InferenceBench

Qwen/Qwen2.5-VL-7B-Instruct

vllm unknown on 8x NVIDIA H100 80GB HBM3 (bf16) · seed 0 · run 019ebfa8-fa25-780a-8f0f-4aea97838607 · signed

Headline metrics

TTFT P50 (ms) 79.79
TTFT P99 (ms)
Throughput (tok/s)
$/M tokens
J/token 33.93
Power avg (W) 625
Power peak (W) 687
WER mean
J / audio s

All metrics

accuracy 0.8000
accuracy_p05 0.2000
accuracy_p50 1.00
accuracy_p95 1.00
energy_joules_total 1,221
joules_per_token 33.93
n_ok 5.00
n_samples 5.00
ok_rate 1.00
power_avg_w 625
power_peak_w 687
tokens_out_total 36.00
total_p50_ms 79.79
ttft_p50_ms 79.79

Provenance

Model revisionunknown00
Providervllm
Hardware fingerprintc353425077b20b36c71d9c32d9e5d49096478f4a0f7d6078c1dfc83c408ccf50
Datasetbuiltin-ocr-mini · 7b1c3328af108dd2aada94d9d3c0c2c7f71b97f740627d481f9c926c2253c160
Timestamp2026-06-13T06:26:22.885810+00:00
Envelope JSONe70f880e40df.json

Verify this result

bench verify /bench/leaderboard/envelopes/e70f880e40df.json

bench verify re-downloads this envelope, recomputes the canonical content hash, and validates the Sigstore signature against the embedded certificate + Rekor entry.