Qi Yu focused on improving tensor device compatibility in the FlagOpen/FlagGems repository, addressing a critical bug that affected tensor operations across different hardware environments. Using Python and PyTorch, Qi Yu implemented a fix that ensured tensors execute on the correct device, reducing device-specific failures and runtime errors. The solution enhanced reliability and performance for machine learning workflows, particularly in multi-threaded contexts, by refining device placement logic and supporting concurrent tensor operations. Qi Yu also documented the changes to facilitate broader deployment and future scaling. The work demonstrated a deep understanding of cross-device computation and robust testing practices within the codebase.