diff --git a/vllm/utils/system_utils.py b/vllm/utils/system_utils.py
index aa14b3951..98632c542 100644
--- a/vllm/utils/system_utils.py
+++ b/vllm/utils/system_utils.py
@@ -16,6 +16,7 @@ import psutil
 
 import vllm.envs as envs
 from vllm.logger import init_logger
+from vllm.platforms.interface import in_wsl
 from vllm.ray.lazy_utils import is_in_ray_actor
 
 from .platform_utils import cuda_is_initialized, xpu_is_initialized
@@ -132,6 +133,9 @@ def _maybe_force_spawn():
     elif xpu_is_initialized():
         reasons.append("XPU is initialized")
 
+    if in_wsl():
+        reasons.append("WSL is detected and NVML is not compatible with fork")
+
     if reasons:
         logger.warning(
             "We must use the `spawn` multiprocessing start method. "