Closed topshik closed 3 years ago
It looks like the the input_sequence tensor is turning into a zero-dimensional tensor when the running_seqs tensor has a size of 1 at the previous iteration, because of the slicing operation input_sequence = input_sequence[running_seqs]
if input_sequence.size()==torch.Size([]):
input_sequence=input_sequence.unsqueeze(0)
Yes, actually, I have some more fixes to the code, which makes it a lot prettier and more usable. I'll update merge request in a week or too.
LGTM, thanks!
fix: make code torch 1.5.0 compatible fix: PEP8 code style issues del: deprecated torch methods and api
Inference script unreliably falls with error on line 172, needs to be fixed.