Developed and integrated speculative token mapping for the EAGLE worker in the Furion-cn/sglang repository, focusing on enhancing speculative decoding performance and reliability. Refactored token map loading into a dedicated function to improve maintainability and clarity of the codebase. Added comprehensive tests to ensure the correctness of token map usage during generation, emphasizing robust validation and reliability. Leveraged Python and PyTorch to implement model optimization techniques, with additional use of Shell scripting for testing workflows. The work prioritized business value by delivering measurable performance improvements and establishing a foundation for more reliable speculative decoding in future development cycles.