Synchronizing local compiler cache.

.gitattributes

@@ -7695,3 +7695,5 @@ neuronxcc-2.17.194.0+d312836f/MODULE_7ba1f8c4b85db8edc510+613edded/model.neff fi
 neuronxcc-2.17.194.0+d312836f/MODULE_91ae7a54e8a53def9754+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.17.194.0+d312836f/MODULE_a70cbe20d1967baf71c2+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.17.194.0+d312836f/MODULE_e1fde5e210a054b71d3c+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.17.194.0+d312836f/MODULE_06a55ccc21ac03b5c5c1+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.17.194.0+d312836f/MODULE_2fa982b13af96f760a14+613edded/model.neff filter=lfs diff=lfs merge=lfs -text

neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.2.0.dev2/inference/granite/ibm-granite/granite-3.1-8b-instruct/45d32f74efe639efa140.json

@@ -0,0 +1,41 @@
+{
+  "_entry_class": "SingleModelCacheEntry",
+  "_model_id": "ibm-granite/granite-3.1-8b-instruct",
+  "_task": "text-generation",
+  "architectures": [
+    "GraniteForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.1,
+  "attention_multiplier": 0.0078125,
+  "embedding_multiplier": 12.0,
+  "hidden_act": "silu",
+  "hidden_size": 4096,
+  "initializer_range": 0.02,
+  "intermediate_size": 12800,
+  "logits_scaling": 16.0,
+  "max_position_embeddings": 131072,
+  "mlp_bias": false,
+  "model_type": "granite",
+  "neuron": {
+    "auto_cast_type": "bf16",
+    "batch_size": 1,
+    "checkpoint_id": "ibm-granite/granite-3.1-8b-instruct",
+    "checkpoint_revision": "3f05a1d007b2484bbf17593efe110bd5b9d67655",
+    "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": 40,
+  "num_key_value_heads": 8,
+  "residual_multiplier": 0.22,
+  "rms_norm_eps": 1e-05,
+  "rope_scaling": null,
+  "rope_theta": 10000000.0,
+  "tie_word_embeddings": true,
+  "use_cache": true,
+  "vocab_size": 49155
+}

neuronxcc-2.17.194.0+d312836f/MODULE_06a55ccc21ac03b5c5c1+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_06a55ccc21ac03b5c5c1+613edded/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d1fc532afcc31309139e74b625e39424dd983fb50d6cf9d59cb8b592a55346a1
+size 540353

neuronxcc-2.17.194.0+d312836f/MODULE_06a55ccc21ac03b5c5c1+613edded/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e83b06241395a75930ac7109f6aab8af0d50a69fe94522d0b0fe999b84ac49df
+size 6339584

neuronxcc-2.17.194.0+d312836f/MODULE_2fa982b13af96f760a14+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_2fa982b13af96f760a14+613edded/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4737f322531c6922fe19ff589d53ff7d6ccdda5bf62bb63bd34658452549441d
+size 427968

neuronxcc-2.17.194.0+d312836f/MODULE_2fa982b13af96f760a14+613edded/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:844d78412b315509918c06edb70c1b51b901534989c0e6fb858cb901c064ece1
+size 9698304