github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-07 16:40:08 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #10583] close qwen3 reasoning #69022

New Issue
Closed
opened 2026-05-04 16:51:55 -05:00 by GiteaMirror · 9 comments
GiteaMirror commented 2026-05-04 16:51:55 -05:00
Owner
Copy Link

Originally created by @niubiqianshui on GitHub (May 6, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/10583

How to turn off the reasoning mode of Qwen3-235B-A22B when I use Docker to run it

Originally created by @niubiqianshui on GitHub (May 6, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/10583 How to turn off the reasoning mode of Qwen3-235B-A22B when I use Docker to run it
GiteaMirror added the feature request label 2026-05-04 16:51:55 -05:00
GiteaMirror commented 2026-05-04 16:52:48 -05:00
Author
Owner
Copy Link

@garychanchan commented on GitHub (May 6, 2025):

+1 目前有参数可以关闭深度思考嘛？

<!-- gh-comment-id:2853481162 --> @garychanchan commented on GitHub (May 6, 2025): +1 目前有参数可以关闭深度思考嘛？
GiteaMirror commented 2026-05-04 16:52:51 -05:00
Author
Owner
Copy Link

@wawameizhangya commented on GitHub (May 6, 2025):

m

<!-- gh-comment-id:2853509548 --> @wawameizhangya commented on GitHub (May 6, 2025): m
GiteaMirror commented 2026-05-04 16:52:56 -05:00
Author
Owner
Copy Link

@niubiqianshui commented on GitHub (May 6, 2025):

curl http://127.0.0.1:11456/api/generate -d '{ "model": "qwen3:235b-a22b", "prompt": "Why is the sky blue? /no_think","stream":false }'

<!-- gh-comment-id:2853571367 --> @niubiqianshui commented on GitHub (May 6, 2025): curl http://127.0.0.1:11456/api/generate -d '{ "model": "qwen3:235b-a22b", "prompt": "Why is the sky blue? /no_think","stream":false }'
GiteaMirror commented 2026-05-04 16:52:59 -05:00
Author
Owner
Copy Link

@sunisstar commented on GitHub (May 6, 2025):

curl http://127.0.0.1:11456/api/generate -d '{ "model": "qwen3:235b-a22b", "prompt": "Why is the sky blue? /no_think","stream":false }'

but a tag is still remained, a parameter will prevent from this.

<!-- gh-comment-id:2853832985 --> @sunisstar commented on GitHub (May 6, 2025): > curl http://127.0.0.1:11456/api/generate -d '{ "model": "qwen3:235b-a22b", "prompt": "Why is the sky blue? /no_think","stream":false }' but a <think> </think> tag is still remained, a parameter will prevent from this.
GiteaMirror commented 2026-05-04 16:53:01 -05:00
Author
Owner
Copy Link

@niubiqianshui commented on GitHub (May 6, 2025):

curl http://127.0.0.1:11456/api/generate -d '{ "model": "qwen3:235b-a22b", "prompt": "Why is the sky blue? /no_think","stream":false }'

but a tag is still remained, a parameter will prevent from this.

what parameter?

<!-- gh-comment-id:2853837922 --> @niubiqianshui commented on GitHub (May 6, 2025): > > curl http://127.0.0.1:11456/api/generate -d '{ "model": "qwen3:235b-a22b", "prompt": "Why is the sky blue? /no_think","stream":false }' > > but a tag is still remained, a parameter will prevent from this. what parameter?
GiteaMirror commented 2026-05-04 16:53:03 -05:00
Author
Owner
Copy Link

@y-tor commented on GitHub (May 6, 2025):

@ehartford shared this model file:

https://gist.github.com/ehartford/a465d24116b7df07170efe69116084a3

This is a clever way of forcing the model to not generate the think tag in the output.

<!-- gh-comment-id:2854059208 --> @y-tor commented on GitHub (May 6, 2025): @ehartford shared this model file: https://gist.github.com/ehartford/a465d24116b7df07170efe69116084a3 This is a clever way of forcing the model to not generate the think tag in the output.
GiteaMirror commented 2026-05-04 16:53:05 -05:00
Author
Owner
Copy Link

@xlorne commented on GitHub (May 6, 2025):

I am using OpenAI’s standard protocol and have developed the implementation based on SpringBoot-AI. By dynamically appending /no_think in the prompt, I can enable or disable the thinking mode. As for the returned tag data, it can be filtered using an advisor.

package com.codingapi.agent.advisor;

import lombok.NonNull;
import lombok.extern.slf4j.Slf4j;
import org.springframework.ai.chat.client.advisor.api.*;
import org.springframework.ai.chat.messages.AssistantMessage;
import org.springframework.ai.chat.model.ChatResponse;
import org.springframework.ai.chat.model.Generation;
import reactor.core.publisher.Flux;

import java.util.ArrayList;
import java.util.List;

@Slf4j
public class Qwen3ThinkFilterAdvisor implements CallAroundAdvisor, StreamAroundAdvisor {

    private final boolean thinkEnabled;

    private boolean qwen3Model = false;

    public Qwen3ThinkFilterAdvisor(boolean thinkEnabled) {
        this.thinkEnabled = thinkEnabled;
    }

    private void checkModelName(AdvisedRequest advisedRequest) {
        String modelName = advisedRequest.chatModel().getDefaultOptions().getModel();
        this.qwen3Model = modelName != null && modelName.contains("qwen3");
    }

    private List<Generation> filterGenerations(List<Generation> generations) {
        if (thinkEnabled) {
            return generations;
        }
        List<Generation> generationList = new ArrayList<>();
        for (Generation generation : generations) {
            AssistantMessage assistantMessage = generation.getOutput();
            String text = assistantMessage.getText();
            if (text.contains("<think>")) {
                text = text.replaceAll("(?s)<think>.*?</think>", "");
                text = text.replaceAll("(?m)^[ \\t]*\\r?\\n", "");
            }
            AssistantMessage responseMessage = new AssistantMessage(text,
                    assistantMessage.getMetadata(),
                    assistantMessage.getToolCalls(),
                    assistantMessage.getMedia());
            generationList.add(new Generation(responseMessage));
        }
        return generationList;
    }


    private AdvisedRequest filterRequest(AdvisedRequest advisedRequest) {
        if (thinkEnabled) {
            return advisedRequest;
        }
        String question = advisedRequest.userText();
        question = question + "/no_think";
        return AdvisedRequest.from(advisedRequest)
                .userText(question)
                .build();
    }

    @Override
    @NonNull
    public AdvisedResponse aroundCall(@NonNull AdvisedRequest advisedRequest, CallAroundAdvisorChain chain) {
        this.checkModelName(advisedRequest);
        if (qwen3Model) {
            AdvisedRequest request = this.filterRequest(advisedRequest);
            AdvisedResponse response = chain.nextAroundCall(request);
            assert response.response() != null;
            List<Generation> generations = this.filterGenerations(response.response().getResults());
            return AdvisedResponse.builder()
                    .response(ChatResponse.builder()
                            .generations(generations)
                            .build())
                    .adviseContext(advisedRequest.adviseContext())
                    .build();
        }else {
            return chain.nextAroundCall(advisedRequest);
        }
    }

    @Override
    @NonNull
    public Flux<AdvisedResponse> aroundStream(@NonNull AdvisedRequest advisedRequest, StreamAroundAdvisorChain chain) {
        this.checkModelName(advisedRequest);
        if (qwen3Model) {
            AdvisedRequest request = this.filterRequest(advisedRequest);
            return chain.nextAroundStream(request)
                    .map(advisedResponse -> {
                        assert advisedResponse.response() != null;
                        List<Generation> generations = this.filterGenerations(advisedResponse.response().getResults());
                        return AdvisedResponse.builder()
                                .response(ChatResponse.builder()
                                        .generations(generations)
                                        .build())
                                .adviseContext(advisedRequest.adviseContext())
                                .build();
                    });
        }else {
            return chain.nextAroundStream(advisedRequest);
        }
    }

    @NonNull
    @Override
    public String getName() {
        return "qwen3ThinkFilter";
    }

    @Override
    public int getOrder() {
        return 0;
    }
}

https://github.com/xlorne/springboot-agent

<!-- gh-comment-id:2854067602 --> @xlorne commented on GitHub (May 6, 2025): I am using OpenAI’s standard protocol and have developed the implementation based on SpringBoot-AI. By dynamically appending /no_think in the prompt, I can enable or disable the thinking mode. As for the returned <think></think> tag data, it can be filtered using an advisor. ``` package com.codingapi.agent.advisor; import lombok.NonNull; import lombok.extern.slf4j.Slf4j; import org.springframework.ai.chat.client.advisor.api.*; import org.springframework.ai.chat.messages.AssistantMessage; import org.springframework.ai.chat.model.ChatResponse; import org.springframework.ai.chat.model.Generation; import reactor.core.publisher.Flux; import java.util.ArrayList; import java.util.List; @Slf4j public class Qwen3ThinkFilterAdvisor implements CallAroundAdvisor, StreamAroundAdvisor { private final boolean thinkEnabled; private boolean qwen3Model = false; public Qwen3ThinkFilterAdvisor(boolean thinkEnabled) { this.thinkEnabled = thinkEnabled; } private void checkModelName(AdvisedRequest advisedRequest) { String modelName = advisedRequest.chatModel().getDefaultOptions().getModel(); this.qwen3Model = modelName != null && modelName.contains("qwen3"); } private List<Generation> filterGenerations(List<Generation> generations) { if (thinkEnabled) { return generations; } List<Generation> generationList = new ArrayList<>(); for (Generation generation : generations) { AssistantMessage assistantMessage = generation.getOutput(); String text = assistantMessage.getText(); if (text.contains("<think>")) { text = text.replaceAll("(?s)<think>.*?</think>", ""); text = text.replaceAll("(?m)^[ \\t]*\\r?\\n", ""); } AssistantMessage responseMessage = new AssistantMessage(text, assistantMessage.getMetadata(), assistantMessage.getToolCalls(), assistantMessage.getMedia()); generationList.add(new Generation(responseMessage)); } return generationList; } private AdvisedRequest filterRequest(AdvisedRequest advisedRequest) { if (thinkEnabled) { return advisedRequest; } String question = advisedRequest.userText(); question = question + "/no_think"; return AdvisedRequest.from(advisedRequest) .userText(question) .build(); } @Override @NonNull public AdvisedResponse aroundCall(@NonNull AdvisedRequest advisedRequest, CallAroundAdvisorChain chain) { this.checkModelName(advisedRequest); if (qwen3Model) { AdvisedRequest request = this.filterRequest(advisedRequest); AdvisedResponse response = chain.nextAroundCall(request); assert response.response() != null; List<Generation> generations = this.filterGenerations(response.response().getResults()); return AdvisedResponse.builder() .response(ChatResponse.builder() .generations(generations) .build()) .adviseContext(advisedRequest.adviseContext()) .build(); }else { return chain.nextAroundCall(advisedRequest); } } @Override @NonNull public Flux<AdvisedResponse> aroundStream(@NonNull AdvisedRequest advisedRequest, StreamAroundAdvisorChain chain) { this.checkModelName(advisedRequest); if (qwen3Model) { AdvisedRequest request = this.filterRequest(advisedRequest); return chain.nextAroundStream(request) .map(advisedResponse -> { assert advisedResponse.response() != null; List<Generation> generations = this.filterGenerations(advisedResponse.response().getResults()); return AdvisedResponse.builder() .response(ChatResponse.builder() .generations(generations) .build()) .adviseContext(advisedRequest.adviseContext()) .build(); }); }else { return chain.nextAroundStream(advisedRequest); } } @NonNull @Override public String getName() { return "qwen3ThinkFilter"; } @Override public int getOrder() { return 0; } } ``` https://github.com/xlorne/springboot-agent
GiteaMirror commented 2026-05-04 16:53:06 -05:00
Author
Owner
Copy Link

@cycloarcane commented on GitHub (May 6, 2025):

/think and /no_think can be used as a toggle in either any message as others have said, or in a system prompt in whatever framework you are using.

You can toggle as many times as you want in the same conversation. The model will go by the latest toggle used.

<!-- gh-comment-id:2854752810 --> @cycloarcane commented on GitHub (May 6, 2025): /think and /no_think can be used as a toggle in either any message as others have said, or in a system prompt in whatever framework you are using. You can toggle as many times as you want in the same conversation. The model will go by the latest toggle used.
GiteaMirror commented 2026-05-04 16:53:09 -05:00
Author
Owner
Copy Link

@niubiqianshui commented on GitHub (May 7, 2025):

/

<!-- gh-comment-id:2856695301 --> @niubiqianshui commented on GitHub (May 7, 2025): /
Sign in to join this conversation.
No Branch/Tag Specified
Branches Tags
Labels
Clear labels
amd
api
app
bug
build
cli
cloud
compatibility
context-length
create
docker
documentation
embeddings
feature request
feedback wanted
good first issue
gpt-oss
gpu
harmony
help wanted
image
install
intel
js
launch
linux
macos
memory
mlx
model
needs more info
networking
nvidia
ollama.com
performance
pull-request
Mirrored from GitHub Pull Request
 python
question
registry
rendering
thinking
tools
top
vulkan
windows
wsl
No Label feature request
Milestone
No items
No Milestone
Projects
Clear projects
No project
Assignees
Clear assignees
GiteaMirror ninjasurge
No Assignees
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: github-starred/ollama#69022
Delete Branch "%!s()"

Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?