Closed inoryy closed 7 years ago
This looks to be pytorch bug. torch.manual_seed shouldn't initialize cuda driver unnecessarily. If this is causing break for you just #comment the torch.manual_seed in main.py. Thanks for letting me know on this
This bug is fixed in pytorch now so closing
I'm not sure how you've avoided the error, but without these fixes all but one workers crash with the error in title when running the example
python main.py --env Pong-v0 --workers 32
.This is a fix as described in https://github.com/pytorch/pytorch/issues/2517#issuecomment-325039259