sglang/docs/advanced_features/separate_reasoning.ipynb

Reasoning Parser

SGLang supports parsing reasoning content out from "normal" content for reasoning models such as DeepSeek R1.

Supported Models & Parsers

Model	Reasoning tags	Parser	Notes
DeepSeek‑R1 series	<think> … </think>	deepseek-r1	Supports all variants (R1, R1-0528, R1-Distill)
Standard Qwen3 models	<think> … </think>	qwen3	Supports enable_thinking parameter
Qwen3-Thinking models	<think> … </think>	qwen3 or qwen3-thinking	Always generates thinking content
Kimi models	◁think▷ … ◁/think▷	kimi	Uses special thinking delimiters

Model-Specific Behaviors

DeepSeek-R1 Family:

• DeepSeek-R1: No <think> start tag, jumps directly to thinking content
• DeepSeek-R1-0528: Generates both <think> start and </think> end tags
• Both are handled by the same deepseek-r1 parser

Qwen3 Family:

• Standard Qwen3 (e.g., Qwen3-2507): Use qwen3 parser, supports enable_thinking in chat templates
• Qwen3-Thinking (e.g., Qwen3-235B-A22B-Thinking-2507): Use qwen3 or qwen3-thinking parser, always thinks

Kimi:

• Kimi: Uses special ◁think▷ and ◁/think▷ tags

Usage

Launching the Server

Specify the --reasoning-parser option.

In [ ]:

import requests
from openai import OpenAI
from sglang.test.doc_patch import launch_server_cmd
from sglang.utils import wait_for_server, print_highlight, terminate_process

server_process, port = launch_server_cmd(
    "python3 -m sglang.launch_server --model-path deepseek-ai/DeepSeek-R1-Distill-Qwen-7B --host 0.0.0.0 --reasoning-parser deepseek-r1"
)

wait_for_server(f"http://localhost:{port}")

Note that --reasoning-parser defines the parser used to interpret responses.

OpenAI Compatible API

Using the OpenAI compatible API, the contract follows the DeepSeek API design established with the release of DeepSeek-R1:

• reasoning_content: The content of the CoT.
• content: The content of the final answer.

In [ ]:

# Initialize OpenAI-like client
client = OpenAI(api_key="None", base_url=f"http://0.0.0.0:{port}/v1")
model_name = client.models.list().data[0].id

messages = [
    {
        "role": "user",
        "content": "What is 1+3?",
    }
]

Non-Streaming Request

In [ ]:

response_non_stream = client.chat.completions.create(
    model=model_name,
    messages=messages,
    temperature=0.6,
    top_p=0.95,
    stream=False,  # Non-streaming
    extra_body={"separate_reasoning": True},
)
print_highlight("==== Reasoning ====")
print_highlight(response_non_stream.choices[0].message.reasoning_content)

print_highlight("==== Text ====")
print_highlight(response_non_stream.choices[0].message.content)

Streaming Request

In [ ]:

response_stream = client.chat.completions.create(
    model=model_name,
    messages=messages,
    temperature=0.6,
    top_p=0.95,
    stream=True,  # Non-streaming
    extra_body={"separate_reasoning": True},
)

reasoning_content = ""
content = ""
for chunk in response_stream:
    if chunk.choices[0].delta.content:
        content += chunk.choices[0].delta.content
    if chunk.choices[0].delta.reasoning_content:
        reasoning_content += chunk.choices[0].delta.reasoning_content

print_highlight("==== Reasoning ====")
print_highlight(reasoning_content)

print_highlight("==== Text ====")
print_highlight(content)

Optionally, you can buffer the reasoning content to the last reasoning chunk (or the first chunk after the reasoning content).

In [ ]:

response_stream = client.chat.completions.create(
    model=model_name,
    messages=messages,
    temperature=0.6,
    top_p=0.95,
    stream=True,  # Non-streaming
    extra_body={"separate_reasoning": True, "stream_reasoning": False},
)

reasoning_content = ""
content = ""
for chunk in response_stream:
    if chunk.choices[0].delta.content:
        content += chunk.choices[0].delta.content
    if chunk.choices[0].delta.reasoning_content:
        reasoning_content = chunk.choices[0].delta.reasoning_content

print_highlight("==== Reasoning ====")
print_highlight(reasoning_content)

print_highlight("==== Text ====")
print_highlight(content)

The reasoning separation is enable by default when specify . To disable it, set the separate_reasoning option to False in request.

In [ ]:

response_non_stream = client.chat.completions.create(
    model=model_name,
    messages=messages,
    temperature=0.6,
    top_p=0.95,
    stream=False,  # Non-streaming
    extra_body={"separate_reasoning": False},
)

print_highlight("==== Original Output ====")
print_highlight(response_non_stream.choices[0].message.content)

SGLang Native API

In [ ]:

from transformers import AutoTokenizer

tokenizer = AutoTokenizer.from_pretrained("deepseek-ai/DeepSeek-R1-Distill-Qwen-7B")
input = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True,
)

gen_url = f"http://localhost:{port}/generate"
gen_data = {
    "text": input,
    "sampling_params": {
        "skip_special_tokens": False,
        "max_new_tokens": 1024,
        "temperature": 0.6,
        "top_p": 0.95,
    },
}
gen_response = requests.post(gen_url, json=gen_data).json()["text"]

print_highlight("==== Original Output ====")
print_highlight(gen_response)

parse_url = f"http://localhost:{port}/separate_reasoning"
separate_reasoning_data = {
    "text": gen_response,
    "reasoning_parser": "deepseek-r1",
}
separate_reasoning_response_json = requests.post(
    parse_url, json=separate_reasoning_data
).json()
print_highlight("==== Reasoning ====")
print_highlight(separate_reasoning_response_json["reasoning_text"])
print_highlight("==== Text ====")
print_highlight(separate_reasoning_response_json["text"])

In [ ]:

terminate_process(server_process)

Offline Engine API

In [ ]:

import sglang as sgl
from sglang.srt.reasoning_parser import ReasoningParser
from sglang.utils import print_highlight

llm = sgl.Engine(model_path="deepseek-ai/DeepSeek-R1-Distill-Qwen-7B")
tokenizer = AutoTokenizer.from_pretrained("deepseek-ai/DeepSeek-R1-Distill-Qwen-7B")
input = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True,
)
sampling_params = {
    "max_new_tokens": 1024,
    "skip_special_tokens": False,
    "temperature": 0.6,
    "top_p": 0.95,
}
result = llm.generate(prompt=input, sampling_params=sampling_params)

generated_text = result["text"]  # Assume there is only one prompt

print_highlight("==== Original Output ====")
print_highlight(generated_text)

parser = ReasoningParser("deepseek-r1")
reasoning_text, text = parser.parse_non_stream(generated_text)
print_highlight("==== Reasoning ====")
print_highlight(reasoning_text)
print_highlight("==== Text ====")
print_highlight(text)

In [ ]:

llm.shutdown()

Supporting New Reasoning Model Schemas

For future reasoning models, you can implement the reasoning parser as a subclass of BaseReasoningFormatDetector in python/sglang/srt/reasoning_parser.py and specify the reasoning parser for new reasoning model schemas accordingly.

class DeepSeekR1Detector(BaseReasoningFormatDetector):
    """
    Detector for DeepSeek-R1 family models.
    
    Supported models:
      - DeepSeek-R1: Always generates thinking content without <think> start tag
      - DeepSeek-R1-0528: Generates thinking content with <think> start tag
    
    This detector handles both patterns automatically.
    """

    def __init__(self, stream_reasoning: bool = True):
        super().__init__("<think>", "</think>", force_reasoning=True, stream_reasoning=stream_reasoning)


class Qwen3Detector(BaseReasoningFormatDetector):
    """
    Detector for standard Qwen3 models that support enable_thinking parameter.
    
    These models can switch between thinking and non-thinking modes:
      - enable_thinking=True: Generates <think>...</think> tags
      - enable_thinking=False: No thinking content generated
    """

    def __init__(self, stream_reasoning: bool = True):
        super().__init__("<think>", "</think>", force_reasoning=False, stream_reasoning=stream_reasoning)


class Qwen3ThinkingDetector(BaseReasoningFormatDetector):
    """
    Detector for Qwen3-Thinking models (e.g., Qwen3-235B-A22B-Thinking-2507).
    
    These models always generate thinking content without <think> start tag.
    They do not support the enable_thinking parameter.
    """

    def __init__(self, stream_reasoning: bool = True):
        super().__init__("<think>", "</think>", force_reasoning=True, stream_reasoning=stream_reasoning)


class ReasoningParser:
    """
    Parser that handles both streaming and non-streaming scenarios.
    
    Usage:
      # For standard Qwen3 models with enable_thinking support
      parser = ReasoningParser("qwen3")
      
      # For Qwen3-Thinking models that always think
      parser = ReasoningParser("qwen3-thinking")
    """

    DetectorMap: Dict[str, Type[BaseReasoningFormatDetector]] = {
        "deepseek-r1": DeepSeekR1Detector,
        "qwen3": Qwen3Detector,
        "qwen3-thinking": Qwen3ThinkingDetector,
        "kimi": KimiDetector,
    }

    def __init__(self, model_type: str = None, stream_reasoning: bool = True):
        if not model_type:
            raise ValueError("Model type must be specified")

        detector_class = self.DetectorMap.get(model_type.lower())
        if not detector_class:
            raise ValueError(f"Unsupported model type: {model_type}")

        self.detector = detector_class(stream_reasoning=stream_reasoning)

    def parse_non_stream(self, full_text: str) -> Tuple[str, str]:
        """Returns (reasoning_text, normal_text)"""
        ret = self.detector.detect_and_parse(full_text)
        return ret.reasoning_text, ret.normal_text

    def parse_stream_chunk(self, chunk_text: str) -> Tuple[str, str]:
        """Returns (reasoning_text, normal_text) for the current chunk"""
        ret = self.detector.parse_streaming_increment(chunk_text)
        return ret.reasoning_text, ret.normal_text