問題文
A retrieval flow returns twenty candidate chunks and the team wants the language model to see the three that answer the question best. Which step addresses that requirement?
選択肢
- Reduce the target accuracy of the search and pass the first three rows to the language model.
- Rerank the candidates with the embedding model and pass the top three to the language model.
- Rerank the candidates with a reranking model and pass the top three to the language model.
- Reduce the chunk size of the source and pass the first three rows to the language model.