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 17, 03:00 UTC
[Core] fix del of communicator (#3702)
youkaichao committed
Mar 28, 2024 at 17:24 UTC
0267fef52a7fade4bd5de05caea0bc8fc8418bf7
1 file changed
+3
-1
vllm/model_executor/parallel_utils/pynccl.py
+3
-1
@@ -252,5 +252,7 @@ class NCCLCommunicator:
252
assert result == 0
253
254
def __del__(self):
255
- dist.destroy_process_group()
255
+ # `dist` module might have been already destroyed
256
+ if hasattr(dist, 'destroy_process_group'):
257
+ dist.destroy_process_group()
258
_c_ncclCommDestroy(self.comm)