[BugFix]: fix engine timeout due to request abort (vllm-project#6255)

Signed-off-by: yatta zhang <[email protected]> Signed-off-by: zhangyuntao.dev <[email protected]> Co-authored-by: zhangyuntao.dev <[email protected]>
adityagoel14 · Jul 11, 2024 · 546b101 · 546b101
1 parent 3963a53
commit 546b101
Showing 1 changed file with 3 additions and 1 deletion.
diff --git a/vllm/engine/async_llm_engine.py b/vllm/engine/async_llm_engine.py
@@ -553,11 +553,13 @@ async def engine_step(self, virtual_engine: int) -> bool:
             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: