Add torch.cat for the case that concatenation happens in the feature dim

rwth-i6 / pytorch-to-returnn

Make PyTorch code runnable within RETURNN

3 stars 6 forks source link

Add torch.cat for the case that concatenation happens in the feature dim #30

Closed lucasoethe closed 3 years ago

lucasoethe commented 3 years ago

In case concatenation of tensors happens in the feature dimension we can take advantage of the fact that the RETURNN copy layer concatenates it's inputs in the feature dimension.