[Misc] Copy HF_TOKEN env var to Ray workers (#21406)

Signed-off-by: Rui Qiao <ruisearch42@gmail.com>

vllm/executor/ray_distributed_executor.py

@@ -58,6 +58,9 @@ class RayDistributedExecutor(DistributedExecutorBase):
         "VLLM_HOST_IP", "VLLM_HOST_PORT", "LOCAL_RANK", "CUDA_VISIBLE_DEVICES"
     }
 
+    # These non-vLLM env vars are copied from the driver to workers
+    ADDITIONAL_ENV_VARS = {"HF_TOKEN", "HUGGING_FACE_HUB_TOKEN"}
+
     uses_ray: bool = True
 
     def _init_executor(self) -> None:
@@ -326,7 +329,8 @@ class RayDistributedExecutor(DistributedExecutorBase):
         # Environment variables to copy from driver to workers
         env_vars_to_copy = get_env_vars_to_copy(
             exclude_vars=self.WORKER_SPECIFIC_ENV_VARS,
-            additional_vars=set(current_platform.additional_env_vars),
+            additional_vars=set(current_platform.additional_env_vars).union(
+                self.ADDITIONAL_ENV_VARS),
             destination="workers")
 
         # Copy existing env vars to each worker's args

vllm/ray/ray_env.py

@@ -43,6 +43,8 @@ def get_env_vars_to_copy(exclude_vars: Optional[set[str]] = None,
         exclude_vars: A set of vllm defined environment variables to exclude
             from copying.
         additional_vars: A set of additional environment variables to copy.
+            If a variable is in both exclude_vars and additional_vars, it will
+            be excluded.
         destination: The destination of the environment variables.
     Returns:
         A set of environment variables to copy.
@@ -52,10 +54,9 @@ def get_env_vars_to_copy(exclude_vars: Optional[set[str]] = None,
 
     env_vars_to_copy = {
         v
-        for v in envs.environment_variables
+        for v in set(envs.environment_variables).union(additional_vars)
         if v not in exclude_vars and v not in RAY_NON_CARRY_OVER_ENV_VARS
     }
-    env_vars_to_copy.update(additional_vars)
 
     to_destination = " to " + destination if destination is not None else ""