Maxime Burchi
Maxime Burchi
Hi @snakers4! Yes I had a similar problem with 4 GPU devices where the rnnt loss was properly computed on the first devices but 0 on the others. I don't...
Yes, this means that logits / target lengths tensors do not match the logits / target tensors. If you have logits lengths longer than your logits tensor for instance.
Because I used the targets lengths instead of logits lengths, stupid error
Hi, ctcdecode doesn't seem to install successfully with newer PyTorch versions. The installation is successful with PyTorch 2.0.0.