Synchronizing local compiler cache.

neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.2.0.dev1/inference/mistral/optimum/mistral-1.1b-testing/26e0edb6fea0d3bc9667.json

@@ -0,0 +1 @@
+{"_entry_class": "SingleModelCacheEntry", "_model_id": "optimum/mistral-1.1b-testing", "_task": "text-generation", "architectures": ["MistralForCausalLM"], "attention_dropout": 0.0, "head_dim": 64, "hidden_act": "silu", "hidden_size": 2048, "initializer_range": 0.02, "intermediate_size": 5632, "max_position_embeddings": 32768, "model_type": "mistral", "neuron": {"auto_cast_type": "bf16", "batch_size": 4, "checkpoint_id": "optimum/mistral-1.1b-testing", "checkpoint_revision": "ce03bc8d47dbd2c173ff65f3a8de1325ba724195", "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": 22, "num_key_value_heads": 4, "rms_norm_eps": 1e-05, "rope_theta": 1000000.0, "sliding_window": null, "tie_word_embeddings": false, "use_cache": true, "vocab_size": 32000}