861c3b/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..6a2752c25c9f7e1092e2b55c010694e168610ce3
--- /dev/null
+++ b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_e6929db0dbb95c6feea7+43861c3b/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-2.21.18209.0+043b1bf7/MODULE_e6929db0dbb95c6feea7+43861c3b/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_e6929db0dbb95c6feea7+43861c3b/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..9fa024e45029f1d818342829a2902249b6066242
--- /dev/null
+++ b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_e6929db0dbb95c6feea7+43861c3b/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8d7a71c557708c67d7b97ce971921d55c087d0d601c298e26ff13806257c1bd3
+size 103424
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_e6929db0dbb95c6feea7+43861c3b/wrapped_neff.hlo b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_e6929db0dbb95c6feea7+43861c3b/wrapped_neff.hlo
new file mode 100644
index 0000000000000000000000000000000000000000..a3d884ca186ca288ee706f217fc44d1e31f620c4
--- /dev/null
+++ b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_e6929db0dbb95c6feea7+43861c3b/wrapped_neff.hlo
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:63b1380065d230fbcf4eafbf57a7beb1145de2cb0631d2c0002a1605242ef9ab
+size 104320
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_fdc7edd8c402dcf7638f+8d4db955/compile_flags.json b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_fdc7edd8c402dcf7638f+8d4db955/compile_flags.json
new file mode 100644
index 0000000000000000000000000000000000000000..1a70bc33bfe4e2dcc0bdfc65559f9b5c1482466c
--- /dev/null
+++ b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_fdc7edd8c402dcf7638f+8d4db955/compile_flags.json
@@ -0,0 +1 @@
+["--target=trn1", "--enable-saturate-infinity", "--auto-cast=none", "--model-type=transformer", "-O1", "--logfile=/tmp/nxdi_test_9cc2228d-aace-4644-94e9-1a6b0084770c/compiler_workdir/Llama3RotaryEmbedding/_tp0_bk0/log-neuron-cc.txt", "--enable-internal-neff-wrapper"]
\ No newline at end of file
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_fdc7edd8c402dcf7638f+8d4db955/model.done b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_fdc7edd8c402dcf7638f+8d4db955/model.done
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/neuronxcc-2.21.18209.0+043b1bf7/MODULE_fdc7edd8c402dcf7638f+8d4db955/model.hlo_module.pb b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_fdc7edd8c402dcf7638f+8d4db955/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..f1e2473c54cffac949ecf31d3b6ec636c290545b
--- /dev/null
+++ b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_fdc7edd8c402dcf7638f+8d4db955/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-2.21.18209.0+043b1bf7/MODULE_fdc7edd8c402dcf7638f+8d4db955/model.neff b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_fdc7edd8c402dcf7638f+8d4db955/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..c473bf4cedcbbebfee1f0f4c2bd6e7828627aeb9
Binary files /dev/null and b/neuronxcc-2.21.18209.0+043b1bf7/MODULE_fdc7edd8c402dcf7638f+8d4db955/model.neff differ