perevalov / qa_chatbots_exercises Goto Github PK

I managed to setup my NEL and Retriever component for the IR-based question answering in exercise 7. Unfortunately, I ran into some issues because the RoBERTa-For-QA input and output requirements are unknown. Maybe you could add some additional information in the exercise description regarding the following points:

1. What input type should the Retriever annotate so that the RoBERTa-For-QA component can retrieve and process it?
   String / List of Token / List of Strings (for multiple subtexts)?

2. BERT can only process 512 Tokens at a time. Will the tokenization be perfomed by the RoBERTa-For-QA component? And is the cut-off done automatically or do we need to take care of this?

3. After the annotation is created by the RoBERTa-For-QA component, will it also return the answer to the General Chatbot frontend, or do we need to add another component here?

Use https://github.com/sebastian-hofstaetter/teaching as reference