[GH-ISSUE #5119] [FEAT]: In anythingllm, unable to find the parameter configuration to turn off the model's thinking ability. #4927

Closed
opened 2026-06-05 14:50:55 -04:00 by yindo · 3 comments
Owner

Originally created by @ouy160 on GitHub (Mar 3, 2026).
Original GitHub issue: https://github.com/Mintplex-Labs/anything-llm/issues/5119

What would you like to see?

By using the command line, you can disable the model's reasoning capabilities, but in anythingllm, I am unable to find the parameter configuration to turn off the model's thinking ability.

Image
Originally created by @ouy160 on GitHub (Mar 3, 2026). Original GitHub issue: https://github.com/Mintplex-Labs/anything-llm/issues/5119 ### What would you like to see? By using the command line, you can disable the model's reasoning capabilities, but in anythingllm, I am unable to find the parameter configuration to turn off the model's thinking ability. <img width="703" height="616" alt="Image" src="https://github.com/user-attachments/assets/1c5c8659-dbbc-465b-a941-304000d3a290" />
yindo added the enhancementfeature request labels 2026-06-05 14:50:55 -04:00
yindo closed this issue 2026-06-05 14:50:55 -04:00
Author
Owner

@kalle07 commented on GitHub (Mar 6, 2026):

ALLM only call the models , nothing moer!
dont know ollama, if you find the jinja template you can somtimes turm off thinking by edit.

<!-- gh-comment-id:4011838499 --> @kalle07 commented on GitHub (Mar 6, 2026): ALLM only call the models , nothing moer! dont know ollama, if you find the jinja template you can somtimes turm off thinking by edit.
Author
Owner

@ppil commented on GitHub (Mar 10, 2026):

On both Qwen3.5, and GLM-4.7, I can turn thinking off via llama.cpp, but the API call from anythingllm still invokes thought. I think this is due to other parameters (mainly presence_penalty) not honoring the backend's configuration/default, since API request params take precedence. If that's the issue being raised here, then #1572 would be a dupe.

Update: I was using oobabooga as an orchestrator for llama.cpp, and hitting that openai compatible api not realizing it handles parameters and arguments differently for API vs web ui requests. They recently added explicit API parameter configuration, which can indeed disable thinking and solved the issue here. Likewise, thinking was disabled correctly if I did hit llama-server directly.

<!-- gh-comment-id:4028558791 --> @ppil commented on GitHub (Mar 10, 2026): On both Qwen3.5, and GLM-4.7, I can turn thinking off via llama.cpp, but the API call from anythingllm still invokes thought. I think this is due to other parameters (mainly presence_penalty) not honoring the backend's configuration/default, since API request params take precedence. If that's the issue being raised here, then #1572 would be a dupe. Update: I was using oobabooga as an orchestrator for llama.cpp, and hitting that openai compatible api not realizing it handles parameters and arguments differently for API vs web ui requests. They recently added explicit API parameter configuration, which can indeed disable thinking and solved the issue here. Likewise, thinking was disabled correctly if I did hit llama-server directly.
Author
Owner

@fairol1988 commented on GitHub (Mar 18, 2026):

you have to put /no_think into the system prompt of your workspace. It worked for me.

<!-- gh-comment-id:4087338948 --> @fairol1988 commented on GitHub (Mar 18, 2026): you have to put /no_think into the system prompt of your workspace. It worked for me.
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: Mintplex-Labs/anything-llm#4927