What is your usecase? If you want to search for relevant info in your documents and get relevant info, and you want to avoid hallucination, you might avoid the

What is your usecase? If you want to search for relevant info in your documents and get relevant info, and you want to avoid hallucination, you might avoid the text generation altogether. Instead you can extract text embeddings from your documents, put them in a vector DB, and then you have a super search. You can convert your search query to an embedding, search the DB and keep the e.g. 10 closest matches.