config.json

{
    "architectures": [
        "LlamaForCausalLM"
    ],
    "hidden_size": 4096,
    "num_attention_heads": 32,
    "num_hidden_layers": 32,
    "intermediate_size": 14336,
    "max_position_embeddings": 131072,
    "vocab_size": 128256,
    "rope_scaling": {
        "type": "linear",
        "factor": 8.0
    },
    "torch_dtype": "bfloat16",
    "model_type": "llama"
}