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 16, 20:45 UTC
[Bugfix] [Easy] Fixed a bug in the multiprocessing GPU executor. (#6770)
Anthony Platanios committed
Jul 26, 2024 at 00:25 UTC
084a01fd3544557990f8af8af6fd3c1185bae848
1 file changed
+4
-2
vllm/executor/multiproc_gpu_executor.py
+4
-2
@@ -1,6 +1,7 @@
1
import asyncio
2
import os
3
import signal
4
+import threading
5
import weakref
6
from functools import partial
7
from typing import Any, List, Optional
@@ -115,8 +116,9 @@ class MultiprocessingGPUExecutor(DistributedGPUExecutor):
116
if executor := ref():
117
executor.shutdown()
118
118
- signal.signal(signal.SIGINT, shutdown)
119
- signal.signal(signal.SIGTERM, shutdown)
119
+ if threading.current_thread() is threading.main_thread():
120
+ signal.signal(signal.SIGINT, shutdown)
121
+ signal.signal(signal.SIGTERM, shutdown)
122
123
self.driver_worker = self._create_worker(
124
distributed_init_method=distributed_init_method)