Synchronizing local compiler cache.

neuronxcc-2.26.6360.0+6f180f47/0_REGISTRY/0.4.7.dev0/granite/ibm-granite/granite-3.1-2b-instruct/da74b3645f29f4218169.json

@@ -0,0 +1,59 @@
+{
+  "_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": 1,
+    "capacity_factor": null,
+    "checkpoint_id": "ibm-granite/granite-3.1-2b-instruct",
+    "checkpoint_revision": "bbc2aed595bd38bd770263dc3ab831db9794441d",
+    "continuous_batching": false,
+    "ep_degree": 1,
+    "fused_qkv": true,
+    "glu_mlp": true,
+    "local_ranks_size": 2,
+    "max_batch_size": 1,
+    "max_context_length": 8192,
+    "max_topk": 256,
+    "n_active_tokens": 8192,
+    "neuronxcc_version": "2.26.6360.0+6f180f47",
+    "on_device_sampling": true,
+    "optimum_neuron_version": "0.4.7.dev0",
+    "output_logits": false,
+    "pp_degree": 1,
+    "prefill_chunk_size": 1024,
+    "sequence_length": 8192,
+    "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-2.26.6360.0+6f180f47/MODULE_7a6ac7459529a5b76291+84337dd9/model.neff

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a4d4353f3bb135d21573bbbb950f89a1a32e59131a06f6f7b8afaf332c1dff23
+oid sha256:634a2f6c42311b11694195d480f8c3f27a552513c737cfe711acca248b6086ee
 size 4834304

neuronxcc-2.26.6360.0+6f180f47/MODULE_7a6ac7459529a5b76291+84337dd9/wrapped_neff.hlo

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0b7a078178d3b6993833b94b5f15e76313127302e105891e212645bc41c7867b
+oid sha256:0520ea7ba0d39aa2e8256b0b8840489326bee239eaf0e10570774baf464de1b7
 size 5018483

neuronxcc-2.26.6360.0+6f180f47/MODULE_fa9815b97c745d5c65d8+12dd9119/model.hlo_module.pb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:40158c087d2ce9afc5a73742fc21a438239f57a64125f5773ee132bbf4e126a1
+oid sha256:8af8f8d25c0383d63e605535a98a09036d1bb4f105bffa641ce8d073e008cfd0
 size 1471344

neuronxcc-2.26.6360.0+6f180f47/MODULE_fa9815b97c745d5c65d8+12dd9119/model.neff

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1f17226073933fb85118a600ba23e945a07811c301cb6f0f2ce20700bace1c9a
+oid sha256:6db38b4a1d1b734ba92c4d003186737eab7036b1e2afae44e12248433e5efc51
 size 10374144