Round score
0.7872
+78.72%
Prompts scored
24 / 32
median_e2e_speedup
Below tolerance
8
answered too little to score
Zeroed total
8
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 | +81.78% | 0.712s | 0.130s | 42 | — |
| hf-001 | +82.93% | 0.760s | 0.130s | 45 | — |
| hf-003 | — | — | — |
32 requests
| Request | TTFT | Total | Tokens | Slowest gap | Stream |
|---|---|---|---|---|---|
| hf-000 | 41.57ms | 0.130s | 42 | 14.92ms | — |
| hf-001 | 41.59ms | 0.130s | 45 | 15.26ms | — |
| hf-002 | 40.41ms | 0.127s |
verdict
evidence
min logprob
num prompts
mean logprob
coverage ratio
quantile logprob
num positions scored
role
| 29 |
| candidate output below tolerance |
| hf-002 | — | — | — | 51 | candidate output below tolerance |
| hf-004 | +83.58% | 0.730s | 0.120s | 42 | — |
| hf-005 | — | — | — | 21 | candidate output below tolerance |
| hf-006 | +76.30% | 0.694s | 0.164s | 42 | — |
| hf-007 | +75.11% | 0.704s | 0.175s | 46 | — |
| hf-009 | — | — | — | 28 | candidate output below tolerance |
| hf-008 | — | — | — | 48 | candidate output below tolerance |
| hf-010 | +78.95% | 0.605s | 0.127s | 40 | — |
| hf-012 | +74.48% | 0.580s | 0.148s | 37 | — |
| hf-011 | +82.34% | 0.903s | 0.159s | 52 | — |
| hf-013 | +79.58% | 0.497s | 0.101s | 28 | — |
| hf-014 | — | — | — | 31 | candidate output below tolerance |
| hf-015 | +80.24% | 0.717s | 0.142s | 41 | — |
| hf-016 | +79.97% | 0.739s | 0.148s | 43 | — |
| hf-017 | +80.13% | 0.742s | 0.147s | 43 | — |
| hf-018 | +81.93% | 0.704s | 0.127s | 42 | — |
| hf-019 | — | — | — | 48 | candidate output below tolerance |
| hf-020 | +80.05% | 0.662s | 0.132s | 41 | — |
| hf-021 | +81.38% | 0.717s | 0.134s | 47 | — |
| hf-023 | +77.60% | 0.764s | 0.171s | 52 | — |
| hf-022 | +83.05% | 0.687s | 0.116s | 45 | — |
| hf-024 | +78.49% | 0.701s | 0.151s | 46 | — |
| hf-026 | +68.47% | 0.359s | 0.113s | 23 | — |
| hf-025 | +72.53% | 0.591s | 0.162s | 42 | — |
| hf-029 | +72.38% | 0.325s | 0.090s | 20 | — |
| hf-028 | +79.77% | 0.708s | 0.143s | 41 | — |
| hf-027 | +82.43% | 0.812s | 0.143s | 54 | — |
| hf-030 | +79.30% | 0.586s | 0.121s | 41 | — |
| hf-031 | — | — | — | 17 | candidate output below tolerance |
| 51 |
| 14.59ms |
| — |
| hf-003 | 44.92ms | 0.105s | 29 | 15.73ms | — |
| hf-004 | 46.30ms | 0.120s | 42 | 15.04ms | — |
| hf-005 | 42.34ms | 0.087s | 21 | 14.82ms | — |
| hf-006 | 46.21ms | 0.164s | 42 | 15.87ms | — |
| hf-007 | 43.37ms | 0.175s | 46 | 14.88ms | — |
| hf-008 | 39.95ms | 0.186s | 48 | 14.90ms | — |
| hf-009 | 33.29ms | 0.136s | 28 | 14.87ms | — |
| hf-010 | 39.80ms | 0.127s | 40 | 14.90ms | — |
| hf-011 | 42.13ms | 0.174s | 69 | 15.18ms | — |
| hf-012 | 45.57ms | 0.194s | 57 | 15.53ms | — |
| hf-013 | 42.80ms | 0.117s | 42 | 15.46ms | — |
| hf-014 | 32.78ms | 0.135s | 31 | 15.16ms | — |
| hf-015 | 39.57ms | 0.171s | 52 | 17.69ms | — |
| hf-016 | 32.13ms | 0.148s | 43 | 14.83ms | — |
| hf-017 | 31.19ms | 0.147s | 43 | 15.03ms | — |
| hf-018 | 39.60ms | 0.127s | 43 | 15.62ms | — |
| hf-019 | 41.51ms | 0.145s | 48 | 14.94ms | — |
| hf-020 | 30.72ms | 0.132s | 41 | 14.92ms | — |
| hf-021 | 45.76ms | 0.134s | 47 | 14.87ms | — |
| hf-022 | 29.61ms | 0.116s | 48 | 14.76ms | — |
| hf-023 | 39.52ms | 0.171s | 53 | 15.72ms | — |
| hf-024 | 33.73ms | 0.151s | 46 | 14.88ms | — |
| hf-025 | 45.15ms | 0.162s | 42 | 14.82ms | — |
| hf-026 | 39.97ms | 0.143s | 37 | 15.18ms | — |
| hf-027 | 40.77ms | 0.157s | 63 | 15.39ms | — |
| hf-028 | 41.04ms | 0.158s | 48 | 14.92ms | — |
| hf-029 | 45.66ms | 0.148s | 45 | 14.96ms | — |
| hf-030 | 33.41ms | 0.136s | 43 | 15.11ms | — |
| hf-031 | 42.93ms | 0.087s | 17 | 14.83ms | — |
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