inference/research
verifiedmedium · 0.75updated 2026-10-15

Argonne MI250 vLLM Llama-3-70B 4x b1

33.0 tok/s aggregate (batch*(in+out)/lat)

Fields

id
benchmarks/argonne-mi250-vllm-llama3-70b-fp16-4x-b1-128io-tps
Metric
tps_aggregate
Value
33
Unit
tok/s aggregate (batch*(in+out)/lat)
Model
meta-llama/Meta-Llama-3-70B
Methodology
THIRD-PARTY MEASURED (Argonne LLM-Inference-Bench arXiv:2411.00136, All_results.csv). vLLM, FP16/BF16, 128in/128out, AGGREGATE over batch. 4 tiles bs1 lat 7.75s. CSV 'AMD MI250 GPU' = Polaris MI250. 70B needs 4 MI250 tiles.
Measured by
third_party
Reproducible
yes

Cross-references