Worked on the pytorch-labs/tritonbench repository to enhance benchmarking accuracy by updating the Triton vector_add kernel. Addressed a correctness issue by modifying the kernel to return its output tensor directly, replacing the previous intermediate return value. This change enabled direct accuracy checks and seamless integration with Helion benchmarking workflows, improving the reliability of performance measurements. The update was implemented and documented in a single commit, focusing on kernel development and performance optimization using Python. Through this targeted bug fix, ensured that the actual kernel output is accessible for evaluation, laying the groundwork for more robust end-to-end benchmarking within the project.