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, 05:45 UTC
[Core] Cleanup startup logging a bit (#10961)
Signed-off-by: Russell Bryant <rbryant@redhat.com>
Russell Bryant committed
Dec 6, 2024 at 21:30 UTC
69d357ba125a8c4243c25d7d9162f1c93cfddd1f
3 files changed
+6
-5
vllm/engine/arg_utils.py
+1
@@ -433,6 +433,7 @@ class EngineArgs:
433
'capping to sliding window size')
434
parser.add_argument('--use-v2-block-manager',
435
action='store_true',
436
+ default=True,
437
help='[DEPRECATED] block manager v1 has been '
438
'removed and SelfAttnBlockSpaceManager (i.e. '
439
'block manager v2) is now the default. '
vllm/entrypoints/openai/api_server.py
+4
-4
@@ -175,8 +175,8 @@ async def build_async_engine_client_from_engine_args(
175
176
# Select random path for IPC.
177
ipc_path = get_open_zmq_ipc_path()
178
- logger.info("Multiprocessing frontend to use %s for IPC Path.",
179
- ipc_path)
178
+ logger.debug("Multiprocessing frontend to use %s for IPC Path.",
179
+ ipc_path)
180
181
# Start RPCServer in separate process (holds the LLMEngine).
182
# the current process might have CUDA context,
@@ -249,8 +249,8 @@ def mount_metrics(app: FastAPI):
249
250
prometheus_multiproc_dir_path = os.getenv("PROMETHEUS_MULTIPROC_DIR", None)
251
if prometheus_multiproc_dir_path is not None:
252
- logger.info("vLLM to use %s as PROMETHEUS_MULTIPROC_DIR",
253
- prometheus_multiproc_dir_path)
252
+ logger.debug("vLLM to use %s as PROMETHEUS_MULTIPROC_DIR",
253
+ prometheus_multiproc_dir_path)
254
registry = CollectorRegistry()
255
multiprocess.MultiProcessCollector(registry)
256
vllm/plugins/__init__.py
+1
-1
@@ -57,7 +57,7 @@ def load_general_plugins():
57
58
discovered_plugins = entry_points(group='vllm.general_plugins')
59
if len(discovered_plugins) == 0:
60
- logger.info("No plugins found.")
60
+ logger.debug("No plugins found.")
61
return
62
logger.info("Available plugins:")
63
for plugin in discovered_plugins: