During March 2025, this developer contributed to the Furion-cn/sglang repository by implementing speculative token mapping for the EAGLE worker, focusing on model optimization and speculative decoding. They refactored token map loading into a dedicated function, improving code maintainability and reliability. Using Python and PyTorch, they ensured the new feature integrated seamlessly with existing generation workflows. To validate correctness, they developed targeted tests that verified token map usage during generation, emphasizing robust testing practices. The work addressed performance bottlenecks in speculative decoding, resulting in more efficient token map handling. Overall, the contribution demonstrated technical depth and attention to reliability.