hard20 pts · first answer countsragchunking

Parent retrieval floods the prompt

With small-to-big retrieval, the top 8 small chunks map to parent sections of up to 6,000 tokens each. Costs jump and answers get worse. What should you change?

A.Return all parents but raise the temperature to read them fasterB.Remove the small chunks and search only the large parentsC.Increase top_k to 20 to find the most relevant parentD.Dedupe parents, cap parent size, and keep only the best few within a token budget

Discussion 0

Join free to post in the discussion.

Parent retrieval floods the prompt

Discussion 0

Related questions