Round score
0.4051
+40.51%
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 | +54.29% | 0.793s | 0.362s | 44 | — |
| hf-001 | +45.40% | 0.773s | 0.422s | 42 | — |
| hf-002 | — | — | — |
32 requests
| Request | TTFT | Total | Tokens | Slowest gap | Stream |
|---|---|---|---|---|---|
| hf-000 | 159.00ms | 0.362s | 44 | 97.22ms | — |
| hf-001 | 168.72ms | 0.422s | 42 | 86.92ms | — |
| hf-002 | 135.84ms | 0.222s |
verdict
evidence
min logprob
num prompts
mean logprob
coverage ratio
quantile logprob
num positions scored
role
| 25 |
| candidate output below tolerance |
| hf-003 | +45.99% | 0.780s | 0.421s | 40 | — |
| hf-004 | +46.90% | 0.807s | 0.429s | 41 | — |
| hf-006 | -1.46% | 0.790s | 0.801s | 36 | — |
| hf-007 | +33.83% | 0.909s | 0.602s | 40 | — |
| hf-005 | — | — | — | 54 | candidate output below tolerance |
| hf-008 | +36.30% | 0.941s | 0.599s | 42 | — |
| hf-009 | +55.92% | 0.905s | 0.399s | 42 | — |
| hf-010 | +52.48% | 0.881s | 0.419s | 44 | — |
| hf-012 | +42.13% | 0.734s | 0.424s | 42 | — |
| hf-011 | — | — | — | 42 | candidate output below tolerance |
| hf-013 | +46.18% | 0.750s | 0.404s | 44 | — |
| hf-014 | +52.20% | 0.773s | 0.369s | 46 | — |
| hf-015 | +51.30% | 0.761s | 0.370s | 41 | — |
| hf-016 | +34.81% | 0.573s | 0.374s | 28 | — |
| hf-017 | +65.33% | 1.108s | 0.384s | 49 | — |
| hf-018 | +52.91% | 0.983s | 0.463s | 45 | — |
| hf-019 | — | — | — | 45 | candidate output below tolerance |
| hf-020 | — | — | — | 43 | candidate output below tolerance |
| hf-023 | +0.33% | 0.414s | 0.413s | 23 | — |
| hf-021 | +10.13% | 0.905s | 0.813s | 47 | — |
| hf-022 | +22.04% | 0.870s | 0.678s | 46 | — |
| hf-024 | +26.66% | 0.771s | 0.565s | 42 | — |
| hf-025 | +40.76% | 0.770s | 0.456s | 42 | — |
| hf-026 | +43.20% | 1.013s | 0.575s | 55 | — |
| hf-027 | +48.95% | 0.914s | 0.467s | 47 | — |
| hf-028 | — | — | — | 21 | candidate output below tolerance |
| hf-029 | +6.17% | 0.842s | 0.790s | 50 | — |
| hf-030 | +40.25% | 0.674s | 0.403s | 43 | — |
| hf-031 | +45.82% | 0.587s | 0.318s | 42 | — |
| 25 |
| 23.96ms |
| — |
| hf-003 | 151.32ms | 0.421s | 44 | 115.15ms | — |
| hf-004 | 174.45ms | 0.429s | 41 | 116.92ms | — |
| hf-005 | 186.69ms | 0.763s | 54 | 137.69ms | — |
| hf-006 | 205.16ms | 0.801s | 36 | 139.57ms | — |
| hf-007 | 186.75ms | 0.602s | 40 | 139.06ms | — |
| hf-008 | 271.20ms | 0.599s | 43 | 118.26ms | — |
| hf-009 | 189.13ms | 0.399s | 42 | 90.98ms | — |
| hf-010 | 149.08ms | 0.466s | 48 | 110.92ms | — |
| hf-011 | 170.84ms | 0.426s | 42 | 116.63ms | — |
| hf-012 | 181.27ms | 0.424s | 42 | 91.95ms | — |
| hf-013 | 158.88ms | 0.404s | 44 | 92.72ms | — |
| hf-014 | 145.85ms | 0.391s | 48 | 90.97ms | — |
| hf-015 | 144.49ms | 0.370s | 41 | 94.70ms | — |
| hf-016 | 151.61ms | 0.415s | 44 | 96.17ms | — |
| hf-017 | 150.06ms | 0.424s | 65 | 84.84ms | — |
| hf-018 | 149.33ms | 0.463s | 45 | 85.26ms | — |
| hf-019 | 137.22ms | 0.378s | 45 | 88.93ms | — |
| hf-020 | 153.15ms | 0.579s | 43 | 144.63ms | — |
| hf-021 | 234.29ms | 0.813s | 47 | 144.74ms | — |
| hf-022 | 253.61ms | 0.823s | 58 | 118.91ms | — |
| hf-023 | 194.53ms | 0.611s | 48 | 121.73ms | — |
| hf-024 | 189.79ms | 0.565s | 42 | 117.75ms | — |
| hf-025 | 182.84ms | 0.547s | 43 | 90.70ms | — |
| hf-026 | 165.35ms | 0.575s | 55 | 114.63ms | — |
| hf-027 | 151.20ms | 0.467s | 47 | 113.98ms | — |
| hf-028 | 177.62ms | 0.397s | 21 | 85.69ms | — |
| hf-029 | 152.46ms | 0.897s | 85 | 131.97ms | — |
| hf-030 | 189.81ms | 0.403s | 43 | 96.73ms | — |
| hf-031 | 154.30ms | 0.318s | 42 | 24.52ms | — |
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