当前位置：网站首页>unhandled system error, NCCL version 2.7.8

unhandled system error, NCCL version 2.7.8

2022-04-23 07:28:00 【wujpbb7】

stay Run on the host based on DDP Of pytorch The training procedure is OK ,

Get into docker Post run , appear "unhandled system error, NCCL version 2.7.8" Error of .

resolvent ：

stay python -m torch.distributed.launch --nproc_per_node=4 ... with NCCL_DEBUG=INFO

You can see ：

s215:623:649 [3] include/shm.h:48 NCCL WARN Error while creating shared memory segment nccl-shm-send-404da1ec128dc62d-0-3-2 (size 4104)

Get into docker when , close --ipc=host that will do .

版权声明
本文为[wujpbb7]所创，转载请带上原文链接，感谢
https://yzsam.com/2022/04/202204230611550178.html