[Core] Support fully transparent sleep mode (#11743)

Signed-off-by: youkaichao <youkaichao@gmail.com>

vllm/executor/executor_base.py

@@ -193,6 +193,17 @@ class ExecutorBase(ABC):
     def stop_profile(self) -> None:
         self.collective_rpc("stop_profile")
 
+    def sleep(self, level: int = 1):
+        if self.cache_config.enable_prefix_caching:
+            # TODO: support sleep with prefix caching
+            # by resetting the prefix cache state,
+            # after https://github.com/vllm-project/vllm/pull/12284
+            raise ValueError("Cannot sleep when prefix caching is enabled.")
+        self.collective_rpc("sleep", kwargs=dict(level=level))
+
+    def wake_up(self):
+        self.collective_rpc("wake_up")
+
     def save_sharded_state(
         self,
         path: str,