Round score
-0.0994
-9.94%
Prompts scored
27 / 32
median_e2e_speedup
Below tolerance
5
answered too little to score
Zeroed total
5
scored 0.0 without measuring
The round score is the scoring rule applied to the prompts below. Both engines are compared at the same output token count, so a candidate that stops early cannot buy speed by answering less: a prompt under the tolerance bar scores nothing and shows its gate reason.
32 prompts
| Prompt | Speedup | Baseline | Candidate | Tokens | Gate |
|---|---|---|---|---|---|
| hf-001 | -58.30% | 1.627s | 2.576s | 25 | — |
| hf-000 | -68.50% | 2.748s | 4.631s | 40 | — |
| hf-003 | — | — | — |
32 requests
| Request | TTFT | Total | Tokens | Slowest gap | Stream |
|---|---|---|---|---|---|
| hf-000 | 448.17ms | 5.228s | 48 | 230.39ms | — |
| hf-001 | 471.55ms | 6.823s | 96 | 227.83ms | — |
| hf-002 | 493.62ms | 2.379s |
verdict
evidence
min logprob
num prompts
mean logprob
coverage ratio
quantile logprob
num positions scored
role
| 30 |
| candidate output below tolerance |
| hf-005 | -134.71% | 1.790s | 4.202s | 41 | — |
| hf-004 | -121.52% | 1.991s | 4.410s | 42 | — |
| hf-002 | — | — | — | 25 | candidate output below tolerance |
| hf-006 | -62.62% | 2.468s | 4.014s | 41 | — |
| hf-007 | -59.07% | 3.212s | 5.109s | 43 | — |
| hf-008 | -62.85% | 3.020s | 4.917s | 43 | — |
| hf-011 | -46.88% | 2.958s | 4.345s | 42 | — |
| hf-009 | -44.71% | 3.371s | 4.878s | 53 | — |
| hf-012 | -3.22% | 4.107s | 4.239s | 43 | — |
| hf-013 | -33.23% | 3.019s | 4.022s | 41 | — |
| hf-014 | -10.01% | 3.653s | 4.018s | 48 | — |
| hf-023 | -9.69% | 2.086s | 2.288s | 42 | — |
| hf-017 | -9.37% | 3.101s | 3.391s | 43 | — |
| hf-020 | -10.39% | 2.541s | 2.806s | 41 | — |
| hf-015 | — | — | — | 47 | candidate output below tolerance |
| hf-022 | -11.56% | 2.233s | 2.492s | 46 | — |
| hf-019 | -8.89% | 2.872s | 3.127s | 52 | — |
| hf-024 | -8.95% | 1.920s | 2.092s | 43 | — |
| hf-025 | -9.81% | 1.728s | 1.897s | 43 | — |
| hf-021 | -9.74% | 2.611s | 2.865s | 62 | — |
| hf-027 | -9.88% | 1.434s | 1.575s | 42 | — |
| hf-010 | — | — | — | 42 | candidate output below tolerance |
| hf-018 | -8.29% | 3.266s | 3.537s | 72 | — |
| hf-029 | -10.78% | 1.062s | 1.177s | 40 | — |
| hf-030 | -11.53% | 0.906s | 1.010s | 42 | — |
| hf-026 | -8.66% | 1.737s | 1.887s | 53 | — |
| hf-031 | -13.13% | 0.821s | 0.929s | 52 | — |
| hf-016 | -7.24% | 3.817s | 4.094s | 89 | — |
| hf-028 | — | — | — | 44 | candidate output below tolerance |
| 25 |
| 213.73ms |
| — |
| hf-003 | 506.99ms | 2.452s | 30 | 218.30ms | — |
| hf-004 | 423.18ms | 4.410s | 42 | 234.88ms | — |
| hf-005 | 352.76ms | 4.202s | 41 | 228.10ms | — |
| hf-006 | 163.54ms | 4.014s | 41 | 215.26ms | — |
| hf-007 | 441.57ms | 5.109s | 43 | 219.07ms | — |
| hf-008 | 451.25ms | 4.917s | 43 | 217.68ms | — |
| hf-009 | 464.33ms | 4.928s | 57 | 222.71ms | — |
| hf-010 | 368.19ms | 4.528s | 42 | 220.66ms | — |
| hf-011 | 173.98ms | 4.345s | 42 | 221.43ms | — |
| hf-012 | 366.94ms | 4.359s | 52 | 233.91ms | — |
| hf-013 | 160.06ms | 4.022s | 41 | 220.37ms | — |
| hf-014 | 452.04ms | 4.018s | 48 | 218.34ms | — |
| hf-015 | 459.20ms | 3.820s | 47 | 219.35ms | — |
| hf-016 | 458.87ms | 4.188s | 98 | 228.84ms | — |
| hf-017 | 489.47ms | 3.391s | 43 | 221.87ms | — |
| hf-018 | 480.00ms | 3.537s | 72 | 218.44ms | — |
| hf-019 | 491.14ms | 3.229s | 61 | 223.81ms | — |
| hf-020 | 500.85ms | 2.806s | 41 | 228.02ms | — |
| hf-021 | 444.31ms | 2.865s | 62 | 215.03ms | — |
| hf-022 | 233.96ms | 2.492s | 46 | 224.33ms | — |
| hf-023 | 467.81ms | 2.316s | 44 | 216.57ms | — |
| hf-024 | 382.56ms | 2.092s | 43 | 218.01ms | — |
| hf-025 | 189.36ms | 1.897s | 43 | 214.74ms | — |
| hf-026 | 461.38ms | 1.887s | 53 | 215.57ms | — |
| hf-027 | 468.88ms | 1.575s | 42 | 218.37ms | — |
| hf-028 | 502.09ms | 1.429s | 44 | 214.55ms | — |
| hf-029 | 491.70ms | 1.177s | 40 | 206.44ms | — |
| hf-030 | 507.88ms | 1.065s | 47 | 19.52ms | — |
| hf-031 | 326.82ms | 0.929s | 52 | 22.47ms | — |
evidence
requests per s
sla goodput ratio
output tokens per s
p99 e2e ms rel range
p99 itl ms rel range
p99 ttft ms rel range
Image digest
Patch hash