This is a PyTorch version of fairseq, a sequence-to-sequence learning toolkit from Facebook AI Research. The original authors of this reimplementation are (in no particular order) Sergey Edunov, Myle ...
**add a new model named len_pre_transformer.py. It takes advantages of transfomer's encoder and get rid of the decoder. As for predictign sentence length, it stacks three fully connected layers above ...