Synchronizing local compiler cache.

.gitattributes

@@ -7847,3 +7847,5 @@ neuronxcc-2.17.194.0+d312836f/MODULE_8bdbf5372d7324466d56+613edded/model.neff fi
 neuronxcc-2.17.194.0+d312836f/MODULE_a64f774f616f6a5de9a2+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.17.194.0+d312836f/MODULE_2f829978f87e39981797+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.17.194.0+d312836f/MODULE_a43b49c96d7e4fd61861+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.17.194.0+d312836f/MODULE_ccd30e9fc51070b1c812+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.17.194.0+d312836f/MODULE_d1581937ee5c6df81b5c+613edded/model.neff filter=lfs diff=lfs merge=lfs -text

neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev2/inference/llama/princeton-nlp/Sheared-LLaMA-1.3B/6845a4ab255499aced61.json

@@ -0,0 +1,39 @@
+{
+  "_entry_class": "SingleModelCacheEntry",
+  "_model_id": "princeton-nlp/Sheared-LLaMA-1.3B",
+  "_task": "text-generation",
+  "architectures": [
+    "LlamaForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "head_dim": 128,
+  "hidden_act": "silu",
+  "hidden_size": 2048,
+  "initializer_range": 0.02,
+  "intermediate_size": 5504,
+  "max_position_embeddings": 4096,
+  "mlp_bias": false,
+  "model_type": "llama",
+  "neuron": {
+    "auto_cast_type": "fp16",
+    "batch_size": 1,
+    "checkpoint_id": "princeton-nlp/Sheared-LLaMA-1.3B",
+    "checkpoint_revision": "a4b76938edbf571ea7d7d9904861cbdca08809b4",
+    "compiler_type": "neuronx-cc",
+    "compiler_version": "2.17.194.0+d312836f",
+    "num_cores": 2,
+    "sequence_length": 4096,
+    "task": "text-generation"
+  },
+  "num_attention_heads": 16,
+  "num_hidden_layers": 24,
+  "num_key_value_heads": 16,
+  "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_ccd30e9fc51070b1c812+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_ccd30e9fc51070b1c812+613edded/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:494e7de87c126456900de7981d44fe1b945a3af3dc0ad8e4c247346aa14cb8d4
+size 306860

neuronxcc-2.17.194.0+d312836f/MODULE_ccd30e9fc51070b1c812+613edded/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4e97c73dbe5e3b14efabdd52690ac4a59a878213b65f3bf7dce0478eb5885de0
+size 30577664

neuronxcc-2.17.194.0+d312836f/MODULE_d1581937ee5c6df81b5c+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_d1581937ee5c6df81b5c+613edded/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d18c4882bffd239c0190a8e127d12fecc6e3d9faf8afd0f433c0ca67839afbf2
+size 231305

neuronxcc-2.17.194.0+d312836f/MODULE_d1581937ee5c6df81b5c+613edded/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ef16afe691e77cfbec19b976cf4ea50a182280483b433865e7dd3183bb8d7d9a
+size 2182144