A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 16, 09:30 UTC
Revert "[Bugfix] Use host argument to bind to interface (#9798)" (#9852)
Kevin H. Luu committed
Oct 30, 2024 at 15:44 UTC
890ca3607208a10514e65cfdf182bdd4125baef6
2 files changed
+2
-2
vllm/entrypoints/openai/api_server.py
+1
-1
@@ -544,7 +544,7 @@ async def run_server(args, **uvicorn_kwargs) -> None:
544
# This avoids race conditions with ray.
545
# see https://github.com/vllm-project/vllm/issues/8204
546
sock = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
547
- sock.bind((args.host, args.port))
547
+ sock.bind(("", args.port))
548
549
def signal_handler(*_) -> None:
550
# Interrupt server on sigterm while initializing
vllm/entrypoints/openai/cli_args.py
+1
-1
@@ -77,7 +77,7 @@ class PromptAdapterParserAction(argparse.Action):
77
def make_arg_parser(parser: FlexibleArgumentParser) -> FlexibleArgumentParser:
78
parser.add_argument("--host",
79
type=nullable_str,
80
- default="0.0.0.0",
80
+ default=None,
81
help="host name")
82
parser.add_argument("--port", type=int, default=8000, help="port number")
83
parser.add_argument(