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
[bench] Fix benchmark/serve.py to ignore unavailable results (#22382)
Signed-off-by: Linkun <github@lkchen.net>
lkchen committed
Aug 7, 2025 at 23:15 UTC
808a7b69df479b6b3a16181711cac7ca28a9b941
1 file changed
+1
-1
vllm/benchmarks/serve.py
+1
-1
@@ -665,7 +665,7 @@ def save_to_pytorch_benchmark_format(args: argparse.Namespace,
665
pt_records = convert_to_pytorch_benchmark_format(
666
args=args,
667
metrics={k: [results[k]]
668
- for k in metrics},
668
+ for k in metrics if k in results},
669
extra_info={
670
k: results[k]
671
for k in results if k not in metrics and k not in ignored_metrics