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 15, 13:30 UTC
Raise error for data-parallel with benchmark_throughput (#16737)
Signed-off-by: Kartik Ramesh <kartikx2000@gmail.com> Co-authored-by: Simon Mo <simon.mo@hey.com>
Kartik Ramesh committed
Apr 21, 2025 at 10:51 UTC
3b34fd5273580942bb573f511a4fc3d2522d67f3
1 file changed
+7
benchmarks/benchmark_throughput.py
+7
@@ -523,6 +523,13 @@ def validate_args(args):
523
raise ValueError(
524
"Tokenizer must be the same as the model for MII backend.")
525
526
+ # --data-parallel is not supported currently.
527
+ # https://github.com/vllm-project/vllm/issues/16222
528
+ if args.data_parallel_size > 1:
529
+ raise ValueError(
530
+ "Data parallel is not supported in offline benchmark, \
531
+ please use benchmark serving instead")
532
+
533
534
if __name__ == "__main__":
535
parser = FlexibleArgumentParser(description="Benchmark the throughput.")