Argonne LLM-Inference-Bench: Gaudi2 x1, DeepSpeed, Llama-2-7B BF16, 128in/128out, batch 1
291.7 tok/s aggregate (batch 1 = single request)
Fields
id
benchmarks/argonne-gaudi2-deepspeed-llama2-7b-bf16-b1-128io-tpsMetric
tps_aggregate
Value
291.7
Unit
tok/s aggregate (batch 1 = single request)
Model
meta-llama/Llama-2-7b-hf
Methodology
THIRD-PARTY MEASURED (Argonne LLM-Inference-Bench, arXiv:2411.00136, All_results.csv). FP16/BF16 weights, 128 in / 128 out. Throughput = batch*(in+out)/end-to-end latency = AGGREGATE over the batch, NOT per-request decode tok/s. Batch 1, 1x Gaudi2 HPU (96GB HBM2e). Latency 0.878s.
Measured by
third_party
Reproducible
no
notes
High batch-1 throughput for a single HPU reflects DeepSpeed's fused kernels on Gaudi2; not comparable to llama.cpp-style single-user decode.
Notes
High batch-1 throughput for a single HPU reflects DeepSpeed's fused kernels on Gaudi2; not comparable to llama.cpp-style single-user decode.