The Keep Alive setting in Open-webui v0.5.12 is NOT changing the time the model stays alive.
Expected Behavior:
After setting Keep ALive to 0m or 0 (tried both), i expect the ollama model to stop immediately after processing question
Actual Behavior:
Instead the model (phi4 in this case) just keeps counting down from the default 5m
Description
Being able to immediate drop the LLM model from Vram is important, because if you want to then use the GPU Vram to generate an image with say FLUX dev, you NEED all that Vram free.
Reproduction Details
One a windows 11 system, running open-webui with docker
Ensure no model is running by checking "ollama ps"
Set Keep Alive to 0m under Settings/General/Advanced Parameters/KeepAlive and click Save.
Ask locally hosted model a question, in this case Phi4
run "ollama ps"
notice ollama ps reports Phi4 is still running and will shut down in 4m
System
CPU: AMD Ryzen 9 5950x
RAM: 128GB DDR 4
GPU: RTX 4090
SSD: 2Tb NVMe
Windows 11
Open-webui v0.5.12 running in docker container for windows.
Additional Information
Again i want to emphasize how important it is for us to be able to manage the be able to manage our Vram usage, and quickly unload models, to move on with our workflow.
Thank you.
Alternatively, is there a way to just modify the keep alive defult code in docker, or ollama, if its not possible to fix this setting?
Originally created by @KyleF0X on GitHub (Feb 15, 2025).
Original GitHub issue: https://github.com/open-webui/open-webui/issues/10048
# Bug Report
The Keep Alive setting in Open-webui v0.5.12 is NOT changing the time the model stays alive.
## Expected Behavior:
After setting Keep ALive to 0m or 0 (tried both), i expect the ollama model to stop immediately after processing question
## Actual Behavior:
Instead the model (phi4 in this case) just keeps counting down from the default 5m
## Description
Being able to immediate drop the LLM model from Vram is important, because if you want to then use the GPU Vram to generate an image with say FLUX dev, you NEED all that Vram free.
## Reproduction Details
One a windows 11 system, running open-webui with docker
1. Ensure no model is running by checking "ollama ps"
2. Set Keep Alive to 0m under Settings/General/Advanced Parameters/KeepAlive and click Save.
3. Ask locally hosted model a question, in this case Phi4
4. run "ollama ps"
5. notice ollama ps reports Phi4 is still running and will shut down in 4m
System
CPU: AMD Ryzen 9 5950x
RAM: 128GB DDR 4
GPU: RTX 4090
SSD: 2Tb NVMe
Windows 11
Open-webui v0.5.12 running in docker container for windows.
## Additional Information
Again i want to emphasize how important it is for us to be able to manage the be able to manage our Vram usage, and quickly unload models, to move on with our workflow.
Thank you.
Alternatively, is there a way to just modify the keep alive defult code in docker, or ollama, if its not possible to fix this setting?
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
Originally created by @KyleF0X on GitHub (Feb 15, 2025).
Original GitHub issue: https://github.com/open-webui/open-webui/issues/10048
Bug Report
The Keep Alive setting in Open-webui v0.5.12 is NOT changing the time the model stays alive.
Expected Behavior:
After setting Keep ALive to 0m or 0 (tried both), i expect the ollama model to stop immediately after processing question
Actual Behavior:
Instead the model (phi4 in this case) just keeps counting down from the default 5m
Description
Being able to immediate drop the LLM model from Vram is important, because if you want to then use the GPU Vram to generate an image with say FLUX dev, you NEED all that Vram free.
Reproduction Details
One a windows 11 system, running open-webui with docker
System
CPU: AMD Ryzen 9 5950x
RAM: 128GB DDR 4
GPU: RTX 4090
SSD: 2Tb NVMe
Windows 11
Open-webui v0.5.12 running in docker container for windows.
Additional Information
Again i want to emphasize how important it is for us to be able to manage the be able to manage our Vram usage, and quickly unload models, to move on with our workflow.
Thank you.
Alternatively, is there a way to just modify the keep alive defult code in docker, or ollama, if its not possible to fix this setting?