[Bug]: Cannot turn on/off thinking of qwen 3.6 27B with llama.cpp

[sunzx]

Issue Checklist

• I understand that issues are for feedback and problem solving, not for complaining in the comment section, and will provide as much information as possible to help solve the problem.
• My issue is not listed in the FAQ.
• I've looked at pinned issues and searched for existing Open Issues, Closed Issues, and Discussions, no similar issue or discussion was found.
• I've filled in short, clear headings so that developers can quickly identify a rough idea of what to expect when flipping through the list of issues. And not "a suggestion", "stuck", etc.
• I've confirmed that I am using the latest version of Cherry Studio.

Platform

Windows

Version

v1.9.4

Bug Description

The thinking mode of Qwen 3.6 model hosted by llama.cpp is always ON regardless of the "Reasoning effort" setting in a conversion.

With --chat-template-kwargs '{"enable_thinking":false}' in llama-server command line, the thinking mode can be turned off, but "Reasoning Effort" setting cannot turn it on.

Changing model name to "qwen3.6:27b", changing provider type between "OpenAI" and "Anthropic" do not help.

Steps To Reproduce

1. Host Qwen 3.6 27B model with llama-server (model is ggufbench/Qwen3.6-27B-4bpw-16GB-VRAM). For example:

llama-server -m Qwen3.6-27B-4bpw-16GB-VRAM.gguf \
        -c 262144 \
        -ngl all -ctk q5_0 -ctv q5_0 \
        --temp 0.6 --top-p 0.95 --top-k 20 --min-p 0 \
        --host 0.0.0.0 \
        --port 22345

2. Config Cherry Studio to use this model;
3. Turning on or off "Reasoning effort" has no effect.

Expected Behavior

Reasoning can be turned on/off with the "Reasoning effort" setting. i.e., add

"chat_template_kwargs":{"enable_thinking":false}

or

"chat_template_kwargs":{"enable_thinking":true}

in request payload.

Relevant Log Output

Additional Context

Cherry studio adds "enable_thinking":false or enable_thinking: true in request payload by default. It works on Aliyun Bailian API, but not works on llama.cpp.

Add "chat_template_kwargs" custom parameter with JSON {"enable_thinking": false} in Model Settings can turn off thinking.

[sunzx]

llama.cpp ignores "enable_thinking":false in request payload, so "enable_thinking":false and "chat_template_kwargs":{"enable_thinking":false} may be added in request payload simultaneously for compatibility.

[sunzx]

workaround: use "nvidia" provider and config API Host to llama.cpp address.
ref: reasoning.ts#126

[github-actions]

This issue has been inactive for a prolonged period and will be closed automatically in 10 days.
该问题已长时间处于闲置状态，10 天后将自动关闭。