Worked on enhancing user experience and model deployment flexibility across the Furion-cn/sglang and kvcache-ai/sglang repositories. Delivered a targeted documentation update in Markdown for Furion-cn/sglang, adding troubleshooting guidance to the README to help users resolve server launch issues by verifying model weight downloads and retrying as needed. In kvcache-ai/sglang, implemented a feature in Python that introduced precision control for DiT models, allowing selection among fp32, fp16, and bf16 data types. Refactored the torch dtype context manager to streamline data processing and improve inference reliability, demonstrating a focus on practical deep learning and machine learning workflows.