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
19.87k stars 1.98k forks source link

[Bug]: Whole System Broke after parsing a 100 line table #2860

Open yw2598 opened 3 days ago

yw2598 commented 3 days ago

Is there an existing issue for the same bug?

Branch name

main

Commit ID

current

Other environment information

Ubuntu 2204 Docker V0.12

Actual behavior

After parsing the a 100 line table, chat will only respond one message even if I disable the table document or change to a different chat. This will only be resolved after a whole deletion of every knowledge base.

Expected behavior

No response

Steps to reproduce

Parse a table with 100 rows, 

sample row: 
编号 定义 速度 备注 
F0-01 低压端子 3 none

Additional information

No response

KevinHuSh commented 3 days ago

Could you specify the error message?

yw2598 commented 3 days ago

There is no error message, but in the chat dialogue, no matter what I input, regardless of the knowledge base, embedding model, or llm model used, the assistant will only reply with the same message, a message contains a dozens of rows for the table I parsed, and even reference to documents that has been deleted. Also, I add keywords to every chunk of the parsed table, the keywords are product codes, such as RS32, ragflow seems cannot detect error codes or product codes, which are combinations or letters and numbers, is this a problem with embedding or something else?

KevinHuSh commented 2 days ago

What about using General as the parsing method.

yw2598 commented 2 days ago

it seams like once the excel is parsed, there is no other way to delete it, except deleting all the knowledge bases. I will try General method with a fresh build docker image on my another machine to avoid contaminating my running server.