[Core/Bugfix] pass VLLM_ATTENTION_BACKEND to ray workers (vllm-project#8172)

Author: SolitaryThinker

vllm/executor/ray_gpu_executor.py

@@ -242,6 +242,9 @@ def sort_by_driver_then_worker_ip(worker):
             VLLM_INSTANCE_ID,
             "VLLM_TRACE_FUNCTION":
             str(envs.VLLM_TRACE_FUNCTION),
+            **({
+                "VLLM_ATTENTION_BACKEND": envs.VLLM_ATTENTION_BACKEND
+            } if envs.VLLM_ATTENTION_BACKEND is not None else {})
         }, ) for (node_id, _) in worker_node_and_gpu_ids]
 
         self._env_vars_for_all_workers = (