diff --git a/.gitattributes b/.gitattributes
index e85259d1a1b571aa9845cf0138928bd04e04877a..af255522bfb91229497980890805dc9b3b575e3f 100644
--- a/.gitattributes
+++ b/.gitattributes
@@ -2837,3 +2837,15 @@ neuronxcc-2.13.66.0+6dfecc895/MODULE_b2ff307fe8e711bcff87+2c2d707e/model.neff fi
 neuronxcc-2.13.66.0+6dfecc895/MODULE_c62ad0accecd75d9e90c+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.13.66.0+6dfecc895/MODULE_db946049e0191d823a61+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.13.66.0+6dfecc895/MODULE_ebd8aa830ea881614d11+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.66.0+6dfecc895/MODULE_006a2b0612d6873eeb14+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.66.0+6dfecc895/MODULE_1d9dbf58321f64ca26bf+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.66.0+6dfecc895/MODULE_3230e71983c51166ba38+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.66.0+6dfecc895/MODULE_33b1f5d697aafe44fc58+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.66.0+6dfecc895/MODULE_3efc2d85c6bd2f6fb618+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.66.0+6dfecc895/MODULE_4338ee87917dbff33f5c+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.66.0+6dfecc895/MODULE_6508011bf68b0a3af4a4+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.66.0+6dfecc895/MODULE_7ff43e7c899f3ddc1ce1+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.66.0+6dfecc895/MODULE_8562f75b7b365525939a+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.66.0+6dfecc895/MODULE_8c86a13c91862ed950e8+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.66.0+6dfecc895/MODULE_f217be79d7cba418065f+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.66.0+6dfecc895/MODULE_f6f59d536472808f0be5+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-2-13b-chat-hf/cb6add0ea39952474cf6.json b/neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.22/inference/llama/meta-llama/Llama-2-13b-chat-hf/cb6add0ea39952474cf6.json
new file mode 100644
index 0000000000000000000000000000000000000000..5806ec4bdb550c0dd819d878ff8da6b5349d9924
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.22/inference/llama/meta-llama/Llama-2-13b-chat-hf/cb6add0ea39952474cf6.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": 5120, "initializer_range": 0.02, "intermediate_size": 13824, "max_position_embeddings": 4096, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 1, "checkpoint_id": "meta-llama/Llama-2-13b-chat-hf", "checkpoint_revision": "a2cb7a712bb6e5e736ca7f8cd98167f81a0b5bd8", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 12, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 40, "num_hidden_layers": 40, "num_key_value_heads": 40, "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_006a2b0612d6873eeb14+2c2d707e/compile_flags.txt b/neuronxcc-2.13.66.0+6dfecc895/MODULE_006a2b0612d6873eeb14+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_006a2b0612d6873eeb14+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_006a2b0612d6873eeb14+2c2d707e/model.hlo_module.pb b/neuronxcc-2.13.66.0+6dfecc895/MODULE_006a2b0612d6873eeb14+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..4516b656e6913abaf5a61287a60466471c31eb14
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_006a2b0612d6873eeb14+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6cf5f6929761b153bdd8ec671775096d63e66fde19267ffadb137c7abd05e4b4
+size 365748
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_006a2b0612d6873eeb14+2c2d707e/model.neff b/neuronxcc-2.13.66.0+6dfecc895/MODULE_006a2b0612d6873eeb14+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..978c62e5e3d8e50f0908e539b89994deac7bfed8
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_006a2b0612d6873eeb14+2c2d707e/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e757f9a405a715e971c1925c71cbfd7705947e084b4c94d08d5cb04d6392b72e
+size 3769344
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_1d9dbf58321f64ca26bf+2c2d707e/compile_flags.txt b/neuronxcc-2.13.66.0+6dfecc895/MODULE_1d9dbf58321f64ca26bf+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_1d9dbf58321f64ca26bf+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_1d9dbf58321f64ca26bf+2c2d707e/model.hlo_module.pb b/neuronxcc-2.13.66.0+6dfecc895/MODULE_1d9dbf58321f64ca26bf+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..076eb444b2b13da6931bfa28b65abf12204d0653
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_1d9dbf58321f64ca26bf+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b12667ab0698d566d36f4b5ab535530ec4c6b16d91c14b1f4b60b5af7ad5f41c
+size 374555
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_1d9dbf58321f64ca26bf+2c2d707e/model.neff b/neuronxcc-2.13.66.0+6dfecc895/MODULE_1d9dbf58321f64ca26bf+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..c1160e966076ca8d5f594e32044ebe2a75ecd807
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_1d9dbf58321f64ca26bf+2c2d707e/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b326ebc65d719fca1075fd0461e67a86a204ac3ad6b15a0c043b19f0ea190421
+size 2663424
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_3230e71983c51166ba38+2c2d707e/compile_flags.txt b/neuronxcc-2.13.66.0+6dfecc895/MODULE_3230e71983c51166ba38+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_3230e71983c51166ba38+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_3230e71983c51166ba38+2c2d707e/model.hlo_module.pb b/neuronxcc-2.13.66.0+6dfecc895/MODULE_3230e71983c51166ba38+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..85c384517eea4c227ba3f29c74d1c0fe9502cc7d
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_3230e71983c51166ba38+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ea25b05a9d522255e601a95f0fc5a9be39170f243277fb46f111441d672de75c
+size 374555
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_3230e71983c51166ba38+2c2d707e/model.neff b/neuronxcc-2.13.66.0+6dfecc895/MODULE_3230e71983c51166ba38+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..136d4378b0b1ca34c51428ef7cd1ef9109dcd7e9
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_3230e71983c51166ba38+2c2d707e/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ae64a8645c86a222d127b5f0d3ea38a5f2dccabf525b856c07300babff7ad711
+size 2960384
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_33b1f5d697aafe44fc58+2c2d707e/compile_flags.txt b/neuronxcc-2.13.66.0+6dfecc895/MODULE_33b1f5d697aafe44fc58+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_33b1f5d697aafe44fc58+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_33b1f5d697aafe44fc58+2c2d707e/model.hlo_module.pb b/neuronxcc-2.13.66.0+6dfecc895/MODULE_33b1f5d697aafe44fc58+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..8fe327e8bad842cf012d0456c04f89479efaf45b
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_33b1f5d697aafe44fc58+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5a7c0239f2ab37cc9b311d00ae4e2c8dd6745f7107f795db3a95fbf3d0ed9d82
+size 370299
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_33b1f5d697aafe44fc58+2c2d707e/model.neff b/neuronxcc-2.13.66.0+6dfecc895/MODULE_33b1f5d697aafe44fc58+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..45870cac7db97a940f8e51127515e0b71ddb9027
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_33b1f5d697aafe44fc58+2c2d707e/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f88f1cdb92c04b2e4e55ef9cec60332d659cd907eb8a373c280c74bbc2c32893
+size 5182464
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_3efc2d85c6bd2f6fb618+2c2d707e/compile_flags.txt b/neuronxcc-2.13.66.0+6dfecc895/MODULE_3efc2d85c6bd2f6fb618+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_3efc2d85c6bd2f6fb618+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_3efc2d85c6bd2f6fb618+2c2d707e/model.hlo_module.pb b/neuronxcc-2.13.66.0+6dfecc895/MODULE_3efc2d85c6bd2f6fb618+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..10bb452e0fad6b72e1ea5ba59bb875a5e927cc12
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_3efc2d85c6bd2f6fb618+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2363c55a3b77b81a3af5eedb5345b2b82bb82bc3492fab54d1d5ec182403e578
+size 374555
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_3efc2d85c6bd2f6fb618+2c2d707e/model.neff b/neuronxcc-2.13.66.0+6dfecc895/MODULE_3efc2d85c6bd2f6fb618+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..82118832221037f95dfd2076e25f2c66301ec431
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_3efc2d85c6bd2f6fb618+2c2d707e/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:70e010e06cf308b68b1e3ec35c9433206a01dd104f4981cfa27a9768c4983df3
+size 3298304
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_4338ee87917dbff33f5c+2c2d707e/compile_flags.txt b/neuronxcc-2.13.66.0+6dfecc895/MODULE_4338ee87917dbff33f5c+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_4338ee87917dbff33f5c+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_4338ee87917dbff33f5c+2c2d707e/model.hlo_module.pb b/neuronxcc-2.13.66.0+6dfecc895/MODULE_4338ee87917dbff33f5c+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..ef15505dba259bebd00e6b77ad2d1814880083e0
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_4338ee87917dbff33f5c+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:58855406abc6bff25a240cbe93a8eddfe5dc1d43a9dbaad1fee0646637de106f
+size 370299
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_4338ee87917dbff33f5c+2c2d707e/model.neff b/neuronxcc-2.13.66.0+6dfecc895/MODULE_4338ee87917dbff33f5c+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..a32900eb902009d7e4874b8316cb417fe31cd9b5
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_4338ee87917dbff33f5c+2c2d707e/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d1b545e90be3bd46f4b641c3dd1ee31a023b89708ae5b12c9bde0fb407ce61a4
+size 9769984
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_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_6508011bf68b0a3af4a4+2c2d707e/compile_flags.txt b/neuronxcc-2.13.66.0+6dfecc895/MODULE_6508011bf68b0a3af4a4+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_6508011bf68b0a3af4a4+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_6508011bf68b0a3af4a4+2c2d707e/model.hlo_module.pb b/neuronxcc-2.13.66.0+6dfecc895/MODULE_6508011bf68b0a3af4a4+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..a96d82a576ba08d5179e94747340b8804869baad
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_6508011bf68b0a3af4a4+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:79096725e6d9ed8363b860e92ae3e66ba975b7e8578846f5f96765eef566d41e
+size 374555
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_6508011bf68b0a3af4a4+2c2d707e/model.neff b/neuronxcc-2.13.66.0+6dfecc895/MODULE_6508011bf68b0a3af4a4+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..00ee37617ee8540da2f74001aa160e0a5e2eee86
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_6508011bf68b0a3af4a4+2c2d707e/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:312368790769801027e7603b38e74507996f0d370b503a22063bba1bf6363208
+size 2796544
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_7ff43e7c899f3ddc1ce1+2c2d707e/compile_flags.txt b/neuronxcc-2.13.66.0+6dfecc895/MODULE_7ff43e7c899f3ddc1ce1+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_7ff43e7c899f3ddc1ce1+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_7ff43e7c899f3ddc1ce1+2c2d707e/model.hlo_module.pb b/neuronxcc-2.13.66.0+6dfecc895/MODULE_7ff43e7c899f3ddc1ce1+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..0f45d65fa1555aad989079d6250286fceae08248
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_7ff43e7c899f3ddc1ce1+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:35e64d0aa6e6ec01fd5c96c3783b37a46c2eb4a449450f023b2246bf5daa367c
+size 370299
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_7ff43e7c899f3ddc1ce1+2c2d707e/model.neff b/neuronxcc-2.13.66.0+6dfecc895/MODULE_7ff43e7c899f3ddc1ce1+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..bf154007526e02425f661b69336bcce2be1f6dfc
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_7ff43e7c899f3ddc1ce1+2c2d707e/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1cd529ba89d877862df3df7eaab3feebc483bc24cc49bea5e54a7dc142361300
+size 20030464
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_8562f75b7b365525939a+2c2d707e/compile_flags.txt b/neuronxcc-2.13.66.0+6dfecc895/MODULE_8562f75b7b365525939a+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_8562f75b7b365525939a+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_8562f75b7b365525939a+2c2d707e/model.hlo_module.pb b/neuronxcc-2.13.66.0+6dfecc895/MODULE_8562f75b7b365525939a+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..92b57aa088cd70c311ee75ea7158845e56b33242
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_8562f75b7b365525939a+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:09637ea11ae6545f05c3db7ea84e9741039bded3f1e2f2d86752f6aac74c65f8
+size 370299
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_8562f75b7b365525939a+2c2d707e/model.neff b/neuronxcc-2.13.66.0+6dfecc895/MODULE_8562f75b7b365525939a+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..fae9134f7ec90943ef9a75552782b78459093ddd
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_8562f75b7b365525939a+2c2d707e/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6aa368e36e598bf7e66d8188165469ef5a8be08105ccef96db0ffc224c0cc522
+size 4281344
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_8c86a13c91862ed950e8+2c2d707e/compile_flags.txt b/neuronxcc-2.13.66.0+6dfecc895/MODULE_8c86a13c91862ed950e8+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_8c86a13c91862ed950e8+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_8c86a13c91862ed950e8+2c2d707e/model.hlo_module.pb b/neuronxcc-2.13.66.0+6dfecc895/MODULE_8c86a13c91862ed950e8+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..cde76882467ff1561a4992e3831f2605f4ab62b0
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_8c86a13c91862ed950e8+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d8bda12e438ae4ea0439de451c05c19f63f6911b0ee0b8f2478b7434f802ef67
+size 374555
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_8c86a13c91862ed950e8+2c2d707e/model.neff b/neuronxcc-2.13.66.0+6dfecc895/MODULE_8c86a13c91862ed950e8+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..4e675ff37dc475dc92718697cf33080403c59c5f
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_8c86a13c91862ed950e8+2c2d707e/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:37365dbda345c3aad0e3ffdf0ae6db863f1d724c9a8bae8a6f0d3dd29a947831
+size 2653184
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_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_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_f217be79d7cba418065f+2c2d707e/compile_flags.txt b/neuronxcc-2.13.66.0+6dfecc895/MODULE_f217be79d7cba418065f+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_f217be79d7cba418065f+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_f217be79d7cba418065f+2c2d707e/model.hlo_module.pb b/neuronxcc-2.13.66.0+6dfecc895/MODULE_f217be79d7cba418065f+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..011fa419a9d34a3ecf9899fd9f753860302ef508
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_f217be79d7cba418065f+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:68c6b8f316592f7e978194b9265caf9eaaac25c5e4f6bd82c3b1ffd21a7502fa
+size 374555
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_f217be79d7cba418065f+2c2d707e/model.neff b/neuronxcc-2.13.66.0+6dfecc895/MODULE_f217be79d7cba418065f+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..2e39a58d94f3a6b8a16669b636eae72565dc8735
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_f217be79d7cba418065f+2c2d707e/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7fe84ad614fc908cb4885861bc0c721f1e54c15392f10005056a0509bdc1c811
+size 2704384
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_f6f59d536472808f0be5+2c2d707e/compile_flags.txt b/neuronxcc-2.13.66.0+6dfecc895/MODULE_f6f59d536472808f0be5+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_f6f59d536472808f0be5+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_f6f59d536472808f0be5+2c2d707e/model.hlo_module.pb b/neuronxcc-2.13.66.0+6dfecc895/MODULE_f6f59d536472808f0be5+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..67c1b060dd188fcd52df02c75ab552ea1c687eb1
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_f6f59d536472808f0be5+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a0546f57e080fe6c147a65a4470e80f7465dec8d9611f81e321be6de0d8ea491
+size 370299
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_f6f59d536472808f0be5+2c2d707e/model.neff b/neuronxcc-2.13.66.0+6dfecc895/MODULE_f6f59d536472808f0be5+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..8e458275f1d99042466f8ad0bfb5b788e5ce5432
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_f6f59d536472808f0be5+2c2d707e/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:228a6922e1214d0135ce1c4646d53f8d6b3f5fa70bf2a32d70d0852695551fad
+size 3462144
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