Closed chinge55 closed 1 year ago
In case of CharRNN:
Modifying the code, i.e sending the lengths
to cpu seems to do the trick. Maybe something updated in the newer versions of pytorch ?
In the file line 31:
moses/char_rnn/model.py
x = rnn_utils.pack_padded_sequence(x, lengths.cpu(), batch_first=True)
Two Models:
Error: