EngineX-Hygon/sglang
5
0
Fork 0
Code Issues Pull Requests Actions 7 Projects Releases Wiki Activity

Fix flush cache (#5590)

Browse Source
This commit is contained in:
Lianmin Zheng
2025-04-20 22:56:40 -07:00
committed by GitHub
parent 97cb762bb6
commit eef9433b46
2 changed files with 2 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
python/sglang/srt/entrypoints/http_server.py
View File

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

2
python/sglang/srt/managers/tokenizer_manager.py
View File

@@ -716,7 +716,7 @@ class TokenizerManager:
pass pass
async def flush_cache(self) -> FlushCacheReqOutput: 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): def abort_request(self, rid: str):
if rid not in self.rid_to_state: if rid not in self.rid_to_state: