Only enable commentary channel for GPT-OSS when really necessary

vllm/entrypoints/harmony_utils.py

@@ -44,8 +44,16 @@ def get_system_message(
     start_date: Optional[str] = None,
     browser_description: Optional[str] = None,
     python_description: Optional[str] = None,
+    has_tools: bool = False,
 ) -> Message:
-    sys_msg_content = SystemContent.new()
+    needs_commentary = (has_tools or browser_description is not None
+                        or python_description is not None)
+    if needs_commentary:
+        sys_msg_content = SystemContent.new().with_required_channels(
+            ["analysis", "commentary", "final"])
+    else:
+        sys_msg_content = SystemContent.new().with_required_channels(
+            ["analysis", "final"])
     if model_identity is not None:
         sys_msg_content = sys_msg_content.with_model_identity(model_identity)
     if reasoning_effort is not None:

vllm/entrypoints/openai/serving_chat.py

@@ -1455,7 +1455,8 @@ def _make_request_with_harmony(
         sys_msg = get_system_message(
             reasoning_effort=request.reasoning_effort,
             browser_description=None,
-            python_description=None)
+            python_description=None,
+            has_tools = bool(request.tools))
         messages.append(sys_msg)
 
         # Add developer message.

vllm/entrypoints/openai/serving_responses.py

@@ -629,6 +629,7 @@ def _construct_input_messages_with_harmony(
                 python_description=self.tool_server.get_tool_description(
                     "python") if enable_code_interpreter
                 and self.tool_server is not None else None,
+                has_tools=bool(request.tools),
             )
             messages.append(sys_msg)
             dev_msg = get_developer_message(request.instructions,