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 7, 06:30 UTC
[compile] Reduce log spam from compile. (#36044)
Signed-off-by: zhxchen17 <zhxchen17@fb.com>
Zhengxu Chen committed
Mar 4, 2026 at 19:48 UTC
2ed4722e26864a212fbd7a48ae663d97318a8887
1 file changed
+5
-2
vllm/compilation/decorators.py
+5
-2
@@ -582,7 +582,6 @@ def _support_torch_compile(
582
self.aot_compiled_fn and self._aot_compilation_path and self._aot_cache_dir
583
)
584
585
- logger.info("saving AOT compiled function to %s", self._aot_compilation_path)
585
try:
586
os.makedirs(self._aot_cache_dir, exist_ok=True)
587
# File saving should be atomic, so we will save to a temporary location
@@ -590,7 +589,11 @@ def _support_torch_compile(
589
tmp_file = f"{self._aot_compilation_path}.{os.getpid()}.tmp"
590
self.aot_compiled_fn.save_compiled_function(tmp_file)
591
os.replace(tmp_file, self._aot_compilation_path)
593
- logger.info("saved AOT compiled function to %s", self._aot_compilation_path)
592
+ logger.info_once(
593
+ "saved AOT compiled function to %s",
594
+ self._aot_compilation_path,
595
+ scope="local",
596
+ )
597
except Exception as e:
598
logger.warning(
599
"unable to save AOT compiled function to %s: %s",