Feat: add support for thinking mode via chat_template_kwargs.enable_t… (#5551)

Co-authored-by: shuaills <shishuaiuoe@gmail.com>
Co-authored-by: Chayenne <zhaochen20@outlook.com>
Co-authored-by: Lianmin Zheng <lianminzheng@gmail.com>
Co-authored-by: Yineng Zhang <me@zhyncs.com>

python/sglang/srt/openai_api/adapter.py

@@ -1001,6 +1001,11 @@ def v1_chat_generate_request(
                         tokenize=True,
                         add_generation_prompt=True,
                         tools=tools,
+                        **(
+                            request.chat_template_kwargs
+                            if request.chat_template_kwargs
+                            else {}
+                        ),
                     )
                 except:
                     #  This except branch will be triggered when the chosen model
@@ -1012,6 +1017,11 @@ def v1_chat_generate_request(
                         tokenize=True,
                         add_generation_prompt=True,
                         tools=tools,
+                        **(
+                            request.chat_template_kwargs
+                            if request.chat_template_kwargs
+                            else {}
+                        ),
                     )
 
                 if assistant_prefix:
@@ -1245,16 +1255,34 @@ def v1_chat_generate_response(
         tool_calls = None
         text = ret_item["text"]
 
+        enable_thinking = True
         if isinstance(request, list):
             tool_choice = request[idx].tool_choice
             tools = request[idx].tools
             separate_reasoning = request[idx].separate_reasoning
+
+            if (
+                request[idx].chat_template_kwargs
+                and request[idx].chat_template_kwargs.get("enable_thinking") is not None
+            ):
+                enable_thinking = request[idx].chat_template_kwargs.get(
+                    "enable_thinking", True
+                )
         else:
             tool_choice = request.tool_choice
             tools = request.tools
             separate_reasoning = request.separate_reasoning
 
-        if reasoning_parser and separate_reasoning:
+            if (
+                request.chat_template_kwargs
+                and request.chat_template_kwargs.get("enable_thinking") is not None
+            ):
+                enable_thinking = request.chat_template_kwargs.get(
+                    "enable_thinking", True
+                )
+
+        reasoning_text = None
+        if reasoning_parser and separate_reasoning and enable_thinking:
             try:
                 parser = ReasoningParser(
                     model_type=reasoning_parser, stream_reasoning=False
@@ -1266,8 +1294,6 @@ def v1_chat_generate_response(
                     HTTPStatus.BAD_REQUEST,
                     "Failed to parse reasoning related info to json format!",
                 )
-        else:
-            reasoning_text = None
 
         if tool_choice != "none" and tools:
             parser = FunctionCallParser(tools, tool_call_parser)

python/sglang/srt/openai_api/protocol.py

@@ -361,6 +361,7 @@ class ChatCompletionRequest(BaseModel):
     session_params: Optional[Dict] = None
     separate_reasoning: bool = True
     stream_reasoning: bool = True
+    chat_template_kwargs: Optional[Dict] = None
 
     # For PD disaggregation
     bootstrap_host: Optional[str] = None

python/sglang/srt/reasoning_parser.py

@@ -117,6 +117,29 @@ class DeepSeekR1Detector(BaseReasoningFormatDetector):
         # https://github.com/sgl-project/sglang/pull/3202#discussion_r1950153599
 
 
+class Qwen3Detector(BaseReasoningFormatDetector):
+    """
+    Detector for Qwen3 model.
+    Assumes reasoning format:
+      (<think>)*(.*)</think>
+    Returns all the text before the </think> tag as `reasoning_text`
+    and the rest of the text as `normal_text`.
+
+    Args:
+        stream_reasoning (bool): If False, accumulates reasoning content until the end tag.
+            If True, streams reasoning content as it arrives.
+    """
+
+    def __init__(self, stream_reasoning: bool = True):
+        # Qwen3 is assumed to be reasoning until `</think>` token
+        super().__init__(
+            "<think>",
+            "</think>",
+            force_reasoning=True,
+            stream_reasoning=stream_reasoning,
+        )
+
+
 class ReasoningParser:
     """
     Parser that handles both streaming and non-streaming scenarios for extracting
@@ -129,7 +152,8 @@ class ReasoningParser:
     """
 
     DetectorMap: Dict[str, BaseReasoningFormatDetector] = {
-        "deepseek-r1": DeepSeekR1Detector
+        "deepseek-r1": DeepSeekR1Detector,
+        "qwen3": Qwen3Detector,
     }
 
     def __init__(self, model_type: str = None, stream_reasoning: bool = True):