implementation different from paper?

[sebastiangonsal]

Is the implementation different from the preprint on arxiv? The training was supposed to be in 2 stages, with only encoder training happening first and then encoder, decoder trained together.

From the code it looks like only the second stage is done.

[codeAC29]

It is being trained the way it is written in the paper. First encoder is trained and then encoder-decoder.