Developed a modular model conversion workflow for the PaddlePaddle/PaddleNLP repository, focusing on maintainability and extensibility for future integrations such as Qwen2. Leveraged Python and deep learning expertise to refactor the codebase, introducing modular components and reorganizing files to align with codestyle standards. Enhanced the project’s documentation and analysis artifacts to clarify the conversion process and support onboarding. Expanded unit and integration test coverage to ensure stability and compatibility, while cleaning up dependencies to streamline continuous integration. This work accelerated model conversion, improved code quality, and established a foundation for faster feature delivery and easier collaboration among contributors.