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 14, 05:30 UTC
Optimize logger init performance by using module-level constants (#22373)
Signed-off-by: zitian.zhao <zitian.zhao@tencentmusic.com>
ZiTian.Zhao committed
Aug 7, 2025 at 11:32 UTC
14bcf93a6a59072fd5bc542d0ad73c54546cef5c
1 file changed
+9
-7
vllm/logger.py
+9
-7
@@ -102,6 +102,14 @@ class _VllmLogger(Logger):
102
_print_warning_once(self, msg, *args)
103
104
105
+# Pre-defined methods mapping to avoid repeated dictionary creation
106
+_METHODS_TO_PATCH = {
107
+ "debug_once": _print_debug_once,
108
+ "info_once": _print_info_once,
109
+ "warning_once": _print_warning_once,
110
+}
111
+
112
+
113
def _configure_vllm_root_logger() -> None:
114
logging_config = dict[str, Any]()
115
@@ -144,13 +152,7 @@ def init_logger(name: str) -> _VllmLogger:
152
153
logger = logging.getLogger(name)
154
147
- methods_to_patch = {
148
- "debug_once": _print_debug_once,
149
- "info_once": _print_info_once,
150
- "warning_once": _print_warning_once,
151
- }
152
-
153
- for method_name, method in methods_to_patch.items():
155
+ for method_name, method in _METHODS_TO_PATCH.items():
156
setattr(logger, method_name, MethodType(method, logger))
157
158
return cast(_VllmLogger, logger)