During October 2025, Hyouk Lee contributed to the fzyzcjy/triton repository by developing a GPU-optimized data layout to enhance matrix multiplication (MatMul) performance on A100 (Hopper) GPUs while maintaining compatibility with Ampere architectures. Leveraging CUDA and Python, Hyouk implemented the MXFP4 Hopper layout optimization, aligning layout naming conventions to support both Hopper and Ampere hardware. This work improved kernel throughput on critical MatMul paths and streamlined cross-architecture GPU layout patterns, facilitating future hardware optimizations. The project focused on machine learning kernel performance and maintainability, demonstrating depth in GPU programming and performance optimization within the Triton kernel ecosystem.