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 17, 04:15 UTC
allow user chose log level by --log-level instead of fixed 'info'. (#3109)
Co-authored-by: zixiao <shunli.dsl@alibaba-inc.com> Co-authored-by: Simon Mo <simon.mo@hey.com>
Allen.Dou committed
Mar 2, 2024 at 07:28 UTC
29e70e3e88698feca9509cf07fcf06b12163f1c3
3 files changed
+9
-2
vllm/config.py
+1
-1
@@ -317,7 +317,7 @@ class CacheConfig:
317
self.num_cpu_blocks = None
318
319
def metrics_info(self):
320
- # convert cache_config to dict(key: str, value:str) for prometheus metrics info
320
+ # convert cache_config to dict(key: str, value: str) for prometheus metrics info
321
return {key: str(value) for key, value in self.__dict__.items()}
322
323
def _verify_args(self) -> None:
vllm/engine/metrics.py
+1
@@ -23,6 +23,7 @@ class Metrics:
23
if hasattr(collector, "_name") and "vllm" in collector._name:
24
REGISTRY.unregister(collector)
25
26
+ # Config Information
27
self.info_cache_config = Info(
28
name='vllm:cache_config',
29
documentation='information of cache_config')
vllm/entrypoints/openai/api_server.py
+7
-1
@@ -62,6 +62,12 @@ def parse_args():
62
description="vLLM OpenAI-Compatible RESTful API server.")
63
parser.add_argument("--host", type=str, default=None, help="host name")
64
parser.add_argument("--port", type=int, default=8000, help="port number")
65
+ parser.add_argument(
66
+ "--uvicorn-log-level",
67
+ type=str,
68
+ default="info",
69
+ choices=['debug', 'info', 'warning', 'error', 'critical', 'trace'],
70
+ help="log level for uvicorn")
71
parser.add_argument("--allow-credentials",
72
action="store_true",
73
help="allow credentials")
@@ -245,7 +251,7 @@ if __name__ == "__main__":
251
uvicorn.run(app,
252
host=args.host,
253
port=args.port,
248
- log_level="info",
254
+ log_level=args.uvicorn_log_level,
255
timeout_keep_alive=TIMEOUT_KEEP_ALIVE,
256
ssl_keyfile=args.ssl_keyfile,
257
ssl_certfile=args.ssl_certfile)