diff --git a/vllm/engine/async_llm_engine.py b/vllm/engine/async_llm_engine.py
index 9b4ef48b..f3c8d69e 100644
--- a/vllm/engine/async_llm_engine.py
+++ b/vllm/engine/async_llm_engine.py
@@ -553,11 +553,13 @@ class AsyncLLMEngine:
             request_outputs = await self.engine.step_async(virtual_engine)
 
         # Put the outputs into the corresponding streams.
+        finished = True
         for request_output in request_outputs:
             self._request_tracker.process_request_output(
                 request_output, verbose=self.log_requests)
+            finished = finished and request_output.finished
 
-        return len(request_outputs) > 0
+        return not finished
 
     async def _engine_abort(self, request_ids: Iterable[str]):
         if self.engine_use_ray: