Very large loss output by BCEWithLogitsLoss

nlp
1

@lais823 and I are having very large loss output by BCEWithLogitsLoss. We are working with a vision and language problem which also has attention modules.

We also have followed the guidelines of http://karpathy.github.io/2019/04/25/recipe/ for training a neural network but we are unsure how to trace back and fix the problem at this point.

What are some suggestions you could provide us with?

2

Hey Mona, could you please paste your model code (especially the forward call) and training loop? This should make it easier for other to help.