Round score
0.2513
+25.13%
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-000 | +27.79% | 1.967s | 1.420s | 43 | — |
| hf-001 | +21.48% | 1.816s | 1.426s | 41 | — |
| hf-005 | — | — | — |
32 requests
| Request | TTFT | Total | Tokens | Slowest gap | Stream |
|---|---|---|---|---|---|
| hf-000 | 147.37ms | 1.420s | 43 | 133.42ms | — |
| hf-001 | 167.24ms | 1.426s | 41 | 133.68ms | — |
| hf-002 | 158.31ms | 1.823s |
verdict
evidence
min logprob
num prompts
mean logprob
coverage ratio
quantile logprob
num positions scored
role
| 24 |
| candidate output below tolerance |
| hf-003 | +25.85% | 2.425s | 1.798s | 49 | — |
| hf-007 | +17.11% | 1.397s | 1.158s | 29 | — |
| hf-002 | +24.41% | 2.411s | 1.823s | 52 | — |
| hf-004 | +20.93% | 2.021s | 1.598s | 43 | — |
| hf-006 | +29.61% | 2.232s | 1.571s | 41 | — |
| hf-008 | +34.41% | 2.397s | 1.572s | 42 | — |
| hf-012 | +48.02% | 1.961s | 1.019s | 28 | — |
| hf-009 | +38.62% | 2.589s | 1.589s | 43 | — |
| hf-011 | +44.41% | 2.559s | 1.422s | 40 | — |
| hf-010 | +45.59% | 2.631s | 1.431s | 41 | — |
| hf-016 | +37.69% | 1.586s | 0.988s | 25 | — |
| hf-013 | +34.78% | 2.420s | 1.578s | 42 | — |
| hf-017 | +29.35% | 1.636s | 1.155s | 28 | — |
| hf-018 | +37.11% | 1.633s | 1.027s | 28 | — |
| hf-021 | +29.43% | 1.171s | 0.826s | 23 | — |
| hf-015 | +34.69% | 2.427s | 1.585s | 42 | — |
| hf-022 | +35.36% | 1.576s | 1.019s | 28 | — |
| hf-020 | +21.71% | 2.003s | 1.568s | 41 | — |
| hf-019 | +27.75% | 2.187s | 1.580s | 42 | — |
| hf-025 | +17.27% | 1.463s | 1.210s | 33 | — |
| hf-028 | +7.01% | 0.910s | 0.846s | 24 | — |
| hf-023 | +14.86% | 1.909s | 1.625s | 46 | — |
| hf-027 | — | — | — | 42 | candidate output below tolerance |
| hf-014 | — | — | — | 48 | candidate output below tolerance |
| hf-024 | +9.93% | 1.810s | 1.630s | 46 | — |
| hf-029 | +3.43% | 1.067s | 1.030s | 46 | — |
| hf-031 | +9.19% | 0.807s | 0.733s | 42 | — |
| hf-030 | — | — | — | 42 | candidate output below tolerance |
| hf-026 | — | — | — | 46 | candidate output below tolerance |
| 52 |
| 133.60ms |
| — |
| hf-003 | 172.83ms | 1.798s | 49 | 133.72ms | — |
| hf-004 | 172.94ms | 1.598s | 43 | 133.67ms | — |
| hf-005 | 175.20ms | 0.977s | 24 | 132.93ms | — |
| hf-006 | 180.35ms | 1.585s | 42 | 133.43ms | — |
| hf-007 | 168.91ms | 1.575s | 42 | 133.53ms | — |
| hf-008 | 173.41ms | 1.572s | 42 | 133.14ms | — |
| hf-009 | 178.07ms | 1.589s | 43 | 133.20ms | — |
| hf-010 | 166.37ms | 1.431s | 41 | 132.96ms | — |
| hf-011 | 171.82ms | 1.422s | 40 | 133.11ms | — |
| hf-012 | 174.23ms | 1.019s | 28 | 132.74ms | — |
| hf-013 | 170.82ms | 1.578s | 42 | 132.92ms | — |
| hf-014 | 177.11ms | 1.784s | 48 | 134.07ms | — |
| hf-015 | 171.85ms | 1.585s | 42 | 135.57ms | — |
| hf-016 | 174.86ms | 1.814s | 50 | 133.61ms | — |
| hf-017 | 181.03ms | 1.598s | 43 | 135.41ms | — |
| hf-018 | 168.18ms | 1.027s | 28 | 133.63ms | — |
| hf-019 | 172.54ms | 1.580s | 42 | 134.23ms | — |
| hf-020 | 178.74ms | 1.594s | 43 | 133.67ms | — |
| hf-021 | 165.43ms | 0.960s | 24 | 133.52ms | — |
| hf-022 | 169.44ms | 1.426s | 40 | 133.01ms | — |
| hf-023 | 174.87ms | 2.096s | 71 | 133.32ms | — |
| hf-024 | 171.96ms | 1.869s | 63 | 134.80ms | — |
| hf-025 | 174.29ms | 1.529s | 47 | 132.94ms | — |
| hf-026 | 179.06ms | 1.400s | 46 | 132.64ms | — |
| hf-027 | 167.52ms | 1.215s | 42 | 132.83ms | — |
| hf-028 | 171.92ms | 0.846s | 24 | 131.69ms | — |
| hf-029 | 172.85ms | 1.030s | 46 | 132.86ms | — |
| hf-030 | 170.00ms | 0.856s | 42 | 133.65ms | — |
| hf-031 | 175.40ms | 0.797s | 47 | 16.16ms | — |
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