Train_loss and valid_loss all nan

I train Movie Lens full dataset but train_loss and valid_loss all nan.
I think my error in DataLoader but I don’t know how to fix it.

Can you post your dataloader code? to start debugging this, I would start by doing:

x,y = learn.dls.one_batch()
out = learn.model(x)
learn.loss_func(out, y)

and see what each of those steps looks like. I bet that will help you narrow down the issue. You could also try running on CPU to see if the issue is gpu related

image
I still can count my loss

I found my error :smiley:
timestamp column have na in it.
Thanks a lot.

1 Like