Worked on enhancing reliability and backend interoperability for the sgLang ecosystem, focusing on Python-based asynchronous programming and backend development. Addressed three bugs across bytedance-iaas/sglang and sgl-project/sglang, including improvements to pause_generation scheduling to prevent request leaks and blocking, and ensured whitespace was preserved in GLM tool call arguments for accurate formatting. Added regression and unit tests to validate these critical flows. Delivered a feature in yhyang201/sglang by integrating HybridLinearAttnBackend support into TboAttnBackend, handling mrope_positions and forwarding calls to the primary backend. The work emphasized robust testing, deep learning integration, and reliable backend performance.