During September 2025, Bremer M. developed quantization and dequantization operators for FP32 to MX4 and vice versa within the pytorch-labs/tritonbench repository. The work focused on enabling low-precision inference by implementing 4-bit quantization workflows, using Python and leveraging GPU computing for performance. Bremer designed benchmarking scaffolding with fbgemm_gpu to generate inputs and measure the impact of quantization, providing a foundation for future optimization and deployment efficiency. The feature addressed internal issue #446 and demonstrated a thorough understanding of quantization concepts, PyTorch extension development, and performance instrumentation, resulting in a well-integrated and extensible solution for low-precision model evaluation.