Synchronizing local compiler cache.

neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/granite/ibm-granite/granite-3.1-2b-instruct/7cf1a953f534c2649a7d.json

@@ -0,0 +1 @@
+{"architectures": ["GraniteForCausalLM"], "attention_bias": false, "attention_dropout": 0.1, "attention_multiplier": 0.015625, "bos_token_id": 0, "embedding_multiplier": 12.0, "eos_token_id": 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": "07eae41278b4cd918dbe32da5d77c27e1379f751", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 2, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 40, "num_key_value_heads": 8, "pad_token_id": 0, "residual_multiplier": 0.22, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 5000000.0, "tie_word_embeddings": true, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 49155}

neuronxcc-2.16.372.0+4a9b2326/MODULE_4f54951d7f1d19705651+613edded/model.neff

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:911cf7985e6f6e7e47492e0286df6ec23219f565be8b5372caedab0da030c4f8
+oid sha256:cc2463ff4bce6db92fe22300077697d0ef0231b57a5616d6ce18801b20c26e24
 size 5202944

neuronxcc-2.16.372.0+4a9b2326/MODULE_a923e74850212465e6ab+613edded/model.neff

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:55c0e38c3b3fc17ffb6d5593c4f47a2835c04d6939c47888b47a83340eabe995
+oid sha256:cc98e835e7a916dc965854006f724dd15b18c7b0369eff4bd0ed06a9f2eaf886
 size 3687424