hi

[zmxsss]

If my loss is one-dimensional, do I need to become two-dimensional? use loss = loss.view(-1, batch_size)

[zmxsss]

my loss.shape [1280.]

[ddkang]

With our implementation of loss dropping, we require a batch dimension. Doing that should work.

It should be a simple code change to deal with one-dimensional losses though!

[zmxsss]

my code:
loss = self.loss_fn(tgt_word_logprobs, tgt_word_labels)
loss=loss.view(-1,bs)
loss = loss.mean(dim=0)
aa = self.dropper(loss)
loss *= aa
loss = loss.mean()
loss = loss.view(-1)
when run 6 epoch , error: RuntimeError: shape '[-1, 64]' is invalid for input of size 120, do you know why?

[ddkang]

I don't think you need the mean if it's already one dimensional.