I have searched the existing issues and discussions.
Problem Description
Hi Kime-Dev-72B uses its own thinking on/off tokens. "◁think▷" and "◁/think▷" any guidance how I can get open webui to parse out the thinking correctly?
Desired Solution you'd like
A solution for recognizing Kimi-Dev-72B thinking parts of the response
Alternatives Considered
No response
Additional Context
No response
Originally created by @fernandaspets on GitHub (Jul 13, 2025).
Original GitHub issue: https://github.com/open-webui/open-webui/issues/15682
Originally assigned to: @tjbck on GitHub.
### Check Existing Issues
- [x] I have searched the existing issues and discussions.
### Problem Description
Hi Kime-Dev-72B uses its own thinking on/off tokens. "◁think▷" and "◁/think▷" any guidance how I can get open webui to parse out the thinking correctly?
### Desired Solution you'd like
A solution for recognizing Kimi-Dev-72B thinking parts of the response
### Alternatives Considered
_No response_
### Additional Context
_No response_
<!-- gh-comment-id:3066817226 -->
@rgaricano commented on GitHub (Jul 13, 2025):
You can use an adapted filter function to this particular,
e.g. editing https://openwebui.com/f/jcarlosroldan/deepseek_r1_think_formatter and replacing "`<think>`" by "`◁think▷`" & "`</think>`" by "`◁/think▷`"
It would still be nice to see this officially implemented, because the plugin only replaces the content with the collapsible message after the model is done responding. (as far as I know)
<!-- gh-comment-id:3074797968 -->
@qingy1337 commented on GitHub (Jul 15, 2025):
It would still be nice to see this officially implemented, because the plugin only replaces the content with the collapsible message after the model is done responding. (as far as I know)
@martinoturrina commented on GitHub (Jul 16, 2025):
In general I think it would be nice if this is configurable per model
<!-- gh-comment-id:3077560717 -->
@martinoturrina commented on GitHub (Jul 16, 2025):
In general I think it would be nice if this is configurable per model
In general I think it would be nice if this is configurable per model
Yeah I think you could have a toggle/checkbox for "Thinking Model" and if you enable that, it allows you to put in two strings for the start_think and end_think, something like that?
<!-- gh-comment-id:3077657006 -->
@qingy1337 commented on GitHub (Jul 16, 2025):
> In general I think it would be nice if this is configurable per model
Yeah I think you could have a toggle/checkbox for "Thinking Model" and if you enable that, it allows you to put in two strings for the start_think and end_think, something like that?
@chrisoutwright commented on GitHub (Aug 9, 2025):
It seems to stop streaming at this token (but actually will then show it when done all at once)
<!-- gh-comment-id:3171909685 -->
@chrisoutwright commented on GitHub (Aug 9, 2025):
It seems to stop streaming at this token (but actually will then show it when done all at once)
<img width="731" height="281" alt="Image" src="https://github.com/user-attachments/assets/3e7c2b6d-3b4f-46f2-a578-90d45160071f" />
@chrisoutwright commented on GitHub (Aug 9, 2025):
actually it will print all at once later and not stream the result any more, it seems to stop .. and then all appears when done..
first:
and then:
<!-- gh-comment-id:3171911894 -->
@chrisoutwright commented on GitHub (Aug 9, 2025):
actually it will print all at once later and not stream the result any more, it seems to stop .. and then all appears when done..
first:
<img width="1390" height="974" alt="Image" src="https://github.com/user-attachments/assets/42a938ce-73b4-4967-8a9a-26c91573f591" />
and then:
<img width="1321" height="924" alt="Image" src="https://github.com/user-attachments/assets/68548c00-cb6f-42e4-92ec-ec340226c4fc" />
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 @fernandaspets on GitHub (Jul 13, 2025).
Original GitHub issue: https://github.com/open-webui/open-webui/issues/15682
Originally assigned to: @tjbck on GitHub.
Check Existing Issues
Problem Description
Hi Kime-Dev-72B uses its own thinking on/off tokens. "◁think▷" and "◁/think▷" any guidance how I can get open webui to parse out the thinking correctly?
Desired Solution you'd like
A solution for recognizing Kimi-Dev-72B thinking parts of the response
Alternatives Considered
No response
Additional Context
No response
@rgaricano commented on GitHub (Jul 13, 2025):
You can use an adapted filter function to this particular,
e.g. editing https://openwebui.com/f/jcarlosroldan/deepseek_r1_think_formatter and replacing "
<think>" by "◁think▷" & "</think>" by "◁/think▷"@qingy1337 commented on GitHub (Jul 15, 2025):
It would still be nice to see this officially implemented, because the plugin only replaces the content with the collapsible message after the model is done responding. (as far as I know)
@martinoturrina commented on GitHub (Jul 16, 2025):
In general I think it would be nice if this is configurable per model
@qingy1337 commented on GitHub (Jul 16, 2025):
Yeah I think you could have a toggle/checkbox for "Thinking Model" and if you enable that, it allows you to put in two strings for the start_think and end_think, something like that?
@tjbck commented on GitHub (Jul 16, 2025):
Addressed with
83c09f15acin dev!@chrisoutwright commented on GitHub (Aug 9, 2025):
It seems to stop streaming at this token (but actually will then show it when done all at once)

@chrisoutwright commented on GitHub (Aug 9, 2025):
actually it will print all at once later and not stream the result any more, it seems to stop .. and then all appears when done..
first:
and then: