Worked on stability improvements for PyTorch benchmarking tools by addressing type handling in command-line argument parsing. Focused on the --iter flag in both the pytorch/benchmark and pytorch-labs/tritonbench repositories, ensuring it was explicitly typed as an integer within Python scripts. This adjustment reduced type-related runtime errors and improved the reliability and reproducibility of benchmark results, particularly for long-running performance tests. Leveraged skills in Python scripting and argument parsing to enforce consistent flag handling across repositories, resulting in smoother benchmarking pipelines and easier maintenance. The work emphasized debugging discipline and contributed to cleaner, more predictable code paths for performance evaluation workflows.