update

Signed-off-by: shen-shanshan <467638484@qq.com>

Author: shen-shanshan

vllm/v1/engine/core.py

@@ -249,13 +249,10 @@ def step_with_batch_queue(self) -> Optional[EngineCoreOutputs]:
         return engine_core_outputs
 
     def shutdown(self):
+        self.structured_output_manager.clear_backend()
         if self.model_executor:
             self.model_executor.shutdown()
 
-        so_backend = self.structured_output_manager.backend
-        if so_backend is not None:
-            so_backend.destroy()
-
     def profile(self, is_start: bool = True):
         self.model_executor.profile(is_start)
 

vllm/v1/structured_output/__init__.py

@@ -107,3 +107,7 @@ def grammar_bitmask(
         # np.ndarray, because that is much more efficient for serialization
         # and deserialization when sending this to the GPU workers.
         return bitmask_tensor.numpy()
+
+    def clear_backend(self) -> None:
+        if self.backend is not None:
+            self.backend.destroy()

vllm/v1/structured_output/backend_types.py

@@ -91,5 +91,5 @@ def allocate_token_bitmask(self, max_num_seqs: int):
     @abstractmethod
     def destroy(self):
         """
-        Clear objects in the backend to avoid nanobind leaked.
+        Backend-specific cleanup.
         """