Description
Is there an existing issue for the same bug?
- I have checked the existing issues.
RAGFlow workspace code commit ID
g555c7067
RAGFlow image version
g555c7067（v0.17.0-29）
Other environment information
Actual behavior
When I use the "Table" mode to parse Excel spreadsheets, I've noticed that the chat model doesn't intervene. It directly outputs data, and the responses are not relevant to the questions.
For example, when I asked about "broccoli", the output was "xishengcai" (which is incorrect).
I've confirmed that "broccoli" exists in the data, and the "search test" can accurately locate it.
I suspect that during keyword searching, only the character "西" (xi) is matched (even though my prompt emphasizes the need for an exact match), and it's clear that the chat model is not involved at all.
For comparison, I tried parsing the same Excel file using the "General" mode, with all other settings (including the prompt) kept the same. The response is shown in the figure below:
It's evident that the chat model intervened in the "General" mode, resulting in much higher quality and more accurate responses. The only difference between the two is the parsing method, yet the gap in responses is significant. I believe this might be a bug, and I would be very grateful if it could be fixed.
Expected behavior
Provide responses of the same high quality as in the 'General' mode.
Steps to reproduce
1. Create a knowledge base and configure the "Table" mode.
2. Import the Excel spreadsheet and parse it.
3. Create a chatbot, select the knowledge base and the deepseek-r1 model.
4. Ask a question.Additional information
No response