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 15, 11:00 UTC
Only log non-default CLI args for online serving (#17803)
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Harry Mellor committed
May 8, 2025 at 06:33 UTC
998eea4a0eaceb40bb3597c0d40ce2cf729dd1fa
2 files changed
+15
-2
vllm/entrypoints/openai/api_server.py
+3
-2
@@ -41,7 +41,8 @@ from vllm.entrypoints.chat_utils import (load_chat_template,
41
resolve_mistral_chat_template)
42
from vllm.entrypoints.launcher import serve_http
43
from vllm.entrypoints.logger import RequestLogger
44
-from vllm.entrypoints.openai.cli_args import (make_arg_parser,
44
+from vllm.entrypoints.openai.cli_args import (log_non_default_args,
45
+ make_arg_parser,
46
validate_parsed_serve_args)
47
# yapf conflicts with isort for this block
48
# yapf: disable
@@ -1040,7 +1041,7 @@ def create_server_socket(addr: tuple[str, int]) -> socket.socket:
1041
1042
async def run_server(args, **uvicorn_kwargs) -> None:
1043
logger.info("vLLM API server version %s", VLLM_VERSION)
1043
- logger.info("args: %s", args)
1044
+ log_non_default_args(args)
1045
1046
if args.tool_parser_plugin and len(args.tool_parser_plugin) > 3:
1047
ToolParserManager.import_tool_parser(args.tool_parser_plugin)
vllm/entrypoints/openai/cli_args.py
+12
@@ -17,8 +17,11 @@ from vllm.entrypoints.chat_utils import (ChatTemplateContentFormatOption,
17
from vllm.entrypoints.openai.serving_models import (LoRAModulePath,
18
PromptAdapterPath)
19
from vllm.entrypoints.openai.tool_parsers import ToolParserManager
20
+from vllm.logger import init_logger
21
from vllm.utils import FlexibleArgumentParser
22
23
+logger = init_logger(__name__)
24
+
25
26
class LoRAParserAction(argparse.Action):
27
@@ -285,6 +288,15 @@ def validate_parsed_serve_args(args: argparse.Namespace):
288
"--tool-call-parser")
289
290
291
+def log_non_default_args(args: argparse.Namespace):
292
+ non_default_args = {}
293
+ parser = make_arg_parser(FlexibleArgumentParser())
294
+ for arg, default in vars(parser.parse_args([])).items():
295
+ if default != getattr(args, arg):
296
+ non_default_args[arg] = getattr(args, arg)
297
+ logger.info("non-default args: %s", non_default_args)
298
+
299
+
300
def create_parser_for_docs() -> FlexibleArgumentParser:
301
parser_for_docs = FlexibleArgumentParser(
302
prog="-m vllm.entrypoints.openai.api_server")