Round score
0.3935
+39.35%
Prompts scored
26 / 32
median_e2e_speedup
Below tolerance
6
answered too little to score
Zeroed total
6
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-000 | +40.18% | 1.967s | 1.177s | 43 | — |
| hf-001 | +34.59% | 1.816s | 1.188s | 41 | — |
| hf-005 | +42.33% | 2.421s | 1.396s |
32 requests
| Request | TTFT | Total | Tokens | Slowest gap | Stream |
|---|---|---|---|---|---|
| hf-000 | 123.09ms | 1.597s | 59 | 107.38ms | — |
| hf-001 | 150.30ms | 1.188s | 41 | 107.68ms | — |
| hf-002 | 149.93ms | 1.535s |
verdict
evidence
min logprob
num prompts
mean logprob
coverage ratio
quantile logprob
num positions scored
role
| 48 |
| — |
| hf-003 | +42.02% | 2.425s | 1.406s | 49 | — |
| hf-007 | +32.52% | 1.397s | 0.942s | 29 | — |
| hf-002 | +36.36% | 2.411s | 1.535s | 52 | — |
| hf-004 | +38.94% | 2.021s | 1.234s | 43 | — |
| hf-006 | +46.13% | 2.232s | 1.202s | 41 | — |
| hf-008 | — | — | — | 21 | candidate output below tolerance |
| hf-012 | +58.06% | 1.961s | 0.822s | 28 | — |
| hf-009 | +52.66% | 2.589s | 1.226s | 43 | — |
| hf-011 | +53.67% | 2.559s | 1.185s | 40 | — |
| hf-010 | +54.10% | 2.631s | 1.208s | 41 | — |
| hf-016 | +51.25% | 1.586s | 0.773s | 25 | — |
| hf-013 | +49.91% | 2.420s | 1.212s | 42 | — |
| hf-017 | +43.13% | 1.636s | 0.930s | 28 | — |
| hf-018 | +42.97% | 1.633s | 0.931s | 28 | — |
| hf-021 | +36.05% | 1.171s | 0.749s | 23 | — |
| hf-015 | +49.80% | 2.427s | 1.219s | 42 | — |
| hf-022 | +41.05% | 1.576s | 0.929s | 28 | — |
| hf-020 | +39.76% | 2.003s | 1.206s | 41 | — |
| hf-019 | +44.17% | 2.187s | 1.221s | 42 | — |
| hf-025 | +32.69% | 1.463s | 0.985s | 33 | — |
| hf-028 | +16.19% | 0.910s | 0.762s | 24 | — |
| hf-023 | +28.24% | 1.909s | 1.370s | 46 | — |
| hf-027 | — | — | — | 43 | candidate output below tolerance |
| hf-014 | — | — | — | 51 | candidate output below tolerance |
| hf-024 | +24.57% | 1.810s | 1.365s | 46 | — |
| hf-029 | — | — | — | 42 | candidate output below tolerance |
| hf-031 | +10.16% | 0.807s | 0.725s | 42 | — |
| hf-030 | — | — | — | 42 | candidate output below tolerance |
| hf-026 | — | — | — | 75 | candidate output below tolerance |
| 52 |
| 108.57ms |
| — |
| hf-003 | 147.25ms | 1.406s | 49 | 107.52ms | — |
| hf-004 | 154.76ms | 1.358s | 45 | 107.66ms | — |
| hf-005 | 149.06ms | 1.534s | 51 | 107.50ms | — |
| hf-006 | 150.53ms | 1.218s | 42 | 109.32ms | — |
| hf-007 | 147.32ms | 0.942s | 29 | 111.19ms | — |
| hf-008 | 153.78ms | 0.635s | 21 | 108.27ms | — |
| hf-009 | 146.33ms | 1.226s | 43 | 108.26ms | — |
| hf-010 | 158.57ms | 1.208s | 41 | 107.81ms | — |
| hf-011 | 153.75ms | 1.185s | 40 | 107.99ms | — |
| hf-012 | 149.82ms | 0.822s | 28 | 107.99ms | — |
| hf-013 | 146.53ms | 1.212s | 42 | 111.48ms | — |
| hf-014 | 153.92ms | 1.538s | 51 | 108.98ms | — |
| hf-015 | 150.04ms | 1.354s | 45 | 108.02ms | — |
| hf-016 | 143.89ms | 1.561s | 53 | 109.44ms | — |
| hf-017 | 158.39ms | 1.230s | 42 | 108.31ms | — |
| hf-018 | 150.82ms | 0.931s | 28 | 112.89ms | — |
| hf-019 | 152.69ms | 1.221s | 42 | 109.15ms | — |
| hf-020 | 150.80ms | 1.222s | 42 | 110.00ms | — |
| hf-021 | 146.97ms | 1.216s | 42 | 109.36ms | — |
| hf-022 | 158.24ms | 0.929s | 28 | 109.44ms | — |
| hf-023 | 152.98ms | 1.370s | 46 | 109.45ms | — |
| hf-024 | 148.26ms | 1.381s | 47 | 109.44ms | — |
| hf-025 | 146.30ms | 1.363s | 46 | 108.07ms | — |
| hf-026 | 141.23ms | 1.677s | 75 | 110.61ms | — |
| hf-027 | 151.66ms | 1.137s | 43 | 107.28ms | — |
| hf-028 | 146.61ms | 1.149s | 51 | 107.20ms | — |
| hf-029 | 143.82ms | 0.933s | 42 | 108.05ms | — |
| hf-030 | 156.02ms | 0.845s | 42 | 107.83ms | — |
| hf-031 | 148.01ms | 0.725s | 42 | 15.18ms | — |
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