search

infiniflow / ragflow

RAGFlow is an open-source RAG (Retrieval-Augmented Generation) engine based on deep document understanding.
https://ragflow.io
Apache License 2.0
22.85k stars 2.24k forks source link

[Bug]: I can get the result by search, but I can't get the answer by chatting with the same knowledge base #2527

Open ihacker2018 opened 1 month ago

ihacker2018 commented 1 month ago

Is there an existing issue for the same bug?

Branch name

last

Commit ID

1397

Other environment information

win10
32G内存
2060s显卡
docker

Actual behavior

知识库训练完毕，搜索中可以根据关键词搜索到结果，聊天中提示知识库中未找到答案。

Expected behavior

知识库训练完毕，搜索中可以根据关键词搜索到结果，聊天中提示知识库中未找到答案。

Steps to reproduce

知识库训练完毕，搜索中可以根据关键词搜索到结果，聊天中提示知识库中未找到答案。

Additional information

无

JinHai-CN commented 1 month ago

We intend to create an international community, so we encourage using English for communication.

wjj28 commented 1 month ago

The demo version at demo.io appears to have the same issue I've encountered. After uploading new files, the knowledge base does not reflect any updates. The Assistant is unable to retrieve or utilize the newly added information, indicating that the system is not processing the new content correctly.

KevinHuSh commented 1 month ago

The demo version at demo.io appears to have the same issue I've encountered. After uploading new files, the knowledge base does not reflect any updates. The Assistant is unable to retrieve or utilize the newly added information, indicating that the system is not processing the new content correctly.

The new file content reflect itself after being correctly parsed, isn't it?

KevinHuSh commented 1 month ago

Is there an existing issue for the same bug?

  • [x] I have checked the existing issues.

Branch name

last

Commit ID

1397

Other environment information

win10
32G内存
2060s显卡
docker

Actual behavior

知识库训练完毕，搜索中可以根据关键词搜索到结果，聊天中提示知识库中未找到答案。

Expected behavior

知识库训练完毕，搜索中可以根据关键词搜索到结果，聊天中提示知识库中未找到答案。

Steps to reproduce

知识库训练完毕，搜索中可以根据关键词搜索到结果，聊天中提示知识库中未找到答案。

Additional information

无

A few suggestions:

ihacker2018 commented 1 month ago

My issue is the same as the one reported by this brother in https://github.com/infiniflow/ragflow/issues/1397, and it hasn't been resolved yet.

KevinHuSh commented 1 month ago

There are two situations, you can add code debugging information, one is to retrieve the chunk but can't answer it, this kind of large model capability problem, change to a large model with strong ability should be able to solve, one is that the relevant chunk information is not retrieved at all, or the retrieved output is other block information, of course, the output will not find the relevant content, the reason may be that RagFlow's own strategy is to retrieve according to the combination of historical problems, which may lead to the weight of the current problem to be reduced. You could click this to distinguish these 2 situation. image

ihacker2018 commented 1 month ago

91e9ce9b0912e9a896ab062e61829ae image The prompt already includes knowledge base content, but the model indicated that it was not found when providing the answer.

KevinHuSh commented 1 month ago

91e9ce9b0912e9a896ab062e61829ae image The prompt already includes knowledge base content, but the model indicated that it was not found when providing the answer.

So, you'd better switch to another LLM to have a try.