Argonne LLM-Inference-Bench: MI250 x4 (TP), vLLM 0.6.1, Qwen2-72B FP16, 128in/128out, batch 1
46.0 tok/s aggregate (batch 1, single request over 4x MI250 TP)
Fields
id
benchmarks/argonne-mi250-vllm-qwen2-72b-fp16-4x-b1-128io-tpsMetric
tps_aggregate
Value
46
Unit
tok/s aggregate (batch 1, single request over 4x MI250 TP)
Model
Qwen/Qwen2-72B
Methodology
THIRD-PARTY MEASURED (Argonne LLM-Inference-Bench, arXiv:2411.00136, All_results.csv). vLLM benchmark_throughput.py, FP16/BF16, 128 in / 128 out. Throughput = batch*(in+out)/end-to-end latency = AGGREGATE over the batch, NOT per-request decode. Batch 1, 4x MI250 TP. Latency 5.566s.
Measured by
third_party
Reproducible
yes