Qwen/Qwen2.5-72B-Instruct

vllm 0.22.1 on 8x NVIDIA H100 80GB HBM3 (bf16) · seed 42 · run 019ebf79-e413-701e-9f78-f51d447e663f · signed

Headline metrics

TTFT P50 (ms)	45.95
TTFT P99 (ms)	55.79
Throughput (tok/s)	125
$/M tokens	—
J/token	15.92
Power avg (W)	1,995
Power peak (W)	2,200
WER mean	—
J / audio s	—

All metrics

compliance_rate	1.00
energy_joules_total	239,866
joules_per_token	15.92
ok_rate	1.00
power_avg_w	1,995
power_peak_w	2,200
req_per_s_all	1.06
req_per_s_passing	1.06
slo_hardware_class	h100
slo_template_resolved	ttft<200ms, tpot<50ms, total<3000ms
throughput_tok_per_s	125
total_p50_ms	2,122
total_p99_ms	2,166
tpot_p50_ms	16.44
tpot_p99_ms	17.07
ttft_p50_ms	45.95
ttft_p99_ms	55.79

Provenance

Model revision	`unknown00`
Provider	vllm
Hardware fingerprint	`c353425077b20b36c71d9c32d9e5d49096478f4a0f7d6078c1dfc83c408ccf50`
Dataset	sharegpt-v3 · `e6e02b2e7cd84474b14e3713525de678e9975b9c27b76e75074c040229dcdd8b`
Timestamp	2026-06-13T05:34:57.043863+00:00
Envelope JSON	f06278fb9eb4.json

Verify this result

bench verify /bench/leaderboard/envelopes/f06278fb9eb4.json

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