[Bugfix]: Fix the incompatibility issue with tool_choice 'required' when Thinking is enabled (vllm-project#19075)

Signed-off-by: chaunceyjiang <[email protected]>

Author: chaunceyjiang

tests/entrypoints/openai/test_completion_with_function_calling.py

@@ -9,7 +9,7 @@
 from ...utils import RemoteOpenAIServer
 
 # any model with a chat template should work here
-MODEL_NAME = "Qwen/Qwen2.5-1.5B-Instruct"
+MODEL_NAME = "Qwen/Qwen3-0.6B"
 
 
 @pytest.fixture(scope="module")

vllm/entrypoints/openai/serving_chat.py

@@ -320,10 +320,13 @@ def _filter_delta_text(delta_text: str,
     def extract_tool_call_required_streaming(
         self,
         previous_text: str,
-        current_text: str,
+        current_text: Optional[str],
         delta_text: str,
         function_name_returned: bool,
     ) -> tuple[Optional[DeltaMessage], bool]:
+        if current_text is None or current_text == "":
+            # if the current text is empty, we cannot parse it
+            return None, function_name_returned
         try:
             obj = partial_json_parser.loads(current_text)
         except partial_json_parser.core.exceptions.MalformedJSON:
@@ -650,10 +653,18 @@ async def chat_completion_stream_generator(
                         current_text = previous_text + delta_text
                         fn_name_returned = function_name_returned[i]
 
+                        if self.reasoning_parser:
+                            _, content = \
+                                reasoning_parser.extract_reasoning_content(
+                                    current_text,
+                                    request
+                                )
+                        else:
+                            content = current_text
                         delta_message, function_name_returned[i] = (
                             self.extract_tool_call_required_streaming(
                                 previous_text=previous_text,
-                                current_text=current_text,
+                                current_text=content,
                                 delta_text=delta_text,
                                 function_name_returned=fn_name_returned))
 
@@ -981,8 +992,9 @@ async def chat_completion_full_generator(
 
                 # the fields of FunctionDefinition are a superset of the
                 # tool call outputs and can be used for parsing
+                assert content is not None
                 tool_calls = TypeAdapter(
-                    list[FunctionDefinition]).validate_json(output.text)
+                    list[FunctionDefinition]).validate_json(content)
                 message = ChatMessage(
                     role=role,
                     content="",