infiniflow / ragflow

RAGFlow is an open-source RAG (Retrieval-Augmented Generation) engine based on deep document understanding.
https://ragflow.io
Apache License 2.0
19.83k stars 1.98k forks source link

[Bug]: display and edit input BUG #900

Open nevermorez opened 4 months ago

nevermorez commented 4 months ago

Is there an existing issue for the same bug?

Branch name

main

Commit ID

c3ba21

Other environment information

OS:CENTOS 8 STREAM
Hardware:RTX4090 64GB RAM X86_64 CUDA 11.8

Actual behavior

BUG1: In the knowledge base operation, the number of chunks does not match the actual number of chunks. In reality, there are 23 chunks, but the knowledge base document page displays 30 chunks. BUG2: In the Chinese UI, the editor cannot input Chinese characters and can only input English characters.

Expected behavior

No response

Steps to reproduce

Started through docker compose with the configuration file: docker-compose-CN.yml, started normally without any errors, and found the above two Bugs.

Additional information

No response

nevermorez commented 4 months ago

Now I have found that the source of BUG1 is that after automatically parsing the document, I manually add text blocks and delete the automatically parsed text blocks. At this point, the number of text blocks is not updated and is the sum of the newly added text blocks and the automatically parsed text blocks. If you click the parse button again, all manually added text blocks will disappear, and the parsing status will be abnormal. I don't know if this is a BUG or if it affects the retrieval hit rate. That is, when the total number of text blocks is increased and deleted without being updated, are the deleted text blocks still present in the vector database?

nevermorez commented 4 months ago

In the search test input box, it is also not possible to directly type in Chinese; only by copying the statement in for testing. English can be typed directly.

KevinHuSh commented 4 months ago

ewly added text blocks and the automatically parsed text blocks. If you click the parse button again, all manually added text blocks will disappear, and the parsing status will be abnormal. I don't know if this is a BUG or if it affects the retrieval hit rate. That is, when the total number of text blocks is increased and deleted without being updated, are the deleted text blocks still present in the vector database?

This number does not effect on the retrieval. It's just a number which has been forgotten to update when manually add chunk.

KevinHuSh commented 4 months ago

In the search test input box, it is also not possible to directly type in Chinese; only by copying the statement in for testing. English can be typed directly.

Do you mean this box? It performs well for Chrome here. image

nevermorez commented 4 months ago

Thank you for taking the time to answer my questions amidst your busy schedule. For BUG1, if it's merely a matter of the quantity not being updated, which does not affect the search function, then it's not a concern. As for BUG2, I've found that input works normally in the Chrome browser, but in Firefox, version 126.0 (64-bit), there are issues with inputting Chinese characters in several locations. The Chinese input method is unable to input characters properly while typing, which should be an issue with browser compatibility. Thank you for your response. b817d9fc20e1d62d8e91dec2710c1910

nevermorez commented 4 months ago

Thank you for taking the time to answer my questions amidst your busy schedule. For BUG1, if it's merely a matter of the quantity not being updated, which does not affect the search function, then it's not a concern. As for BUG2, I've found that input works normally in the Chrome browser, but in Firefox, version 126.0 (64-bit), there are issues with inputting Chinese characters in several locations. The Chinese input method is unable to input characters properly while typing, which should be an issue with browser compatibility. Thank you for your response. b817d9fc20e1d62d8e91dec2710c1910

And this location. 4de88fa17e5a5d75540ea6465459bf82