Snowdar / asv-subtools

An Open Source Tools for Speaker Recognition
Apache License 2.0
587 stars 135 forks source link

多卡GPU运行失败 #65

Open niucheney opened 1 year ago

niucheney commented 1 year ago
1 2

Environment: Pytorch version: 1.10.0 Cuda version: 11.1 nccl version: 2.10.3 driver version: 470.63.01 OS version: Ubuntu 18.04

单卡可以正常训练,多卡失败

matln commented 1 year ago

可能是pytorch版本问题,把 python -m torch.distributed.launch 换成 torchrun 试试

niucheney commented 1 year ago

可能是pytorch版本问题,把 python -m torch.distributed.launch 换成 torchrun 试试

感谢回复,替换成torchrun依然不行,按照https://github.com/Snowdar/asv-subtools/issues/36 提示,我将环境替换为cuda 11.0, Pytorch 1.7.0, 可以进行训练了