Synchronizing local compiler cache.

neuronxcc-2.21.18209.0+043b1bf7/0_REGISTRY/0.4.3.dev0/granite/ibm-granite/granite-3.1-2b-instruct/c76e71ff00ec8c5532ec.json

@@ -0,0 +1,58 @@
+{
+  "_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,
+  "dtype": "bfloat16",
+  "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": {
+    "_serialized_key": "NxDNeuronConfig",
+    "batch_size": 4,
+    "capacity_factor": null,
+    "checkpoint_id": "ibm-granite/granite-3.1-2b-instruct",
+    "checkpoint_revision": "bbc2aed595bd38bd770263dc3ab831db9794441d",
+    "continuous_batching": true,
+    "ep_degree": 1,
+    "fused_qkv": true,
+    "glu_mlp": true,
+    "local_ranks_size": 2,
+    "max_batch_size": 4,
+    "max_context_length": 4096,
+    "max_topk": 256,
+    "n_active_tokens": 4096,
+    "neuronxcc_version": "2.21.18209.0+043b1bf7",
+    "on_device_sampling": true,
+    "optimum_neuron_version": "0.4.3.dev0",
+    "output_logits": false,
+    "pp_degree": 1,
+    "sequence_length": 4096,
+    "speculation_length": 0,
+    "start_rank_id": 0,
+    "target": "trn1",
+    "torch_dtype": "bfloat16",
+    "tp_degree": 2
+  },
+  "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-2.21.18209.0+043b1bf7/MODULE_b3cfe28275e1a02a015c+24129607/model.hlo_module.pb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:649ff83e0cbee13abda84e02c1a114e8b60d780456161e42f68c99c57744691c
+oid sha256:7d98bb4abcef46bc4dfcb539ece861af724c555279e78c4ba6ef27c16f25cecc
 size 739558

neuronxcc-2.21.18209.0+043b1bf7/MODULE_b3cfe28275e1a02a015c+24129607/model.neff

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f71df2ef73789e25f4514066e33be9bb355e9cd3373fc5874333f64d9b8e58ae
+oid sha256:a8f11dc4f21ab9be25f2f8e6fa30afe961d957c06d88794401230105ff3bfe06
 size 26133504

neuronxcc-2.21.18209.0+043b1bf7/MODULE_d4df6648fd68fe444e68+a02c3a36/model.hlo_module.pb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2c2920e5fde77a80d2d66c9ba5addf46e53d032f533c7097cba34e31243f84ed
+oid sha256:4b570926ce0fcfd782ef61d818285f77f7df61321cfef5c4f1085b51b81d2a1a
 size 588406

neuronxcc-2.21.18209.0+043b1bf7/MODULE_d4df6648fd68fe444e68+a02c3a36/model.neff

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:678ec457ae2bad234a72708e006fcf0d66f32e207ee59bfa7791491109372c9d
+oid sha256:0b7e1df7c167307eebb0cb890f0069cfca3b9d6856d75d02599d69d071525bc7
 size 1926144

neuronxcc-2.21.18209.0+043b1bf7/MODULE_d4df6648fd68fe444e68+a02c3a36/wrapped_neff.hlo

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ab18fc1141cf73daac1b35117b41bb341b2c594d66b4a005af15167b73c54c1a
+oid sha256:5891964fa1d2e4d1f4dcf53122d8a8fe7f5d25bfe1b908e9330abcdebb982270
 size 2082478