Worked on the sglang repository to address a critical bug in the PD offloading path of the KV Cache, focusing on stability and correctness under high load. The solution involved implementing the missing get_cpu_copy functionality by adding CPU copy methods to the PagedTokenToKVPoolAllocator, enabling proper KV cache CPU copy operations in offload mode. This targeted fix improved memory management and reduced offload-related failures, ensuring more reliable performance. The work was carried out using Python and leveraged skills in bug fixing, memory management, and performance optimization, demonstrating a methodical approach to resolving complex issues in distributed cache systems.