Synchronizing local compiler cache.

.gitattributes

@@ -4788,3 +4788,5 @@ neuronxcc-2.15.128.0+56dc5a86/MODULE_841428df9a7f7fcd1384/vae_decoder/model.neur
 neuronxcc-2.15.128.0+56dc5a86/MODULE_841428df9a7f7fcd1384/vae_encoder/model.neuron filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.15.128.0+56dc5a86/MODULE_1155707dee9639252808+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.15.128.0+56dc5a86/MODULE_2492d0aac41996a9baa4+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_8cb7017cc6f77827d053+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_f9a354ef4d61702f42da+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text

neuronxcc-2.15.128.0+56dc5a86/0_REGISTRY/0.0.25.dev0/inference/llama/meta-llama/Llama-2-13b-chat-hf/9879d2971bdd2756d40a.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, "mlp_bias": false, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 16, "checkpoint_id": "meta-llama/Llama-2-13b-chat-hf", "checkpoint_revision": "a2cb7a712bb6e5e736ca7f8cd98167f81a0b5bd8", "compiler_type": "neuronx-cc", "compiler_version": "2.15.128.0+56dc5a86", "num_cores": 24, "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}

neuronxcc-2.15.128.0+56dc5a86/MODULE_8cb7017cc6f77827d053+39f12043/compile_flags.json

@@ -0,0 +1 @@
+["--model-type=transformer", "--auto-cast=none", "--execute-repetition=1"]

neuronxcc-2.15.128.0+56dc5a86/MODULE_8cb7017cc6f77827d053+39f12043/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b20c798f5a06aed3f9263da4f68b991009afcb25b5865483f1c1a7120d4660cf
+size 561423

neuronxcc-2.15.128.0+56dc5a86/MODULE_8cb7017cc6f77827d053+39f12043/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e69a1db15326f7a99483fc23ae05e30efba636c6017357c56dc1265e5a46b032
+size 29144064

neuronxcc-2.15.128.0+56dc5a86/MODULE_f9a354ef4d61702f42da+39f12043/compile_flags.json

@@ -0,0 +1 @@
+["--model-type=transformer", "--auto-cast=none", "--execute-repetition=1"]

neuronxcc-2.15.128.0+56dc5a86/MODULE_f9a354ef4d61702f42da+39f12043/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:efa863e3f20829f5f1c1797c2f1fe908b3c8d8a18b5dda713d2b708c95189694
+size 438365

neuronxcc-2.15.128.0+56dc5a86/MODULE_f9a354ef4d61702f42da+39f12043/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cb349abc38fabb747a81bc5113e9792e0facfcc0650560a9ca846e74b334d3e5
+size 5970944