Synchronizing local compiler cache.

.gitattributes

@@ -7898,3 +7898,5 @@ neuronxcc-2.17.194.0+d312836f/MODULE_8ce80c0a7e475086c914+613edded/model.neff fi
 neuronxcc-2.17.194.0+d312836f/MODULE_f992331f929adb7094fb+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.17.194.0+d312836f/MODULE_08e1757d36537b5ba038+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.17.194.0+d312836f/MODULE_5e4561088b45bbe15d94+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.17.194.0+d312836f/MODULE_052ac89b53a3ef1c39e5+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.17.194.0+d312836f/MODULE_d1a07b3f87341b20d662+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-3.2-1B/178748f8e86d0180fe29.json

@@ -0,0 +1,45 @@
+{
+  "_entry_class": "SingleModelCacheEntry",
+  "_model_id": "meta-llama/Llama-3.2-1B",
+  "_task": "text-generation",
+  "architectures": [
+    "LlamaForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "head_dim": 64,
+  "hidden_act": "silu",
+  "hidden_size": 2048,
+  "initializer_range": 0.02,
+  "intermediate_size": 8192,
+  "max_position_embeddings": 131072,
+  "mlp_bias": false,
+  "model_type": "llama",
+  "neuron": {
+    "auto_cast_type": "bf16",
+    "batch_size": 1,
+    "checkpoint_id": "meta-llama/Llama-3.2-1B",
+    "checkpoint_revision": "4e20de362430cd3b72f300e6b0f18e50e7166e08",
+    "compiler_type": "neuronx-cc",
+    "compiler_version": "2.17.194.0+d312836f",
+    "num_cores": 2,
+    "sequence_length": 4096,
+    "task": "text-generation"
+  },
+  "num_attention_heads": 32,
+  "num_hidden_layers": 16,
+  "num_key_value_heads": 8,
+  "pretraining_tp": 1,
+  "rms_norm_eps": 1e-05,
+  "rope_scaling": {
+    "factor": 32.0,
+    "high_freq_factor": 4.0,
+    "low_freq_factor": 1.0,
+    "original_max_position_embeddings": 8192,
+    "rope_type": "llama3"
+  },
+  "rope_theta": 500000.0,
+  "tie_word_embeddings": true,
+  "use_cache": true,
+  "vocab_size": 128256
+}

neuronxcc-2.17.194.0+d312836f/MODULE_052ac89b53a3ef1c39e5+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_052ac89b53a3ef1c39e5+613edded/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a29ffe2abb1b8f6baec07f43ff0b3340ae551cdf07bfbf1e5885cd44f880d67c
+size 165618

neuronxcc-2.17.194.0+d312836f/MODULE_052ac89b53a3ef1c39e5+613edded/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:04250776c734468a9d0329b49afc4ee766a7568e64a81a14fe6e8b6d78a7feb5
+size 1649664

neuronxcc-2.17.194.0+d312836f/MODULE_d1a07b3f87341b20d662+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_d1a07b3f87341b20d662+613edded/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8ddc0b633baed56d8458c17df039b60e27e1b47124a868f3f56f836071b8b3d8
+size 210421

neuronxcc-2.17.194.0+d312836f/MODULE_d1a07b3f87341b20d662+613edded/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:33fa9e496ff1861b09078036afdf6a24ef0498fb47c402257a98e975b4d3cbc2
+size 30587904