quantize_config.json

{
    "per_channel": true,
    "reduce_range": true,
    "quantize_mode": "q4",
    "per_model_config": {}
}