Is there an option to only return the retrieved text without the llm participating in the response?

[bidada-bcr]

Is there an option to only return the retrieved text without the llm participating in the response?
This seems very important, otherwise we have to wait for the LLM to generate for a long time every time we retrieve. It should be more straightforward.

[wangxingjun778]

Do you mean we should directly adopt the results from text retrieval (referred to as 'evidence' in the Sirchmunk context)? However, we typically still need to use an LLM to perform fused reasoning on these fragmented pieces of textual evidence and ultimately generate a summarized result. Therefore, it seems that invoking the LLM remains necessary.