Round score
0.3698
+36.98%
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 | +44.06% | 0.396s | 0.221s | 18 | — |
| hf-000 | +45.97% | 0.740s | 0.400s | 40 | — |
| hf-002 | +26.45% | 0.528s | 0.388s |
32 requests
| Request | TTFT | Total | Tokens | Slowest gap | Stream |
|---|---|---|---|---|---|
| hf-000 | 157.07ms | 0.400s | 40 | 74.06ms | — |
| hf-001 | 151.71ms | 0.221s | 18 | 25.56ms | — |
| hf-002 | 167.96ms | 0.388s |
verdict
evidence
min logprob
num prompts
mean logprob
coverage ratio
quantile logprob
num positions scored
role
| 28 |
| — |
| hf-003 | +51.58% | 0.790s | 0.382s | 43 | — |
| hf-004 | +20.69% | 0.765s | 0.607s | 41 | — |
| hf-005 | +36.76% | 0.933s | 0.590s | 47 | — |
| hf-006 | — | — | — | 18 | candidate output below tolerance |
| hf-007 | +44.52% | 0.778s | 0.431s | 41 | — |
| hf-008 | +44.92% | 0.865s | 0.476s | 42 | — |
| hf-009 | +48.10% | 0.775s | 0.402s | 41 | — |
| hf-010 | +48.80% | 0.746s | 0.382s | 41 | — |
| hf-011 | +49.80% | 0.768s | 0.386s | 46 | — |
| hf-012 | +35.74% | 0.565s | 0.363s | 34 | — |
| hf-013 | +47.72% | 0.739s | 0.386s | 40 | — |
| hf-014 | +30.05% | 0.723s | 0.506s | 42 | — |
| hf-015 | +28.23% | 0.796s | 0.571s | 45 | — |
| hf-017 | -4.42% | 0.486s | 0.508s | 28 | — |
| hf-016 | +37.23% | 0.910s | 0.571s | 51 | — |
| hf-018 | +37.19% | 1.101s | 0.692s | 54 | — |
| hf-019 | +39.81% | 1.002s | 0.603s | 50 | — |
| hf-020 | — | — | — | 21 | candidate output below tolerance |
| hf-022 | +55.67% | 0.612s | 0.271s | 24 | — |
| hf-023 | +56.88% | 1.017s | 0.439s | 46 | — |
| hf-021 | +52.28% | 1.141s | 0.544s | 48 | — |
| hf-024 | — | — | — | 42 | candidate output below tolerance |
| hf-025 | — | — | — | 40 | candidate output below tolerance |
| hf-026 | +10.01% | 0.938s | 0.844s | 42 | — |
| hf-027 | — | — | — | 41 | candidate output below tolerance |
| hf-028 | -16.92% | 0.795s | 0.930s | 40 | — |
| hf-030 | +10.30% | 0.645s | 0.578s | 41 | — |
| hf-029 | +15.09% | 0.913s | 0.776s | 55 | — |
| hf-031 | +47.53% | 0.987s | 0.518s | 75 | — |
| 32 |
| 71.19ms |
| — |
| hf-003 | 142.22ms | 0.382s | 43 | 91.24ms | — |
| hf-004 | 176.46ms | 0.629s | 43 | 136.17ms | — |
| hf-005 | 235.96ms | 0.590s | 47 | 80.29ms | — |
| hf-006 | 158.20ms | 0.230s | 18 | 25.66ms | — |
| hf-007 | 156.73ms | 0.487s | 49 | 95.42ms | — |
| hf-008 | 193.07ms | 0.476s | 42 | 75.64ms | — |
| hf-009 | 149.70ms | 0.423s | 45 | 76.18ms | — |
| hf-010 | 152.86ms | 0.382s | 41 | 75.68ms | — |
| hf-011 | 154.77ms | 0.386s | 46 | 81.26ms | — |
| hf-012 | 159.53ms | 0.387s | 43 | 77.59ms | — |
| hf-013 | 163.24ms | 0.386s | 40 | 78.12ms | — |
| hf-014 | 141.92ms | 0.506s | 42 | 141.61ms | — |
| hf-015 | 236.71ms | 0.571s | 45 | 81.84ms | — |
| hf-016 | 183.97ms | 0.571s | 51 | 152.23ms | — |
| hf-017 | 171.44ms | 0.817s | 74 | 152.35ms | — |
| hf-018 | 261.69ms | 3.458s | 540 | 193.90ms | — |
| hf-019 | 179.02ms | 0.603s | 50 | 91.92ms | — |
| hf-020 | 188.92ms | 0.430s | 21 | 94.43ms | — |
| hf-021 | 177.81ms | 0.544s | 48 | 78.61ms | — |
| hf-022 | 164.15ms | 0.393s | 49 | 74.43ms | — |
| hf-023 | 145.39ms | 0.699s | 73 | 133.33ms | — |
| hf-024 | 227.03ms | 0.709s | 42 | 138.28ms | — |
| hf-025 | 228.76ms | 0.845s | 40 | 111.10ms | — |
| hf-026 | 176.56ms | 1.015s | 51 | 190.29ms | — |
| hf-027 | 223.87ms | 0.970s | 41 | 194.44ms | — |
| hf-028 | 214.31ms | 0.930s | 40 | 193.51ms | — |
| hf-029 | 415.76ms | 0.776s | 56 | 153.86ms | — |
| hf-030 | 372.45ms | 0.578s | 41 | 78.02ms | — |
| hf-031 | 247.45ms | 0.518s | 75 | 29.64ms | — |
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