In June 2025, Zhihan focused on stabilizing the PyTorch CoreML backend within the pytorch/pytorch repository by addressing a critical issue in the model conversion API. Using Python and leveraging expertise in backend development and machine learning, Zhihan introduced support for a convert_to parameter and removed the requirement for a weights directory, enabling more flexible CoreML exports with coremltools. This work resolved compatibility issues with the latest coremltools API, preventing export failures and improving the reliability of iOS deployments. Zhihan’s contributions enhanced the maintainability and traceability of backend changes through clear commit messaging and thorough documentation practices.