Synchronizing local compiler cache.

.gitattributes

@@ -7894,3 +7894,5 @@ neuronxcc-2.17.194.0+d312836f/MODULE_c4162c03a5bca6d7620e/vae_decoder/model.neur
 neuronxcc-2.17.194.0+d312836f/MODULE_c4162c03a5bca6d7620e/vae_encoder/model.neuron filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.17.194.0+d312836f/MODULE_1b6d7576802d84bea017+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.17.194.0+d312836f/MODULE_c578025927b993537ac9+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.17.194.0+d312836f/MODULE_8ce80c0a7e475086c914+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.17.194.0+d312836f/MODULE_f992331f929adb7094fb+613edded/model.neff filter=lfs diff=lfs merge=lfs -text

neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev2/inference/llama/meta-llama/Llama-2-13b-hf/15497d4569cb4aabb3a9.json

@@ -0,0 +1,39 @@
+{
+  "_entry_class": "SingleModelCacheEntry",
+  "_model_id": "meta-llama/Llama-2-13b-hf",
+  "_task": "text-generation",
+  "architectures": [
+    "LlamaForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "head_dim": 128,
+  "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": "bf16",
+    "batch_size": 1,
+    "checkpoint_id": "meta-llama/Llama-2-13b-hf",
+    "checkpoint_revision": "5c31dfb671ce7cfe2d7bb7c04375e44c55e815b1",
+    "compiler_type": "neuronx-cc",
+    "compiler_version": "2.17.194.0+d312836f",
+    "num_cores": 8,
+    "sequence_length": 2048,
+    "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,
+  "use_cache": true,
+  "vocab_size": 32000
+}

neuronxcc-2.17.194.0+d312836f/MODULE_8ce80c0a7e475086c914+613edded/compile_flags.json

@@ -0,0 +1 @@
+["--target=trn1", "--model-type=transformer", "--auto-cast=none", "--execute-repetition=1"]

neuronxcc-2.17.194.0+d312836f/MODULE_8ce80c0a7e475086c914+613edded/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bbc929aa648846520b3ef9b58099acc1123e9ad4d269da79f5ef70c5e02c46c5
+size 385292

neuronxcc-2.17.194.0+d312836f/MODULE_8ce80c0a7e475086c914+613edded/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b63026c82d5340fd211e53fcad2f740596d5587499f7366630faa5ffae29f135
+size 4496384

neuronxcc-2.17.194.0+d312836f/MODULE_f992331f929adb7094fb+613edded/compile_flags.json

@@ -0,0 +1 @@
+["--target=trn1", "--model-type=transformer", "--auto-cast=none", "--execute-repetition=1"]

neuronxcc-2.17.194.0+d312836f/MODULE_f992331f929adb7094fb+613edded/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:74fc068d85127548ff0b21fa2a6dd966a9aab13472ccafcf93e32f4769754c7e
+size 305677

neuronxcc-2.17.194.0+d312836f/MODULE_f992331f929adb7094fb+613edded/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:07914968a0f88cf81b3f89f142da97a699a1234ba7f76667ab4f19e44b95e444
+size 25129984