Cutting RAG inference costs 6x starts with deciding what never reaches the LLM

Cutting RAG inference costs 6x starts with deciding what never reaches the LLM