add support QWQ enable_thinking (#2706)

* add support QWQ enable_thinking

* add stream=True

* fix stream=true

* fix qwen

---------

Co-authored-by: lizexu <lizexu@baidu.com>

fastdeploy/reasoning/qwen3_reasoning_parsers.py

@@ -17,7 +17,7 @@ from collections.abc import Sequence
 from typing import Optional, Union
 
 from fastdeploy.entrypoints.openai.protocol import (ChatCompletionRequest,
-                                              DeltaMessage)
+                                                    DeltaMessage)
 from fastdeploy.reasoning import ReasoningParser, ReasoningParserManager
 
 
@@ -67,47 +67,47 @@ class Qwen3ReasoningParser(ReasoningParser):
         - 'abc' goes to reasoning_content
         - 'xyz' goes to content
         """
-        # Skip single special tokens
         if len(delta_token_ids) == 1 and (delta_token_ids[0] in [
                 self.think_start_token_id, self.think_end_token_id
         ]):
             return "", ""
 
-        if self.think_start_token_id in previous_token_ids:
+        # </think> in delta
-            if self.think_end_token_id in delta_token_ids:
+        if self.think_end_token_id in delta_token_ids:
-                # <think> in previous, </think> in delta,
+            #<think> in delta, </think> in delta, extract reasoning content
-                # extract reasoning content
+            if self.think_start_token_id in delta_token_ids:
+                start_index = delta_text.find(self.think_start_token)
+                end_index = delta_token_ids.find(self.think_end_token)
+                reasoning_content = delta_text[start_index +
+                                               len(self.think_start_token
+                                                   ):end_index]
+                content = delta_text[end_index+len(self.think_end_token):]
+                return reasoning_content, content
+            # <think> in previous, </think> in delta,
+            else:
                 end_index = delta_text.find(self.think_end_token)
                 reasoning_content = delta_text[:end_index]
                 content = delta_text[end_index + len(self.think_end_token):]
                 content = content if content else None
                 return reasoning_content, content
-            elif self.think_end_token_id in previous_token_ids:
+        # </think> in previous reasoning content continues
-                # <think> in previous, </think> in previous,
+        elif self.think_end_token_id in previous_token_ids:
-                # reasoning content continues
-                return "", delta_text
-            else:
-                # <think> in previous, no </think> in previous or delta,
-                # reasoning content continues
-                return delta_text, ""
-        elif self.think_start_token_id in delta_token_ids:
-            if self.think_end_token_id in delta_token_ids:
-                # <think> in delta, </think> in delta, extract reasoning content
-                start_index = delta_text.find(self.think_start_token)
-                end_index = delta_text.find(self.think_end_token)
-                reasoning_content = delta_text[start_index +
-                                               len(self.think_start_token
-                                                   ):end_index]
-                content = delta_text[end_index + len(self.think_end_token):]
-                content = content if content else None
-                return reasoning_content, content
-            else:
-                # <think> in delta, no </think> in delta,
-                # reasoning content continues
-                return delta_text, ""
-        else:
-            # thinking is disabled, just content
             return "", delta_text
+        # <think> in previous
+        elif self.think_start_token_id in previous_token_ids:
+            return delta_text,""
+        # <think> in delta
+        elif self.think_start_token_id in delta_token_ids:
+            start_index=delta_text.find(self.think_start_token)
+            reasoning_content=delta_text[start_index + len(self.think_start_token):]
+            content = ""
+            return reasoning_content, content
+        else:
+            return delta_text, ""
+
+
+
+
 
     def extract_reasoning_content(
             self, model_output: str, request: ChatCompletionRequest
@@ -115,31 +115,47 @@ class Qwen3ReasoningParser(ReasoningParser):
         """
         Extract reasoning content from the model output.
 
-        For text abc</think>xyz:
+        支持两种格式:
-        - 'abc' goes to reasoning_content
+        1. <think>abc</think>xyz - 标准格式
-        - 'xyz' goes to content
+        2. abc</think>xyz - 缺少起始标签的格式
 
         Returns:
             tuple[Optional[str], Optional[str]]: reasoning content and content
         """
 
-        # Check if the model output contains the <think> and </think> tokens.
+        # 检查是否包含结束标签
-        if (self.think_start_token not in model_output
-                or self.think_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.think_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.think_end_token not in model_output:
             return None, model_output
 
-        # Extract reasoning content from the model output.
+        # 检查是否有起始标签
-        reasoning_content, _, content = model_output.partition(
+        if self.think_start_token in model_output:
-            self.think_end_token)
+            # 标准格式：<think>content</think>answer
+            if (self.think_start_token not in model_output
+                    or self.think_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.think_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.think_end_token not in model_output:
+                return None, model_output
 
-        final_content = content or None
+            # Extract reasoning content from the model output.
-        return reasoning_content, final_content
+            reasoning_content, _, content = model_output.partition(
+                self.think_end_token)
+
+            final_content = content or None
+            return reasoning_content, final_content
+        else:
+            # 缺少起始标签的格式：content</think>answer
+            parts = model_output.split(self.think_end_token, 1)
+
+            if len(parts) == 2:
+                reasoning_content = parts[0].strip()
+                final_content = parts[1].strip() if parts[1].strip() else None
+                return reasoning_content, final_content
+
+        return None, model_output