chatglm6b_v2 单机多卡训练found at least two devices, cuda:1 and cuda:0!

Open amwork2020 opened this issue 2 years ago • 5 comments

在3090 24G内存运行chatglm2啥也没改报如下错误： RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:1 and cuda:0! (when checking argument for argument target in method wrapper_CUDA_nll_loss_forward)

Jul 21 '23 08:07 amwork2020

请务必查看我的readme.md里面的07-17日里面写的【注意】⚠️ 截屏2023-07-21 16 13 51

Jul 21 '23 08:07 yuanzhoulvpi2017

如果代码也该好了，依然有这个问题。你再检查一下transformers的版本，建议安装最新版本pip install transformers

Jul 21 '23 08:07 yuanzhoulvpi2017

@yuanzhoulvpi2017 ok了，谢谢！

Jul 21 '23 09:07 amwork2020

同谢，已解决报错

Jul 24 '23 05:07 wuxiulike

改了代码，同时也升级了transformers，为啥还是出现这个问题呢

Oct 07 '23 13:10 SWORD-ZEUS