Add 'none' reasoning effort to ChatCompletionRequest (#36238)

Signed-off-by: Julien Denize <julien.denize@mistral.ai>

vllm/entrypoints/openai/chat_completion/protocol.py

@@ -179,7 +179,7 @@ class ChatCompletionRequest(OpenAIBaseModel):
         | ChatCompletionNamedToolChoiceParam
         | None
     ) = "none"
-    reasoning_effort: Literal["low", "medium", "high"] | None = None
+    reasoning_effort: Literal["none", "low", "medium", "high"] | None = None
     include_reasoning: bool = True
     parallel_tool_calls: bool | None = True
 
@@ -778,3 +778,10 @@ class ChatCompletionRequest(OpenAIBaseModel):
                                 )
 
         return data
+
+    @model_validator(mode="before")
+    @classmethod
+    def set_include_reasoning_for_none_effort(cls, data: Any) -> Any:
+        if data.get("reasoning_effort") == "none":
+            data["include_reasoning"] = False
+        return data

vllm/entrypoints/openai/chat_completion/serving.py

@@ -1893,8 +1893,10 @@ class OpenAIServingChat(OpenAIServing):
         # if the model supports it. TODO: Support browsing.
         assert not self.supports_browsing
         assert not self.supports_code_interpreter
+        if (reasoning_effort := request.reasoning_effort) == "none":
+            raise ValueError(f"Harmony does not support {reasoning_effort=}")
         sys_msg = get_system_message(
-            reasoning_effort=request.reasoning_effort,
+            reasoning_effort=reasoning_effort,
             browser_description=None,
             python_description=None,
             with_custom_tools=should_include_tools,

vllm/entrypoints/serve/render/serving.py

@@ -221,6 +221,9 @@ class OpenAIServingRender:
         # if the model supports it. TODO: Support browsing.
         assert not self.supports_browsing
         assert not self.supports_code_interpreter
+        assert request.reasoning_effort != "none", (
+            "Harmony does not support reasoning_effort='none'"
+        )
         sys_msg = get_system_message(
             reasoning_effort=request.reasoning_effort,
             browser_description=None,