diff --git a/api/core/app/workflow/layers/llm_quota.py b/api/core/app/workflow/layers/llm_quota.py
index cf5a43eac2..be46fbab08 100644
--- a/api/core/app/workflow/layers/llm_quota.py
+++ b/api/core/app/workflow/layers/llm_quota.py
@@ -118,7 +118,7 @@ class LLMQuotaLayer(GraphEngineLayer):
                     model_instance = cast("ParameterExtractorNode", node).model_instance
                 case BuiltinNodeTypes.QUESTION_CLASSIFIER:
                     typed_node: QuestionClassifierNode = cast("QuestionClassifierNode", node)
-                    model_instance = cast(PreparedLLMProtocol, getattr(typed_node, "_model_instance"))
+                    model_instance = cast(PreparedLLMProtocol, typed_node._model_instance)
                 case _:
                     return None
         except AttributeError:
diff --git a/api/core/plugin/impl/model_runtime.py b/api/core/plugin/impl/model_runtime.py
index 51df79b7ea..52c8eece61 100644
--- a/api/core/plugin/impl/model_runtime.py
+++ b/api/core/plugin/impl/model_runtime.py
@@ -254,7 +254,7 @@ class PluginModelRuntime(ModelRuntime):
             stop=list(stop) if stop else None,
             stream=stream,
         )
-    
+
     @overload
     def invoke_llm_with_structured_output(
         self,
@@ -294,10 +294,7 @@ class PluginModelRuntime(ModelRuntime):
         prompt_messages: Sequence[PromptMessage],
         stop: Sequence[str] | None,
         stream: bool,
-    ) -> (
-        LLMResultWithStructuredOutput
-        | Generator[LLMResultChunkWithStructuredOutput, None, None]
-    ): 
+    ) -> LLMResultWithStructuredOutput | Generator[LLMResultChunkWithStructuredOutput, None, None]:
         # TODO: added to pass type check.
         # it is a new method from upstream that is not invoked at all.
         raise NotImplementedError