[refactor] Update reasoning parameter to require_reasoning (#14922)

Signed-off-by: Xinyuan Tong <xinyuantong.cs@gmail.com>

docs/advanced_features/structured_outputs_for_reasoning_models.ipynb

@@ -364,7 +364,7 @@
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "> Note: For native API, as a work-around, you need to set `reasoning` argument to `True` to ensure the model will think before generating the structured output. It's not required for chat-completion API."
+    "> Note: For native API, as a work-around, you need to set `require_reasoning` argument to `True` to ensure the model will think before generating the structured output. It's not required for chat-completion API."
    ]
   },
   {
@@ -414,7 +414,7 @@
     "    f\"http://localhost:{port}/generate\",\n",
     "    json={\n",
     "        \"text\": text,\n",
-    "        \"reasoning\": True,\n",
+    "        \"require_reasoning\": True,\n",
     "        \"sampling_params\": {\n",
     "            \"temperature\": 0,\n",
     "            \"max_new_tokens\": 2048,\n",
@@ -462,7 +462,7 @@
     "    f\"http://localhost:{port}/generate\",\n",
     "    json={\n",
     "        \"text\": text,\n",
-    "        \"reasoning\": True,\n",
+    "        \"require_reasoning\": True,\n",
     "        \"sampling_params\": {\n",
     "            \"temperature\": 0,\n",
     "            \"max_new_tokens\": 2048,\n",
@@ -491,7 +491,7 @@
     "    f\"http://localhost:{port}/generate\",\n",
     "    json={\n",
     "        \"text\": \"Give me the information of the capital of France.\",\n",
-    "        \"reasoning\": True,\n",
+    "        \"require_reasoning\": True,\n",
     "        \"sampling_params\": {\n",
     "            \"max_new_tokens\": 2048,\n",
     "            \"temperature\": 0,\n",
@@ -529,7 +529,7 @@
     "    f\"http://localhost:{port}/generate\",\n",
     "    json={\n",
     "        \"text\": \"Paris is the capital of\",\n",
-    "        \"reasoning\": True,\n",
+    "        \"require_reasoning\": True,\n",
     "        \"sampling_params\": {\n",
     "            \"temperature\": 0,\n",
     "            \"max_new_tokens\": 2048,\n",
@@ -558,7 +558,7 @@
     ")\n",
     "payload = {\n",
     "    \"text\": text,\n",
-    "    \"reasoning\": True,\n",
+    "    \"require_reasoning\": True,\n",
     "    \"sampling_params\": {\n",
     "        \"max_new_tokens\": 2048,\n",
     "        \"structural_tag\": json.dumps(\n",

python/sglang/srt/constrained/base_grammar_backend.py

@@ -155,7 +155,7 @@ class BaseGrammarBackend:
         return self._not_supported("structural_tag", key_string)
 
     def _init_value_dispatch(
-        self, key: Tuple[str, str], reasoning: bool
+        self, key: Tuple[str, str], require_reasoning: bool
     ) -> Optional[BaseGrammarObject]:
         s = time.perf_counter()
         key_type, key_string = key
@@ -179,14 +179,14 @@ class BaseGrammarBackend:
         return grammar
 
     def get_cached_or_future_value(
-        self, key: Tuple[str, str], reasoning: bool
+        self, key: Tuple[str, str], require_reasoning: bool
     ) -> Optional[BaseGrammarObject]:
         value = self.cache.get(key)
         if value:
             copied_value = value.copy()
-            copied_value.maybe_init_reasoning(reasoning)
+            copied_value.maybe_init_reasoning(require_reasoning)
             return copied_value, True
-        value = self.executor.submit(self._init_value_dispatch, key, reasoning)
+        value = self.executor.submit(self._init_value_dispatch, key, require_reasoning)
         return value, False
 
     def set_cache(self, key: Tuple[str, str], value: BaseGrammarObject):

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

@@ -214,7 +214,7 @@ class OpenAIServingChat(OpenAIServingBase):
             return_hidden_states=request.return_hidden_states,
             rid=request.rid,
             extra_key=self._compute_extra_key(request),
-            reasoning=self._get_reasoning_from_request(request),
+            require_reasoning=self._get_reasoning_from_request(request),
             priority=request.priority,
             custom_labels=custom_labels,
             custom_logit_processor=request.custom_logit_processor,

python/sglang/srt/managers/io_struct.py

@@ -215,8 +215,8 @@ class GenerateReqInput(BaseReq, APIServingTimingMixin):
     bootstrap_pair_key: Optional[Union[List[str], str]] = None
     decode_tp_size: Optional[Union[List[Optional[int]], int]] = None
 
-    # For reasoning
-    reasoning: bool = False
+    # Require reasoning for the request (hybrid reasoning model only)
+    require_reasoning: bool = False
 
     # For data parallel rank routing
     data_parallel_rank: Optional[int] = None
@@ -700,8 +700,8 @@ class TokenizedGenerateReqInput(BaseReq):
     bootstrap_pair_key: Optional[str] = None
     decode_tp_size: Optional[int] = None
 
-    # For reasoning
-    reasoning: bool = False
+    # Require reasoning for the request (hybrid reasoning model only)
+    require_reasoning: bool = False
 
     # For data parallel rank routing
     data_parallel_rank: Optional[int] = None

python/sglang/srt/managers/schedule_batch.py

@@ -472,7 +472,7 @@ class Req:
         token_type_ids: List[int] = None,
         session_id: Optional[str] = None,
         custom_logit_processor: Optional[str] = None,
-        reasoning: bool = False,
+        require_reasoning: bool = False,
         return_hidden_states: bool = False,
         eos_token_ids: Optional[Set[int]] = None,
         bootstrap_host: Optional[str] = None,
@@ -518,8 +518,8 @@ class Req:
         # For multi-http worker
         self.http_worker_ipc = http_worker_ipc
 
-        # For reasoning
-        self.reasoning = reasoning
+        # Require reasoning for the request (hybrid reasoning model only)
+        self.require_reasoning = require_reasoning
 
         # Sampling info
         if isinstance(sampling_params.custom_params, dict):

python/sglang/srt/managers/scheduler.py

@@ -1314,7 +1314,7 @@ class Scheduler(
                 lora_id=recv_req.lora_id,
                 input_embeds=recv_req.input_embeds,
                 custom_logit_processor=recv_req.custom_logit_processor,
-                reasoning=recv_req.reasoning,
+                require_reasoning=recv_req.require_reasoning,
                 return_hidden_states=recv_req.return_hidden_states,
                 eos_token_ids=self.model_config.hf_eos_token_id,
                 bootstrap_host=recv_req.bootstrap_host,
@@ -1443,7 +1443,7 @@ class Scheduler(
                     key = ("structural_tag", req.sampling_params.structural_tag)
 
                 value, cache_hit = self.grammar_backend.get_cached_or_future_value(
-                    key, req.reasoning
+                    key, req.require_reasoning
                 )
                 req.grammar = value
 

python/sglang/srt/managers/tokenizer_manager.py

@@ -796,7 +796,7 @@ class TokenizerManager(TokenizerCommunicatorMixin, TokenizerManagerMultiItemMixi
                 input_embeds=input_embeds,
                 session_params=session_params,
                 custom_logit_processor=obj.custom_logit_processor,
-                reasoning=obj.reasoning,
+                require_reasoning=obj.require_reasoning,
                 return_hidden_states=obj.return_hidden_states,
                 data_parallel_rank=obj.data_parallel_rank,
                 priority=obj.priority,