[GH-ISSUE #16525] feat: Add title generation (and other "tasks") to model capabilities #121253

New Issue

GiteaMirror · 2026-05-20T23:10:45-05:00

GiteaMirror commented

2026-05-20 23:10:45 -05:00

Originally created by @Master-Pr0grammer on GitHub (Aug 12, 2025).
Original GitHub issue: https://github.com/open-webui/open-webui/issues/16525

Check Existing Issues

I have searched the existing issues and discussions.

Problem Description

With the recent releases of reasoning models, title generation (and other features like it) have begun to halt all processing on less powerful machines for potentially 10+ minutes before any other prompt can begin to process.

It massively slows everything down, and creates a poor user experience.

Desired Solution you'd like

Since OWUI models already have a capabilities selection on a per model basis, we should add title generation (and potentially most other tasks) to this selection. If the current model doesn't support a given OWUI capability, then OWUI would just default to the selected "task model". If the current model has the capablility, OWUI will just use the current model for that task.

This seems like a really elegant way of handling these tasks, even without the issues i brought up earlier, it just seems like a more straight forward and simpler way to handle this.

Alternatives Considered

There are currently two ways to get around this, but neither option is great:

set a task model
- is slow to unload current model, load in task model, generate title, unload again, then reload first model
- then this model will be used for EVERY task, and you need to go through the slow unload/reload for any task.
add /no_think to OWUI prompts
- no longer works with new models since hybrid models were found to be less efficient.

Additional Context

In looking for similar feature requests, I came across #15459 which lays out a similar problem. This implementation would also solve these issues aswell.

Originally created by @Master-Pr0grammer on GitHub (Aug 12, 2025). Original GitHub issue: https://github.com/open-webui/open-webui/issues/16525 ### Check Existing Issues - [x] I have searched the existing issues and discussions. ### Problem Description With the recent releases of reasoning models, title generation (and other features like it) have begun to halt all processing on less powerful machines for potentially 10+ minutes before any other prompt can begin to process. It massively slows everything down, and creates a poor user experience. ### Desired Solution you'd like Since OWUI models already have a capabilities selection on a per model basis, we should add title generation (and potentially most other tasks) to this selection. If the current model doesn't support a given OWUI capability, then OWUI would just default to the selected "task model". If the current model has the capablility, OWUI will just use the current model for that task. This seems like a really elegant way of handling these tasks, even without the issues i brought up earlier, it just seems like a more straight forward and simpler way to handle this. ### Alternatives Considered There are currently two ways to get around this, but neither option is great: * set a task model - is slow to unload current model, load in task model, generate title, unload again, then reload first model - then this model will be used for EVERY task, and you need to go through the slow unload/reload for any task. * add /no_think to OWUI prompts - no longer works with new models since hybrid models were found to be less efficient. ### Additional Context In looking for similar feature requests, I came across #15459 which lays out a similar problem. This implementation would also solve these issues aswell.

GiteaMirror closed this issue

2026-05-20 23:10:46 -05:00

GiteaMirror commented

2026-05-20 23:10:47 -05:00

@tjbck commented on GitHub (Aug 12, 2025):

#15459

@tjbck commented on GitHub (Aug 12, 2025): #15459

GiteaMirror commented

2026-05-20 23:10:47 -05:00

@Master-Pr0grammer commented on GitHub (Aug 16, 2025):

This is a similar problem, but I just wanted to point out that it's a very different solution, and their solution wouldn't help in my case.

For example, if you have the current model generate the titles, it doesn't matter for most models, except for reasoning models. With the solution I proposed, you would be able to configure it to exclude reasoning models from this capability, defaulting to a smaller, quicker model.

For models where it doesn't matter, it would be best to have the current model do the generation so you don't need to wait for models to unload, load, unload, and reload again.

I think a combination of both of these ideas (mine and # 15459) would work best for most people.

@Master-Pr0grammer commented on GitHub (Aug 16, 2025): This is a similar problem, but I just wanted to point out that it's a very different solution, and their solution wouldn't help in my case. For example, if you have the current model generate the titles, it doesn't matter for most models, except for reasoning models. With the solution I proposed, you would be able to configure it to exclude reasoning models from this capability, defaulting to a smaller, quicker model. For models where it doesn't matter, it would be best to have the current model do the generation so you don't need to wait for models to unload, load, unload, and reload again. I think a combination of both of these ideas (mine and # 15459) would work best for most people.

Sign in to join this conversation.

Branches Tags

1 Participants

Notifications

Due Date

No due date set.

Dependencies

No dependencies set.

Reference: github-starred/open-webui#121253