meta-llama/Llama-3.1-70B-Instruct

vllm 0.21.0 on 8x NVIDIA H100 80GB HBM3 (fp16) · seed 42 · run 019e3b91-fedc-74a2-a660-44ccb0c4352b · signed

Headline metrics

TTFT P50 (ms)	46.74
TTFT P99 (ms)	1,333
Throughput (tok/s)	195
$/M tokens	—
J/token	10.07
Power avg (W)	2,003
Power peak (W)	2,153
WER mean	—
J / audio s	—

All metrics

compliance_rate	0.9714
cost_source	registry:groq
cost_usd_per_million_tokens	0.6400
energy_joules_total	43,248
joules_per_token	10.07
ok_rate	1.00
power_avg_w	2,003
power_peak_w	2,153
req_per_s_all	1.59
req_per_s_passing	1.54
slo_hardware_class	h100
slo_template_resolved	ttft<200ms, tpot<50ms, total<3000ms
throughput_tok_per_s	195
total_p50_ms	2,059
total_p99_ms	2,664
tpot_p50_ms	15.86
tpot_p99_ms	16.10
ttft_p50_ms	46.74
ttft_p99_ms	1,333

Provenance

Model revision	`unknown00`
Provider	vllm
Hardware fingerprint	`550474fc9132129654f5d20c316eaffec99a1f67c0cc0b4641e60efb46ea79e7`
Dataset	builtin-chatbot-short · `6f4b1f68fc3a813baa983cbe70cd9ef57f8c86e6b2e6ccc9aaa2a498e588d510`
Timestamp	2026-05-18T14:51:24.252586+00:00
Envelope JSON	1c20a3df24fe.json

Verify this result

bench verify /bench/leaderboard/envelopes/1c20a3df24fe.json

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