diff --git a/.gitattributes b/.gitattributes
index 1ea9b6da07559dc2cb81b70e982fe3e51a535267..513867ef9a1f6f2f8d1c06db839fbb26378d90c1 100644
--- a/.gitattributes
+++ b/.gitattributes
@@ -2917,3 +2917,15 @@ neuronxcc-2.13.66.0+6dfecc895/MODULE_d79e77e121ae50f260ee+2c2d707e/model.neff fi
 neuronxcc-2.13.66.0+6dfecc895/MODULE_d7e53a79eed722050835+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.13.66.0+6dfecc895/MODULE_f18a2d3eaf5290cb34e5+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.13.66.0+6dfecc895/MODULE_f3708c713d3ecf13d290+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.66.0+6dfecc895/MODULE_005044a54de48516f9d6+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.66.0+6dfecc895/MODULE_03d14695157421a1a680+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.66.0+6dfecc895/MODULE_0d2f6a194ed6bb76fa91+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.66.0+6dfecc895/MODULE_478e625ca58c9ae373d5+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.66.0+6dfecc895/MODULE_a50f1bc59e07df734a75+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.66.0+6dfecc895/MODULE_aa7fd25afda59bdaf747+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.66.0+6dfecc895/MODULE_bd38c2a884aabaec3d62+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.66.0+6dfecc895/MODULE_c7bc5bab6ec8ea2e1063+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.66.0+6dfecc895/MODULE_de4e8d5ed85f8709144f+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.66.0+6dfecc895/MODULE_e291d316cec0d0e9de79+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.66.0+6dfecc895/MODULE_fc0f4892d01029ca630a+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.66.0+6dfecc895/MODULE_ff52385503277deb3b07+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
diff --git a/neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.22/inference/llama/meta-llama/Llama-2-70b-chat-hf/ed5c36192e448b28f071.json b/neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.22/inference/llama/meta-llama/Llama-2-70b-chat-hf/ed5c36192e448b28f071.json
new file mode 100644
index 0000000000000000000000000000000000000000..40fcf622bf04336d964023511f98a7ce1f5b2194
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.22/inference/llama/meta-llama/Llama-2-70b-chat-hf/ed5c36192e448b28f071.json
@@ -0,0 +1 @@
+{"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 1, "eos_token_id": 2, "hidden_act": "silu", "hidden_size": 8192, "initializer_range": 0.02, "intermediate_size": 28672, "max_position_embeddings": 4096, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 1, "checkpoint_id": "meta-llama/Llama-2-70b-chat-hf", "checkpoint_revision": "e9149a12809580e8602995856f8098ce973d1080", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 24, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 64, "num_hidden_layers": 80, "num_key_value_heads": 8, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 10000.0, "tie_word_embeddings": false, "torch_dtype": "float16", "use_cache": true, "vocab_size": 32000}
\ No newline at end of file
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_005044a54de48516f9d6+2c2d707e/compile_flags.txt b/neuronxcc-2.13.66.0+6dfecc895/MODULE_005044a54de48516f9d6+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_005044a54de48516f9d6+2c2d707e/compile_flags.txt
@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none
\ No newline at end of file
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_005044a54de48516f9d6+2c2d707e/model.hlo_module.pb b/neuronxcc-2.13.66.0+6dfecc895/MODULE_005044a54de48516f9d6+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..63fcf1d63591195576d735f3d43cd51aef43c7ea
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_005044a54de48516f9d6+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:17fab5988625f23796c8e9cc83e42e6976ae1b27264e3217bdeda015249c7d08
+size 752116
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_005044a54de48516f9d6+2c2d707e/model.neff b/neuronxcc-2.13.66.0+6dfecc895/MODULE_005044a54de48516f9d6+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..d29e704d88232d33d0dd27f7921acd3db65d4576
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_005044a54de48516f9d6+2c2d707e/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ff1ce7d8fc8c7538000114b555be313971e8f7e67009718e880af0588089df49
+size 8223744
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_03d14695157421a1a680+2c2d707e/compile_flags.txt b/neuronxcc-2.13.66.0+6dfecc895/MODULE_03d14695157421a1a680+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_03d14695157421a1a680+2c2d707e/compile_flags.txt
@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none
\ No newline at end of file
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_03d14695157421a1a680+2c2d707e/model.hlo_module.pb b/neuronxcc-2.13.66.0+6dfecc895/MODULE_03d14695157421a1a680+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..dc3e01d6e44105286019d5d96aa0547ccb1cc0b0
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_03d14695157421a1a680+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:213ce0e6085b3b52f128d6477740453afdca1b9229151338f0a820c6a3e6f4f1
+size 752116
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_03d14695157421a1a680+2c2d707e/model.neff b/neuronxcc-2.13.66.0+6dfecc895/MODULE_03d14695157421a1a680+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..42d6f7266a90f6dfd3efe7dd08e1ce445352743c
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_03d14695157421a1a680+2c2d707e/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8d25eb7a84eff49fc5b3e3bcc50ce3e1ae6fe3b4d9d718889e0e1aba1a82e6e7
+size 8745984
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_0d2f6a194ed6bb76fa91+2c2d707e/compile_flags.txt b/neuronxcc-2.13.66.0+6dfecc895/MODULE_0d2f6a194ed6bb76fa91+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_0d2f6a194ed6bb76fa91+2c2d707e/compile_flags.txt
@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none
\ No newline at end of file
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_0d2f6a194ed6bb76fa91+2c2d707e/model.hlo_module.pb b/neuronxcc-2.13.66.0+6dfecc895/MODULE_0d2f6a194ed6bb76fa91+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..7ee5d8b27261052d90260225060f019ceef2cb7e
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_0d2f6a194ed6bb76fa91+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a5db0dbc808ab80bd8c96cfe000f1464cb03df6909d290c97b53b3fdfcb3f80a
+size 752116
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_0d2f6a194ed6bb76fa91+2c2d707e/model.neff b/neuronxcc-2.13.66.0+6dfecc895/MODULE_0d2f6a194ed6bb76fa91+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..e8694775db7600207fff292e1517512a9b4f0ac0
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_0d2f6a194ed6bb76fa91+2c2d707e/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f1d83cd0236a354e2d78e7d31f22441f02739c2f7ca57eddf6d1d497789f6120
+size 8346624
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_21b1b0223e4fb6f5af41+2c2d707e/model.neff b/neuronxcc-2.13.66.0+6dfecc895/MODULE_21b1b0223e4fb6f5af41+2c2d707e/model.neff
index e5344bb9d83957bc4f131ad0dafed18d1e2fd02b..bb2b42032c90ac6ef775ea99034b71d38c682f62 100644
--- a/neuronxcc-2.13.66.0+6dfecc895/MODULE_21b1b0223e4fb6f5af41+2c2d707e/model.neff
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_21b1b0223e4fb6f5af41+2c2d707e/model.neff
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:647d4971554b115074a5d1efb73a71b547215bc89bc5ac3af9683f91af042454
+oid sha256:a3f7daaa297382b89110fd4fa0c5bf14ea42e5008640cb0778fa0653ca455b89
 size 2161664
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_3020655465d2e0990c19+2c2d707e/model.neff b/neuronxcc-2.13.66.0+6dfecc895/MODULE_3020655465d2e0990c19+2c2d707e/model.neff
index fdad8d3612c1affd0275d7578e50d0b782b40082..f8555673c42e344003e13c844c680166bb1d4b94 100644
--- a/neuronxcc-2.13.66.0+6dfecc895/MODULE_3020655465d2e0990c19+2c2d707e/model.neff
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_3020655465d2e0990c19+2c2d707e/model.neff
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:13f64c745785649059fd2bf9f760387b9aa29ed2a436e808878ac9dcc8190b53
+oid sha256:79192a7fb5dd85b7939598c4263bfe21a439ab9aca19bbcbf1d9958e35233b94
 size 2427904
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_426cd9312d26b5e0bf95+2c2d707e/model.neff b/neuronxcc-2.13.66.0+6dfecc895/MODULE_426cd9312d26b5e0bf95+2c2d707e/model.neff
index 70bc41c9691afe9dc4facf40df773db54b47879b..3d936dde022569714b63e0c42924e1a046cbea4e 100644
--- a/neuronxcc-2.13.66.0+6dfecc895/MODULE_426cd9312d26b5e0bf95+2c2d707e/model.neff
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_426cd9312d26b5e0bf95+2c2d707e/model.neff
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7f087f3482370f029c3910b2ebce46016ca33dec12fa2b2b11dbfab59ba8acb3
+oid sha256:eeb5c95149c2da77bb7df221e50263ac22b5912e47376f52d99463f7be71256b
 size 2960384
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_44563e64a3dddda1db4d+2c2d707e/model.neff b/neuronxcc-2.13.66.0+6dfecc895/MODULE_44563e64a3dddda1db4d+2c2d707e/model.neff
index 01c74dc5245b7d51af210da8f3d9132fb6e28b6f..4de9ea6678affe73a24f3c2313cc2f849d27ced5 100644
--- a/neuronxcc-2.13.66.0+6dfecc895/MODULE_44563e64a3dddda1db4d+2c2d707e/model.neff
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_44563e64a3dddda1db4d+2c2d707e/model.neff
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:04731055ba71bc04552825fc869ee2fc20dea0efe8e56b5ce5aac95584ec347a
+oid sha256:65c582bb7ae78446c9aeeed5f800d4fb7130776361aac6f59d849f4a4e4462f7
 size 2212864
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_478e625ca58c9ae373d5+2c2d707e/compile_flags.txt b/neuronxcc-2.13.66.0+6dfecc895/MODULE_478e625ca58c9ae373d5+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_478e625ca58c9ae373d5+2c2d707e/compile_flags.txt
@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none
\ No newline at end of file
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_478e625ca58c9ae373d5+2c2d707e/model.hlo_module.pb b/neuronxcc-2.13.66.0+6dfecc895/MODULE_478e625ca58c9ae373d5+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..329473f58ced3979c229dcb9a7ea233ba7456e5d
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_478e625ca58c9ae373d5+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dbc024633722c68e8c62700fc28a645badd7c92777979a06479bf91a7f7474a2
+size 734223
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_478e625ca58c9ae373d5+2c2d707e/model.neff b/neuronxcc-2.13.66.0+6dfecc895/MODULE_478e625ca58c9ae373d5+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..9dbfa6c393cb49fdf47629b5fffb938f4abae99c
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_478e625ca58c9ae373d5+2c2d707e/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e52c56b90df846627b2c7071a497d066f6154e076138560341b68cd6f9c63114
+size 11899904
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_47eae16823191f315893+2c2d707e/model.neff b/neuronxcc-2.13.66.0+6dfecc895/MODULE_47eae16823191f315893+2c2d707e/model.neff
index 36fc45a156a35b383e802996bfd802f162eb5d6a..59c31b514ef097bac022de78897cd21cee5668c1 100644
--- a/neuronxcc-2.13.66.0+6dfecc895/MODULE_47eae16823191f315893+2c2d707e/model.neff
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_47eae16823191f315893+2c2d707e/model.neff
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:036d173fe684811ba65e05a405f930106563744240bfaf8d18160825adc5d652
+oid sha256:82aab6bd8004360ed66a862ee39f686b45f8f41f2ae8ac873f6f236c42d8cc37
 size 2274304
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_62cafa382ec618cff287+2c2d707e/model.neff b/neuronxcc-2.13.66.0+6dfecc895/MODULE_62cafa382ec618cff287+2c2d707e/model.neff
index c88ea9732819cd7f173533ccb2d84ab62ddb5b56..7accb9368b98669c97990f53af457f508d6f65d6 100644
--- a/neuronxcc-2.13.66.0+6dfecc895/MODULE_62cafa382ec618cff287+2c2d707e/model.neff
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_62cafa382ec618cff287+2c2d707e/model.neff
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:93573985c21c0950c0e93f6ce3d7ee45f5b1d47d12563f33f4074715219a26db
+oid sha256:f89ee6bb0d91b3eacec8b4dd613ed048c099f9f9b2facb8558a9a9f142037be4
 size 2171904
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_85a47887d4fc41e26097+2c2d707e/model.neff b/neuronxcc-2.13.66.0+6dfecc895/MODULE_85a47887d4fc41e26097+2c2d707e/model.neff
index daed5900b9075b7349285d213d297cb0e85d5e95..fb41c1168460e382657145bd692dd1823bc263c9 100644
--- a/neuronxcc-2.13.66.0+6dfecc895/MODULE_85a47887d4fc41e26097+2c2d707e/model.neff
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_85a47887d4fc41e26097+2c2d707e/model.neff
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f2d7bae14c521ed9e9e00d82c5ce0bd21d31dc79e87002453acf7f25aca20edc
+oid sha256:c2b62324205c04b98c089db2cb17a7caaebbaa6267f8527765a93cdc80d15b05
 size 2376704
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_8f8b48a08a0c68320dbf+2c2d707e/model.neff b/neuronxcc-2.13.66.0+6dfecc895/MODULE_8f8b48a08a0c68320dbf+2c2d707e/model.neff
index 191973652b8fd8b261e2a0d4e4acd321c9960773..dd17a073116bb56937dc14c7d23d80e9888f4d19 100644
--- a/neuronxcc-2.13.66.0+6dfecc895/MODULE_8f8b48a08a0c68320dbf+2c2d707e/model.neff
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_8f8b48a08a0c68320dbf+2c2d707e/model.neff
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0a21d384b81dc7841dfc18d2e632b49f4db793655dcd88ab78fae91fa14d02ac
+oid sha256:23f6fe17a222d4fa73f1a1592ec61556935b3e79b88cff351b570e3bc9d0f57b
 size 6626304
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_98b888cee6ab04043987+2c2d707e/model.neff b/neuronxcc-2.13.66.0+6dfecc895/MODULE_98b888cee6ab04043987+2c2d707e/model.neff
index 9fb4221e1dd7827cc93e5c9532be5bb84320753c..6a435dc35057083f9bc291413efbcc70a022cb9e 100644
--- a/neuronxcc-2.13.66.0+6dfecc895/MODULE_98b888cee6ab04043987+2c2d707e/model.neff
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_98b888cee6ab04043987+2c2d707e/model.neff
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d52de0f620c1a03cc0f067ce052ab1e5f231b9a95e1d1efd2f89e087dad2d0db
+oid sha256:7de9b5fe530dee36c5158fbe867c603ca44c22ba93f96e789e7bac61f5b82594
 size 2683904
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_a50f1bc59e07df734a75+2c2d707e/compile_flags.txt b/neuronxcc-2.13.66.0+6dfecc895/MODULE_a50f1bc59e07df734a75+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_a50f1bc59e07df734a75+2c2d707e/compile_flags.txt
@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none
\ No newline at end of file
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_a50f1bc59e07df734a75+2c2d707e/model.hlo_module.pb b/neuronxcc-2.13.66.0+6dfecc895/MODULE_a50f1bc59e07df734a75+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..d60fe498d81cca044293021eb8834dd3077db939
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_a50f1bc59e07df734a75+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d3e85b97a887c3ad128cdd3fca48532041d55c5f46e778f53f394b607a27cf33
+size 743294
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_a50f1bc59e07df734a75+2c2d707e/model.neff b/neuronxcc-2.13.66.0+6dfecc895/MODULE_a50f1bc59e07df734a75+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..b9405028ed1270793c29c67c5e95fdf9ebdc69b7
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_a50f1bc59e07df734a75+2c2d707e/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:120e020e9e227177e06731c98a62f8243f509213e8b8c8c4294a8120ce21b18f
+size 4752384
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_aa7fd25afda59bdaf747+2c2d707e/compile_flags.txt b/neuronxcc-2.13.66.0+6dfecc895/MODULE_aa7fd25afda59bdaf747+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_aa7fd25afda59bdaf747+2c2d707e/compile_flags.txt
@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none
\ No newline at end of file
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_aa7fd25afda59bdaf747+2c2d707e/model.hlo_module.pb b/neuronxcc-2.13.66.0+6dfecc895/MODULE_aa7fd25afda59bdaf747+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..5018fc9b3f4a768a068b98acac8a7d1c0df178c2
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_aa7fd25afda59bdaf747+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:18941b0dea69d5452feed647eb88902d39edc86c8eb27f3b8b6b479eec87582b
+size 752116
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_aa7fd25afda59bdaf747+2c2d707e/model.neff b/neuronxcc-2.13.66.0+6dfecc895/MODULE_aa7fd25afda59bdaf747+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..4ad67cc9172ce624b12dc7dc367f3e05cdd8a888
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_aa7fd25afda59bdaf747+2c2d707e/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9d8ad8abb9be99298a40f26d5bc418437d6e722ba13171ca7f0658fa7ec1b173
+size 8489984
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_bd38c2a884aabaec3d62+2c2d707e/compile_flags.txt b/neuronxcc-2.13.66.0+6dfecc895/MODULE_bd38c2a884aabaec3d62+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_bd38c2a884aabaec3d62+2c2d707e/compile_flags.txt
@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none
\ No newline at end of file
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_bd38c2a884aabaec3d62+2c2d707e/model.hlo_module.pb b/neuronxcc-2.13.66.0+6dfecc895/MODULE_bd38c2a884aabaec3d62+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..f344b024271e45e6be25fa2b07858f57a25317b1
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_bd38c2a884aabaec3d62+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f6445a12c039fbfe162979d6c639dc1a79c698d7fff5a282d5eb8c97caa01876
+size 752116
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_bd38c2a884aabaec3d62+2c2d707e/model.neff b/neuronxcc-2.13.66.0+6dfecc895/MODULE_bd38c2a884aabaec3d62+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..237bd844e2efe88f05d51e56ed10e7219f97cd00
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_bd38c2a884aabaec3d62+2c2d707e/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0cdd4eccbec0ffd1c51733308df86e50ed9c9457c1daa112b4eedda4330a2169
+size 9339904
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_c7bc5bab6ec8ea2e1063+2c2d707e/compile_flags.txt b/neuronxcc-2.13.66.0+6dfecc895/MODULE_c7bc5bab6ec8ea2e1063+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_c7bc5bab6ec8ea2e1063+2c2d707e/compile_flags.txt
@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none
\ No newline at end of file
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_c7bc5bab6ec8ea2e1063+2c2d707e/model.hlo_module.pb b/neuronxcc-2.13.66.0+6dfecc895/MODULE_c7bc5bab6ec8ea2e1063+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..8155cc455b487d843520416352f8d318a27e1553
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_c7bc5bab6ec8ea2e1063+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7a1c8eee12bb20857c2b8b583a9fa0a84a2e2edf6e5e88b0c4bc2e54678ffc4c
+size 743294
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_c7bc5bab6ec8ea2e1063+2c2d707e/model.neff b/neuronxcc-2.13.66.0+6dfecc895/MODULE_c7bc5bab6ec8ea2e1063+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..ced50767ceae659a4d588720bafb589c186dc1d8
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_c7bc5bab6ec8ea2e1063+2c2d707e/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4f76006ff13403691e7ad997d85ccc3d7fdb481dbddb080c2851fcb3851e71de
+size 14941184
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_cda75e05b1d55681d8eb+2c2d707e/model.neff b/neuronxcc-2.13.66.0+6dfecc895/MODULE_cda75e05b1d55681d8eb+2c2d707e/model.neff
index eaf8c1156b62fac3458e8a63f58921b5bb71d70d..54403e72ad69eb2dde17b16b0ecf93a12965891e 100644
--- a/neuronxcc-2.13.66.0+6dfecc895/MODULE_cda75e05b1d55681d8eb+2c2d707e/model.neff
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_cda75e05b1d55681d8eb+2c2d707e/model.neff
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fe8959429da487468dd65b82b1662733a18d5085f86938874f0dec838e7e6a7b
+oid sha256:a5b4a2b6a3d071aa86bb62b8550da5069129c9d6d904827e2b3ba851bece4924
 size 27638784
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_d66f5802e5ad890e3919+2c2d707e/model.neff b/neuronxcc-2.13.66.0+6dfecc895/MODULE_d66f5802e5ad890e3919+2c2d707e/model.neff
index b82324a7eeae5404d946ba4d99e9cc9ecb821ef2..71c9e6dfac5461763184a5b2f80fdbbaabcac212 100644
--- a/neuronxcc-2.13.66.0+6dfecc895/MODULE_d66f5802e5ad890e3919+2c2d707e/model.neff
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_d66f5802e5ad890e3919+2c2d707e/model.neff
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1341e072b22a545183876855a68e6e7447f5b889e09a44a4b910aefcb73d992e
+oid sha256:a277f2c607a99e6ffe5fe4020145efe62fcc3585f49eec2edcc9bc251a172ca9
 size 13865984
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_de4e8d5ed85f8709144f+2c2d707e/compile_flags.txt b/neuronxcc-2.13.66.0+6dfecc895/MODULE_de4e8d5ed85f8709144f+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_de4e8d5ed85f8709144f+2c2d707e/compile_flags.txt
@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none
\ No newline at end of file
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_de4e8d5ed85f8709144f+2c2d707e/model.hlo_module.pb b/neuronxcc-2.13.66.0+6dfecc895/MODULE_de4e8d5ed85f8709144f+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..197e8fe7e0b1cde31c795e11804618beb8cf3025
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_de4e8d5ed85f8709144f+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:657824b7fd140d64fade75b94a89f0b571840ac90c38e3e4f05a0798b375be91
+size 752116
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_de4e8d5ed85f8709144f+2c2d707e/model.neff b/neuronxcc-2.13.66.0+6dfecc895/MODULE_de4e8d5ed85f8709144f+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..ceb2b0b4ce1fe4b0067167c296f2b238e9e51a7d
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_de4e8d5ed85f8709144f+2c2d707e/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a6e62cf9471721494f695d0287f1a4ad8220137b156b9b282e5d423a3f629c80
+size 8244224
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_e291d316cec0d0e9de79+2c2d707e/compile_flags.txt b/neuronxcc-2.13.66.0+6dfecc895/MODULE_e291d316cec0d0e9de79+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_e291d316cec0d0e9de79+2c2d707e/compile_flags.txt
@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none
\ No newline at end of file
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_e291d316cec0d0e9de79+2c2d707e/model.hlo_module.pb b/neuronxcc-2.13.66.0+6dfecc895/MODULE_e291d316cec0d0e9de79+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..c996fb3f72093b37d1e5ca2de4db4f8b9d9610c0
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_e291d316cec0d0e9de79+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c8caf017becab8743e035dd9933c01e1b6a023ebe211fce6dae2c4b4bcf84b77
+size 743294
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_e291d316cec0d0e9de79+2c2d707e/model.neff b/neuronxcc-2.13.66.0+6dfecc895/MODULE_e291d316cec0d0e9de79+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..35397e6c6cd5a7731f8685df09e704773b0303e7
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_e291d316cec0d0e9de79+2c2d707e/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bd165faa2f4d99ac146fd58cc4c1eedfde0dde1bd8e71838e0763144eab570b4
+size 11367424
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_ea017e5effcd92c4e6ff+2c2d707e/model.neff b/neuronxcc-2.13.66.0+6dfecc895/MODULE_ea017e5effcd92c4e6ff+2c2d707e/model.neff
index f9b9443ccce25ff3f722e986a7fe3da2d6ad0d44..67e1279c2a2dad789e38fb7116fe5ffa94a757f2 100644
--- a/neuronxcc-2.13.66.0+6dfecc895/MODULE_ea017e5effcd92c4e6ff+2c2d707e/model.neff
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_ea017e5effcd92c4e6ff+2c2d707e/model.neff
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a272b8c214312f8d6fc5e9ab9085a3f961b155f888b4d5ba7a4c8d30da5701d3
+oid sha256:697cb25356bbcc11de484b4afb82bf2d36f35ade7cfe21635fab57d33c6f1da5
 size 3503104
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_f2145a17aa8d689f7038+2c2d707e/model.neff b/neuronxcc-2.13.66.0+6dfecc895/MODULE_f2145a17aa8d689f7038+2c2d707e/model.neff
index f36d1c3c529b97ae6f8e3fbb3f9423a801724987..f3c868e17a2539a4b135a40ad47f5d08d6638472 100644
--- a/neuronxcc-2.13.66.0+6dfecc895/MODULE_f2145a17aa8d689f7038+2c2d707e/model.neff
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_f2145a17aa8d689f7038+2c2d707e/model.neff
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8a3c0cea92a6bc0003fc88804f657a009a14701935c81fd6b81f9de5e201a579
+oid sha256:d148f449dda8e52cae65644cb3ef26740b217f4b9a074abe6ce4eadfd4dec357
 size 2673664
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_f74f07a635721468c2ae+2c2d707e/model.neff b/neuronxcc-2.13.66.0+6dfecc895/MODULE_f74f07a635721468c2ae+2c2d707e/model.neff
index 8f750a0f68529bf56782e8f22547d932e7b5b7be..59c8643afa1badfc192ad19b88783ca8de65f54d 100644
--- a/neuronxcc-2.13.66.0+6dfecc895/MODULE_f74f07a635721468c2ae+2c2d707e/model.neff
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_f74f07a635721468c2ae+2c2d707e/model.neff
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:68a99502d6060a605216bd6dd9c4ebe325432f745e4a3f1794432ebb7ed556db
+oid sha256:4209b3e1de8a9d02d5bcf8c076761bdfc757d5b54504e530091320a695210ffd
 size 4711424
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_fc0f4892d01029ca630a+2c2d707e/compile_flags.txt b/neuronxcc-2.13.66.0+6dfecc895/MODULE_fc0f4892d01029ca630a+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_fc0f4892d01029ca630a+2c2d707e/compile_flags.txt
@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none
\ No newline at end of file
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_fc0f4892d01029ca630a+2c2d707e/model.hlo_module.pb b/neuronxcc-2.13.66.0+6dfecc895/MODULE_fc0f4892d01029ca630a+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..eabe020f21092bbad9251e8696d28ba5a949c2c1
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_fc0f4892d01029ca630a+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4e8c194ca31579866ddfcae54bcd89447cd503acce943b0db20b2d83a047abc4
+size 743294
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_fc0f4892d01029ca630a+2c2d707e/model.neff b/neuronxcc-2.13.66.0+6dfecc895/MODULE_fc0f4892d01029ca630a+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..c163ffa3713c64aeda1048a21cb584a8b5e4402b
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_fc0f4892d01029ca630a+2c2d707e/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:196d7275a9ba1650906b209b5a9ce06433163c5eee13a55509bfb8158da66b2f
+size 27720704
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_ff52385503277deb3b07+2c2d707e/compile_flags.txt b/neuronxcc-2.13.66.0+6dfecc895/MODULE_ff52385503277deb3b07+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_ff52385503277deb3b07+2c2d707e/compile_flags.txt
@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none
\ No newline at end of file
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_ff52385503277deb3b07+2c2d707e/model.hlo_module.pb b/neuronxcc-2.13.66.0+6dfecc895/MODULE_ff52385503277deb3b07+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..d9bbc251acfa0fe4dbb3aec82677c4ea5da2420c
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_ff52385503277deb3b07+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:19d088abd8c10743c8277e61fe3398d8b91aa53a02cfe3df29c2707e7ce2fece
+size 743294
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_ff52385503277deb3b07+2c2d707e/model.neff b/neuronxcc-2.13.66.0+6dfecc895/MODULE_ff52385503277deb3b07+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..37a47310262815ffe7ce0960819113b2cf836f43
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_ff52385503277deb3b07+2c2d707e/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f5f7f9d8c44da6438f5ccd74e4fbe9aca20c937cd183b224d70c8b9c463dfe50
+size 12237824