diff --git a/vllm/engine/async_llm_engine.py b/vllm/engine/async_llm_engine.py index c7591945be243..2885aab9f3161 100644 --- a/vllm/engine/async_llm_engine.py +++ b/vllm/engine/async_llm_engine.py @@ -53,7 +53,7 @@ class AsyncStream: self._queue.put_nowait(item) def finish(self) -> None: - self._queue.put_nowait(StopIteration) + self._queue.put_nowait(StopAsyncIteration()) self._finished = True @property @@ -65,9 +65,7 @@ class AsyncStream: async def __anext__(self) -> RequestOutput: result = await self._queue.get() - if result is StopIteration: - raise StopAsyncIteration - elif isinstance(result, Exception): + if isinstance(result, Exception): raise result return result