mirror of
https://git.datalinker.icu/vllm-project/vllm.git
synced 2025-12-21 07:55:01 +08:00
3c62d28bb9
Signed-off-by: Yan Lu <luyan@nvidia.com> Co-authored-by: Michael Goin <mgoin64@gmail.com> Signed-off-by: yewentao256 <zhyanwentao@126.com>
73 lines
2.7 KiB
Python
73 lines
2.7 KiB
Python
# SPDX-License-Identifier: Apache-2.0
|
|
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
|
|
|
|
from typing import Optional, Union
|
|
|
|
from vllm.entrypoints.openai.protocol import (ChatCompletionRequest,
|
|
ResponsesRequest)
|
|
from vllm.reasoning.abs_reasoning_parsers import ReasoningParserManager
|
|
from vllm.reasoning.basic_parsers import BaseThinkingReasoningParser
|
|
|
|
|
|
@ReasoningParserManager.register_module("qwen3")
|
|
class Qwen3ReasoningParser(BaseThinkingReasoningParser):
|
|
"""
|
|
Reasoning parser for the Qwen3 model.
|
|
|
|
The Qwen3 model uses <think>...</think> tokens to denote reasoning text
|
|
within its output. The model provides a strict switch to disable reasoning
|
|
output via the 'enable_thinking=False' parameter. This parser extracts the
|
|
reasoning content enclosed by <think> and </think> tokens from the model's
|
|
output.
|
|
"""
|
|
|
|
@property
|
|
def start_token(self) -> str:
|
|
"""The token that starts reasoning content."""
|
|
return "<think>"
|
|
|
|
@property
|
|
def end_token(self) -> str:
|
|
"""The token that ends reasoning content."""
|
|
return "</think>"
|
|
|
|
def extract_reasoning_content(
|
|
self, model_output: str, request: Union[ChatCompletionRequest,
|
|
ResponsesRequest]
|
|
) -> tuple[Optional[str], Optional[str]]:
|
|
"""
|
|
Extract reasoning content from the model output.
|
|
|
|
Qwen3 has stricter requirements - it needs both start and end tokens
|
|
to be present, unlike other models that work with just the end token.
|
|
|
|
For text <think>abc</think>xyz:
|
|
- 'abc' goes to reasoning_content
|
|
- 'xyz' goes to content
|
|
|
|
Returns:
|
|
tuple[Optional[str], Optional[str]]: reasoning content and content
|
|
"""
|
|
|
|
# Check if the model output contains both <think> and </think> tokens.
|
|
if (self.start_token not in model_output
|
|
or self.end_token not in model_output):
|
|
return None, model_output
|
|
|
|
# Check if the <think> is present in the model output, remove it
|
|
# if it is present.
|
|
model_output_parts = model_output.partition(self.start_token)
|
|
model_output = model_output_parts[2] if model_output_parts[
|
|
1] else model_output_parts[0]
|
|
|
|
# Check if the model output contains the </think> tokens.
|
|
# If the end token is not found, return the model output as is.
|
|
if self.end_token not in model_output:
|
|
return None, model_output
|
|
|
|
# Extract reasoning content from the model output.
|
|
reasoning_content, _, content = model_output.partition(self.end_token)
|
|
|
|
final_content = content or None
|
|
return reasoning_content, final_content
|