[Core] Asynchronous Output Processor (#7049)

Co-authored-by: Alexander Matveev <alexm@neuralmagic.com>

vllm/executor/distributed_gpu_executor.py

@@ -64,8 +64,9 @@ class DistributedGPUExecutor(GPUExecutor):
                           num_cpu_blocks=num_cpu_blocks)
 
     def execute_model(
-            self,
-            execute_model_req: ExecuteModelRequest) -> List[SamplerOutput]:
+        self,
+        execute_model_req: ExecuteModelRequest,
+    ) -> List[SamplerOutput]:
         if self.parallel_worker_tasks is None:
             self.parallel_worker_tasks = self._run_workers(
                 "start_worker_execution_loop",
@@ -188,7 +189,7 @@ class DistributedGPUExecutorAsync(DistributedGPUExecutor, ExecutorAsyncBase):
     @abstractmethod
     async def _driver_execute_model_async(
         self,
-        execute_model_req: Optional[ExecuteModelRequest] = None
+        execute_model_req: Optional[ExecuteModelRequest] = None,
     ) -> List[SamplerOutput]:
         """Execute the model asynchronously in the driver worker.
 

vllm/executor/gpu_executor.py

@@ -176,5 +176,5 @@ class GPUExecutorAsync(GPUExecutor, ExecutorAsyncBase):
         execute_model_req: ExecuteModelRequest,
     ) -> List[Union[SamplerOutput, PoolerOutput]]:
         output = await make_async(self.driver_worker.execute_model
-                                  )(execute_model_req=execute_model_req, )
+                                  )(execute_model_req=execute_model_req)
         return output