Breaking the issue solves half of it. Chaining them makes it even higher.
20 hours in the past
There isn’t a greater means than question modification to enhance LLMs.
In certainly one of my current posts, I mentioned 5 question translation methods and the way they enhance the retrieval course of in RAG apps. One approach was Question decomposition.
This improbable approach creates sub-questions to assemble a extra detailed reply to our preliminary question. These sub-questions will then be used within the retrieval course of. The ultimate LLM takes every query and reply pair as context to generate a complete reply to our preliminary question.
This publish discusses two different selling methods we regularly mix with question decomposition for higher outcomes.
The primary approach is recursive answering, which entails producing subquestions in bulk however answering them recursively. The second approach is followup questioning. As you may need guessed, we reply the query…