Worked on stabilizing GPU-accelerated vector operations in the FlagOpen/FlagGems repository, focusing on improving the reliability of tensor computations. Addressed a bug in the GPU worker path by implementing tensor size validation and refining data type handling within vector normalization and summation routines. This approach prevented size-mismatch and type-related errors, enhancing the correctness of GPU-based data processing workflows. Utilized Python for development, leveraging skills in GPU programming and tensor operations to ensure robust handling of computational edge cases. The work demonstrated attention to detail in error prevention and contributed to the overall stability of GPU computations without introducing new features.