Oh., I’ve just read from NLLLoss documentation, that NLLLoss is implemented differently than what I’ve known before:
loss(x, class) = -x[class]
it doesn’t use log function. So, does it means I can’t use Softmax and NLLLoss together?
2 Likes