Round score
0.6725
+67.25%
Prompts scored
25 / 32
median_e2e_speedup
Below tolerance
7
answered too little to score
Zeroed total
7
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 | +77.25% | 0.793s | 0.180s | 44 | — |
| hf-001 | +75.10% | 0.786s | 0.196s | 43 | — |
| hf-002 | — | — | — |
32 requests
| Request | TTFT | Total | Tokens | Slowest gap | Stream |
|---|---|---|---|---|---|
| hf-000 | 77.40ms | 0.180s | 44 | 22.84ms | — |
| hf-001 | 46.00ms | 0.196s | 43 | 19.59ms | — |
| hf-002 | 42.04ms | 0.112s |
verdict
evidence
min logprob
num prompts
mean logprob
coverage ratio
quantile logprob
num positions scored
role
| 25 |
| candidate output below tolerance |
| hf-003 | +75.10% | 0.780s | 0.194s | 40 | — |
| hf-004 | +79.61% | 0.807s | 0.165s | 41 | — |
| hf-006 | — | — | — | 21 | candidate output below tolerance |
| hf-007 | +79.65% | 0.921s | 0.188s | 41 | — |
| hf-005 | — | — | — | 54 | candidate output below tolerance |
| hf-008 | +70.62% | 0.941s | 0.276s | 42 | — |
| hf-009 | +79.49% | 0.905s | 0.186s | 42 | — |
| hf-010 | — | — | — | 33 | candidate output below tolerance |
| hf-012 | +63.38% | 0.734s | 0.269s | 42 | — |
| hf-011 | — | — | — | 42 | candidate output below tolerance |
| hf-013 | +76.10% | 0.750s | 0.179s | 44 | — |
| hf-014 | +79.21% | 0.773s | 0.161s | 46 | — |
| hf-015 | +70.07% | 0.761s | 0.228s | 41 | — |
| hf-016 | +69.32% | 0.573s | 0.176s | 28 | — |
| hf-017 | +65.88% | 1.108s | 0.378s | 49 | — |
| hf-018 | +59.55% | 0.983s | 0.398s | 45 | — |
| hf-019 | +64.45% | 1.113s | 0.396s | 54 | — |
| hf-020 | +68.03% | 0.986s | 0.315s | 50 | — |
| hf-023 | +28.24% | 0.414s | 0.297s | 23 | — |
| hf-021 | +78.22% | 0.905s | 0.197s | 47 | — |
| hf-022 | +68.68% | 0.870s | 0.273s | 46 | — |
| hf-024 | +64.00% | 0.771s | 0.277s | 42 | — |
| hf-025 | — | — | — | 23 | candidate output below tolerance |
| hf-026 | +66.48% | 1.000s | 0.335s | 54 | — |
| hf-027 | +71.49% | 0.914s | 0.261s | 47 | — |
| hf-028 | — | — | — | 24 | candidate output below tolerance |
| hf-029 | +61.12% | 0.778s | 0.302s | 45 | — |
| hf-030 | +69.96% | 0.674s | 0.203s | 43 | — |
| hf-031 | +52.41% | 0.587s | 0.279s | 42 | — |
| 25 |
| 18.47ms |
| — |
| hf-003 | 45.43ms | 0.194s | 40 | 18.94ms | — |
| hf-004 | 50.37ms | 0.165s | 41 | 19.91ms | — |
| hf-005 | 72.49ms | 0.227s | 54 | 21.23ms | — |
| hf-006 | 110.69ms | 0.271s | 21 | 49.39ms | — |
| hf-007 | 71.29ms | 0.188s | 41 | 19.97ms | — |
| hf-008 | 100.61ms | 0.276s | 42 | 50.32ms | — |
| hf-009 | 75.78ms | 0.186s | 42 | 18.69ms | — |
| hf-010 | 99.83ms | 0.322s | 33 | 50.17ms | — |
| hf-011 | 100.58ms | 0.219s | 42 | 21.34ms | — |
| hf-012 | 97.65ms | 0.269s | 42 | 50.16ms | — |
| hf-013 | 73.00ms | 0.179s | 44 | 18.37ms | — |
| hf-014 | 36.79ms | 0.179s | 48 | 18.46ms | — |
| hf-015 | 120.29ms | 0.228s | 41 | 24.19ms | — |
| hf-016 | 66.92ms | 0.213s | 43 | 18.65ms | — |
| hf-017 | 54.42ms | 0.378s | 49 | 83.59ms | — |
| hf-018 | 153.88ms | 0.398s | 45 | 61.27ms | — |
| hf-019 | 131.56ms | 0.396s | 54 | 81.78ms | — |
| hf-020 | 154.58ms | 0.358s | 65 | 51.54ms | — |
| hf-021 | 78.96ms | 0.596s | 176 | 88.22ms | — |
| hf-022 | 69.23ms | 0.473s | 73 | 93.15ms | — |
| hf-023 | 164.56ms | 0.405s | 45 | 52.52ms | — |
| hf-024 | 100.64ms | 0.277s | 42 | 50.30ms | — |
| hf-025 | 101.53ms | 0.205s | 23 | 23.40ms | — |
| hf-026 | 134.35ms | 0.335s | 54 | 52.06ms | — |
| hf-027 | 80.22ms | 0.261s | 47 | 49.27ms | — |
| hf-028 | 88.61ms | 0.205s | 24 | 20.71ms | — |
| hf-029 | 114.53ms | 0.302s | 45 | 50.13ms | — |
| hf-030 | 102.10ms | 0.203s | 43 | 21.03ms | — |
| hf-031 | 137.17ms | 0.279s | 42 | 27.24ms | — |
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