Worked on the Tencent/ncnn repository to address a serialization issue in the Model Writer, focusing on improving how models with optional weights are handled. Using C++ and applying software engineering principles, implemented a targeted bug fix that ensures bias data is serialized only when relevant, preventing unnecessary data from being included in model artifacts. This adjustment reduced serialization artifacts and improved the reliability and compatibility of models for downstream tools that consume these artifacts. The work demonstrated attention to detail in model serialization, enhancing the overall quality and correctness of exported models while maintaining interoperability for users relying on conditional weights.