llm_config.json

{
    "hidden_size": 4096,
    "layer_nums": 32,
    "attention_mask": "float",
    "key_value_shape": [
        2,
        1,
        0,
        32,
        128
    ],
    "prompt_template": "[INST]%s[/INST]",
    "is_visual": false,
    "is_single": true
}