Synchronizing local compiler cache.

.gitattributes

@@ -4626,3 +4626,5 @@ neuronxcc-2.15.128.0+56dc5a86/MODULE_d2c3463a941b252fd0ea+39f12043/model.neff fi
 neuronxcc-2.15.128.0+56dc5a86/MODULE_daa82e2e4fbe45ef3961+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.15.128.0+56dc5a86/MODULE_db7587d7de3a81400318+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.15.128.0+56dc5a86/MODULE_fa80be5dfd7ec00287f6+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_1883e2969a3bcb906f06+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_a09912919c199504f90f+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/3fed786cbc1cbe29d152.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": 8, "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_1883e2969a3bcb906f06+39f12043/compile_flags.json

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

neuronxcc-2.15.128.0+56dc5a86/MODULE_1883e2969a3bcb906f06+39f12043/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5b00db9e6cc6a1141a897aa2b1a7a74e4b5301116072571e6d2930d68d8b4096
+size 561423

neuronxcc-2.15.128.0+56dc5a86/MODULE_1883e2969a3bcb906f06+39f12043/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4f048edceb4d844d81097c08567e0c00c82afbe523609215b6bf69d008ba2d0f
+size 29144064

neuronxcc-2.15.128.0+56dc5a86/MODULE_a09912919c199504f90f+39f12043/compile_flags.json

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

neuronxcc-2.15.128.0+56dc5a86/MODULE_a09912919c199504f90f+39f12043/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5d38e723a1ddcb231bce38ebb737384771c2fbb57a0fbd4ede6b75a8b9577394
+size 438365

neuronxcc-2.15.128.0+56dc5a86/MODULE_a09912919c199504f90f+39f12043/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cde95dbeeec692da1af50e2e129a327b8f1129f2f7f8b99e313554e15b168f90
+size 4209664