Synchronizing local compiler cache.

.gitattributes

@@ -7641,3 +7641,5 @@ neuronxcc-2.17.194.0+d312836f/MODULE_a6cfe103ba38d7ced324+613edded/model.neff fi
 neuronxcc-2.17.194.0+d312836f/MODULE_d02bc244b8a319b871c1+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.17.194.0+d312836f/MODULE_34d371c764b2edea8dcd+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.17.194.0+d312836f/MODULE_3925d01728915c012cb3+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.17.194.0+d312836f/MODULE_20223e04b1bfd1a5fe7e+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.17.194.0+d312836f/MODULE_4f54951d7f1d19705651+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-3.1-2b-instruct/9e7df2d93403be65f917.json

@@ -0,0 +1,41 @@
+{
+  "_entry_class": "SingleModelCacheEntry",
+  "_model_id": "ibm-granite/granite-3.1-2b-instruct",
+  "_task": "text-generation",
+  "architectures": [
+    "GraniteForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.1,
+  "attention_multiplier": 0.015625,
+  "embedding_multiplier": 12.0,
+  "hidden_act": "silu",
+  "hidden_size": 2048,
+  "initializer_range": 0.02,
+  "intermediate_size": 8192,
+  "logits_scaling": 8.0,
+  "max_position_embeddings": 131072,
+  "mlp_bias": false,
+  "model_type": "granite",
+  "neuron": {
+    "auto_cast_type": "bf16",
+    "batch_size": 4,
+    "checkpoint_id": "ibm-granite/granite-3.1-2b-instruct",
+    "checkpoint_revision": "374ef54e020a3ce208c65e96d6213922a87d8952",
+    "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": 5000000.0,
+  "tie_word_embeddings": true,
+  "use_cache": true,
+  "vocab_size": 49155
+}

neuronxcc-2.17.194.0+d312836f/MODULE_20223e04b1bfd1a5fe7e+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_20223e04b1bfd1a5fe7e+613edded/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d81644f4779d98919b332c9d597e63a220025de21c6fd1b5f839ac842f3ca41b
+size 595152

neuronxcc-2.17.194.0+d312836f/MODULE_20223e04b1bfd1a5fe7e+613edded/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9f717a5d2ad6f62d3cddbdc3a281f806e102d59b10455ffd0014d5752365f104
+size 3851264

neuronxcc-2.17.194.0+d312836f/MODULE_4f54951d7f1d19705651+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_4f54951d7f1d19705651+613edded/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:20af142240a176cb7747f1f12b209b0951c70b2402261da25df5366d3c222b75
+size 466083

neuronxcc-2.17.194.0+d312836f/MODULE_4f54951d7f1d19705651+613edded/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:db295628273938396444c46fc97864b434a25583ae9bbbf53396e5cee5a72748
+size 5377024