Round score
0.8169
+81.69%
Prompts scored
23 / 32
median_e2e_speedup
Below tolerance
9
answered too little to score
Zeroed total
9
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 | +76.19% | 0.543s | 0.129s | 31 | — |
| hf-000 | — | — | — | 38 | candidate output below tolerance |
| hf-002 | — | — | — |
32 requests
| Request | TTFT | Total | Tokens | Slowest gap | Stream |
|---|---|---|---|---|---|
| hf-000 | 54.61ms | 0.171s | 38 | 15.00ms | — |
| hf-001 | 40.93ms | 0.129s | 31 | 15.00ms | — |
| hf-002 | 43.29ms | 0.087s |
verdict
evidence
min logprob
num prompts
mean logprob
coverage ratio
quantile logprob
num positions scored
role
| 24 |
| candidate output below tolerance |
| hf-004 | +79.82% | 0.767s | 0.155s | 41 | — |
| hf-003 | +81.81% | 1.151s | 0.209s | 66 | — |
| hf-005 | — | — | — | 43 | candidate output below tolerance |
| hf-007 | +83.99% | 0.970s | 0.155s | 49 | — |
| hf-006 | — | — | — | 18 | candidate output below tolerance |
| hf-008 | — | — | — | 27 | candidate output below tolerance |
| hf-009 | +82.61% | 0.958s | 0.167s | 48 | — |
| hf-010 | +84.47% | 0.925s | 0.144s | 46 | — |
| hf-011 | +82.67% | 0.795s | 0.138s | 40 | — |
| hf-013 | +84.76% | 0.807s | 0.123s | 42 | — |
| hf-015 | +78.66% | 0.516s | 0.110s | 22 | — |
| hf-012 | — | — | — | 43 | candidate output below tolerance |
| hf-014 | +82.91% | 0.812s | 0.139s | 42 | — |
| hf-016 | +84.65% | 0.526s | 0.081s | 24 | — |
| hf-017 | +83.28% | 0.790s | 0.132s | 41 | — |
| hf-018 | +85.64% | 0.900s | 0.129s | 48 | — |
| hf-021 | +78.29% | 0.329s | 0.072s | 17 | — |
| hf-019 | +81.56% | 0.724s | 0.133s | 41 | — |
| hf-020 | +82.64% | 0.715s | 0.124s | 41 | — |
| hf-022 | — | — | — | 42 | candidate output below tolerance |
| hf-023 | — | — | — | 33 | candidate output below tolerance |
| hf-024 | +82.24% | 0.706s | 0.125s | 40 | — |
| hf-025 | +82.20% | 0.772s | 0.137s | 43 | — |
| hf-026 | +82.13% | 0.759s | 0.136s | 42 | — |
| hf-027 | +82.98% | 0.762s | 0.130s | 41 | — |
| hf-028 | +83.88% | 0.798s | 0.129s | 44 | — |
| hf-029 | +78.42% | 0.697s | 0.150s | 42 | — |
| hf-030 | +80.61% | 0.705s | 0.137s | 50 | — |
| hf-031 | — | — | — | 23 | candidate output below tolerance |
| 24 |
| 14.73ms |
| — |
| hf-003 | 47.25ms | 0.209s | 66 | 15.31ms | — |
| hf-004 | 53.41ms | 0.155s | 41 | 14.80ms | — |
| hf-005 | 34.82ms | 0.122s | 43 | 14.65ms | — |
| hf-006 | 43.71ms | 0.073s | 18 | 14.79ms | — |
| hf-007 | 39.00ms | 0.155s | 49 | 15.08ms | — |
| hf-008 | 43.00ms | 0.145s | 27 | 14.71ms | — |
| hf-009 | 47.22ms | 0.167s | 51 | 16.21ms | — |
| hf-010 | 42.01ms | 0.144s | 46 | 14.87ms | — |
| hf-011 | 48.99ms | 0.138s | 40 | 15.15ms | — |
| hf-012 | 37.22ms | 0.124s | 43 | 15.03ms | — |
| hf-013 | 47.78ms | 0.123s | 42 | 15.58ms | — |
| hf-014 | 36.94ms | 0.139s | 42 | 14.66ms | — |
| hf-015 | 37.47ms | 0.154s | 43 | 14.69ms | — |
| hf-016 | 37.93ms | 0.095s | 29 | 14.70ms | — |
| hf-017 | 44.22ms | 0.147s | 46 | 14.87ms | — |
| hf-018 | 41.63ms | 0.129s | 48 | 15.26ms | — |
| hf-019 | 44.78ms | 0.147s | 42 | 15.10ms | — |
| hf-020 | 37.20ms | 0.124s | 41 | 14.71ms | — |
| hf-021 | 42.06ms | 0.072s | 17 | 14.78ms | — |
| hf-022 | 49.78ms | 0.184s | 42 | 15.52ms | — |
| hf-023 | 47.86ms | 0.150s | 33 | 14.76ms | — |
| hf-024 | 37.91ms | 0.125s | 40 | 15.06ms | — |
| hf-025 | 35.65ms | 0.137s | 43 | 14.97ms | — |
| hf-026 | 46.84ms | 0.136s | 42 | 15.58ms | — |
| hf-027 | 41.71ms | 0.130s | 41 | 15.09ms | — |
| hf-028 | 40.81ms | 0.143s | 48 | 14.92ms | — |
| hf-029 | 34.56ms | 0.150s | 42 | 14.85ms | — |
| hf-030 | 35.26ms | 0.137s | 50 | 14.62ms | — |
| hf-031 | 35.91ms | 0.123s | 23 | 15.07ms | — |
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