Yes, as explained in the error message, the use case of sigmoid + nn.BCELoss might be numerically unstable so you should remove the sigmoid and use nn.BCEWithLogitsLoss.
1 Like
Alright, understood! Thank you very much 