[Core] Support reset_prefix_cache (#12284)

vllm/entrypoints/llm.py

@@ -1132,6 +1132,9 @@ class LLM:
     def stop_profile(self) -> None:
         self.llm_engine.stop_profile()
 
+    def reset_prefix_cache(self) -> bool:
+        return self.llm_engine.reset_prefix_cache()
+
     def sleep(self, level: int = 1):
         """
         Put the engine to sleep. The engine should not process any requests.
@@ -1150,6 +1153,7 @@ class LLM:
             where previous model weights are not needed. It reduces CPU memory 
             pressure.
         """
+        self.reset_prefix_cache()
         self.llm_engine.sleep(level=level)
 
     def wake_up(self):

vllm/entrypoints/openai/api_server.py

@@ -518,6 +518,18 @@ TASK_HANDLERS: Dict[str, Dict[str, tuple]] = {
     },
 }
 
+if envs.VLLM_SERVER_DEV_MODE:
+
+    @router.post("/reset_prefix_cache")
+    async def reset_prefix_cache(raw_request: Request):
+        """
+        Reset the prefix cache. Note that we currently do not check if the
+        prefix cache is successfully reset in the API server.
+        """
+        logger.info("Resetting prefix cache...")
+        await engine_client(raw_request).reset_prefix_cache()
+        return Response(status_code=200)
+
 
 @router.post("/invocations")
 async def invocations(raw_request: Request):