Round score
-1.0562
weighted tier speedup after failure cap and deduction
Eligible requests
29
trusted baseline exclusions removed
Failed eligible requests
2
each request counted once
Raw weighted speedup
-104.93%
before failure cap and deduction
Raw weighted speedup
Eligible speedup
Request failures cap speed credit at zero: min(-1.049288, 0) = -1.049288. Negative speedups remain negative.
Failed requests
Deduction
Final score
Each duration is the median across repetitions, from the group start until the tier's last request finishes. This includes client queueing. Weighted speedup = sum of weight × (1 − candidate / baseline).
| Input tier | Eligible requests | Weight | Baseline | Candidate | Speedup | Contribution |
|---|
The round score uses weighted completion times for all eligible work in each tier. Per-request speedups below are diagnostics, not score contributions. Only trusted baseline exclusions are removed; candidate failures remain in the eligible count and incur the configured failure penalty.
29 prompts
| Prompt | Diagnostic speedup | Baseline | Candidate | Aligned output | Diagnostic / failure |
|---|---|---|---|---|---|
| hf-001 | — | — | — | 5120 | Per-token timing unavailable (not a request failure) |
| hf-030 | — | — | — | 3262 | Per-token timing unavailable (not a request failure) |
| hf-022 | — | — | — |
29 requests
| Request | TTFT | Total | Tokens | Slowest gap | Stream |
|---|---|---|---|---|---|
| hf-000 | 1.200s | 58.151s | 3499 | 15.280s | 2383 gaps for 3499 tokens |
| hf-001 | 8.655s | 67.387s | 5120 | 7.823s | 2830 gaps for 5120 tokens |
| hf-002 | 3.311s | 64.679s |
Occupied client slots include protocol completion and final drain. Mean occupancy is weighted by elapsed time; it can be below the requested cap.
| Repetition | Group start | Requested | Effective cap | Peak | Mean |
|---|---|---|---|---|---|
| 1 | 0.000s | C32 | 29 | 29 | 15.58 |
| 2 | 0.000s | C32 | 29 | 29 | 16.96 |
| 3 | 0.000s | C32 | 29 | 29 | 17.24 |
Each latency-median response is compared with the lowest valid opening-baseline ratio for the same prompt. Drops above the recorded limit fail; absolute checks also apply. These checks detect repetition, not overall writing quality.
| Prompt | Candidate char-16 |
|---|
| 16k | 8 | 25% | 121.650s | 243.656s | -100.29% | -25.07% |
|---|
| 32k | 8 | 25% | 116.561s | 242.630s | -108.16% | -27.04% |
|---|
| 64k | 7 | 25% | 121.318s | 249.665s | -105.79% | -26.45% |
|---|
| 128k | 6 | 25% | 120.339s | 247.264s | -105.47% | -26.37% |
|---|
| 4240 |
| Per-token timing unavailable (not a request failure) |
| hf-006 | — | — | — | 3326 | Per-token timing unavailable (not a request failure) |
| hf-027 | — | — | — | 3420 | Per-token timing unavailable (not a request failure) |
| hf-017 | — | — | — | 5120 | Per-token timing unavailable (not a request failure) |
| hf-016 | — | — | — | 5120 | Per-token timing unavailable (not a request failure) |
| hf-013 | — | — | — | 5120 | Per-token timing unavailable (not a request failure) |
| hf-000 | — | — | — | 3121 | Per-token timing unavailable (not a request failure) |
| hf-021 | — | — | — | 5120 | Per-token timing unavailable (not a request failure) |
| hf-020 | — | — | — | 5120 | Per-token timing unavailable (not a request failure) |
| hf-003 | — | — | — | 5120 | Per-token timing unavailable (not a request failure) |
| hf-005 | — | — | — | 3377 | Per-token timing unavailable (not a request failure) |
| hf-011 | — | — | — | 4306 | Per-token timing unavailable (not a request failure) |
| hf-029 | — | — | — | 5120 | Per-token timing unavailable (not a request failure) |
| hf-009 | — | — | — | 4208 | candidate output below tolerance |
| hf-018 | — | — | — | 5120 | Per-token timing unavailable (not a request failure) |
| hf-019 | — | — | — | 5120 | Per-token timing unavailable (not a request failure) |
| hf-002 | — | — | — | 5120 | Per-token timing unavailable (not a request failure) |
| hf-007 | — | — | — | 5120 | Per-token timing unavailable (not a request failure) |
| hf-023 | — | — | — | 5120 | Per-token timing unavailable (not a request failure) |
| hf-028 | — | — | — | 4636 | Per-token timing unavailable (not a request failure) |
| hf-012 | — | — | — | 5120 | Per-token timing unavailable (not a request failure) |
| hf-026 | — | — | — | 5120 | Per-token timing unavailable (not a request failure) |
| hf-010 | — | — | — | 5120 | Per-token timing unavailable (not a request failure) |
| hf-025 | — | — | — | 5120 | Per-token timing unavailable (not a request failure) |
| hf-031 | — | — | — | 5120 | Per-token timing unavailable (not a request failure) |
| hf-014 | — | — | — | 5120 | Per-token timing unavailable (not a request failure) |
| hf-024 | — | — | — | 4546 | candidate output below tolerance |
| 5120 |
| 13.116s |
| 2726 gaps for 5120 tokens |
| hf-003 | 16.408s | 53.046s | 5120 | 3.076s | 2038 gaps for 5120 tokens |
| hf-005 | 2.603s | 38.466s | 3377 | 13.790s | 1376 gaps for 3377 tokens |
| hf-006 | 16.402s | 48.264s | 5120 | 22.79ms | 1969 gaps for 5120 tokens |
| hf-007 | 9.363s | 64.364s | 5120 | 7.112s | 2821 gaps for 5120 tokens |
| hf-009 | 51.064s | 90.218s | 4208 | 2.411s | 2308 gaps for 4208 tokens |
| hf-010 | 12.232s | 71.027s | 5120 | 4.195s | 2932 gaps for 5120 tokens |
| hf-011 | 69.729s | 122.923s | 5120 | 10.855s | 2599 gaps for 5120 tokens |
| hf-012 | 67.281s | 124.646s | 5120 | 5.992s | 2916 gaps for 5120 tokens |
| hf-013 | 67.292s | 105.041s | 5120 | 5.594s | 2242 gaps for 5120 tokens |
| hf-014 | 55.007s | 107.020s | 5120 | 5.594s | 2967 gaps for 5120 tokens |
| hf-016 | 93.896s | 132.162s | 5120 | 10.855s | 1968 gaps for 5120 tokens |
| hf-017 | 119.398s | 159.102s | 5120 | 6.155s | 1935 gaps for 5120 tokens |
| hf-018 | 122.009s | 176.174s | 5120 | 4.646s | 2963 gaps for 5120 tokens |
| hf-019 | 136.653s | 184.734s | 5120 | 4.646s | 2682 gaps for 5120 tokens |
| hf-020 | 112.637s | 144.741s | 5120 | 2.888s | 1923 gaps for 5120 tokens |
| hf-021 | 158.621s | 208.743s | 5120 | 6.168s | 2709 gaps for 5120 tokens |
| hf-022 | 117.343s | 151.387s | 5120 | 2.071s | 1857 gaps for 5120 tokens |
| hf-023 | 136.926s | 187.089s | 5120 | 5.414s | 2779 gaps for 5120 tokens |
| hf-024 | 162.438s | 215.824s | 4546 | 6.168s | 2827 gaps for 4546 tokens |
| hf-025 | 212.226s | 249.635s | 5120 | 1.281s | 2854 gaps for 5120 tokens |
| hf-026 | 198.680s | 240.751s | 5120 | 3.592s | 2880 gaps for 5120 tokens |
| hf-027 | 213.391s | 242.599s | 5120 | 113.92ms | 2140 gaps for 5120 tokens |
| hf-028 | 137.049s | 203.201s | 5120 | 6.261s | 3056 gaps for 5120 tokens |
| hf-029 | 204.949s | 247.237s | 5120 | 4.249s | 2869 gaps for 5120 tokens |
| hf-030 | 176.139s | 227.995s | 5120 | 6.261s | 2802 gaps for 5120 tokens |
| hf-031 | 199.407s | 243.626s | 5120 | 2.863s | 3198 gaps for 5120 tokens |
| Baseline char-16 |
|---|
| Drop |
|---|
| Limit |
|---|
| Result |
|---|
| hf-000 | 0.8319 | 0.6532 | -0.1787 | 0.1000 | Passed |
| hf-001 | 0.7570 | 0.7253 | -0.0317 | 0.1000 | Passed |
| hf-002 | 0.7000 | 0.6755 | -0.0245 | 0.1000 | Passed |
| hf-003 | 0.6633 | 0.6417 | -0.0216 | 0.1000 | Passed |
| hf-005 | 0.8441 | 0.8205 | -0.0237 | 0.1000 | Passed |
| hf-006 | 0.8295 | 0.9420 | 0.1125 | 0.1000 | Failed distinct 16-gram ratio 0.829 is more than 0.10 below baseline 0.942 over 23562 chars |
| hf-007 | 0.6784 | 0.5829 | -0.0955 | 0.1000 | Passed |
| hf-009 | 0.8774 | 0.7753 | -0.1021 | 0.1000 | Passed |
| hf-010 | 0.7674 | 0.7463 | -0.0211 | 0.1000 | Passed |
| hf-011 | 0.6350 | 0.6971 | 0.0622 | 0.1000 | Passed |
| hf-012 | 0.7960 | 0.7874 | -0.0085 | 0.1000 | Passed |
| hf-013 | 0.8920 | 0.8404 | -0.0516 | 0.1000 | Passed |
| hf-014 | 0.6638 | 0.7632 | 0.0994 | 0.1000 | Passed |
| hf-016 | 0.7997 | 0.8207 | 0.0211 | 0.1000 | Passed |
| hf-017 | 0.7970 | 0.7281 | -0.0690 | 0.1000 | Passed |
| hf-018 | 0.7050 | 0.6505 | -0.0545 | 0.1000 | Passed |
| hf-019 | 0.6095 | 0.7296 | 0.1201 | 0.1000 | Failed distinct 16-gram ratio 0.610 is more than 0.10 below baseline 0.730 over 23767 chars |
| hf-020 | 0.9120 | 0.8612 | -0.0508 | 0.1000 | Passed |
| hf-021 | 0.7861 | 0.7980 | 0.0119 | 0.1000 | Passed |
| hf-022 | 0.8657 | 0.8596 | -0.0060 | 0.1000 | Passed |
| hf-023 | 0.7648 | 0.7329 | -0.0320 | 0.1000 | Passed |
| hf-024 | 0.7063 | 0.7422 | 0.0359 | 0.1000 | Passed |
| hf-025 | 0.6574 | 0.7315 | 0.0741 | 0.1000 | Passed |
| hf-026 | 0.8069 | 0.8155 | 0.0086 | 0.1000 | Passed |
| hf-027 | 0.8256 | 0.5957 | -0.2299 | 0.1000 | Passed |
| hf-028 | 0.5782 | 0.7612 | 0.1830 | 0.1000 | Failed distinct 16-gram ratio 0.578 is more than 0.10 below baseline 0.761 over 25885 chars |
| hf-029 | 0.7326 | 0.7455 | 0.0129 | 0.1000 | Passed |
| hf-030 | 0.7865 | 0.8648 | 0.0783 | 0.1000 | Passed |
| hf-031 | 0.8622 | 0.8597 | -0.0025 | 0.1000 | Passed |
Actual settings for each measured request. Each prompt keeps its temperature across repetitions and engines. Generation uses seed 0 throughout qualification, warmups and measured repetitions; fixed settings do not guarantee identical output text.
| Prompt | Repetition | Temperature | Top p | Seed |
|---|---|---|---|---|
| hf-005 | 1 | 0.760143 | 1 | 0 |
| hf-006 | 1 | 0.8152 | 1 | 0 |
| hf-003 | 1 | 0.740817 | 1 | 0 |
| hf-000 | 1 | 0.101237 | 1 | 0 |
| hf-002 | 1 | 0.426332 | 1 | 0 |
| hf-007 | 1 | 0.412296 | 1 | 0 |
| hf-001 | 1 | 0.444844 | 1 | 0 |
| hf-010 | 1 | 0.319315 | 1 | 0 |
| hf-009 | 1 | 0.602937 | 1 | 0 |
| hf-022 | 1 | 0.985248 | 1 | 0 |
| hf-013 | 1 | 0.726073 | 1 | 0 |
| hf-014 | 1 | 0.248189 | 1 | 0 |
| hf-011 | 1 | 0.548234 | 1 | 0 |
| hf-012 | 1 | 0.350242 | 1 | 0 |
| hf-016 | 1 | 0.940146 | 1 | 0 |
| hf-020 | 1 | 0.921463 | 1 | 0 |
| hf-017 | 1 | 0.913776 | 1 | 0 |
| hf-023 | 1 | 0.485263 | 1 | 0 |
| hf-018 | 1 | 0.221058 | 1 | 0 |
| hf-019 | 1 | 0.382532 | 1 | 0 |
| hf-028 | 1 | 0.165048 | 1 | 0 |
| hf-021 | 1 | 0.504238 | 1 | 0 |
| hf-030 | 1 | 0.476485 | 1 | 0 |
| hf-024 | 1 | 0.121643 | 1 | 0 |
| hf-027 | 1 | 0.901469 | 1 | 0 |
| hf-029 | 1 | 0.316805 | 1 | 0 |
| hf-026 | 1 | 0.240206 | 1 | 0 |
| hf-025 | 1 | 0.298822 | 1 | 0 |
| hf-031 | 1 | 0.106656 | 1 | 0 |
| hf-005 | 2 | 0.760143 | 1 | 0 |
| hf-007 | 2 | 0.412296 | 1 | 0 |
| hf-003 | 2 | 0.740817 | 1 | 0 |
| hf-006 | 2 | 0.8152 | 1 | 0 |
| hf-000 | 2 | 0.101237 | 1 | 0 |
| hf-002 | 2 | 0.426332 | 1 | 0 |
| hf-001 | 2 | 0.444844 | 1 | 0 |
| hf-010 | 2 | 0.319315 | 1 | 0 |
| hf-009 | 2 | 0.602937 | 1 | 0 |
| hf-013 | 2 | 0.726073 | 1 | 0 |
| hf-012 | 2 | 0.350242 | 1 | 0 |
| hf-014 | 2 | 0.248189 | 1 | 0 |
| hf-011 | 2 | 0.548234 | 1 | 0 |
| hf-016 | 2 | 0.940146 | 1 | 0 |
| hf-020 | 2 | 0.921463 | 1 | 0 |
| hf-022 | 2 | 0.985248 | 1 | 0 |
| hf-017 | 2 | 0.913776 | 1 | 0 |
| hf-018 | 2 | 0.221058 | 1 | 0 |
| hf-019 | 2 | 0.382532 | 1 | 0 |
| hf-023 | 2 | 0.485263 | 1 | 0 |
| hf-021 | 2 | 0.504238 | 1 | 0 |
| hf-028 | 2 | 0.165048 | 1 | 0 |
| hf-024 | 2 | 0.121643 | 1 | 0 |
| hf-027 | 2 | 0.901469 | 1 | 0 |
| hf-030 | 2 | 0.476485 | 1 | 0 |
| hf-029 | 2 | 0.316805 | 1 | 0 |
| hf-026 | 2 | 0.240206 | 1 | 0 |
| hf-025 | 2 | 0.298822 | 1 | 0 |
| hf-031 | 2 | 0.106656 | 1 | 0 |
| hf-006 | 3 | 0.8152 | 1 | 0 |
| hf-003 | 3 | 0.740817 | 1 | 0 |
| hf-005 | 3 | 0.760143 | 1 | 0 |
| hf-000 | 3 | 0.101237 | 1 | 0 |
| hf-002 | 3 | 0.426332 | 1 | 0 |
| hf-001 | 3 | 0.444844 | 1 | 0 |
| hf-010 | 3 | 0.319315 | 1 | 0 |
| hf-007 | 3 | 0.412296 | 1 | 0 |
| hf-009 | 3 | 0.602937 | 1 | 0 |
| hf-016 | 3 | 0.940146 | 1 | 0 |
| hf-014 | 3 | 0.248189 | 1 | 0 |
| hf-011 | 3 | 0.548234 | 1 | 0 |
| hf-012 | 3 | 0.350242 | 1 | 0 |
| hf-022 | 3 | 0.985248 | 1 | 0 |
| hf-020 | 3 | 0.921463 | 1 | 0 |
| hf-013 | 3 | 0.726073 | 1 | 0 |
| hf-017 | 3 | 0.913776 | 1 | 0 |
| hf-019 | 3 | 0.382532 | 1 | 0 |
| hf-018 | 3 | 0.221058 | 1 | 0 |
| hf-023 | 3 | 0.485263 | 1 | 0 |
| hf-030 | 3 | 0.476485 | 1 | 0 |
| hf-021 | 3 | 0.504238 | 1 | 0 |
| hf-024 | 3 | 0.121643 | 1 | 0 |
| hf-026 | 3 | 0.240206 | 1 | 0 |
| hf-031 | 3 | 0.106656 | 1 | 0 |
| hf-028 | 3 | 0.165048 | 1 | 0 |
| hf-027 | 3 | 0.901469 | 1 | 0 |
| hf-029 | 3 | 0.316805 | 1 | 0 |
| hf-025 | 3 | 0.298822 | 1 | 0 |
verdict
evidence
min logprob
num prompts
mean logprob
coverage ratio
quantile logprob
num positions scored
role
evidence
request concurrency
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
tier completion max rel range
Image digest
Patch hash