Worked on optimizing and validating the Top-K Top-P sampling kernel in the ROCm/aiter repository, focusing on performance improvements and code quality. The approach involved moving block reduction operations outside of loops to enhance throughput and reduce latency, leveraging GPU programming and performance optimization techniques. Comprehensive statistical and controlled tests were added using Python and C++ to ensure sampling accuracy and efficiency, while the benchmarking suite was expanded to quantify performance gains. The codebase and test suite were refactored for better maintainability, including updates to handle statistical test warnings more gracefully, resulting in a more robust and reliable validation process.