set VLLM_WORKER_MULTIPROC_METHOD=spawn

vllm/envs.py

@@ -50,7 +50,7 @@ if TYPE_CHECKING:
     VLLM_USE_RAY_COMPILED_DAG: bool = False
     VLLM_USE_RAY_COMPILED_DAG_NCCL_CHANNEL: bool = True
     VLLM_USE_RAY_COMPILED_DAG_OVERLAP_COMM: bool = False
-    VLLM_WORKER_MULTIPROC_METHOD: str = "fork"
+    VLLM_WORKER_MULTIPROC_METHOD: str = "spawn"
     VLLM_ASSETS_CACHE: str = os.path.join(VLLM_CACHE_ROOT, "assets")
     VLLM_IMAGE_FETCH_TIMEOUT: int = 5
     VLLM_VIDEO_FETCH_TIMEOUT: int = 30
@@ -406,7 +406,7 @@ environment_variables: dict[str, Callable[[], Any]] = {
     # Use dedicated multiprocess context for workers.
     # Both spawn and fork work
     "VLLM_WORKER_MULTIPROC_METHOD":
-    lambda: os.getenv("VLLM_WORKER_MULTIPROC_METHOD", "fork"),
+    lambda: os.getenv("VLLM_WORKER_MULTIPROC_METHOD", "spawn"),
 
     # Path to the cache for storing downloaded assets
     "VLLM_ASSETS_CACHE":