Worked on the ggml-org/llama.cpp repository to implement an Idle Slot Handling Optimization feature focused on improving server performance and memory efficiency. The approach involved persisting idle slots to RAM, ensuring cache consistency and reducing unnecessary preprocessing during idle workloads. Idle slots were cleared only when a unified KV cache was present, preventing stale cache states and stabilizing VRAM-RAM cache synchronization. The work included code cleanup to enhance maintainability and readability. Utilizing C++ programming and server development skills, the contribution addressed performance optimization by streamlining cache management, resulting in more efficient resource usage and improved stability under varying server conditions.