Qwen3.5 Reasoning is forced?

[air-fuel-ratio]

curl -m 60 http://localhost:8081/v1/chat/completions
-H "Content-Type: application/json"
-d '{"model": "Qwen3.5-4B-Q4_K_M.gguf", "messages": [{"role": "user", "content": "say hello"}], "stream": false}'
{"created":1773284146,"object":"chat.completion","id":"2c5092f8-db39-4c15-aedf-a4ef2c6c8f15","model":"Qwen3.5-4B-Q4_K_M.gguf","choices":[{"index":0,"finish_reason":"stop","message":{"role":"assistant","content":"","reasoning":"(the response is a confirmation of the following statement: "The response is in a system response. The assistant is an AI assistant, and the user is referring to the assistant in a system response. The user is asking about the relationship between the assistant and the user in a system response. The assistant is a system response, and the user is referring to the assistant in a system response. The user is asking about the relationship between the assistant and the user in a system response. The assistant is a system response, and the user is referring to the assistant in a system response. The user is asking about the relationship between the assistant and the user in a system response. The assistant is a system response, and the user is referring to the assistant in a system response." in Chinese."}}],"usage":{"prompt_tokens":12,"completion_tokens":158,"total_tokens":170}}

backend: llama-cpp
description: Imported from https://huggingface.co/bartowski/Qwen_Qwen3.5-4B-GGUF
function:
    grammar:
        disable: true
known_usecases:
    - chat
mmproj: llama-cpp/mmproj/mmproj-Qwen_Qwen3.5-4B-f16.gguf
name: qwen_qwen3.5-4b
options:
    - use_jinja:true
parameters:
    model: llama-cpp/models/Qwen_Qwen3.5-4B-Q4_K_M.gguf
template:
    use_tokenizer_template: true

version: "3.9"
services:
  api:
    #image: localai/localai:latest-gpu-intel
    image: localai/localai:master-gpu-intel
    restart: unless-stopped
    healthcheck:
      test: ["CMD", "curl", "-f", "http://localhost:8080/readyz"]
      interval: 1m
      timeout: 20m
      retries: 5
    ports:
      - 8081:8080
    shm_size: '8gb'
    devices:
      - /dev/dri/card1:/dev/dri/card1
      - /dev/dri/renderD129:/dev/dri/renderD129
    group_add:
      - "993"
    environment:
      - CONTEXT_SIZE=4096
      - GGML_SYCL_TARGET=INTEL
      - OLLAMA_COMPAT=true
      - THREADS=1 # Single thread is best for GPU offload
      - DEBUG=true
    volumes:
      - /etc/localtime:/etc/localtime:ro
      - /opt/localai/models:/models:rw
      - /opt/localai/backends:/backends:rw
      - /opt/localai/config:/configuration:rw

[air-fuel-ratio]

https://huggingface.co/Qwen/Qwen3.5-35B-A3B#instruct-or-non-thinking-mode

Qwen3.5 does not officially support the soft switch of Qwen3, i.e., /think and /nothink

I've tried

reasoning:
    disable: true
enable_thinking:
    disable: true

All without effect, i still see the below

{%- if add_generation_prompt %}
    {{- '<|im_start|>assistant\n' }}
    {%- if enable_thinking is defined and enable_thinking is false %}
        {{- '<think>\n\n</think>\n\n' }}
    {%- else %}
        {{- '<think>\n' }}
    {%- endif %}

I've tried all sorts of methods to pass through the below but it has no effect

chat_response = client.chat.completions.create(
    model="Qwen/Qwen3.5-4B",
    messages=messages,
    max_tokens=32768,
    temperature=0.7,
    top_p=0.8,
    presence_penalty=1.5,
    extra_body={
        "top_k": 20,
        "chat_template_kwargs": {"enable_thinking": False},
    }, 
)

[localai-bot]

Thanks for reporting this issue. To help investigate, could you please provide more details about:

1. What version of LocalAI are you using?
2. Is this behavior unexpected (what would you expect instead)?
3. Can you share the full model configuration you're using?

This will help us understand if this is a bug or expected behavior.

[mudler]

This should be fixed by #8973

[localai-bot]

PR #8973 has been merged with the fix. This issue should now be resolved. Please verify and close if the fix addresses your requirements.

[localai-bot]

PR #8973 has been merged with the fix for the Qwen3.5 reasoning issue. The setting is now correctly placed in . Please verify and close this issue if the fix resolves your problem.