During March 2025, Zhangyub focused on enhancing the robustness of the llama.cpp server cache by addressing a critical issue in token cache integrity. Working within the ggerganov/llama.cpp repository, Zhangyub corrected the cache reuse logic related to llama_kv_cache_seq_rm, ensuring that valid tokens were preserved during session generation. This C++-based fix improved session stability and reduced the risk of token loss, directly supporting long-running server workflows. The solution was validated through regression testing and code review, demonstrating disciplined debugging and server development skills. Zhangyub’s work contributed to the reliability and maintainability goals of the server’s cache subsystem.