Developed ROCm HIP transport support for the Mooncake Python package, enabling HIP as a transport layer for AMD GPUs and ensuring compatibility with RDMA. Integrated HIP transport into both build and runtime selection paths, addressing peer-access issues and decoupling HIP from the NVLink branch to improve stability. Expanded validation and updated wheel packaging to support ROCm HIP usage in vllm-omni, enhancing deployment reliability. The work in the kvcache-ai/Mooncake repository involved C++ and Python development, GPU programming, and unit testing, with attention to code maintainability through clang-format application. The feature delivered robust HIP integration without introducing new bugs.