Fix flush cache (sgl-project#5590)

Author: merrymercy

python/sglang/srt/entrypoints/http_server.py

@@ -25,11 +25,8 @@
 import os
 import threading
 import time
-from ast import Mult
 from http import HTTPStatus
-from typing import AsyncIterator, Callable, Dict, Optional, Union
-
-from sglang.srt.model_executor.model_runner import LocalSerializedTensor
+from typing import AsyncIterator, Callable, Dict, Optional
 
 # Fix a bug of Python threading
 setattr(threading, "_register_atexit", lambda *args, **kwargs: None)
@@ -84,7 +81,6 @@
 from sglang.srt.reasoning_parser import ReasoningParser
 from sglang.srt.server_args import ServerArgs
 from sglang.srt.utils import (
-    MultiprocessingSerializer,
     add_api_key_middleware,
     add_prometheus_middleware,
     delete_directory,

python/sglang/srt/managers/tokenizer_manager.py

@@ -716,7 +716,7 @@ async def _handle_batch_request(
                         pass
 
     async def flush_cache(self) -> FlushCacheReqOutput:
-        return await self.flush_cache_communicator(FlushCacheReqInput())
+        return (await self.flush_cache_communicator(FlushCacheReqInput()))[0]
 
     def abort_request(self, rid: str):
         if rid not in self.rid_to_state: