mirror of
https://github.com/open-webui/open-webui.git
synced 2026-05-07 11:28:35 -05:00
issue: [v0.6.15] Open WebUI container uses VRAM at Idle #5650
Reference in New Issue
Block a user
Delete Branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Originally created by @slavkosky on GitHub (Jun 26, 2025).
Check Existing Issues
Installation Method
Docker
Open WebUI Version
v0.6.15-cuda (ghcr.io/open-webui/open-webui:cuda)
Ollama Version (if applicable)
v0.9.3
Operating System
Windows 11 Pro (10.0.22631 Build 22631)
Browser (if applicable)
Brave 1.80.113 x64 (Chromium: 138.0.7204.49)
Confirmation
README.md.Expected Behavior
Open WebUI uses next to no VRAM when started, and releases any used by models afterwards (OLLAMA_KEEP_ALIVE=0)
Actual Behavior
Starting the Open WebUI Docker container eats VRAM and never releases it. Clicking Save in the Admin Panel > Settings > Documents tab consumes VRAM indefinitely each time the button is clicked. That VRAM is also never released.
Steps to Reproduce
ghcr.io/open-webui/open-webui:cudausing the following run command:Admin Panel > Settings > Documentswindow, click "Save"The only way to release the VRAM is to stop the Open WebUI Docker container. Closing the Ollama server does not free the VRAM, which means it is being consumed directly by OWUI.
Ollama server is running separately and natively on Windows, it is not part of the OWUI container.
Logs & Screenshots
OWUI:
Ollama
server.log:VRAM Usage on Docker container startup:
https://github.com/user-attachments/assets/0953d683-185a-4bb2-965c-c89bb867ed26
Additional Information
VRAM Usage also increases each time I click "Save" in
Admin Panel > Settings > Documentswindow: https://drive.google.com/file/d/16yof8N1mN9VSy94uDIHZf-EynO4fnzNh/view?usp=sharingThe VRAM is never released afterwards until the OWUI docker container is stopped.
@tjbck commented on GitHub (Jun 27, 2025):
Intended behaviour with built-in sentence transformer embedding engine, I'd suggest you to use external embedding engines if you would like to change the current behaviour.