mirror of
https://github.com/open-webui/open-webui.git
synced 2026-11-14 05:02:01 -06:00
[GH-ISSUE #18269] issue: hybrid search/reranker = uploading documents to the chat work > with knowledge base - not working - active: hybrid search/reranker in v.0.6.33 #105542
Reference in New Issue
Block a user
Originally created by @eXt73 on GitHub (Oct 12, 2025).
Original GitHub issue: https://github.com/open-webui/open-webui/issues/18269
Check Existing Issues
Installation Method
Docker
Open WebUI Version
v0.6.33
Ollama Version (if applicable)
0.12.5
Operating System
Kubuntu 25.10
Browser (if applicable)
Brave 1.83.112
Confirmation
README.md.Expected Behavior
The document should be properly embedded in the knowledge base, then searched by embeddinggemma, re-ranked by bge-reranker-v2-m3, and the fragments prepared in this way should be sent to LLM "on the front end".
Actual Behavior
In version 0.6.33, embedding documents in the [upload] chat, as well as hybrid search and reranking, works. However, hybrid search and reranking from the knowledge base [under Qdrant v1.15.5] do not work. This is visible in the logs and in the displayed messages (attached). Tested 5 times – removed the Docker image and volume, and full clean installations from '0'.
Steps to Reproduce
I'm skipping the obvious parts:
With the embeddinggemma:300m embedding model and the hybrid search engine + reranker bge-reranker-v2-m3 active.
Place the document in the knowledge base. Ask the main LLM model for information from the document. The model receives the data, but it looks like it's practically 'full context', and when it goes beyond the context window, we get a mess – there's no hybrid search/ranking.
Logs & Screenshots
When searching from the knowledge base:
When searching from a document uploaded to chat:
Additional Information
No response
@vaclcer commented on GitHub (Oct 13, 2025):
Have you seen https://github.com/open-webui/open-webui/issues/18133 ?
@eXt73 commented on GitHub (Oct 13, 2025):
I haven't seen it - thanks for the info
@eXt73 commented on GitHub (Oct 13, 2025):
I can confirm that the bug has been fixed in the dev branch
@silentoplayz commented on GitHub (Oct 13, 2025):
I'm closing out this issue since OP confirmed the issue has been solved on the
devbranch. It might've also been a duplicate.