{ "quantization": { "group_size": 64, "bits": 8 }, "precision": "int8" }