[Feature] Add --default-chat-template-kwargs server arg (#29579)

python/sglang/srt/entrypoints/openai/serving_chat.py

@@ -171,6 +171,9 @@ class OpenAIServingChat(OpenAIServingBase):
         self.template_manager = template_manager
         self.tool_call_parser = self.tokenizer_manager.server_args.tool_call_parser
         self.reasoning_parser = self.tokenizer_manager.server_args.reasoning_parser
+        self.default_chat_template_kwargs = (
+            self.tokenizer_manager.server_args.default_chat_template_kwargs or {}
+        )
         self._reasoning_detector = None
         if self.reasoning_parser:
             try:
@@ -633,6 +636,15 @@ class OpenAIServingChat(OpenAIServingBase):
         self, request: ChatCompletionRequest, is_multimodal: bool
     ) -> MessageProcessingResult:
         """Process chat messages and apply chat template"""
+        if self.default_chat_template_kwargs:
+            ctk = dict(request.chat_template_kwargs or {})
+            for k, v in self.default_chat_template_kwargs.items():
+                ctk.setdefault(k, v)
+            request.chat_template_kwargs = ctk
+            effort = ctk.get("reasoning_effort")
+            if effort is not None and request.reasoning_effort is None:
+                request.reasoning_effort = effort
+
         # GptOss model needs to keep special tokens for harmony parsing
         if self.is_gpt_oss or self.is_gemma4:
             request.skip_special_tokens = False

python/sglang/srt/server_args.py

@@ -1116,6 +1116,16 @@ class ServerArgs:
         "Return number of cached tokens in usage.prompt_tokens_details for each openai request.",
     ] = False
     reasoning_parser: Optional[str] = None
+    default_chat_template_kwargs: A[
+        Optional[Dict[str, Any]],
+        Arg(
+            help="Default chat template kwargs applied to every request when not "
+            "overridden per-request. Keys must match what the model's chat template "
+            "expects (e.g. enable_thinking, thinking, reasoning_effort). Per-request "
+            "chat_template_kwargs takes precedence.",
+            type_parser=json.loads,
+        ),
+    ] = None
     strip_thinking_cache: A[
         bool,
         "Skip caching reasoning-model output (thinking + answer) in the radix tree on finish; keep only the prompt prefix. Opt-in: changes cache contents.",
@@ -2810,6 +2820,13 @@ class ServerArgs:
         # Validate transcription/ASR-specific server args.
         self._handle_asr_validation()
 
+        if self.default_chat_template_kwargs is not None and not isinstance(
+            self.default_chat_template_kwargs, dict
+        ):
+            raise ValueError(
+                "--default-chat-template-kwargs must decode to a JSON object"
+            )
+
         # Handle deprecated arguments.
         self._handle_deprecated_args()
 

test/registered/unit/entrypoints/openai/test_serving_chat.py

@@ -45,6 +45,7 @@ class _MockTokenizerManager:
             tool_call_parser="hermes",
             reasoning_parser=None,
             stream_response_default_include_usage=False,
+            default_chat_template_kwargs=None,
         )
         # Mock hf_config for _resolve_chat_encoding_spec check
         mock_hf_config = Mock()
@@ -283,6 +284,54 @@ class ServingChatTestCase(unittest.TestCase):
 
         self.assertFalse(adapted.require_reasoning)
 
+    def test_default_chat_template_kwargs_applied_when_request_unset(self):
+        self.template_manager.chat_template_name = None
+        self.template_manager.jinja_template_content_format = "string"
+        self.tm.tokenizer.apply_chat_template.return_value = [1, 2, 3]
+        self.chat.default_chat_template_kwargs = {"enable_thinking": False}
+
+        req = ChatCompletionRequest(
+            model="x",
+            messages=[{"role": "user", "content": "What is 2+2?"}],
+        )
+
+        self.chat._process_messages(req, is_multimodal=False)
+
+        kwargs = self.tm.tokenizer.apply_chat_template.call_args.kwargs
+        self.assertIs(kwargs["enable_thinking"], False)
+
+    def test_default_chat_template_kwargs_overridden_per_request(self):
+        self.template_manager.chat_template_name = None
+        self.template_manager.jinja_template_content_format = "string"
+        self.tm.tokenizer.apply_chat_template.return_value = [1, 2, 3]
+        self.chat.default_chat_template_kwargs = {"enable_thinking": False}
+
+        req = ChatCompletionRequest(
+            model="x",
+            messages=[{"role": "user", "content": "What is 2+2?"}],
+            chat_template_kwargs={"enable_thinking": True},
+        )
+
+        self.chat._process_messages(req, is_multimodal=False)
+
+        kwargs = self.tm.tokenizer.apply_chat_template.call_args.kwargs
+        self.assertIs(kwargs["enable_thinking"], True)
+
+    def test_default_chat_template_kwargs_mirrors_reasoning_effort(self):
+        self.template_manager.chat_template_name = None
+        self.template_manager.jinja_template_content_format = "string"
+        self.tm.tokenizer.apply_chat_template.return_value = [1, 2, 3]
+        self.chat.default_chat_template_kwargs = {"reasoning_effort": "high"}
+
+        req = ChatCompletionRequest(
+            model="x",
+            messages=[{"role": "user", "content": "What is 2+2?"}],
+        )
+
+        self.chat._process_messages(req, is_multimodal=False)
+
+        self.assertEqual(req.reasoning_effort, "high")
+
     def test_kimi_tool_call_keeps_template_default_thinking(self):
         self.template_manager.chat_template_name = None
         self.template_manager.jinja_template_content_format = "string"