📈️ Existing Feature Description
Current RAG, where we pass the LLM retrieved chunks from documents, is useful for answering short fact questions but it falls short trying to respond to some user requests like summarizing a document, comparing documents, collecting information that is in multiple parts of a document.
📈️ Proposed Enhancement Details
The context lengths of LLMs are increasing a lot and the prices dropping so I think a natural step would be to give the option to retrieve the whole document and send it to the LLM along the user query. It provides the LLM with a lot more context to answer questions and simplifies the whole RAG process.
Additional context
No response
📈️ Existing Feature Description
Current RAG, where we pass the LLM retrieved chunks from documents, is useful for answering short fact questions but it falls short trying to respond to some user requests like summarizing a document, comparing documents, collecting information that is in multiple parts of a document.
📈️ Proposed Enhancement Details
The context lengths of LLMs are increasing a lot and the prices dropping so I think a natural step would be to give the option to retrieve the whole document and send it to the LLM along the user query. It provides the LLM with a lot more context to answer questions and simplifies the whole RAG process.
Additional context
No response