weights-mlx/phi-2-4bit/config.json

{
    "quantization": {
        "group_size": 64,
        "bits": 4
    },
    "model_type": "phi2"
}