Configure multi-machine environment variables of NCCL.
()
| 11 | |
| 12 | |
| 13 | def configure_nccl(): |
| 14 | """Configure multi-machine environment variables of NCCL.""" |
| 15 | os.environ["NCCL_LAUNCH_MODE"] = "PARALLEL" |
| 16 | os.environ["NCCL_IB_HCA"] = subprocess.getoutput( |
| 17 | "pushd /sys/class/infiniband/ > /dev/null; for i in mlx5_*; " |
| 18 | "do cat $i/ports/1/gid_attrs/types/* 2>/dev/null " |
| 19 | "| grep v >/dev/null && echo $i ; done; popd > /dev/null" |
| 20 | ) |
| 21 | os.environ["NCCL_IB_GID_INDEX"] = "3" |
| 22 | os.environ["NCCL_IB_TC"] = "106" |
| 23 | |
| 24 | |
| 25 | def configure_module(ulimit_value=8192): |
no outgoing calls
no test coverage detected