NLP project ULMFiT: output as input or seq2seq

Hello everyone,

I’m currently working on a project, in which I need to make several predictions per sentence (after every token). Therefore I was wondering if I could somehow adapt the code from chapter 10 (used for sentiment classification of movie reviews) to create a seq2seq model or a model in which I can use the output of a prediction to make a prediction in the next time step. Any ideas on how to start?

Thanks in advance!