Currently, we run one cycle through the dataset for `QuantizationModifier` for weight-only quantization. This ~doubles the runtime of a GPTQ run