Torch.utils.checkpoint.checkpoint with multi-gpus

autograd
1

Hi,
The program is blocked when I use Torch.utils.checkpoint.checkpoint with multi-GPUs. And when I change to single GPU training, the program works! I also try to remove the checkpoint function and the program also works with multi-GPUs. Does checkpoint function not support multi-GPUs training?

Thanks!

1 Like
2

Same problem here!
do you have any solution by now?

3

what is ‘blocked’?:man_student: