Round score
0.4692
+46.92%
Prompts scored
28 / 32
median_e2e_speedup
Below tolerance
4
answered too little to score
Zeroed total
4
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-003 | +84.02% | 3.983s | 0.636s | 39 | — |
| hf-000 | — | — | — | 28 | candidate output below tolerance |
| hf-001 | +85.16% | 4.261s | 0.633s |
32 requests
| Request | TTFT | Total | Tokens | Slowest gap | Stream |
|---|---|---|---|---|---|
| hf-000 | 225.63ms | 0.779s | 28 | 203.38ms | — |
| hf-001 | 385.83ms | 0.633s | 42 | 145.67ms | — |
| hf-002 | 201.06ms | 0.610s |
verdict
evidence
min logprob
num prompts
mean logprob
coverage ratio
quantile logprob
num positions scored
role
| 42 |
| — |
| hf-002 | +85.55% | 4.223s | 0.610s | 46 | — |
| hf-005 | +77.91% | 4.446s | 0.982s | 41 | — |
| hf-004 | +76.50% | 4.235s | 0.995s | 40 | — |
| hf-006 | +82.39% | 4.447s | 0.783s | 41 | — |
| hf-013 | +78.07% | 3.705s | 0.813s | 22 | — |
| hf-007 | +87.66% | 5.137s | 0.634s | 50 | — |
| hf-009 | +78.95% | 4.832s | 1.017s | 41 | — |
| hf-011 | +72.84% | 4.472s | 1.215s | 42 | — |
| hf-010 | +69.76% | 4.677s | 1.414s | 44 | — |
| hf-014 | +68.15% | 3.920s | 1.249s | 43 | — |
| hf-015 | +60.72% | 3.719s | 1.461s | 41 | — |
| hf-008 | — | — | — | 51 | candidate output below tolerance |
| hf-018 | +47.37% | 3.168s | 1.667s | 42 | — |
| hf-027 | +19.60% | 1.367s | 1.099s | 28 | — |
| hf-020 | +39.83% | 2.782s | 1.674s | 41 | — |
| hf-019 | +37.93% | 3.025s | 1.878s | 42 | — |
| hf-017 | +63.63% | 3.406s | 1.239s | 46 | — |
| hf-022 | +24.40% | 2.433s | 1.839s | 43 | — |
| hf-012 | +72.59% | 4.488s | 1.230s | 59 | — |
| hf-024 | +19.99% | 2.101s | 1.681s | 42 | — |
| hf-021 | — | — | — | 42 | candidate output below tolerance |
| hf-026 | +24.21% | 1.733s | 1.314s | 43 | — |
| hf-025 | +23.80% | 1.944s | 1.481s | 45 | — |
| hf-030 | +39.83% | 0.956s | 0.575s | 41 | — |
| hf-028 | +29.22% | 1.324s | 0.937s | 40 | — |
| hf-029 | +35.16% | 1.169s | 0.758s | 43 | — |
| hf-031 | +46.48% | 0.768s | 0.411s | 41 | — |
| hf-023 | +21.99% | 2.386s | 1.862s | 51 | — |
| hf-016 | — | — | — | 49 | candidate output below tolerance |
| 48 |
| 144.84ms |
| — |
| hf-003 | 200.48ms | 0.986s | 47 | 206.33ms | — |
| hf-004 | 195.16ms | 5.082s | 227 | 218.91ms | — |
| hf-005 | 441.88ms | 1.000s | 45 | 145.08ms | — |
| hf-006 | 386.34ms | 0.800s | 47 | 146.63ms | — |
| hf-007 | 195.00ms | 0.841s | 56 | 207.09ms | — |
| hf-008 | 193.71ms | 1.058s | 51 | 211.94ms | — |
| hf-009 | 199.36ms | 1.017s | 42 | 212.16ms | — |
| hf-010 | 452.05ms | 1.414s | 44 | 210.29ms | — |
| hf-011 | 459.19ms | 1.215s | 42 | 208.93ms | — |
| hf-012 | 398.26ms | 1.230s | 59 | 209.06ms | — |
| hf-013 | 216.08ms | 0.813s | 22 | 207.69ms | — |
| hf-014 | 450.99ms | 1.249s | 43 | 214.18ms | — |
| hf-015 | 394.01ms | 1.461s | 41 | 215.51ms | — |
| hf-016 | 210.34ms | 1.401s | 49 | 212.44ms | — |
| hf-017 | 204.32ms | 1.239s | 46 | 219.63ms | — |
| hf-018 | 456.74ms | 1.667s | 42 | 214.96ms | — |
| hf-019 | 475.29ms | 1.878s | 42 | 211.69ms | — |
| hf-020 | 480.94ms | 1.674s | 41 | 208.81ms | — |
| hf-021 | 422.58ms | 1.678s | 42 | 214.81ms | — |
| hf-022 | 241.65ms | 1.839s | 43 | 212.32ms | — |
| hf-023 | 450.83ms | 1.862s | 51 | 212.84ms | — |
| hf-024 | 461.80ms | 1.756s | 59 | 206.83ms | — |
| hf-025 | 471.83ms | 1.481s | 45 | 204.94ms | — |
| hf-026 | 472.71ms | 1.314s | 43 | 209.66ms | — |
| hf-027 | 479.39ms | 1.138s | 43 | 203.96ms | — |
| hf-028 | 483.13ms | 0.937s | 40 | 202.58ms | — |
| hf-029 | 421.56ms | 0.758s | 43 | 204.15ms | — |
| hf-030 | 238.85ms | 0.594s | 50 | 205.43ms | — |
| hf-031 | 260.96ms | 0.411s | 41 | 21.96ms | — |
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