github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-06 16:11:34 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #14502] Qwen3.5 instruct (Non-Thinking) versions #35168

New Issue
Closed
opened 2026-04-22 19:28:47 -05:00 by GiteaMirror · 14 comments
GiteaMirror commented 2026-04-22 19:28:47 -05:00
Owner
Copy Link

Originally created by @eavelardev on GitHub (Feb 27, 2026).
Original GitHub issue: https://github.com/ollama/ollama/issues/14502

I have been playing with Qwen3.5 model and for simple prompts it takes too much time compared with Qwen3 instruct.

I would like to know how the configuration for Qwen3.5 should be as instruct model, I don't even know if Ollama server support this kind of config for Qwen3.5

chat_response = client.chat.completions.create(
    model="qwen3.5:35b",
    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},
    }, 
)

Instruct versions for Qwen3.5 will help to do a proper comparison against Qwen3 and Qwen3-VL models.

Originally created by @eavelardev on GitHub (Feb 27, 2026). Original GitHub issue: https://github.com/ollama/ollama/issues/14502 I have been playing with Qwen3.5 model and for simple prompts it takes too much time compared with Qwen3 instruct. I would like to know how the configuration for Qwen3.5 should be as instruct model, I don't even know if Ollama server support this kind of config for Qwen3.5 ```py chat_response = client.chat.completions.create( model="qwen3.5:35b", 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}, }, ) ``` Instruct versions for Qwen3.5 will help to do a proper comparison against Qwen3 and Qwen3-VL models.
GiteaMirror commented 2026-04-22 19:28:50 -05:00
Author
Owner
Copy Link

@Da-coding-pro commented on GitHub (Feb 27, 2026):

if you use open-web-ui or the terminal, you could be able to disable thinking because I am pretty sure qwen3.5 has a toggle

<!-- gh-comment-id:3975471623 --> @Da-coding-pro commented on GitHub (Feb 27, 2026): if you use open-web-ui or the terminal, you could be able to disable thinking because I am pretty sure qwen3.5 has a toggle
GiteaMirror commented 2026-04-22 19:28:51 -05:00
Author
Owner
Copy Link

@Igorgro commented on GitHub (Feb 28, 2026):

I tried to disable thinking in open webui using built-in think parameter and adding "enable_thinking": false, but it resulted in weird output:

Image
<!-- gh-comment-id:3977250458 --> @Igorgro commented on GitHub (Feb 28, 2026): I tried to disable thinking in open webui using built-in think parameter and adding "enable_thinking": false, but it resulted in weird output: <img width="2128" height="322" alt="Image" src="https://github.com/user-attachments/assets/922efb05-0d73-4657-a7f1-5517f5bf62af" />
GiteaMirror commented 2026-04-22 19:28:52 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Mar 1, 2026):

$ diff -u 14502.py.orig 14502.py
--- 14502.py.orig       2026-03-01 23:36:21.000416055 +0100
+++ 14502.py    2026-03-01 23:37:01.237286715 +0100
@@ -18,10 +18,7 @@
     temperature=0.7,
     top_p=0.8,
     presence_penalty=1.5,
-    extra_body={
-        "top_k": 20,
-        "chat_template_kwargs": {"enable_thinking": False},
-    }, 
+    reasoning_effort="none",
 )
 
 print(chat_response)
<!-- gh-comment-id:3981193000 --> @rick-github commented on GitHub (Mar 1, 2026): ```diff $ diff -u 14502.py.orig 14502.py --- 14502.py.orig 2026-03-01 23:36:21.000416055 +0100 +++ 14502.py 2026-03-01 23:37:01.237286715 +0100 @@ -18,10 +18,7 @@ temperature=0.7, top_p=0.8, presence_penalty=1.5, - extra_body={ - "top_k": 20, - "chat_template_kwargs": {"enable_thinking": False}, - }, + reasoning_effort="none", ) print(chat_response) ```
GiteaMirror commented 2026-04-22 19:28:53 -05:00
Author
Owner
Copy Link

@starswasps commented on GitHub (Mar 2, 2026):

reasoning_effort="none", doesn't solve a problem. Ollama just stuck and even don't load a model. There are no both: error and response.

from openai import OpenAI

client = OpenAI(
    base_url="http://127.0.0.1:11433/v1",
    api_key="ollama",
)

messages = [
    {"role": "user", "content": "Hello world!"}
]

chat_response = client.chat.completions.create(
    model="qwen3.5:2b",
    messages=messages,
    max_tokens=32768,
    temperature=0.7,
    top_p=0.8,
    presence_penalty=1.5,
    reasoning_effort="none",
)

print(chat_response.choices[0].message.content)
<!-- gh-comment-id:3985641791 --> @starswasps commented on GitHub (Mar 2, 2026): `reasoning_effort="none",` doesn't solve a problem. Ollama just stuck and even don't load a model. There are no both: error and response. ```python from openai import OpenAI client = OpenAI( base_url="http://127.0.0.1:11433/v1", api_key="ollama", ) messages = [ {"role": "user", "content": "Hello world!"} ] chat_response = client.chat.completions.create( model="qwen3.5:2b", messages=messages, max_tokens=32768, temperature=0.7, top_p=0.8, presence_penalty=1.5, reasoning_effort="none", ) print(chat_response.choices[0].message.content) ```
GiteaMirror commented 2026-04-22 19:28:54 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Mar 2, 2026):

$ python3 14502-1.py
Hello! How can I help you today?
<!-- gh-comment-id:3985690439 --> @rick-github commented on GitHub (Mar 2, 2026): ```console $ python3 14502-1.py Hello! How can I help you today? ```
GiteaMirror commented 2026-04-22 19:28:56 -05:00
Author
Owner
Copy Link

@zuohaoda commented on GitHub (Mar 4, 2026):

Has this issue been resolved?

<!-- gh-comment-id:3996982260 --> @zuohaoda commented on GitHub (Mar 4, 2026): Has this issue been resolved?
GiteaMirror commented 2026-04-22 19:28:59 -05:00
Author
Owner
Copy Link

@hcaihao commented on GitHub (Mar 5, 2026):

Has this issue been resolved?

<!-- gh-comment-id:4002118752 --> @hcaihao commented on GitHub (Mar 5, 2026): Has this issue been resolved?
GiteaMirror commented 2026-04-22 19:29:00 -05:00
Author
Owner
Copy Link

@MaigcWind commented on GitHub (Mar 5, 2026):

Hope this can be add in modelfile PARAMETER

<!-- gh-comment-id:4002416340 --> @MaigcWind commented on GitHub (Mar 5, 2026): Hope this can be add in modelfile PARAMETER
GiteaMirror commented 2026-04-22 19:29:01 -05:00
Author
Owner
Copy Link

@eavelardev commented on GitHub (Mar 5, 2026):

Hope this can be add in modelfile PARAMETER

same question

reasoning_effort="none" works but it seems is not a modelfile PARAMETER

It could be a feature to consider

<!-- gh-comment-id:4003121089 --> @eavelardev commented on GitHub (Mar 5, 2026): > Hope this can be add in modelfile PARAMETER same question `reasoning_effort="none"` works but it seems is not a modelfile PARAMETER It could be a feature to consider
GiteaMirror commented 2026-04-22 19:29:02 -05:00
Author
Owner
Copy Link

@ronhuafeng commented on GitHub (Mar 7, 2026):

same question and confirmed reasoning_effort="none" works on OpenRouter.

<!-- gh-comment-id:4016959038 --> @ronhuafeng commented on GitHub (Mar 7, 2026): same question and confirmed `reasoning_effort="none"` works on OpenRouter.
GiteaMirror commented 2026-04-22 19:29:03 -05:00
Author
Owner
Copy Link

@souhailchenaoua commented on GitHub (Mar 8, 2026):

Same issue here. Running qwen3.5:9b on an RTX 4060 laptop, using it for vision/OCR tasks (PDF pages rendered as PNG).
The problem: all output goes into the thinking field and content is always empty when images are involved. Tried everything:

"thinking": False in options — ignored for image inputs
/no_think in prompt — treated as literal text by the model
Recommended VL sampling params (temperature=0.7, top_p=0.80, top_k=20, presence_penalty=1.5) — no effect

For comparison, qwen3-vl:8b-instruct works perfectly for the same task — output always lands in content as expected. So the instruct variant pattern clearly works.
A qwen3.5:9b-instruct tag or proper thinking disable support for vision inputs would be a huge help. This model is otherwise unusable for any image-based task through Ollama right now.

<!-- gh-comment-id:4018876358 --> @souhailchenaoua commented on GitHub (Mar 8, 2026): Same issue here. Running qwen3.5:9b on an RTX 4060 laptop, using it for vision/OCR tasks (PDF pages rendered as PNG). The problem: all output goes into the thinking field and content is always empty when images are involved. Tried everything: "thinking": False in options — ignored for image inputs /no_think in prompt — treated as literal text by the model Recommended VL sampling params (temperature=0.7, top_p=0.80, top_k=20, presence_penalty=1.5) — no effect For comparison, qwen3-vl:8b-instruct works perfectly for the same task — output always lands in content as expected. So the instruct variant pattern clearly works. A qwen3.5:9b-instruct tag or proper thinking disable support for vision inputs would be a huge help. This model is otherwise unusable for any image-based task through Ollama right now.
GiteaMirror commented 2026-04-22 19:29:04 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Mar 8, 2026):

$ curl -s localhost:11434/api/chat -d '{"model":"qwen3.5:9b","stream":false,"messages":[{"role":"user","content":"describe this image","images":["'$(base64 -w0 image1.jpg)'"]}]}' | jq
{
  "model": "qwen3.5:9b",
  "created_at": "2026-03-08T12:26:57.265339279Z",
  "message": {
    "role": "assistant",
    "content": "Based on the image, here is a description:\n\nThe image features a small, adorable white puppy sitting on what appears to be weathered stone steps or a patio. The puppy has thick, fluffy fur and is facing slightly to the right, gazing off into the distance with dark eyes and a tiny black nose. It is wearing a red collar with a gold bell attached to it. The background is blurred, showing horizontal lines that suggest steps leading up to a darker area, keeping the focus entirely on the cute animal.",
    "thinking": "The user wants a description of the provided image.\n\n1.  **Subject:** The main subject is a small, fluffy white puppy. It looks like it could be a Chow Chow, Pomeranian mix, or similar spitz-type breed.\n2.  **Action/Pose:** The puppy is sitting down on its hind legs. Its front paws are planted on the ground. It is facing slightly to the right of the camera.\n3.  **Appearance/Details:**\n    *   **Fur:** Very fluffy, thick, snow-white.\n    *   **Face:** Dark eyes (black/dark brown), small black nose, short white ears that are semi-erect but soft-looking. It has a tiny bit of whisker stubble visible near the mouth.\n    *   **Accessories:** It is wearing a red collar with a gold bell attached to it.\n4.  **Setting/Background:**\n    *   **Ground:** The puppy is sitting on a light-colored, textured surface, possibly stone steps or a patio. The texture looks like weathered concrete or stone with some discoloration.\n    *   **Background:** The background is blurred (shallow depth of field). It shows horizontal lines suggesting steps or ledges behind the puppy. The colors are muted browns and dark tones, indicating shadows or a darker structure further back.\n5.  **Lighting/Mood:** The lighting is soft and natural, likely daylight. The mood is cute, innocent, and peaceful.\n\n**Drafting the description:**\n*   Start with the main subject: A fluffy white puppy sitting on steps.\n*   Describe physical features: Small black nose, dark eyes, red collar with a gold bell.\n*   Describe the pose: Sitting upright, looking off to the side.\n*   Describe the environment: Stone or concrete surface, blurred background.\n\n**Refining the description:**\n\"A heartwarming image features a small, fluffy white puppy sitting on what appears to be stone steps or a patio. The puppy is facing slightly to the right, gazing off-camera with dark, alert eyes and a tiny black nose. Its fur is thick and snowy white, giving it a very soft appearance. Around its neck, it wears a bright red collar with a small gold bell attached. The background is out of focus, showing hints of a brownish structure behind it, which keeps the attention firmly on the adorable animal.\"\n\nThis covers all the key elements. I will present this in a clear, descriptive paragraph."
  },
  "done": true,
  "done_reason": "stop",
  "total_duration": 5473258729,
  "load_duration": 692853672,
  "prompt_eval_count": 202,
  "prompt_eval_duration": 48221143,
  "eval_count": 633,
  "eval_duration": 3905508481
}


$ curl -s localhost:11434/api/chat -d '{"model":"qwen3.5:9b","think":false,"stream":false,"messages":[{"role":"user","content":"describe this image","images":["'$(base64 -w0 image1.jpg)'"]}]}' | jq
{
  "model": "qwen3.5:9b",
  "created_at": "2026-03-08T12:26:10.236134142Z",
  "message": {
    "role": "assistant",
    "content": "This is a heartwarming, close-up photograph of an adorable white puppy sitting on what appears to be a stone or concrete step outdoors.\n\n**Subject:**\n- The puppy is small, fluffy, and predominantly pure white with soft, slightly tousled fur.\n- It has dark, expressive eyes and a tiny black nose.\n- Its ears are floppy but upright at the base, typical of many young puppies.\n- A delicate red collar sits snugly around its neck, adorned with a shiny gold-colored bell — adding a touch of charm and suggesting it’s a cherished pet.\n- The puppy is sitting upright, looking off to the right side of the frame with a calm, curious expression.\n\n**Setting & Background:**\n- The puppy is seated on a light gray, textured stone surface — possibly steps or a patio.\n- The background is softly blurred (shallow depth of field), showing warm brown tones that suggest wooden flooring or furniture indoors just behind the step. This separation keeps focus entirely on the puppy.\n- Natural lighting highlights the fluffiness of its coat and casts gentle shadows, enhancing the three-dimensional feel.\n\n**Mood & Composition:**\n- The image exudes innocence, warmth, and tenderness.\n- The composition centers the puppy perfectly, drawing the viewer’s eye immediately to its sweet gaze and innocent posture.\n- The contrast between the bright white fur and muted background makes the subject pop beautifully.\n\nOverall, this is a classic “cute animal” photo that captures a fleeting moment of pure charm — perfect for evoking smiles or怜爱 in viewers."
  },
  "done": true,
  "done_reason": "stop",
  "total_duration": 3165840139,
  "load_duration": 647152477,
  "prompt_eval_count": 204,
  "prompt_eval_duration": 49787074,
  "eval_count": 327,
  "eval_duration": 2018546555
}
<!-- gh-comment-id:4018960989 --> @rick-github commented on GitHub (Mar 8, 2026): ```console $ curl -s localhost:11434/api/chat -d '{"model":"qwen3.5:9b","stream":false,"messages":[{"role":"user","content":"describe this image","images":["'$(base64 -w0 image1.jpg)'"]}]}' | jq { "model": "qwen3.5:9b", "created_at": "2026-03-08T12:26:57.265339279Z", "message": { "role": "assistant", "content": "Based on the image, here is a description:\n\nThe image features a small, adorable white puppy sitting on what appears to be weathered stone steps or a patio. The puppy has thick, fluffy fur and is facing slightly to the right, gazing off into the distance with dark eyes and a tiny black nose. It is wearing a red collar with a gold bell attached to it. The background is blurred, showing horizontal lines that suggest steps leading up to a darker area, keeping the focus entirely on the cute animal.", "thinking": "The user wants a description of the provided image.\n\n1. **Subject:** The main subject is a small, fluffy white puppy. It looks like it could be a Chow Chow, Pomeranian mix, or similar spitz-type breed.\n2. **Action/Pose:** The puppy is sitting down on its hind legs. Its front paws are planted on the ground. It is facing slightly to the right of the camera.\n3. **Appearance/Details:**\n * **Fur:** Very fluffy, thick, snow-white.\n * **Face:** Dark eyes (black/dark brown), small black nose, short white ears that are semi-erect but soft-looking. It has a tiny bit of whisker stubble visible near the mouth.\n * **Accessories:** It is wearing a red collar with a gold bell attached to it.\n4. **Setting/Background:**\n * **Ground:** The puppy is sitting on a light-colored, textured surface, possibly stone steps or a patio. The texture looks like weathered concrete or stone with some discoloration.\n * **Background:** The background is blurred (shallow depth of field). It shows horizontal lines suggesting steps or ledges behind the puppy. The colors are muted browns and dark tones, indicating shadows or a darker structure further back.\n5. **Lighting/Mood:** The lighting is soft and natural, likely daylight. The mood is cute, innocent, and peaceful.\n\n**Drafting the description:**\n* Start with the main subject: A fluffy white puppy sitting on steps.\n* Describe physical features: Small black nose, dark eyes, red collar with a gold bell.\n* Describe the pose: Sitting upright, looking off to the side.\n* Describe the environment: Stone or concrete surface, blurred background.\n\n**Refining the description:**\n\"A heartwarming image features a small, fluffy white puppy sitting on what appears to be stone steps or a patio. The puppy is facing slightly to the right, gazing off-camera with dark, alert eyes and a tiny black nose. Its fur is thick and snowy white, giving it a very soft appearance. Around its neck, it wears a bright red collar with a small gold bell attached. The background is out of focus, showing hints of a brownish structure behind it, which keeps the attention firmly on the adorable animal.\"\n\nThis covers all the key elements. I will present this in a clear, descriptive paragraph." }, "done": true, "done_reason": "stop", "total_duration": 5473258729, "load_duration": 692853672, "prompt_eval_count": 202, "prompt_eval_duration": 48221143, "eval_count": 633, "eval_duration": 3905508481 } ``` ```console $ curl -s localhost:11434/api/chat -d '{"model":"qwen3.5:9b","think":false,"stream":false,"messages":[{"role":"user","content":"describe this image","images":["'$(base64 -w0 image1.jpg)'"]}]}' | jq { "model": "qwen3.5:9b", "created_at": "2026-03-08T12:26:10.236134142Z", "message": { "role": "assistant", "content": "This is a heartwarming, close-up photograph of an adorable white puppy sitting on what appears to be a stone or concrete step outdoors.\n\n**Subject:**\n- The puppy is small, fluffy, and predominantly pure white with soft, slightly tousled fur.\n- It has dark, expressive eyes and a tiny black nose.\n- Its ears are floppy but upright at the base, typical of many young puppies.\n- A delicate red collar sits snugly around its neck, adorned with a shiny gold-colored bell — adding a touch of charm and suggesting it’s a cherished pet.\n- The puppy is sitting upright, looking off to the right side of the frame with a calm, curious expression.\n\n**Setting & Background:**\n- The puppy is seated on a light gray, textured stone surface — possibly steps or a patio.\n- The background is softly blurred (shallow depth of field), showing warm brown tones that suggest wooden flooring or furniture indoors just behind the step. This separation keeps focus entirely on the puppy.\n- Natural lighting highlights the fluffiness of its coat and casts gentle shadows, enhancing the three-dimensional feel.\n\n**Mood & Composition:**\n- The image exudes innocence, warmth, and tenderness.\n- The composition centers the puppy perfectly, drawing the viewer’s eye immediately to its sweet gaze and innocent posture.\n- The contrast between the bright white fur and muted background makes the subject pop beautifully.\n\nOverall, this is a classic “cute animal” photo that captures a fleeting moment of pure charm — perfect for evoking smiles or怜爱 in viewers." }, "done": true, "done_reason": "stop", "total_duration": 3165840139, "load_duration": 647152477, "prompt_eval_count": 204, "prompt_eval_duration": 49787074, "eval_count": 327, "eval_duration": 2018546555 } ```
GiteaMirror commented 2026-04-22 19:29:05 -05:00
Author
Owner
Copy Link

@andreamoro commented on GitHub (Mar 12, 2026):

Related finding: the generate API completely ignores think: false in options for qwen3.5. The thinking field still populates and consumes the full num_predict budget, leaving response: "".

However, the chat API with think=False as a top-level parameter works correctly — thinking is disabled and the model produces actual output.

Minimal repro and details in #14793.

Workaround: use chat endpoint instead of generate, passing think=False at the top level (not inside options).

<!-- gh-comment-id:4045238106 --> @andreamoro commented on GitHub (Mar 12, 2026): Related finding: the `generate` API completely ignores `think: false` in `options` for qwen3.5. The `thinking` field still populates and consumes the full `num_predict` budget, leaving `response: ""`. However, the `chat` API with `think=False` as a **top-level parameter** works correctly — thinking is disabled and the model produces actual output. Minimal repro and details in #14793. **Workaround**: use `chat` endpoint instead of `generate`, passing `think=False` at the top level (not inside `options`).
GiteaMirror commented 2026-04-22 19:29:07 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Mar 12, 2026):

Workaround: use chat endpoint instead of generate, passing think=False at the top level (not inside options).

Or use think at the top level as documented.

<!-- gh-comment-id:4045431442 --> @rick-github commented on GitHub (Mar 12, 2026): > **Workaround**: use chat endpoint instead of generate, passing think=False at the top level (not inside options). Or use `think` at the top level as [documented](https://github.com/ollama/ollama/blob/main/docs/api.md#generate-a-completion:~:text=as%20llava%29-,think,-%3A%20%28for%20thinking%20models).
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
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#35168
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?