diff --git a/.gitattributes b/.gitattributes
index 10ba579ed9107645354539591b690d743e60933e..c5689448178ec7cc369fe4c6e34d87b514bc9226 100644
--- a/.gitattributes
+++ b/.gitattributes
@@ -2797,3 +2797,15 @@ neuronxcc-2.13.66.0+6dfecc895/MODULE_e8375e827aa65e3c9d6d+2c2d707e/model.neff fi
 neuronxcc-2.13.66.0+6dfecc895/MODULE_ee55eab3eac6b30f23a0+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.13.66.0+6dfecc895/MODULE_91d99978aaab3b136791+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.13.66.0+6dfecc895/MODULE_be50f74b25b299fbb342+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.66.0+6dfecc895/MODULE_3e1ed7f9f637696e3bfd+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.66.0+6dfecc895/MODULE_3ec3b33f65fa40a563b9+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.66.0+6dfecc895/MODULE_96f8e185f66787463fed+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.66.0+6dfecc895/MODULE_9bb14fc87dc287829e65+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.66.0+6dfecc895/MODULE_ae95f5c04c529ba0e54c+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.66.0+6dfecc895/MODULE_b191a0a43fedcfa9ac37+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.66.0+6dfecc895/MODULE_b4732b9a44b6941efdef+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.66.0+6dfecc895/MODULE_b499ff57794bd2b55273+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.66.0+6dfecc895/MODULE_b6f59521083a4c513238+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.66.0+6dfecc895/MODULE_d755a6931d0c1f489fa9+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.66.0+6dfecc895/MODULE_d7606d9adfeb9cb0fd73+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.13.66.0+6dfecc895/MODULE_dfb85ebd0d1b542fb616+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
diff --git a/neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/llama/meta-llama/Llama-2-7b-chat-hf/13cbb9ec2234e7ca3c26.json b/neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/llama/meta-llama/Llama-2-7b-chat-hf/13cbb9ec2234e7ca3c26.json
new file mode 100644
index 0000000000000000000000000000000000000000..cfa3942d30ac8e2a85d90d95d5932040b10f4039
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/llama/meta-llama/Llama-2-7b-chat-hf/13cbb9ec2234e7ca3c26.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": 4096, "initializer_range": 0.02, "intermediate_size": 11008, "max_position_embeddings": 4096, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 8, "checkpoint_id": "meta-llama/Llama-2-7b-chat-hf", "checkpoint_revision": "92011f62d7604e261f748ec0cfe6329f31193e33", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 24, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 32, "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+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/llama/meta-llama/Llama-2-7b-chat-hf/22d316bd8b3cdfb5514b.json b/neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/llama/meta-llama/Llama-2-7b-chat-hf/22d316bd8b3cdfb5514b.json
new file mode 100644
index 0000000000000000000000000000000000000000..0c514701ab0939842c076dcab78db0f705f793ae
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/llama/meta-llama/Llama-2-7b-chat-hf/22d316bd8b3cdfb5514b.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": 4096, "initializer_range": 0.02, "intermediate_size": 11008, "max_position_embeddings": 4096, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 16, "checkpoint_id": "meta-llama/Llama-2-7b-chat-hf", "checkpoint_revision": "92011f62d7604e261f748ec0cfe6329f31193e33", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 24, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 32, "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+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/llama/meta-llama/Llama-2-7b-chat-hf/2e95f36ce181921d7a7b.json b/neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/llama/meta-llama/Llama-2-7b-chat-hf/2e95f36ce181921d7a7b.json
new file mode 100644
index 0000000000000000000000000000000000000000..2f30a090a70bffaafb6c6e181c27057e15d40713
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/llama/meta-llama/Llama-2-7b-chat-hf/2e95f36ce181921d7a7b.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": 4096, "initializer_range": 0.02, "intermediate_size": 11008, "max_position_embeddings": 4096, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 4, "checkpoint_id": "meta-llama/Llama-2-7b-chat-hf", "checkpoint_revision": "92011f62d7604e261f748ec0cfe6329f31193e33", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 24, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 32, "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+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/llama/meta-llama/Llama-2-7b-chat-hf/4d28ec49befb4e87ec23.json b/neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/llama/meta-llama/Llama-2-7b-chat-hf/4d28ec49befb4e87ec23.json
new file mode 100644
index 0000000000000000000000000000000000000000..7fd0f77083436b761e2017a2804a8db5c8e3eece
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/llama/meta-llama/Llama-2-7b-chat-hf/4d28ec49befb4e87ec23.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": 4096, "initializer_range": 0.02, "intermediate_size": 11008, "max_position_embeddings": 4096, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 32, "checkpoint_id": "meta-llama/Llama-2-7b-chat-hf", "checkpoint_revision": "92011f62d7604e261f748ec0cfe6329f31193e33", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 24, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 32, "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+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/llama/meta-llama/Llama-2-7b-chat-hf/7a9166d6f50e7c014961.json b/neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/llama/meta-llama/Llama-2-7b-chat-hf/7a9166d6f50e7c014961.json
new file mode 100644
index 0000000000000000000000000000000000000000..d9b942a9250234ebf7d1d3dc8eecaafa9c9096e8
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/llama/meta-llama/Llama-2-7b-chat-hf/7a9166d6f50e7c014961.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": 4096, "initializer_range": 0.02, "intermediate_size": 11008, "max_position_embeddings": 4096, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 32, "checkpoint_id": "meta-llama/Llama-2-7b-chat-hf", "checkpoint_revision": "92011f62d7604e261f748ec0cfe6329f31193e33", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 32, "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+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/llama/meta-llama/Llama-2-7b-chat-hf/d3acdf9ff1795080fd1c.json b/neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/llama/meta-llama/Llama-2-7b-chat-hf/d3acdf9ff1795080fd1c.json
new file mode 100644
index 0000000000000000000000000000000000000000..137de61fc0a048bd16b7f5490141116caea96f5c
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/llama/meta-llama/Llama-2-7b-chat-hf/d3acdf9ff1795080fd1c.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": 4096, "initializer_range": 0.02, "intermediate_size": 11008, "max_position_embeddings": 4096, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 1, "checkpoint_id": "meta-llama/Llama-2-7b-chat-hf", "checkpoint_revision": "92011f62d7604e261f748ec0cfe6329f31193e33", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 24, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 32, "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+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/llama/meta-llama/Llama-2-7b-chat-hf/ff6b3f58fac88c2f96c4.json b/neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/llama/meta-llama/Llama-2-7b-chat-hf/ff6b3f58fac88c2f96c4.json
new file mode 100644
index 0000000000000000000000000000000000000000..f88abb5045d1730cb5418d567a9f36b3ea71d2eb
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/llama/meta-llama/Llama-2-7b-chat-hf/ff6b3f58fac88c2f96c4.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": 4096, "initializer_range": 0.02, "intermediate_size": 11008, "max_position_embeddings": 4096, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 16, "checkpoint_id": "meta-llama/Llama-2-7b-chat-hf", "checkpoint_revision": "92011f62d7604e261f748ec0cfe6329f31193e33", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 32, "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+6dfecc895/0_REGISTRY/0.0.22/inference/llama/m-a-p/OpenCodeInterpreter-DS-6.7B/331556dfa1f920d3e73b.json b/neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.22/inference/llama/m-a-p/OpenCodeInterpreter-DS-6.7B/331556dfa1f920d3e73b.json
new file mode 100644
index 0000000000000000000000000000000000000000..f254670bb54383cb000ba6bc6f01fa092e0f97ab
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.22/inference/llama/m-a-p/OpenCodeInterpreter-DS-6.7B/331556dfa1f920d3e73b.json
@@ -0,0 +1 @@
+{"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 32013, "eos_token_id": 32021, "hidden_act": "silu", "hidden_size": 4096, "initializer_range": 0.02, "intermediate_size": 11008, "max_position_embeddings": 16384, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 1, "checkpoint_id": "m-a-p/OpenCodeInterpreter-DS-6.7B", "checkpoint_revision": "60b89884df814590abd76757a6db4a527cbdfc91", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 2, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 32, "pretraining_tp": 1, "rms_norm_eps": 1e-06, "rope_scaling": {"factor": 4.0, "type": "linear"}, "rope_theta": 100000, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": false, "vocab_size": 32256}
\ No newline at end of file
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 bb2b42032c90ac6ef775ea99034b71d38c682f62..e5344bb9d83957bc4f131ad0dafed18d1e2fd02b 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:a3f7daaa297382b89110fd4fa0c5bf14ea42e5008640cb0778fa0653ca455b89
+oid sha256:647d4971554b115074a5d1efb73a71b547215bc89bc5ac3af9683f91af042454
 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 f8555673c42e344003e13c844c680166bb1d4b94..fdad8d3612c1affd0275d7578e50d0b782b40082 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 sha256:79192a7fb5dd85b7939598c4263bfe21a439ab9aca19bbcbf1d9958e35233b94
+oid sha256:13f64c745785649059fd2bf9f760387b9aa29ed2a436e808878ac9dcc8190b53
 size 2427904
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_3e1ed7f9f637696e3bfd+2c2d707e/compile_flags.txt b/neuronxcc-2.13.66.0+6dfecc895/MODULE_3e1ed7f9f637696e3bfd+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_3e1ed7f9f637696e3bfd+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_3e1ed7f9f637696e3bfd+2c2d707e/model.hlo_module.pb b/neuronxcc-2.13.66.0+6dfecc895/MODULE_3e1ed7f9f637696e3bfd+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..a389888246278bc44a6a243ebd9f6beb510bc73c
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_3e1ed7f9f637696e3bfd+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e8743a25d8fa47dfd555ac6636faf6ba57c8e3a00101c2a339bc319d2729bbb7
+size 295835
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_3e1ed7f9f637696e3bfd+2c2d707e/model.neff b/neuronxcc-2.13.66.0+6dfecc895/MODULE_3e1ed7f9f637696e3bfd+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..d919f064393ceb50c6411ce0db106d3f35e18b63
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_3e1ed7f9f637696e3bfd+2c2d707e/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:13189effc6a775bffecd02cbd33a57d1cdd6c16fac47df7477be57b4eba65592
+size 7558144
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_3ec3b33f65fa40a563b9+2c2d707e/compile_flags.txt b/neuronxcc-2.13.66.0+6dfecc895/MODULE_3ec3b33f65fa40a563b9+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_3ec3b33f65fa40a563b9+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_3ec3b33f65fa40a563b9+2c2d707e/model.hlo_module.pb b/neuronxcc-2.13.66.0+6dfecc895/MODULE_3ec3b33f65fa40a563b9+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..b1e365ac287062e55c842812f1394e60436f9bd2
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_3ec3b33f65fa40a563b9+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0f1bae4b411946a9f2f720b0aa42e035289a6aa8c14a44d2b6ae65b33c19cb10
+size 295835
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_3ec3b33f65fa40a563b9+2c2d707e/model.neff b/neuronxcc-2.13.66.0+6dfecc895/MODULE_3ec3b33f65fa40a563b9+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..bff86065110c35d68e751fc8141c9a10ae85b30c
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_3ec3b33f65fa40a563b9+2c2d707e/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6cbf81668a43bc7d761e2bfc161d78c20a9eee6528f2ea04bccf73588e643cf0
+size 22703104
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 3d936dde022569714b63e0c42924e1a046cbea4e..70bc41c9691afe9dc4facf40df773db54b47879b 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:eeb5c95149c2da77bb7df221e50263ac22b5912e47376f52d99463f7be71256b
+oid sha256:7f087f3482370f029c3910b2ebce46016ca33dec12fa2b2b11dbfab59ba8acb3
 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 4de9ea6678affe73a24f3c2313cc2f849d27ced5..01c74dc5245b7d51af210da8f3d9132fb6e28b6f 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 sha256:65c582bb7ae78446c9aeeed5f800d4fb7130776361aac6f59d849f4a4e4462f7
+oid sha256:04731055ba71bc04552825fc869ee2fc20dea0efe8e56b5ce5aac95584ec347a
 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 59c31b514ef097bac022de78897cd21cee5668c1..36fc45a156a35b383e802996bfd802f162eb5d6a 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 sha256:82aab6bd8004360ed66a862ee39f686b45f8f41f2ae8ac873f6f236c42d8cc37
+oid sha256:036d173fe684811ba65e05a405f930106563744240bfaf8d18160825adc5d652
 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 7accb9368b98669c97990f53af457f508d6f65d6..c88ea9732819cd7f173533ccb2d84ab62ddb5b56 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:f89ee6bb0d91b3eacec8b4dd613ed048c099f9f9b2facb8558a9a9f142037be4
+oid sha256:93573985c21c0950c0e93f6ce3d7ee45f5b1d47d12563f33f4074715219a26db
 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 fb41c1168460e382657145bd692dd1823bc263c9..daed5900b9075b7349285d213d297cb0e85d5e95 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:c2b62324205c04b98c089db2cb17a7caaebbaa6267f8527765a93cdc80d15b05
+oid sha256:f2d7bae14c521ed9e9e00d82c5ce0bd21d31dc79e87002453acf7f25aca20edc
 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 dd17a073116bb56937dc14c7d23d80e9888f4d19..191973652b8fd8b261e2a0d4e4acd321c9960773 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 sha256:23f6fe17a222d4fa73f1a1592ec61556935b3e79b88cff351b570e3bc9d0f57b
+oid sha256:0a21d384b81dc7841dfc18d2e632b49f4db793655dcd88ab78fae91fa14d02ac
 size 6626304
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_96f8e185f66787463fed+2c2d707e/compile_flags.txt b/neuronxcc-2.13.66.0+6dfecc895/MODULE_96f8e185f66787463fed+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_96f8e185f66787463fed+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_96f8e185f66787463fed+2c2d707e/model.hlo_module.pb b/neuronxcc-2.13.66.0+6dfecc895/MODULE_96f8e185f66787463fed+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..56c69981244065330db0d6be04a51fcaa5fd4ffd
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_96f8e185f66787463fed+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fe8f01b04cd8a1d606700468bebcd0eb6f96db45a50649e5ad62259a6562b425
+size 299226
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_96f8e185f66787463fed+2c2d707e/model.neff b/neuronxcc-2.13.66.0+6dfecc895/MODULE_96f8e185f66787463fed+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..fc9766e4235984fd6b846a850c197a3f72e98f9d
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_96f8e185f66787463fed+2c2d707e/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b336c24236e1dfa9bd896d22b414acda24ab5916f7c901ec053cbccb634be7d7
+size 7619584
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 6a435dc35057083f9bc291413efbcc70a022cb9e..9fb4221e1dd7827cc93e5c9532be5bb84320753c 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 sha256:7de9b5fe530dee36c5158fbe867c603ca44c22ba93f96e789e7bac61f5b82594
+oid sha256:d52de0f620c1a03cc0f067ce052ab1e5f231b9a95e1d1efd2f89e087dad2d0db
 size 2683904
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_9bb14fc87dc287829e65+2c2d707e/compile_flags.txt b/neuronxcc-2.13.66.0+6dfecc895/MODULE_9bb14fc87dc287829e65+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_9bb14fc87dc287829e65+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_9bb14fc87dc287829e65+2c2d707e/model.hlo_module.pb b/neuronxcc-2.13.66.0+6dfecc895/MODULE_9bb14fc87dc287829e65+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..622011640d7676eee20f07e7a5e508edf53ca8f2
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_9bb14fc87dc287829e65+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bd9e692e7fa4500d1f748245897939999a19a80d85251a870f6434870a8ad0df
+size 292188
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_9bb14fc87dc287829e65+2c2d707e/model.neff b/neuronxcc-2.13.66.0+6dfecc895/MODULE_9bb14fc87dc287829e65+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..e2dc0dd6bed37841ed97b65a05d49e6755c77e0e
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_9bb14fc87dc287829e65+2c2d707e/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:74ca385010ef60da87591ed67b2bf0cb14d7ade86d19d898b5dbd0072f3dd314
+size 8244224
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_ae95f5c04c529ba0e54c+2c2d707e/compile_flags.txt b/neuronxcc-2.13.66.0+6dfecc895/MODULE_ae95f5c04c529ba0e54c+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_ae95f5c04c529ba0e54c+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_ae95f5c04c529ba0e54c+2c2d707e/model.hlo_module.pb b/neuronxcc-2.13.66.0+6dfecc895/MODULE_ae95f5c04c529ba0e54c+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..0deadf57394b54db93a072d47cc601d580a8ff8c
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_ae95f5c04c529ba0e54c+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9b7d4461930a706c0e869e1fef8dd1893f9cd94148d031a111dad8203810ba73
+size 295835
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_ae95f5c04c529ba0e54c+2c2d707e/model.neff b/neuronxcc-2.13.66.0+6dfecc895/MODULE_ae95f5c04c529ba0e54c+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..31b6cb8b09adf07a058f42ad580e706935735c65
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_ae95f5c04c529ba0e54c+2c2d707e/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:98e5d0eb6e657c9436a51cace5f98e74739e52ed88253fbb01992e1cf71e6a65
+size 11797504
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_b191a0a43fedcfa9ac37+2c2d707e/compile_flags.txt b/neuronxcc-2.13.66.0+6dfecc895/MODULE_b191a0a43fedcfa9ac37+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_b191a0a43fedcfa9ac37+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_b191a0a43fedcfa9ac37+2c2d707e/model.hlo_module.pb b/neuronxcc-2.13.66.0+6dfecc895/MODULE_b191a0a43fedcfa9ac37+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..948397e7e89a100f61ead81a8c61e137f96aa085
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_b191a0a43fedcfa9ac37+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f7fe178d2259e44d8d9d276fc36af347476547a8db17241452324a09d41522bd
+size 299226
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_b191a0a43fedcfa9ac37+2c2d707e/model.neff b/neuronxcc-2.13.66.0+6dfecc895/MODULE_b191a0a43fedcfa9ac37+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..842c3a3f8ed0eab7941192cb1b5769afd1c03a88
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_b191a0a43fedcfa9ac37+2c2d707e/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:249fcc83430fde3576ac11565ad385c769e3b10a08373e62172a37b74699d9d4
+size 7527424
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_b4732b9a44b6941efdef+2c2d707e/compile_flags.txt b/neuronxcc-2.13.66.0+6dfecc895/MODULE_b4732b9a44b6941efdef+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_b4732b9a44b6941efdef+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_b4732b9a44b6941efdef+2c2d707e/model.hlo_module.pb b/neuronxcc-2.13.66.0+6dfecc895/MODULE_b4732b9a44b6941efdef+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..cad866d7e0e20dc703b8886dfa895a7eaf8cc1ee
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_b4732b9a44b6941efdef+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6c44f4d5631c45cf4a8a6296908f0efd7cdd54f94714dc231032e11a67f984ee
+size 299226
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_b4732b9a44b6941efdef+2c2d707e/model.neff b/neuronxcc-2.13.66.0+6dfecc895/MODULE_b4732b9a44b6941efdef+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..13e3dc5a88cdcbf2eac0e74b9172f30a26930c47
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_b4732b9a44b6941efdef+2c2d707e/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d725ff016018114eb5aec88e291d4713684d793088d30a9ab867dc92fbd2d27d
+size 7465984
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_b499ff57794bd2b55273+2c2d707e/compile_flags.txt b/neuronxcc-2.13.66.0+6dfecc895/MODULE_b499ff57794bd2b55273+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_b499ff57794bd2b55273+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_b499ff57794bd2b55273+2c2d707e/model.hlo_module.pb b/neuronxcc-2.13.66.0+6dfecc895/MODULE_b499ff57794bd2b55273+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..342fc2ca8769f6cf29150e0f46373b9872fa1a73
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_b499ff57794bd2b55273+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a0ada93cc2f76efd713423ddf84c469756fc9fddf967972335d05577ebbbd718
+size 299226
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_b499ff57794bd2b55273+2c2d707e/model.neff b/neuronxcc-2.13.66.0+6dfecc895/MODULE_b499ff57794bd2b55273+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..50f60dc5f12561da7017822260805360a164fb31
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_b499ff57794bd2b55273+2c2d707e/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:db322df1ef44b4c4b9e419e054c1467490e5b4d1729ce6f22970579cca7c1141
+size 8817664
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_b6f59521083a4c513238+2c2d707e/compile_flags.txt b/neuronxcc-2.13.66.0+6dfecc895/MODULE_b6f59521083a4c513238+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_b6f59521083a4c513238+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_b6f59521083a4c513238+2c2d707e/model.hlo_module.pb b/neuronxcc-2.13.66.0+6dfecc895/MODULE_b6f59521083a4c513238+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..e81db0d3c5ea3b451d9c037e8c12a6b5a818b5fd
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_b6f59521083a4c513238+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ac3b9a41c228570b74c97f52255b89874d4732c03353ccf4c2bc78893d07cee9
+size 295835
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_b6f59521083a4c513238+2c2d707e/model.neff b/neuronxcc-2.13.66.0+6dfecc895/MODULE_b6f59521083a4c513238+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..e3e74450cfafc5ec636b57d1b62f5bac37a9ce26
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_b6f59521083a4c513238+2c2d707e/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bd3e760be4320598a5613eadf3d945e954397a57cf40f499fc87ecc31c08d595
+size 12196864
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 54403e72ad69eb2dde17b16b0ecf93a12965891e..eaf8c1156b62fac3458e8a63f58921b5bb71d70d 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:a5b4a2b6a3d071aa86bb62b8550da5069129c9d6d904827e2b3ba851bece4924
+oid sha256:fe8959429da487468dd65b82b1662733a18d5085f86938874f0dec838e7e6a7b
 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 71c9e6dfac5461763184a5b2f80fdbbaabcac212..b82324a7eeae5404d946ba4d99e9cc9ecb821ef2 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:a277f2c607a99e6ffe5fe4020145efe62fcc3585f49eec2edcc9bc251a172ca9
+oid sha256:1341e072b22a545183876855a68e6e7447f5b889e09a44a4b910aefcb73d992e
 size 13865984
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_d755a6931d0c1f489fa9+2c2d707e/compile_flags.txt b/neuronxcc-2.13.66.0+6dfecc895/MODULE_d755a6931d0c1f489fa9+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_d755a6931d0c1f489fa9+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_d755a6931d0c1f489fa9+2c2d707e/model.hlo_module.pb b/neuronxcc-2.13.66.0+6dfecc895/MODULE_d755a6931d0c1f489fa9+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..23ee94b293714f8ca7fd8ddafc030f36f5c0645d
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_d755a6931d0c1f489fa9+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4d9af0548fad9748470f9b58d8700fd218701ac0a580c3c7da465e5709d82295
+size 299226
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_d755a6931d0c1f489fa9+2c2d707e/model.neff b/neuronxcc-2.13.66.0+6dfecc895/MODULE_d755a6931d0c1f489fa9+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..0c13aa4ed8d53ea0b1cc589737149572e506787d
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_d755a6931d0c1f489fa9+2c2d707e/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:eb33cc8cb31aeb2d1b454d3393b88a0883251dc50f570589c1526185b13c135c
+size 8182784
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_d7606d9adfeb9cb0fd73+2c2d707e/compile_flags.txt b/neuronxcc-2.13.66.0+6dfecc895/MODULE_d7606d9adfeb9cb0fd73+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_d7606d9adfeb9cb0fd73+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_d7606d9adfeb9cb0fd73+2c2d707e/model.hlo_module.pb b/neuronxcc-2.13.66.0+6dfecc895/MODULE_d7606d9adfeb9cb0fd73+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..6e2b84348a5b4339d8ddeb439b006b0ae02e24b2
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_d7606d9adfeb9cb0fd73+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7db5f3a3e9656821ad74ee22a5afdabeb5bdddaac59be9084243ce6cdeccf379
+size 295835
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_d7606d9adfeb9cb0fd73+2c2d707e/model.neff b/neuronxcc-2.13.66.0+6dfecc895/MODULE_d7606d9adfeb9cb0fd73+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..d312c57fe6676a66afb0b57114935ea791875478
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_d7606d9adfeb9cb0fd73+2c2d707e/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:db032650fbd3171bf66a1d70477345f32d115de3e247552aeae774b12c7c2560
+size 8909824
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_dfb85ebd0d1b542fb616+2c2d707e/compile_flags.txt b/neuronxcc-2.13.66.0+6dfecc895/MODULE_dfb85ebd0d1b542fb616+2c2d707e/compile_flags.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fd72d102654c531414d7f703987c9729e8cb438
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_dfb85ebd0d1b542fb616+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_dfb85ebd0d1b542fb616+2c2d707e/model.hlo_module.pb b/neuronxcc-2.13.66.0+6dfecc895/MODULE_dfb85ebd0d1b542fb616+2c2d707e/model.hlo_module.pb
new file mode 100644
index 0000000000000000000000000000000000000000..d1bd75ee4b5acb8877beebaf1376f394763fafc1
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_dfb85ebd0d1b542fb616+2c2d707e/model.hlo_module.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9b015cc999a73433fc038b271429a0f1e8de4e7862baabc7c31cd52b99e8465f
+size 299226
diff --git a/neuronxcc-2.13.66.0+6dfecc895/MODULE_dfb85ebd0d1b542fb616+2c2d707e/model.neff b/neuronxcc-2.13.66.0+6dfecc895/MODULE_dfb85ebd0d1b542fb616+2c2d707e/model.neff
new file mode 100644
index 0000000000000000000000000000000000000000..0cca5a8b30aa8821b127d145a143ddddf8f5624c
--- /dev/null
+++ b/neuronxcc-2.13.66.0+6dfecc895/MODULE_dfb85ebd0d1b542fb616+2c2d707e/model.neff
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d5a3a48d7adbdc04a31d04b30c34fd0d745ce18ca61231f8717bed5e4a307bdc
+size 7814144
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 67e1279c2a2dad789e38fb7116fe5ffa94a757f2..f9b9443ccce25ff3f722e986a7fe3da2d6ad0d44 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 sha256:697cb25356bbcc11de484b4afb82bf2d36f35ade7cfe21635fab57d33c6f1da5
+oid sha256:a272b8c214312f8d6fc5e9ab9085a3f961b155f888b4d5ba7a4c8d30da5701d3
 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 f3c868e17a2539a4b135a40ad47f5d08d6638472..f36d1c3c529b97ae6f8e3fbb3f9423a801724987 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:d148f449dda8e52cae65644cb3ef26740b217f4b9a074abe6ce4eadfd4dec357
+oid sha256:8a3c0cea92a6bc0003fc88804f657a009a14701935c81fd6b81f9de5e201a579
 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 59c8643afa1badfc192ad19b88783ca8de65f54d..8f750a0f68529bf56782e8f22547d932e7b5b7be 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 sha256:4209b3e1de8a9d02d5bcf8c076761bdfc757d5b54504e530091320a695210ffd
+oid sha256:68a99502d6060a605216bd6dd9c4ebe325432f745e4a3f1794432ebb7ed556db
 size 4711424