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, 19:00 UTC
[compile] Improve error message during artifacts load failure. (#35115)
Signed-off-by: zhxchen17 <zhxchen17@fb.com>
Zhengxu Chen committed
Feb 25, 2026 at 01:01 UTC
5cc29cfb8bdf1fadf3abb57556966785a8501285
1 file changed
+7
-2
vllm/compilation/decorators.py
+7
-2
@@ -449,10 +449,15 @@ def _support_torch_compile(
449
self.was_aot_compile_fn_loaded_from_disk = True
450
except Exception as e:
451
if os.path.exists(aot_compilation_path):
452
+ if isinstance(e, EOFError):
453
+ message = "Compile cache file corrupted."
454
+ else:
455
+ message = str(e)
456
logger.warning(
453
- "Cannot load aot compilation from path %s, error: %s",
457
+ "Compiling model again due to a load failure from %s, "
458
+ "reason: %s",
459
aot_compilation_path,
455
- str(e),
460
+ message,
461
)
462
if envs.VLLM_FORCE_AOT_LOAD:
463
raise e