Wesun focused on improving the reliability of metric calculations in the pytorch/torchrec repository by addressing robustness issues in the Tower QPS metric update process. Using Python and leveraging skills in data processing and tensor manipulation, Wesun enhanced the fused mode to handle tensor conversion errors more gracefully, reducing the risk of crashes during production monitoring. The work centered on strengthening error handling paths, ensuring that QPS metric updates could withstand edge cases without failure. Although no new features were released, Wesun’s targeted bug fix contributed to more stable and dependable monitoring, supporting better capacity planning and operational resilience in machine learning workflows.