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 9, 21:48 UTC
[Bugfix] Force using spawn multiprocess method when it's the WSL platform (#32749)
Signed-off-by: Yanwen Lin <lyw1124278064@gmail.com>
Yanwen Lin committed
Jan 21, 2026 at 01:35 UTC
6bb2bc71e299c2f0d9ecb1fcebe54aa64e658307
1 file changed
+4
vllm/utils/system_utils.py
+4
@@ -16,6 +16,7 @@ import psutil
16
17
import vllm.envs as envs
18
from vllm.logger import init_logger
19
+from vllm.platforms.interface import in_wsl
20
from vllm.ray.lazy_utils import is_in_ray_actor
21
22
from .platform_utils import cuda_is_initialized, xpu_is_initialized
@@ -132,6 +133,9 @@ def _maybe_force_spawn():
133
elif xpu_is_initialized():
134
reasons.append("XPU is initialized")
135
136
+ if in_wsl():
137
+ reasons.append("WSL is detected and NVML is not compatible with fork")
138
+
139
if reasons:
140
logger.warning(
141
"We must use the `spawn` multiprocessing start method. "