{ "model_type": "llama", "hidden_size": 4096, "num_hidden_layers": 32, "num_attention_heads": 32, "intermediate_size": 16384, "vocab_size": 32000 }