[Perf] Optimize CUDA IPC for multimodal transfer by caching IPC pool handles#21418

mickqian merged 1 commit intosgl-project:mainsgl-project/sglang:mainfrom

saatwiknagpal:feature/cuda-ipc-pool-handle-cachesaatwiknagpal/sglang:feature/cuda-ipc-pool-handle-cacheCopy head branch name to clipboard