Worked on improving tokenizer reliability in the yhyang201/sglang repository by addressing a bug related to the preservation of cached token details in multi-tokenizer output. Focused on maintaining the output structure when caching token details, which reduced inconsistencies and downstream processing errors. The solution involved careful software debugging and unit testing to ensure that the fix aligned with repository standards and did not introduce new issues. Utilized Python for both development and testing, emphasizing maintainability and data quality. This work contributed to more reliable tokenizer results and supported consistent data handling across the codebase, enhancing overall robustness of the system.