/no_think does not work

#5
by mdpi-ai - opened

hi, I notice with this model the /no_think that works on Qwen/Qwen3-32B-AWQ does not work as intended.
also I notice it does not start it's reply with .
I am curious if is it a problem with the base model or is it a problem only with this quantized version?

{
  "messages": [
    {
      "content": "/no_think What is the capital of Switzerland?",
      "role": "user"
    }
  ],
  "model": "QuantTrio/Qwen3-VL-30B-A3B-Instruct-AWQ"
}

reply

 {
 "choices": [
    {
      "index": 0,
      "message": {
        "role": "assistant",
        "content": "Okay, the user is asking for the capital of Switzerland. Hmm, this seems straightforward, but let me double-check because sometimes people confuse it with other cities like Zurich or Geneva. I remember that Bern is the capital, but I should verify to be absolutely sure. \n\nThe user didn't provide any extra context, so they probably just want a quick factual answer. No signs of confusion or deeper needs here—just a simple geography question. Maybe they're studying, traveling, or just curious. \n\nI'll confirm: Switzerland's capital is Bern, not Zurich (which is a major city but not the capital) or Geneva (which hosts the UN). Yep, Bern is correct. The answer should be short and clear since the user didn't ask for details. \n\nWait—should I mention why it's not Zurich? Nah, they didn't ask for explanation, so just the answer. But if they follow up, I can elaborate on the \"why\" later. \n\nFinal answer: Bern. Simple and precise.\n</think>\n\nThe capital of Switzerland is **Bern**.  \n\nWhile Zurich and Geneva are major Swiss cities (Zurich is the largest city and Geneva is known for international organizations), **Bern** is the official federal capital, where the Swiss government and parliament are located.  \n\nThis is a common point of confusion, but Bern has been the capital since 1848. 🇨🇭",
        "refusal": null,
        "annotations": null,
        "audio": null,
        "function_call": null,
        "tool_calls": [],
        "reasoning_content": null
      },
      "logprobs": null,
      "finish_reason": "stop",
      "stop_reason": null,
      "token_ids": null
    }
  ],
}

This 30B-VL model is an Instruct model, not a Thinking one.
Please note that starting from the Qwen3 2507 release, the Instruct and Thinking models were split into two separate variants.
Since then, the no_think flag in prompts is no longer effective.

For this model in particular, regardless of whether your prompt includes think or no_think,
the output will not contain tags.

If you want to use the Thinking mode, please refer to
👉 QuantTrio/Qwen3-VL-30B-A3B-Thinking-AWQ

Sign up or log in to comment