davidjaw
commented
2 days ago It appears to be related to using a newer version of PyTorch. A similar problem has been discussed in this issue.
While there may be a more detailed solution in the discussion thread, the simplest workaround is to downgrade your PyTorch version:
- PyTorch 2.0.1
- Torchvision 0.15.2
This should resolve the compatibility issue you're experiencing.
with out modify any code，train on first epoch will meet this problem even change batch size to 2