github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-06 08:02:14 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #10496] Can the empty <think> tags be removed in non-thinking mode for the Qwen3 series models? #6905

New Issue
Closed
opened 2026-04-12 18:47:29 -05:00 by GiteaMirror · 8 comments
GiteaMirror commented 2026-04-12 18:47:29 -05:00
Owner
Copy Link

Originally created by @somnifex on GitHub (Apr 30, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/10496

In the description of Qwen3, due to compatibility issues, even though the "think" mode is not used, the model's output still includes an empty tag. However, for Ollama, as a downstream application of the base model, removing this part of the tag might be more user-friendly for end users

Originally created by @somnifex on GitHub (Apr 30, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/10496 In the description of Qwen3, due to compatibility issues, even though the "think" mode is not used, the model's output still includes an empty tag. However, for Ollama, as a downstream application of the base model, removing this part of the tag might be more user-friendly for end users
GiteaMirror added the feature request label 2026-04-12 18:47:29 -05:00
GiteaMirror commented 2026-04-12 18:47:32 -05:00
Author
Owner
Copy Link

@ChenDianWzh commented on GitHub (Apr 30, 2025):

from langchain_ollama import ChatOllama
llm = ChatOllama(
model="qwen3:14b", temperature=0.6, base_url="",
)
print(llm.invoke("Hello, world!").content)

I'd like to know if there are any parameters that can be used to implement control thinking

<!-- gh-comment-id:2841048183 --> @ChenDianWzh commented on GitHub (Apr 30, 2025): from langchain_ollama import ChatOllama llm = ChatOllama( model="qwen3:14b", temperature=0.6, base_url="", ) print(llm.invoke("Hello, world!").content) I'd like to know if there are any parameters that can be used to implement control thinking
GiteaMirror commented 2026-04-12 18:47:33 -05:00
Author
Owner
Copy Link

@smileyboy2019 commented on GitHub (Apr 30, 2025):

自定义参数可以吗

<!-- gh-comment-id:2841403533 --> @smileyboy2019 commented on GitHub (Apr 30, 2025): 自定义参数可以吗
GiteaMirror commented 2026-04-12 18:47:33 -05:00
Author
Owner
Copy Link

@regmibijay commented on GitHub (May 1, 2025):

You can use /no_think either infront of system prompt or user query to bypass thinking meanwhile

<!-- gh-comment-id:2844670836 --> @regmibijay commented on GitHub (May 1, 2025): You can use `/no_think` either infront of system prompt or user query to bypass thinking meanwhile
GiteaMirror commented 2026-04-12 18:47:35 -05:00
Author
Owner
Copy Link

@lasseedfast commented on GitHub (May 1, 2025):

#10482

<!-- gh-comment-id:2844691892 --> @lasseedfast commented on GitHub (May 1, 2025): #10482
GiteaMirror commented 2026-04-12 18:47:35 -05:00
Author
Owner
Copy Link

@somnifex commented on GitHub (May 1, 2025):

You can use /no_think either infront of system prompt or user query to bypass thinking meanwhile

This is an effective method for eliminating the thought process, but the consequences of doing so are the core issue: even without utilizing a thinking mode, an empty <think> tag remains in the output. This tag exists within the model for compatibility purposes, but its presence can lead to unpredictable results in downstream applications – for example, inappropriate tags in a workflow could cause contextual anomalies, and certain text processing tasks may not yield directly usable answers. As an application layer built on foundational models, and potentially an upstream layer for user-facing applications, proper handling of these empty tags can simplify matters considerably.

<!-- gh-comment-id:2844697509 --> @somnifex commented on GitHub (May 1, 2025): > You can use `/no_think` either infront of system prompt or user query to bypass thinking meanwhile This is an effective method for eliminating the thought process, but the consequences of doing so are the core issue: even without utilizing a thinking mode, an empty `<think>` tag remains in the output. This tag exists within the model for compatibility purposes, but its presence can lead to unpredictable results in downstream applications – for example, inappropriate tags in a workflow could cause contextual anomalies, and certain text processing tasks may not yield directly usable answers. As an application layer built on foundational models, and potentially an upstream layer for user-facing applications, proper handling of these empty tags can simplify matters considerably.
GiteaMirror commented 2026-04-12 18:47:36 -05:00
Author
Owner
Copy Link

@190679163 commented on GitHub (May 14, 2025):


import openai
import re

client = openai.OpenAI(
    base_url="http://localhost:11434/v1",
    api_key="ollama",
)

def chat_with_ollama(query: str):
    response = client.chat.completions.create(
        model="qwen3:32b",
        messages=[{"role": "user", "content": query}],
        stream=True,
        temperature=0,
        seed=42
    )

    buffer = ""
    in_think = False  # Flag to check if inside <think> and </no_think>
    print("AI:", end=" ", flush=True)
    for chunk in response:
        if hasattr(chunk, "choices") and chunk.choices:
            delta = chunk.choices[0].delta
            if hasattr(delta, "content") and delta.content:
                content = delta.content

                # Check if entering <think> block
                if not in_think:
                    idx = content.find("<think>")
                    if idx != -1:
                        # Print content before <think>
                        print(content[:idx], end="", flush=True)
                        in_think = True
                        # Discard content after <think>
                        buffer = ""
                        continue
                    else:
                        print(content, end="", flush=True)
                
                if in_think:
                    # Check if </no_think> is encountered
                    end_idx = content.find("</think>")
                    if end_idx != -1:
                        # Exit <think> block
                        in_think = False
                        # Skip </no_think> and the content after it
                        after = content[end_idx + len("</think>"):]
                        # Remove leading \n\n
                        after = re.sub(r'^\n\n', '', after)
                        print(after, end="", flush=True)
                        continue
                    # Otherwise discard content inside <think>
                    continue
    print()  # Newline after the response

if __name__ == "__main__":
    chat_with_ollama("how to use ollama")


import subprocess
import json
import re

def chat_with_ollama(query: str):
    # 使用 curl 命令发起请求
    curl_command = [
        "curl",
        "-X", "POST",
        "http://localhost:11434/v1/chat/completions",
        "-H", "Content-Type: application/json",
        "-H", "Authorization: Bearer ollama",
        "-d", json.dumps({
            "model": "qwen3:32b",
            "messages": [{"role": "user", "content": query}],
            "stream": True,
            "temperature": 0,
            "seed": 42
        })
    ]

    # 使用 subprocess 执行 curl 命令
    process = subprocess.Popen(curl_command, stdout=subprocess.PIPE, stderr=subprocess.PIPE)

    buffer = ""
    in_think = False  # Flag to check if inside <think> and </no_think>
    print("AI:", end=" ", flush=True)

    for line in process.stdout:
        # Decode the line from the curl output
        decoded_line = line.decode("utf-8").strip()

        # Skip empty lines or irrelevant lines
        if not decoded_line or not decoded_line.startswith("data:"):
            continue

        # Remove the "data:" prefix
        decoded_line = decoded_line[5:].strip()

        # Parse the JSON payload
        try:
            chunk = json.loads(decoded_line)
        except json.JSONDecodeError:
            continue

        # Process the chunk content
        if "choices" in chunk and chunk["choices"]:
            delta = chunk["choices"][0].get("delta", {})
            if "content" in delta:
                content = delta["content"]

                # Check if entering <think> block
                if not in_think:
                    idx = content.find("<think>")
                    if idx != -1:
                        # Print content before <think>
                        print(content[:idx], end="", flush=True)
                        in_think = True
                        # Discard content after <think>
                        buffer = ""
                        continue
                    else:
                        print(content, end="", flush=True)

                if in_think:
                    # Check if </no_think> is encountered
                    end_idx = content.find("</think>")
                    if end_idx != -1:
                        # Exit <think> block
                        in_think = False
                        # Skip </no_think> and the content after it
                        after = content[end_idx + len("</think>"):]
                        # Remove leading \n\n
                        after = re.sub(r'^\n\n', '', after)
                        print(after, end="", flush=True)
                        continue
                    # Otherwise discard content inside <think>
                    continue

    print()  # Newline after the response

    # Wait for the process to complete and handle errors
    process.wait()
    if process.returncode != 0:
        stderr = process.stderr.read().decode("utf-8").strip()
        print(f"Error: {stderr}")

if __name__ == "__main__":
    chat_with_ollama("how to use ollama \\no_think")

This is my solution, See if this would help you

<!-- gh-comment-id:2879372596 --> @190679163 commented on GitHub (May 14, 2025): ``` import openai import re client = openai.OpenAI( base_url="http://localhost:11434/v1", api_key="ollama", ) def chat_with_ollama(query: str): response = client.chat.completions.create( model="qwen3:32b", messages=[{"role": "user", "content": query}], stream=True, temperature=0, seed=42 ) buffer = "" in_think = False # Flag to check if inside <think> and </no_think> print("AI:", end=" ", flush=True) for chunk in response: if hasattr(chunk, "choices") and chunk.choices: delta = chunk.choices[0].delta if hasattr(delta, "content") and delta.content: content = delta.content # Check if entering <think> block if not in_think: idx = content.find("<think>") if idx != -1: # Print content before <think> print(content[:idx], end="", flush=True) in_think = True # Discard content after <think> buffer = "" continue else: print(content, end="", flush=True) if in_think: # Check if </no_think> is encountered end_idx = content.find("</think>") if end_idx != -1: # Exit <think> block in_think = False # Skip </no_think> and the content after it after = content[end_idx + len("</think>"):] # Remove leading \n\n after = re.sub(r'^\n\n', '', after) print(after, end="", flush=True) continue # Otherwise discard content inside <think> continue print() # Newline after the response if __name__ == "__main__": chat_with_ollama("how to use ollama") ``` ``` import subprocess import json import re def chat_with_ollama(query: str): # 使用 curl 命令发起请求 curl_command = [ "curl", "-X", "POST", "http://localhost:11434/v1/chat/completions", "-H", "Content-Type: application/json", "-H", "Authorization: Bearer ollama", "-d", json.dumps({ "model": "qwen3:32b", "messages": [{"role": "user", "content": query}], "stream": True, "temperature": 0, "seed": 42 }) ] # 使用 subprocess 执行 curl 命令 process = subprocess.Popen(curl_command, stdout=subprocess.PIPE, stderr=subprocess.PIPE) buffer = "" in_think = False # Flag to check if inside <think> and </no_think> print("AI:", end=" ", flush=True) for line in process.stdout: # Decode the line from the curl output decoded_line = line.decode("utf-8").strip() # Skip empty lines or irrelevant lines if not decoded_line or not decoded_line.startswith("data:"): continue # Remove the "data:" prefix decoded_line = decoded_line[5:].strip() # Parse the JSON payload try: chunk = json.loads(decoded_line) except json.JSONDecodeError: continue # Process the chunk content if "choices" in chunk and chunk["choices"]: delta = chunk["choices"][0].get("delta", {}) if "content" in delta: content = delta["content"] # Check if entering <think> block if not in_think: idx = content.find("<think>") if idx != -1: # Print content before <think> print(content[:idx], end="", flush=True) in_think = True # Discard content after <think> buffer = "" continue else: print(content, end="", flush=True) if in_think: # Check if </no_think> is encountered end_idx = content.find("</think>") if end_idx != -1: # Exit <think> block in_think = False # Skip </no_think> and the content after it after = content[end_idx + len("</think>"):] # Remove leading \n\n after = re.sub(r'^\n\n', '', after) print(after, end="", flush=True) continue # Otherwise discard content inside <think> continue print() # Newline after the response # Wait for the process to complete and handle errors process.wait() if process.returncode != 0: stderr = process.stderr.read().decode("utf-8").strip() print(f"Error: {stderr}") if __name__ == "__main__": chat_with_ollama("how to use ollama \\no_think") ``` This is my solution, See if this would help you
GiteaMirror commented 2026-04-12 18:47:37 -05:00
Author
Owner
Copy Link

@lasseedfast commented on GitHub (May 14, 2025):

I’m using a similar buffer construction, but I hope this can be made a lot more simple in the API in some coming update!

<!-- gh-comment-id:2880013882 --> @lasseedfast commented on GitHub (May 14, 2025): I’m using a similar buffer construction, but I hope this can be made a lot more simple in the API in some coming update!
GiteaMirror commented 2026-04-12 18:47:39 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (May 30, 2025):

https://github.com/ollama/ollama/releases/tag/v0.9.0

<!-- gh-comment-id:2921921351 --> @rick-github commented on GitHub (May 30, 2025): https://github.com/ollama/ollama/releases/tag/v0.9.0
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#6905
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?