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 #12917] qwen3:4b: Can't turn off thinking #70623

New Issue
Closed
opened 2026-05-04 22:18:20 -05:00 by GiteaMirror · 2 comments
GiteaMirror commented 2026-05-04 22:18:20 -05:00
Owner
Copy Link

Originally created by @xy3xy3 on GitHub (Nov 3, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/12917

What is the issue?

According to https://docs.ollama.com/capabilities/thinking#thinking.
I try to disable thinking for qwen3:4b.However, I failed.

Relevant log output

both of the 2 ways from api were failed.


(AgentTest) root@GEM12PRO:/proj/AgentTest# curl http://localhost:11434/api/chat -d '{
  "model": "qwen3:4b",
  "messages": [{
    "role": "user",
    "content": "What is the answer of 1 + 1? /no_think"
  }],
  "stream": false
}'
{"model":"qwen3:4b","created_at":"2025-11-03T05:17:48.869018648Z","message":{"role":"assistant","content":"2","thinking":"Okay, the user asked \"What is the answer of 1 + 1? /no_think\". Hmm, interesting. First, I need to unpack this. The \"/no_think\" part is throwing me off—usually that's a command in some contexts, like in programming or chat interfaces to skip thinking. But the user seems to be testing something here.\n\nLet me think about the math part first. 1 + 1 is a basic arithmetic problem. In standard math, that's 2. But why the \"/no_think\" tag? Maybe they're trying to see if I'll overcomplicate it or if I'll just give the straightforward answer without overthinking. The instruction says \"/no_think\", which might mean they don't want any reasoning—just the answer. \n\nWait, in some contexts, like in AI responses, \"/no_think\" could be a way to suppress the thought process. So the user probably expects me to output the answer without any extra explanation. But I should check if there's a trick here. For example, in binary, 1 + 1 is 10 (which is 2), but that's still 2 in decimal. Or maybe in set theory? No, 1 + 1 is still 2. \n\nThe user might be referencing a meme or a specific context where \"1 + 1\" has a different meaning. Like in some jokes or pop culture? But the simplest answer is 2. Since they said \"/no_think\", I should ignore any complex explanations and just state 2. \n\nAlso, the user might be testing if the AI follows instructions precisely. They want the answer without thinking, so I shouldn't add any reasoning. Let me confirm: the problem is 1 + 1, which is 2. The \"/no_think\" is probably a command to skip the thought process. So the answer is 2.\n"},"done":true,"done_reason":"stop","total_duration":2901768411,"load_duration":110199669,"prompt_eval_count":25,"prompt_eval_duration":20803618,"eval_count":398,"eval_duration":2659200603}(AgentTest) root@GEM12P(AgentTest) root@GEM12PRO:/proj/AgentTest# curl http://localhost:11434/api/chat -d '{
  "model": "qwen3:4b",
  "messages": [{
    "role": "user",
    "content": "What is the answer of 1 + 1?"
  }],
  "think": false,
  "stream": false
}'
{"model":"qwen3:4b","created_at":"2025-11-03T05:18:40.816714332Z","message":{"role":"assistant","content":"Okay, the user is asking \"What is the answer of 1 + 1?\" Hmm, this seems really basic but I should be careful. \n\nFirst thought: In elementary math, 1 + 1 is obviously 2. But maybe the user is testing me? Or is there some context I'm missing? \n\nLet me consider possible angles:\n- Basic arithmetic: Definitely 2. \n- Computer science: In binary, 1 + 1 is 10 (which is 2 in decimal). But the question says \"1 + 1\" without specifying base, so probably decimal.\n- Philosophy? Like \"what is the answer to life\" but no, the question is literal.\n- Wordplay? \"1\" as in \"one\" but that's redundant.\n\nWait, the user might be a kid learning math? Or someone checking if I'm smart? The phrasing \"the answer\" is a bit odd though - usually we say \"what is 1 + 1\". \n\n*checks mental notes* \nIn all standard mathematical contexts (decimal, binary, integer arithmetic), 1 + 1 = 2. Even in set theory, the union of two singletons is size 2. \n\nShould I mention binary just in case? But the question doesn't specify base... most people would expect decimal. \n\nUser seems to be in a hurry or testing me - the question is so simple. Maybe they're bored? Or a non-native English speaker? \n\n*decision*: I'll go with the straightforward answer: 2. But I'll add a tiny bit about binary in case they're curious, while emphasizing decimal is default. \n\nAlso... is there any cultural context where this is different? Like in some languages? Probably not. \n\nFinal answer structure:\n1. State clearly: 2\n2. Briefly explain why (basic addition)\n3. Optional: mention binary as interesting but not relevant here\n4. Keep it friendly since the question seems simple\n\n*mental note*: Don't overcomplicate it. User probably just wants confirmation they're not crazy for asking such a basic thing.\n\u003c/think\u003e\n\nThe answer to **1 + 1** in standard arithmetic (base 10, which is the most common context) is **2**.  \n\n### Why?\n- In basic mathematics, adding two units (1 + 1) results in **2**.\n- This holds true in all standard number systems (e.g., decimal, binary), but **base 10 is the default** unless specified otherwise.\n\n### Quick clarification:\n- **Binary (base 2)**: 1 + 1 = **10** (which equals 2 in decimal).  \n  *(But since your question didn’t specify binary, we use base 10 here).*\n- **Real-world context**: If you have 1 apple and add another apple, you have **2 apples**.\n\nIf you meant something else (e.g., a trick question, programming, or philosophy), feel free to clarify! For now, **1 + 1 = 2** is the straightforward answer. 😊"},"done":true,"done_reason":"stop","total_duration":4692553237,"load_duration":93419349,"prompt_eval_count":21,"prompt_eval_duration":17713872,"eval_count":639,"eval_duration":4403562275}(AgentTest) root@GEM12PRO:/proj/AgentTest# 


`/set` is also failed.


ollama run qwen3:4b
>>> /set nothink
Set 'nothink' mode.
>>> 1+1=?
Okay, the user asked "1+1=?" Let me think about how to approach this.

First, I should consider the context. The question is very simple, but maybe they want to know the basic arithmetic answer. In 
standard math, 1 plus 1 equals 2. But sometimes people might have different contexts, like in binary where 1+1 is 10 (which is 2 in 
decimal). Wait, the user didn't specify the base. Let me check.

The question is written as "1+1=?" without any context. In most cases, especially in elementary math, this refers to base 10. So 
the answer should be 2. But maybe they're testing for something else? Like in programming, but the question is in plain English.

Wait, maybe the user is a child learning addition, so the straightforward answer is 2. Let me confirm. In base 10, 1+1=2. In base 
2, it's 10, but the question doesn't mention base 2. So the answer is 2.

I should make sure there's no trick here. Sometimes people might think of other operations, but the question is "1+1", which is 
standard addition. So the answer is 2. I think that's it.
</think>

The answer to $1 + 1$ in standard base-10 arithmetic is **2**. 

This is the most straightforward interpretation unless specified otherwise (e.g., binary systems use different rules, but the 
question doesn’t indicate that context). 

**Answer:** 2

OS

WSL2

GPU

Nvidia

CPU

AMD

Ollama version

0.12.9

Originally created by @xy3xy3 on GitHub (Nov 3, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/12917 ### What is the issue? According to https://docs.ollama.com/capabilities/thinking#thinking. I try to disable thinking for qwen3:4b.However, I failed. ### Relevant log output ```shell both of the 2 ways from api were failed. (AgentTest) root@GEM12PRO:/proj/AgentTest# curl http://localhost:11434/api/chat -d '{ "model": "qwen3:4b", "messages": [{ "role": "user", "content": "What is the answer of 1 + 1? /no_think" }], "stream": false }' {"model":"qwen3:4b","created_at":"2025-11-03T05:17:48.869018648Z","message":{"role":"assistant","content":"2","thinking":"Okay, the user asked \"What is the answer of 1 + 1? /no_think\". Hmm, interesting. First, I need to unpack this. The \"/no_think\" part is throwing me off—usually that's a command in some contexts, like in programming or chat interfaces to skip thinking. But the user seems to be testing something here.\n\nLet me think about the math part first. 1 + 1 is a basic arithmetic problem. In standard math, that's 2. But why the \"/no_think\" tag? Maybe they're trying to see if I'll overcomplicate it or if I'll just give the straightforward answer without overthinking. The instruction says \"/no_think\", which might mean they don't want any reasoning—just the answer. \n\nWait, in some contexts, like in AI responses, \"/no_think\" could be a way to suppress the thought process. So the user probably expects me to output the answer without any extra explanation. But I should check if there's a trick here. For example, in binary, 1 + 1 is 10 (which is 2), but that's still 2 in decimal. Or maybe in set theory? No, 1 + 1 is still 2. \n\nThe user might be referencing a meme or a specific context where \"1 + 1\" has a different meaning. Like in some jokes or pop culture? But the simplest answer is 2. Since they said \"/no_think\", I should ignore any complex explanations and just state 2. \n\nAlso, the user might be testing if the AI follows instructions precisely. They want the answer without thinking, so I shouldn't add any reasoning. Let me confirm: the problem is 1 + 1, which is 2. The \"/no_think\" is probably a command to skip the thought process. So the answer is 2.\n"},"done":true,"done_reason":"stop","total_duration":2901768411,"load_duration":110199669,"prompt_eval_count":25,"prompt_eval_duration":20803618,"eval_count":398,"eval_duration":2659200603}(AgentTest) root@GEM12P(AgentTest) root@GEM12PRO:/proj/AgentTest# curl http://localhost:11434/api/chat -d '{ "model": "qwen3:4b", "messages": [{ "role": "user", "content": "What is the answer of 1 + 1?" }], "think": false, "stream": false }' {"model":"qwen3:4b","created_at":"2025-11-03T05:18:40.816714332Z","message":{"role":"assistant","content":"Okay, the user is asking \"What is the answer of 1 + 1?\" Hmm, this seems really basic but I should be careful. \n\nFirst thought: In elementary math, 1 + 1 is obviously 2. But maybe the user is testing me? Or is there some context I'm missing? \n\nLet me consider possible angles:\n- Basic arithmetic: Definitely 2. \n- Computer science: In binary, 1 + 1 is 10 (which is 2 in decimal). But the question says \"1 + 1\" without specifying base, so probably decimal.\n- Philosophy? Like \"what is the answer to life\" but no, the question is literal.\n- Wordplay? \"1\" as in \"one\" but that's redundant.\n\nWait, the user might be a kid learning math? Or someone checking if I'm smart? The phrasing \"the answer\" is a bit odd though - usually we say \"what is 1 + 1\". \n\n*checks mental notes* \nIn all standard mathematical contexts (decimal, binary, integer arithmetic), 1 + 1 = 2. Even in set theory, the union of two singletons is size 2. \n\nShould I mention binary just in case? But the question doesn't specify base... most people would expect decimal. \n\nUser seems to be in a hurry or testing me - the question is so simple. Maybe they're bored? Or a non-native English speaker? \n\n*decision*: I'll go with the straightforward answer: 2. But I'll add a tiny bit about binary in case they're curious, while emphasizing decimal is default. \n\nAlso... is there any cultural context where this is different? Like in some languages? Probably not. \n\nFinal answer structure:\n1. State clearly: 2\n2. Briefly explain why (basic addition)\n3. Optional: mention binary as interesting but not relevant here\n4. Keep it friendly since the question seems simple\n\n*mental note*: Don't overcomplicate it. User probably just wants confirmation they're not crazy for asking such a basic thing.\n\u003c/think\u003e\n\nThe answer to **1 + 1** in standard arithmetic (base 10, which is the most common context) is **2**. \n\n### Why?\n- In basic mathematics, adding two units (1 + 1) results in **2**.\n- This holds true in all standard number systems (e.g., decimal, binary), but **base 10 is the default** unless specified otherwise.\n\n### Quick clarification:\n- **Binary (base 2)**: 1 + 1 = **10** (which equals 2 in decimal). \n *(But since your question didn’t specify binary, we use base 10 here).*\n- **Real-world context**: If you have 1 apple and add another apple, you have **2 apples**.\n\nIf you meant something else (e.g., a trick question, programming, or philosophy), feel free to clarify! For now, **1 + 1 = 2** is the straightforward answer. 😊"},"done":true,"done_reason":"stop","total_duration":4692553237,"load_duration":93419349,"prompt_eval_count":21,"prompt_eval_duration":17713872,"eval_count":639,"eval_duration":4403562275}(AgentTest) root@GEM12PRO:/proj/AgentTest# `/set` is also failed. ollama run qwen3:4b >>> /set nothink Set 'nothink' mode. >>> 1+1=? Okay, the user asked "1+1=?" Let me think about how to approach this. First, I should consider the context. The question is very simple, but maybe they want to know the basic arithmetic answer. In standard math, 1 plus 1 equals 2. But sometimes people might have different contexts, like in binary where 1+1 is 10 (which is 2 in decimal). Wait, the user didn't specify the base. Let me check. The question is written as "1+1=?" without any context. In most cases, especially in elementary math, this refers to base 10. So the answer should be 2. But maybe they're testing for something else? Like in programming, but the question is in plain English. Wait, maybe the user is a child learning addition, so the straightforward answer is 2. Let me confirm. In base 10, 1+1=2. In base 2, it's 10, but the question doesn't mention base 2. So the answer is 2. I should make sure there's no trick here. Sometimes people might think of other operations, but the question is "1+1", which is standard addition. So the answer is 2. I think that's it. </think> The answer to $1 + 1$ in standard base-10 arithmetic is **2**. This is the most straightforward interpretation unless specified otherwise (e.g., binary systems use different rules, but the question doesn’t indicate that context). **Answer:** 2 ``` ### OS WSL2 ### GPU Nvidia ### CPU AMD ### Ollama version 0.12.9
GiteaMirror added the bug label 2026-05-04 22:18:20 -05:00
GiteaMirror commented 2026-05-04 22:18:23 -05:00
Author
Owner
Copy Link

@xy3xy3 commented on GitHub (Nov 3, 2025):

The 4b 30b and 235b for qwen3 had a 2507 update which split the models into thinking and instruct variants. The default on ollama is now thinking, you can click "view all" on the model page and you will find all the models including the instruct variants. So for your usage ollama run qwen3:4b-instruct-2507-q4_K_M
https://ollama.com/library/qwen3/tags
This page has all the available models.

<!-- gh-comment-id:3478963867 --> @xy3xy3 commented on GitHub (Nov 3, 2025): The 4b 30b and 235b for qwen3 had a 2507 update which split the models into thinking and instruct variants. The default on ollama is now thinking, you can click "view all" on the model page and you will find all the models including the instruct variants. So for your usage ollama run qwen3:4b-instruct-2507-q4_K_M https://ollama.com/library/qwen3/tags This page has all the available models.
GiteaMirror commented 2026-05-04 22:18:24 -05:00
Author
Owner
Copy Link

@xy3xy3 commented on GitHub (Nov 3, 2025):

Now I use qwen3:4b-instruct and success.

curl http://localhost:11434/api/chat -d '{
  "model": "qwen3:4b-instruct",
  "messages": [{
    "role": "user",
    "content": "What is the answer of 1 + 1?"
  }],
  "stream": false
}'
{"model":"qwen3:4b-instruct","created_at":"2025-11-03T05:33:47.149505368Z","message":{"role":"assistant","content":"The answer to 1 + 1 is **2**."},"done":true,"done_reason":"stop","total_duration":3241137160,"load_duration":3106878899,"prompt_eval_count":19,"prompt_eval_duration":18088408,"eval_count":14,"eval_duration":94230297}
<!-- gh-comment-id:3478965134 --> @xy3xy3 commented on GitHub (Nov 3, 2025): Now I use qwen3:4b-instruct and success. ``` curl http://localhost:11434/api/chat -d '{ "model": "qwen3:4b-instruct", "messages": [{ "role": "user", "content": "What is the answer of 1 + 1?" }], "stream": false }' {"model":"qwen3:4b-instruct","created_at":"2025-11-03T05:33:47.149505368Z","message":{"role":"assistant","content":"The answer to 1 + 1 is **2**."},"done":true,"done_reason":"stop","total_duration":3241137160,"load_duration":3106878899,"prompt_eval_count":19,"prompt_eval_duration":18088408,"eval_count":14,"eval_duration":94230297} ```
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#70623
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?