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
Report error log after vllm bench serve (#31808)
Signed-off-by: Elvir Crncevic <elvircrn@gmail.com>
Elvir Crnčević committed
Jan 6, 2026 at 21:24 UTC
dba95378a66884c889b5dc9428ea68285a908658
1 file changed
+6
vllm/benchmarks/serve.py
+6
@@ -412,6 +412,12 @@ def calculate_metrics(
412
# Find the time range across all successful requests
413
successful_outputs = [output for output in outputs if output.success]
414
failed_outputs = [output for output in outputs if not output.success]
415
+
416
+ if len(failed_outputs) > 0:
417
+ print("Failed requests during benchmark run detected (capping to 10):")
418
+ for i, err in enumerate(failed_outputs[:10]):
419
+ print(f"Error {i}: {err.error}")
420
+
421
if successful_outputs:
422
min_start_time = min(output.start_time for output in successful_outputs)
423
max_end_time = max(