optimum-neuron-cache
/
neuronxcc-2.12.68.0+4480452af
/0_REGISTRY
/0.0.20
/inference
/mistral
/NousResearch
/Genstruct-7B
/b5451de080074ed7734f.json
{"vocab_size": 32001, "max_position_embeddings": 32768, "hidden_size": 4096, "intermediate_size": 14336, "num_hidden_layers": 32, "num_attention_heads": 32, "sliding_window": 4096, "num_key_value_heads": 8, "hidden_act": "silu", "initializer_range": 0.02, "rms_norm_eps": 1e-05, "use_cache": false, "rope_theta": 10000.0, "attention_dropout": 0.0, "torch_dtype": "bfloat16", "tie_word_embeddings": false, "architectures": ["MistralForCausalLM"], "bos_token_id": 1, "eos_token_id": 2, "model_type": "mistral", "neuron": {"task": "text-generation", "batch_size": 1, "num_cores": 2, "auto_cast_type": "bf16", "sequence_length": 4096, "compiler_type": "neuronx-cc", "compiler_version": "2.12.68.0+4480452af", "checkpoint_id": "NousResearch/Genstruct-7B", "checkpoint_revision": "532dc4500e621ee725672a92a78c18a9a319179c"}} |