Worked on performance optimization for the AveragedModel component in the pytorch/pytorch repository, focusing on accelerating training loops for large-scale machine learning models. The approach involved moving the check for n_averaged outside the update_parameters loop, which reduced per-iteration overhead and improved execution efficiency. Additionally, integrated stochastic weight averaging (SWA) enhancements by avoiding unnecessary stream synchronization, further streamlining resource usage. The work demonstrated strong skills in Python, low-level PyTorch code optimization, and performance profiling. These changes resulted in faster execution times and more efficient workflows for AveragedModel, supporting maintainability and scalability in machine learning model training environments.