Train_loss and valid_loss becom nan from the second epoch, fastai2

Part 1 (2019)
3

Many thanks.
Yes, I check the GPU memory. It’s nearly full. It work normally after I reduce the batch size.

1 Like