Export Reviews, Discussions, Author Feedback and Meta-Reviews

In sequence to sequence model for each n, you do not really have to train a different model. You could just train a different softmax for each n and the rest of the parameters can be shared for all the problems of different lengths.