Synchronizing local compiler cache.

.gitattributes

@@ -4566,3 +4566,5 @@ neuronxcc-2.15.128.0+56dc5a86/MODULE_66f90436f7bcc687b201+39f12043/model.neff fi
 neuronxcc-2.15.128.0+56dc5a86/MODULE_74f17a0c13d0e6181fd7+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.15.128.0+56dc5a86/MODULE_70b50d6b61b82f10a10e+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.15.128.0+56dc5a86/MODULE_8c05151ad4846522984e+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_b4b7a7746d9eb631054a+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_baccad05d368f0a8ce1a+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/ea299c342bed8af1409d.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": 4, "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": 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}

neuronxcc-2.15.128.0+56dc5a86/MODULE_b4b7a7746d9eb631054a+39f12043/compile_flags.json

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

neuronxcc-2.15.128.0+56dc5a86/MODULE_b4b7a7746d9eb631054a+39f12043/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3ac6c73139a2b8dc493684dc9f91b62f568425ba124fc9655a5b756acfe8c363
+size 437405

neuronxcc-2.15.128.0+56dc5a86/MODULE_b4b7a7746d9eb631054a+39f12043/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5a7af513dbfded10d6283bf45d133dfb22ddd19ad3c61b550d1af223e7328fcf
+size 4916224

neuronxcc-2.15.128.0+56dc5a86/MODULE_baccad05d368f0a8ce1a+39f12043/compile_flags.json

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

neuronxcc-2.15.128.0+56dc5a86/MODULE_baccad05d368f0a8ce1a+39f12043/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1b4d6342c3578f2ab9b757ab405f59130cfb03a9df38dac434dfd7e32929049a
+size 560463

neuronxcc-2.15.128.0+56dc5a86/MODULE_baccad05d368f0a8ce1a+39f12043/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4f7c84d65e8d8435dec52ea08ad90901fa0e9664baf541f94018d167b97faeeb
+size 2151424