Round score
0.3153
+31.53%
Prompts scored
26 / 32
median_e2e_speedup
Below tolerance
6
answered too little to score
Zeroed total
6
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-003 | +47.25% | 3.983s | 2.101s | 39 | — |
| hf-000 | +55.92% | 4.257s | 1.876s | 42 | — |
| hf-001 | +55.87% | 4.261s | 1.880s |
32 requests
| Request | TTFT | Total | Tokens | Slowest gap | Stream |
|---|---|---|---|---|---|
| hf-000 | 427.89ms | 1.876s | 42 | 237.16ms | — |
| hf-001 | 431.65ms | 1.880s | 42 | 237.66ms | — |
| hf-002 | 433.58ms | 2.300s |
verdict
evidence
min logprob
num prompts
mean logprob
coverage ratio
quantile logprob
num positions scored
role
| 42 |
| — |
| hf-002 | +45.54% | 4.223s | 2.300s | 46 | — |
| hf-005 | +52.55% | 4.446s | 2.110s | 41 | — |
| hf-004 | +50.16% | 4.235s | 2.111s | 40 | — |
| hf-006 | +57.04% | 4.447s | 1.911s | 41 | — |
| hf-013 | +58.20% | 3.705s | 1.549s | 22 | — |
| hf-007 | +50.54% | 5.137s | 2.541s | 50 | — |
| hf-009 | +60.11% | 4.832s | 1.927s | 41 | — |
| hf-011 | +56.48% | 4.472s | 1.946s | 42 | — |
| hf-010 | +54.04% | 4.677s | 2.150s | 44 | — |
| hf-014 | +40.08% | 3.920s | 2.349s | 43 | — |
| hf-015 | +31.32% | 3.735s | 2.565s | 42 | — |
| hf-008 | +31.75% | 5.195s | 3.545s | 59 | — |
| hf-018 | +37.88% | 3.159s | 1.963s | 41 | — |
| hf-027 | +7.41% | 1.367s | 1.266s | 28 | — |
| hf-020 | +28.65% | 2.782s | 1.985s | 41 | — |
| hf-019 | +27.62% | 3.025s | 2.189s | 42 | — |
| hf-017 | — | — | — | 20 | candidate output below tolerance |
| hf-022 | +17.83% | 2.433s | 1.999s | 43 | — |
| hf-012 | +37.99% | 4.488s | 2.783s | 59 | — |
| hf-024 | +11.83% | 2.101s | 1.852s | 42 | — |
| hf-021 | — | — | — | 42 | candidate output below tolerance |
| hf-026 | +15.29% | 1.733s | 1.468s | 43 | — |
| hf-025 | — | — | — | 20 | candidate output below tolerance |
| hf-030 | — | — | — | 25 | candidate output below tolerance |
| hf-028 | +19.66% | 1.356s | 1.089s | 43 | — |
| hf-029 | +23.03% | 1.169s | 0.899s | 43 | — |
| hf-031 | +28.99% | 0.768s | 0.545s | 41 | — |
| hf-023 | — | — | — | 44 | candidate output below tolerance |
| hf-016 | — | — | — | 60 | candidate output below tolerance |
| 48 |
| 236.97ms |
| — |
| hf-003 | 435.73ms | 2.310s | 47 | 237.03ms | — |
| hf-004 | 434.17ms | 2.310s | 42 | 237.60ms | — |
| hf-005 | 436.65ms | 2.110s | 41 | 240.34ms | — |
| hf-006 | 441.13ms | 1.911s | 42 | 238.57ms | — |
| hf-007 | 475.98ms | 2.751s | 56 | 216.66ms | — |
| hf-008 | 591.92ms | 5.035s | 107 | 225.34ms | — |
| hf-009 | 485.83ms | 1.927s | 42 | 208.85ms | — |
| hf-010 | 491.79ms | 2.150s | 44 | 212.67ms | — |
| hf-011 | 500.73ms | 1.946s | 42 | 209.62ms | — |
| hf-012 | 507.91ms | 2.783s | 59 | 215.60ms | — |
| hf-013 | 511.84ms | 1.549s | 22 | 216.79ms | — |
| hf-014 | 662.82ms | 2.349s | 43 | 220.08ms | — |
| hf-015 | 464.86ms | 3.686s | 93 | 219.97ms | — |
| hf-016 | 467.57ms | 2.993s | 60 | 218.06ms | — |
| hf-017 | 476.08ms | 1.318s | 20 | 214.35ms | — |
| hf-018 | 486.26ms | 1.963s | 41 | 216.94ms | — |
| hf-019 | 506.37ms | 2.189s | 42 | 218.46ms | — |
| hf-020 | 510.75ms | 1.985s | 41 | 224.66ms | — |
| hf-021 | 519.77ms | 1.990s | 42 | 214.72ms | — |
| hf-022 | 534.93ms | 2.203s | 45 | 212.31ms | — |
| hf-023 | 584.93ms | 2.139s | 44 | 218.24ms | — |
| hf-024 | 557.23ms | 1.852s | 42 | 213.87ms | — |
| hf-025 | 563.78ms | 1.403s | 20 | 216.87ms | — |
| hf-026 | 575.60ms | 1.468s | 43 | 209.81ms | — |
| hf-027 | 585.96ms | 1.301s | 43 | 208.00ms | — |
| hf-028 | 591.29ms | 1.089s | 43 | 207.84ms | — |
| hf-029 | 604.29ms | 0.899s | 43 | 196.84ms | — |
| hf-030 | 602.55ms | 0.688s | 25 | 18.96ms | — |
| hf-031 | 418.98ms | 0.545s | 42 | 20.79ms | — |
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