config.json

{
    "llm_model": "llm.mnn",
    "llm_weight": "llm.mnn.weight",
    "backend_type": "auto",
    "thread_num": 8,
    "precision": "low",
    "memory": "low",
    "max_new_tokens": 2048,
    "sampler_type": "mixed",
    "mixed_samplers": [
        "penalty",
        "topK",
        "topP",
        "min_p",
        "temperature"
    ],
    "penalty": 1.05,
    "temperature": 0.7,
    "topP": 0.6,
    "topK": 20,
    "min_p": 0
}