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 10, 16:30 UTC
make 500: InternalServerError more informative (#20610)
Signed-off-by: Masataro Asai <guicho2.71828@gmail.com>
Masataro Asai committed
Jan 6, 2026 at 12:36 UTC
142c4d173896b02e6b73e2dd05493c1f180c5977
1 file changed
+11
-3
vllm/v1/engine/async_llm.py
+11
-3
@@ -461,9 +461,9 @@ class AsyncLLM(EngineClient):
461
raise
462
463
# Request validation error.
464
- except ValueError:
464
+ except ValueError as e:
465
if self.log_requests:
466
- logger.info("Request %s failed (bad request).", request_id)
466
+ logger.info("Request %s failed (bad request): %s.", request_id, e)
467
raise
468
469
# Unexpected error in the generate() task (possibly recoverable).
@@ -471,7 +471,15 @@ class AsyncLLM(EngineClient):
471
if q is not None:
472
await self.abort(q.request_id, internal=True)
473
if self.log_requests:
474
- logger.info("Request %s failed.", request_id)
474
+ try:
475
+ s = f"{e.__class__.__name__}: {e}"
476
+ except Exception as e2:
477
+ s = (
478
+ f"{e.__class__.__name__}: "
479
+ + "error during printing an exception of class"
480
+ + e2.__class__.__name__
481
+ )
482
+ logger.info("Request %s failed due to %s.", request_id, s)
483
raise EngineGenerateError() from e
484
485
def _run_output_handler(self):