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, 06:45 UTC
[Fix] Fix comm test (#1691)
Zhuohan Li committed
Nov 16, 2023 at 16:28 UTC
20d0699d49a730661434f8374ba495714a92f953
1 file changed
+2
-1
tests/distributed/test_comm_ops.py
+2
-1
@@ -2,7 +2,7 @@
2
3
Run `pytest tests/distributed/test_comm_ops.py --forked`.
4
"""
5
-from multiprocessing import Process
5
+from multiprocessing import Process, set_start_method
6
7
import pytest
8
import torch
@@ -70,6 +70,7 @@ def all_gather_test_worker(tensor_parallel_size: int, rank: int,
70
@pytest.mark.parametrize("test_target",
71
[all_reduce_test_worker, all_gather_test_worker])
72
def test_multi_process_tensor_parallel(tensor_parallel_size, test_target):
73
+ set_start_method("spawn", force=True)
74
distributed_init_port = get_open_port()
75
processes = []
76
for rank in range(tensor_parallel_size):