fix: skip check unstreamed tool arg tokens when tool call name is present

Signed-off-by: Sungjae Lee <[email protected]>

Author: llsj14

vllm/entrypoints/openai/serving_chat.py

@@ -1729,7 +1729,10 @@ def _should_check_for_unstreamed_tool_arg_tokens(
         """
         Check to see if we should check for unstreamed tool arguments tokens.
         This is only applicable when auto tool parsing is enabled, the delta
-        is a tool call with arguments.
+        is a tool call with arguments. We only enter this branch when arguments
+        are being streamed incrementally (i.e., when name is not present).
+        This branch is skipped if the complete tool call (including name) is present
+        or will be streamed.
         """
 
         return bool(
@@ -1743,6 +1746,9 @@ def _should_check_for_unstreamed_tool_arg_tokens(
             and delta_message.tool_calls[0]
             and delta_message.tool_calls[0].function
             and delta_message.tool_calls[0].function.arguments is not None
+            # Only enter this branch when name is not already streamed
+            # (i.e., when arguments are being streamed incrementally)
+            and not delta_message.tool_calls[0].function.name
         )
 
     def _make_request_with_harmony(