xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-06-05 17:42:15 +08:00
Code Issues Packages Projects Releases Wiki Activity

run format.sh

Browse Source 
Signed-off-by: clark <panf2333@gmail.com>
This commit is contained in:
clark 2025-01-12 21:33:46 +08:00
parent 27c1afe88b
commit 8fa9df7987
2 changed files with 5 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
benchmarks/disagg_benchmarks/zmq/test_request.py
View File

@ -61,6 +61,7 @@ async def test_connect_completions(session):
except aiohttp.ClientError as e: except aiohttp.ClientError as e:
print(f"Error: {e}") print(f"Error: {e}")
def is_json(data): def is_json(data):
try: try:
json.loads(data) json.loads(data)
@ -68,6 +69,7 @@ def is_json(data):
except ValueError: except ValueError:
return False return False
def extract_data(responseText): def extract_data(responseText):
reply = "" reply = ""
if responseText == "": if responseText == "":

4
vllm/entrypoints/disagg_connector.py
View File

@ -113,6 +113,7 @@ async def generate_stream_response(fisrt_reply: str,
async for _, reply in generator: async for _, reply in generator:
yield reply yield reply
async def prefill(route: str, header: dict, original_request_data: dict): async def prefill(route: str, header: dict, original_request_data: dict):
logger.info("start prefill") logger.info("start prefill")
generator = execute_task_async(route, header, original_request_data, generator = execute_task_async(route, header, original_request_data,
@ -124,7 +125,8 @@ async def prefill(route: str, header: dict, original_request_data: dict):
response.status_code = 500 response.status_code = 500
return response return response
return True return True
async def decode(route: str, header: dict, original_request_data: dict): async def decode(route: str, header: dict, original_request_data: dict):
logger.info("start decode") logger.info("start decode")
generator = execute_task_async(route, header, original_request_data, generator = execute_task_async(route, header, original_request_data,