- first used elastic search to build index of our database usingsquad dataset.
- It is fast and has great advantages like it allows parallel processing.
- It is an open-domain open-book question answering model, reader-retriever model to be more specific.
- Retriever will take the question and search the related contexts using the external database.
- Then reader will get the context and question from retriever and it will then returns the start and end indexes of the answer.
Reader model - deepset/bert-base-cased-squad2
Retriever model
- Query embedding model - facebook/dpr-question_encoder-single-nq-base
- Passage embedding model - facebook/dpr-ctx_encoder-single-nq-base