github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-07 00:22:43 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #11032] Can't Disable Think Mode of Qwen3 and DeepSeek #33036

New Issue
Closed
opened 2026-04-22 15:11:38 -05:00 by GiteaMirror · 17 comments
GiteaMirror commented 2026-04-22 15:11:38 -05:00
Owner
Copy Link

Originally created by @zhangiser on GitHub (Jun 10, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/11032

What is the issue?

I'm use ollama on win11,the version is 0.9.0

C:\Users\zhangiser>ollama run qwen3:30b-a3b
>>> /set nothink
warning: model "qwen3:30b-a3b" does not support thinking output
Set 'nothink' mode.
>>> who are you?
<think>
Okay, the user asked, "who are you?" I need to respond clearly. First, I should introduce myself as Qwen, a
large-scale language model developed by Alibaba. I should mention my main functions like answering questions,
creating text, coding, etc. Also, highlight my multilingual support. Keep it friendly and open for more questions.
Make sure the response is concise and covers the key points without being too technical. Check for any errors and
ensure the tone is helpful and approachable.
</think>

Hello! I'm Qwen, a large-scale language model developed by Alibaba. I can answer questions, create text, code, and
more. I support multiple languages, including Chinese and English. How can I assist you today? 😊

When set from CLI

C:\Users\zhangiser>ollama run qwen3:30b-a3b --think=false
>>> who are you?
<think>
Okay, the user asked, "who are you?" I need to respond clearly. First, I should introduce myself as Qwen, a
large-scale language model developed by Alibaba. I should mention my main functions like answering questions,
creating text, coding, etc. Also, highlight my multilingual support. Keep it friendly and open for more questions.
Make sure the response is concise and covers the key points without being too technical. Check for any errors and
ensure the tone is helpful and approachable.
</think>

Hello! I'm Qwen, a large-scale language model developed by Alibaba. I can answer questions, create text, code, and
more. I support multiple languages, including Chinese and English. How can I assist you today? 😊

I've also tried DeepSeek,but get same problem.
Anyone knows why?

C:\Users\zhangiser>ollama list
NAME                                    ID              SIZE      MODIFIED
qwen3:30b-a3b                           2ee832bc15b5    18 GB     5 weeks ago
huihui_ai/qwen2.5-1m-abliterated:14b    3bf3d8d5e063    9.0 GB    2 months ago
qwen2.5-coder:32b                       4bd6cbf2d094    19 GB     2 months ago
qwen2.5-coder:14b                       3028237cc8c5    9.0 GB    4 months ago
deepseek-r1:32b                         38056bbcbb2d    19 GB     4 months ago
deepseek-r1:14b                         ea35dfe18182    9.0 GB    4 months ago
deepseek-coder-v2:16b                   63fb193b3a9b    8.9 GB    4 months ago

Should i re-download the model ?

Originally created by @zhangiser on GitHub (Jun 10, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/11032 ### What is the issue? I'm use ollama on win11,the version is 0.9.0 ```shell C:\Users\zhangiser>ollama run qwen3:30b-a3b >>> /set nothink warning: model "qwen3:30b-a3b" does not support thinking output Set 'nothink' mode. >>> who are you? <think> Okay, the user asked, "who are you?" I need to respond clearly. First, I should introduce myself as Qwen, a large-scale language model developed by Alibaba. I should mention my main functions like answering questions, creating text, coding, etc. Also, highlight my multilingual support. Keep it friendly and open for more questions. Make sure the response is concise and covers the key points without being too technical. Check for any errors and ensure the tone is helpful and approachable. </think> Hello! I'm Qwen, a large-scale language model developed by Alibaba. I can answer questions, create text, code, and more. I support multiple languages, including Chinese and English. How can I assist you today? 😊 ``` When set from CLI ```shell C:\Users\zhangiser>ollama run qwen3:30b-a3b --think=false >>> who are you? <think> Okay, the user asked, "who are you?" I need to respond clearly. First, I should introduce myself as Qwen, a large-scale language model developed by Alibaba. I should mention my main functions like answering questions, creating text, coding, etc. Also, highlight my multilingual support. Keep it friendly and open for more questions. Make sure the response is concise and covers the key points without being too technical. Check for any errors and ensure the tone is helpful and approachable. </think> Hello! I'm Qwen, a large-scale language model developed by Alibaba. I can answer questions, create text, code, and more. I support multiple languages, including Chinese and English. How can I assist you today? 😊 ``` I've also tried DeepSeek,but get same problem. Anyone knows why? ```shell C:\Users\zhangiser>ollama list NAME ID SIZE MODIFIED qwen3:30b-a3b 2ee832bc15b5 18 GB 5 weeks ago huihui_ai/qwen2.5-1m-abliterated:14b 3bf3d8d5e063 9.0 GB 2 months ago qwen2.5-coder:32b 4bd6cbf2d094 19 GB 2 months ago qwen2.5-coder:14b 3028237cc8c5 9.0 GB 4 months ago deepseek-r1:32b 38056bbcbb2d 19 GB 4 months ago deepseek-r1:14b ea35dfe18182 9.0 GB 4 months ago deepseek-coder-v2:16b 63fb193b3a9b 8.9 GB 4 months ago ``` Should i re-download the model ?
GiteaMirror added the bug label 2026-04-22 15:11:38 -05:00
GiteaMirror commented 2026-04-22 15:11:43 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Jun 10, 2025):

Should i re-download the model ?

Yes.

<!-- gh-comment-id:2958018686 --> @rick-github commented on GitHub (Jun 10, 2025): > Should i re-download the model ? Yes.
GiteaMirror commented 2026-04-22 15:11:44 -05:00
Author
Owner
Copy Link

@tonysprite commented on GitHub (Jun 10, 2025):

Please choose model name with "*chat" suffix,
eg:deepseek-r1-7b-Chat

<!-- gh-comment-id:2958147584 --> @tonysprite commented on GitHub (Jun 10, 2025): Please choose model name with "*chat" suffix, eg:deepseek-r1-7b-Chat
GiteaMirror commented 2026-04-22 15:11:44 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Jun 10, 2025):

The are no thinking deepseek models in the ollama library with a chat suffix.

<!-- gh-comment-id:2958165509 --> @rick-github commented on GitHub (Jun 10, 2025): The are no thinking deepseek models in the ollama library with a chat suffix.
GiteaMirror commented 2026-04-22 15:11:45 -05:00
Author
Owner
Copy Link

@tonysprite commented on GitHub (Jun 10, 2025):

I tested two conversations and found that deepseek-R1:1.5B always ran in chat model . Mybe you can try it.

ollama run deepseek-R1:1.5B

who are you?

Greetings! I'm DeepSeek-R1, an artificial intelligence assistant created
by DeepSeek. I'm at your service and would be delighted to assist you with
any inquiries or tasks you may have.

/set nothink
Unknown command '/set nothink'. Type /? for help
who are you?

Greetings! I'm DeepSeek-R1, an artificial intelligence assistant created
by DeepSeek. I'm at your service and would be delighted to assist you with
any inquiries or tasks you may have.

<!-- gh-comment-id:2958287682 --> @tonysprite commented on GitHub (Jun 10, 2025): I tested two conversations and found that deepseek-R1:1.5B always ran in chat model . Mybe you can try it. ollama run deepseek-R1:1.5B >>> who are you? <think> </think> Greetings! I'm DeepSeek-R1, an artificial intelligence assistant created by DeepSeek. I'm at your service and would be delighted to assist you with any inquiries or tasks you may have. >>> /set nothink Unknown command '/set nothink'. Type /? for help >>> who are you? <think> </think> Greetings! I'm DeepSeek-R1, an artificial intelligence assistant created by DeepSeek. I'm at your service and would be delighted to assist you with any inquiries or tasks you may have.
GiteaMirror commented 2026-04-22 15:11:46 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Jun 10, 2025):

/set nothink
Unknown command '/set nothink'. Type /? for help

Upgrade ollama.

<!-- gh-comment-id:2958295629 --> @rick-github commented on GitHub (Jun 10, 2025): > > > > /set nothink > > > > Unknown command '/set nothink'. Type /? for help Upgrade ollama.
GiteaMirror commented 2026-04-22 15:11:47 -05:00
Author
Owner
Copy Link

@tonysprite commented on GitHub (Jun 10, 2025):

tks. Just now I pulled the latest ollama and built the ollam command. It worked as below:

/set think
warning: model "deepseek-R1:1.5B" does not support thinking output # As you see, the version actually dose not support think mode.
Set 'think' mode.
who are you ?
error: registry.ollama.ai/library/deepseek-R1:1.5B does not support thinking
/set nothink
warning: model "deepseek-R1:1.5B" does not support thinking output
Set 'nothink' mode.
who are you?

<!-- gh-comment-id:2958412713 --> @tonysprite commented on GitHub (Jun 10, 2025): tks. Just now I pulled the latest ollama and built the ollam command. It worked as below: >>> /set think warning: model "deepseek-R1:1.5B" does not support thinking output # As you see, the version actually dose not support think mode. Set 'think' mode. >>> who are you ? error: registry.ollama.ai/library/deepseek-R1:1.5B does not support thinking >>> /set nothink warning: model "deepseek-R1:1.5B" does not support thinking output Set 'nothink' mode. >>> who are you? <think>
GiteaMirror commented 2026-04-22 15:11:48 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Jun 10, 2025):

warning: model "deepseek-R1:1.5B" does not support thinking output

https://github.com/ollama/ollama/issues/11032#issuecomment-2958018686

<!-- gh-comment-id:2958423188 --> @rick-github commented on GitHub (Jun 10, 2025): > warning: model "deepseek-R1:1.5B" does not support thinking output https://github.com/ollama/ollama/issues/11032#issuecomment-2958018686
GiteaMirror commented 2026-04-22 15:11:49 -05:00
Author
Owner
Copy Link

@pdevine commented on GitHub (Jun 10, 2025):

Going to close the issue as answered.

<!-- gh-comment-id:2960178308 --> @pdevine commented on GitHub (Jun 10, 2025): Going to close the issue as answered.
GiteaMirror commented 2026-04-22 15:11:51 -05:00
Author
Owner
Copy Link

@rcspam commented on GitHub (Sep 21, 2025):

Why close this isssue given that Qwen3: 30B still does not run in Nothinking mode ?

<!-- gh-comment-id:3315869042 --> @rcspam commented on GitHub (Sep 21, 2025): Why close this isssue given that Qwen3: 30B still does not run in Nothinking mode ?
GiteaMirror commented 2026-04-22 15:11:52 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Sep 21, 2025):

https://ollama.com/library/qwen3:30b-a3b-instruct-2507-q4_K_M

<!-- gh-comment-id:3315873259 --> @rick-github commented on GitHub (Sep 21, 2025): https://ollama.com/library/qwen3:30b-a3b-instruct-2507-q4_K_M
GiteaMirror commented 2026-04-22 15:11:54 -05:00
Author
Owner
Copy Link

@rcspam commented on GitHub (Sep 21, 2025):

It's coolIs but is there a qwen3:30b that allows you to alternate Think and Thinking mode with the /set parameters or from the commande line !!

<!-- gh-comment-id:3315893759 --> @rcspam commented on GitHub (Sep 21, 2025): It's coolIs but is there a qwen3:30b that allows you to alternate Think and Thinking mode with the /set parameters or from the commande line !!
GiteaMirror commented 2026-04-22 15:11:55 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Sep 21, 2025):

https://ollama.com/library/qwen3:30b-a3b-q4_K_M

<!-- gh-comment-id:3315897230 --> @rick-github commented on GitHub (Sep 21, 2025): https://ollama.com/library/qwen3:30b-a3b-q4_K_M
GiteaMirror commented 2026-04-22 15:11:56 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Sep 25, 2025):

https://github.com/ollama/ollama/issues/11032#issuecomment-3315897230

<!-- gh-comment-id:3333541571 --> @rick-github commented on GitHub (Sep 25, 2025): https://github.com/ollama/ollama/issues/11032#issuecomment-3315897230
GiteaMirror commented 2026-04-22 15:11:57 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Sep 25, 2025):

It's a qwen3:30b that allows you to alternate Think and Thinking mode with the /set parameters or from the commande line.

$ ollama run qwen3:30b-a3b-q4_K_M 
>>> hello
Thinking...
Okay, the user just said "hello". I need to respond appropriately. Since it's a greeting, I should be friendly and welcoming. Let me make sure to 
acknowledge their greeting and offer assistance. Maybe say something like "Hello! How can I assist you today?" That's polite and opens the door for 
them to ask questions or share what they need. I should keep it simple and positive. No need for any complex phrases. Just a straightforward, friendly 
response.
...done thinking.

Hello! How can I assist you today? 😊

>>> /set nothink
Set 'nothink' mode.
>>> hello
Hello again! How can I help you today? 😊

>>> /set think
Set 'think' mode.
>>> hello
Thinking...
Okay, the user has sent "hello" multiple times. Let me check the history. The first message was "hello", then the assistant replied with "Hello! How 
can I assist you today? 😊". The user sent "hello" again, and the assistant replied with "Hello again! How can I help you today? 😊". Now the user is 
sending "hello" once more.

I need to figure out the appropriate response. The user might be testing the system, trying to get a different response, or maybe they're not sure how 
to proceed. Since the previous responses were friendly and open-ended, I should maintain that tone. Maybe the user is looking for a more specific 
answer or a different kind of greeting. Alternatively, they might be experiencing technical issues where the messages aren't being properly received.

I should consider the user's intent. If they're just saying hello, perhaps they want to continue the conversation. The assistant's role is to be 
helpful and engaging. Maybe the user is looking for a more personalized response. However, since the user hasn't provided any specific questions or 
topics, the assistant should keep the response open and inviting. It's also possible that the user is using a different device or platform where the 
previous interactions aren't showing up correctly. 

In any case, the assistant should respond politely and encourage the user to ask a question or share what they need help with. Repeating the same 
response might not be effective, so maybe varying the greeting slightly could help. But since the user has only said "hello" three times, it's possible 
they're just testing the response. The assistant should remain consistent and friendly, ensuring the user feels welcomed and supported.
...done thinking.

Hello again! 😊 How can I assist you today? Let me know if you have any questions or need help with something specific!
<!-- gh-comment-id:3333564734 --> @rick-github commented on GitHub (Sep 25, 2025): It's a qwen3:30b that allows you to alternate Think and Thinking mode with the /set parameters or from the commande line. ```console $ ollama run qwen3:30b-a3b-q4_K_M >>> hello Thinking... Okay, the user just said "hello". I need to respond appropriately. Since it's a greeting, I should be friendly and welcoming. Let me make sure to acknowledge their greeting and offer assistance. Maybe say something like "Hello! How can I assist you today?" That's polite and opens the door for them to ask questions or share what they need. I should keep it simple and positive. No need for any complex phrases. Just a straightforward, friendly response. ...done thinking. Hello! How can I assist you today? 😊 >>> /set nothink Set 'nothink' mode. >>> hello Hello again! How can I help you today? 😊 >>> /set think Set 'think' mode. >>> hello Thinking... Okay, the user has sent "hello" multiple times. Let me check the history. The first message was "hello", then the assistant replied with "Hello! How can I assist you today? 😊". The user sent "hello" again, and the assistant replied with "Hello again! How can I help you today? 😊". Now the user is sending "hello" once more. I need to figure out the appropriate response. The user might be testing the system, trying to get a different response, or maybe they're not sure how to proceed. Since the previous responses were friendly and open-ended, I should maintain that tone. Maybe the user is looking for a more specific answer or a different kind of greeting. Alternatively, they might be experiencing technical issues where the messages aren't being properly received. I should consider the user's intent. If they're just saying hello, perhaps they want to continue the conversation. The assistant's role is to be helpful and engaging. Maybe the user is looking for a more personalized response. However, since the user hasn't provided any specific questions or topics, the assistant should keep the response open and inviting. It's also possible that the user is using a different device or platform where the previous interactions aren't showing up correctly. In any case, the assistant should respond politely and encourage the user to ask a question or share what they need help with. Repeating the same response might not be effective, so maybe varying the greeting slightly could help. But since the user has only said "hello" three times, it's possible they're just testing the response. The assistant should remain consistent and friendly, ensuring the user feels welcomed and supported. ...done thinking. Hello again! 😊 How can I assist you today? Let me know if you have any questions or need help with something specific! ```
GiteaMirror commented 2026-04-22 15:11:57 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Sep 25, 2025):

import ollama

response = ollama.chat(
  model="qwen3:30b-a3b-q4_K_M",
  messages=[{"role":"user","content":"hello"}],
  think=False)

print(response)
<!-- gh-comment-id:3333586017 --> @rick-github commented on GitHub (Sep 25, 2025): ```python import ollama response = ollama.chat( model="qwen3:30b-a3b-q4_K_M", messages=[{"role":"user","content":"hello"}], think=False) print(response) ```
GiteaMirror commented 2026-04-22 15:11:58 -05:00
Author
Owner
Copy Link

@gowrijsuria commented on GitHub (Oct 12, 2025):

But qwen3:30b isn't still working with the below code.

import ollama

response = ollama.chat(
  model="qwen3:30b",
  messages=[{"role":"user","content":"hello"}],
  think=False)

print(response)
<!-- gh-comment-id:3393866432 --> @gowrijsuria commented on GitHub (Oct 12, 2025): But qwen3:30b isn't still working with the below code. ``` import ollama response = ollama.chat( model="qwen3:30b", messages=[{"role":"user","content":"hello"}], think=False) print(response) ```
GiteaMirror commented 2026-04-22 15:11:59 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Oct 12, 2025):

qwen3:30b is an alias for qwen3:30b-a3b-thinking-2507-q4_K_M which is a thinking-only model - thinking cannot be disabled. If you want a non-thinking model, use qwen3:30b-a3b-instruct-2507-q4_K_M. If you want to use a hybrid model that allows control of thinking, use qwen3:30b-a3b-q4_K_M.

<!-- gh-comment-id:3394316977 --> @rick-github commented on GitHub (Oct 12, 2025): qwen3:30b is an alias for qwen3:30b-a3b-thinking-2507-q4_K_M which is a thinking-only model - thinking cannot be disabled. If you want a non-thinking model, use qwen3:30b-a3b-instruct-2507-q4_K_M. If you want to use a hybrid model that allows control of thinking, use qwen3:30b-a3b-q4_K_M.
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 bug
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#33036
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?