{ "name": "nccl", "version": "2.4.6", "port-version": 2, "description": "Optimized primitives for collective multi-GPU communication.", "supports": "linux & x64", "dependencies": [ "cuda" ] }