During January 2025, Yv Liu focused on improving multiprocessing stability in the pytorch/torchrec repository, specifically addressing compatibility issues between MTIA multiprocessing initialization and the AMD HIP runtime. Yv Liu resolved a critical bug by replacing the forkserver initialization mode with spawn, ensuring that multiprocessing could function reliably across different runtime environments. The solution incorporated runtime-aware logic to dynamically select the appropriate initialization mode, enhancing robustness and adaptability. This work was implemented using Python and leveraged expertise in multiprocessing and unit testing. The contribution demonstrated a targeted, in-depth approach to solving a nuanced compatibility problem rather than broad feature development.