A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 15, 11:00 UTC
[BugFix] Avoid secondary missing `MultiprocExecutor.workers` error (#17811)
Signed-off-by: Nick Hill <nhill@redhat.com>
Nick Hill committed
May 7, 2025 at 14:55 UTC
ed5272cf217cf81f004523a8e7ec67e65aa489c0
1 file changed
+4
-3
vllm/v1/executor/multiproc_executor.py
+4
-3
@@ -258,9 +258,10 @@ class MultiprocExecutor(Executor):
258
self.io_thread_pool.shutdown(wait=False, cancel_futures=True)
259
self.io_thread_pool = None
260
261
- for w in self.workers:
262
- w.worker_response_mq = None
263
- self._ensure_worker_termination([w.proc for w in self.workers])
261
+ if workers := getattr(self, 'workers', None):
262
+ for w in workers:
263
+ w.worker_response_mq = None
264
+ self._ensure_worker_termination([w.proc for w in workers])
265
266
self.rpc_broadcast_mq = None
267