Argonne LLM-Inference-Bench: MI250 x1, vLLM 0.6.1, Mixtral-8x7B FP16, 128in/128out, batch 1
83.1 tok/s aggregate (batch 1, single request)
Fields
id
benchmarks/argonne-mi250-vllm-mixtral-8x7b-fp16-b1-128io-tpsMetric
tps_aggregate
Value
83.1
Unit
tok/s aggregate (batch 1, single request)
Model
mistralai/Mixtral-8x7B-v0.1
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, 1x MI250. Latency 3.081s.
Measured by
third_party
Reproducible
yes
notes
MoE 8x7B fits in 128GB at FP16; only 2 experts active per token.
Notes
MoE 8x7B fits in 128GB at FP16; only 2 experts active per token.