[Question]: Retrieval is very slow, and time is mainly spent on generating answers and tune question.

[lizhao-8202]

Describe your problem

ragflow: v0.15.1-slim chat model: qwen2.5:14b embeding model: bge-m3

Retrieval too slowly，This is the time spent on each step

I hope to optimize the response speed, but I don't know how to start. Can I adjust the docker images startup parameters? If possible, which components and parameters should be adjusted

[stevenguan08]

the same problem for me. It's so frustrating! The retrieval speed is making ragflow unusable.

[KevinHuSh]

Do you use re-rank model here?

[zhouyangit]

the same problem for me. I'm also encountering the same problem. Sometimes the same problem is relatively fast, and sometimes it's very slow. May I ask if it has anything to do with the API I'm using: Currently,
chat api uses the online siliconflow.cn (deepseek_v3), embedding model api: siliconflow.cn (BAAI/bge-m3)

[lizhao-8202]

Do you use re-rank model here?

Not useing re-rank model

[KevinHuSh]

How many chunks do you estimate it has?

[chat19]

Same to me.

Total: 5999.8ms Check LLM: 6.8ms Create retriever: 2.2ms Bind embedding: 68.4ms Bind LLM: 70.3ms Tune question: 2462.6ms Bind reranker: 0.0ms Generate keyword: 0.0ms Retrieval: 1501.1ms Generate answer: 1888.5ms

I have two PDF documents that aren't large. Is there a way to reduce the Tune question time?

[KevinHuSh]

Disable it. @chat19

[lizhao-8202]

现在可以了。感谢