Round score
0.0915
weighted tier speedup after failure cap and deduction
Eligible requests
28
trusted baseline exclusions removed
Failed eligible requests
7
each request counted once
Raw weighted speedup
+11.65%
before failure cap and deduction
Temperature range
Request concurrency
Tier scheduling
Output policy
Thinking
Context limit
The same eligible inputs and output ceilings apply to every engine. Candidates must emit at least 90% of baseline tokens per request; natural output lengths can affect tier completion time. Slots refill until the group drains and stay occupied through protocol completion. Baseline exclusions and final drain can reduce actual concurrency below the configured cap.
Raw weighted speedup
Eligible speedup
Request failures cap speed credit at zero: min(0.116504, 0) = 0.116504. Negative speedups remain negative.
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.
28 prompts
| Prompt | Diagnostic speedup | Baseline | Candidate | Aligned output | Diagnostic / failure |
|---|---|---|---|---|---|
| hf-026 | — | — | 76.928s | 2340 | Per-token timing unavailable (not a request failure) |
| hf-001 | — | — | — | 2820 | Per-token timing unavailable (not a request failure) |
| hf-022 | — | — | 67.812s |
28 requests
| Request | TTFT | Total | Tokens | Slowest gap | Stream |
|---|---|---|---|---|---|
| hf-001 | 695.79ms | 76.752s | 5120 | 38.52ms | 2599 gaps for 5120 tokens |
| hf-002 | 629.82ms | 82.028s | 5120 | 100.13ms | 2617 gaps for 5120 tokens |
| hf-003 | 697.42ms |
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 | 28 | 28 | 21.13 |
| 2 | 0.000s | C32 | 28 | 28 | 22.34 |
| 3 | 0.000s | C32 | 28 | 28 | 22.28 |
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 |
|---|
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 |
|---|---|---|---|---|---|---|
| 2k | 6 | 25% | 103.068s | 90.692s | +12.01% | +3.00% |
| 4k | 7 | 25% | 104.854s | 94.131s | +10.23% | +2.56% |
| 8k | 7 | 25% | 101.621s | 87.950s | +13.45% | +3.36% |
| 16k | 8 | 25% | 99.499s | 88.639s | +10.91% | +2.73% |
| 2132 |
| Per-token timing unavailable (not a request failure) |
| hf-002 | — | — | — | 3187 | Per-token timing unavailable (not a request failure) |
| hf-008 | — | — | — | 4750 | Per-token timing unavailable (not a request failure) |
| hf-031 | — | — | — | 1539 | candidate output below tolerance |
| hf-024 | — | — | — | 3211 | candidate output below tolerance |
| hf-009 | — | — | — | 3986 | Per-token timing unavailable (not a request failure) |
| hf-011 | — | — | — | 2611 | candidate output below tolerance |
| hf-010 | — | — | — | 2745 | Per-token timing unavailable (not a request failure) |
| hf-005 | — | — | — | 4448 | Per-token timing unavailable (not a request failure) |
| hf-018 | — | — | — | 4071 | Per-token timing unavailable (not a request failure) |
| hf-006 | — | — | — | 4654 | Per-token timing unavailable (not a request failure) |
| hf-016 | — | — | — | 5120 | Per-token timing unavailable (not a request failure) |
| hf-029 | — | — | — | 2950 | candidate output below tolerance |
| hf-014 | — | — | — | 3582 | candidate output below tolerance |
| hf-020 | — | — | — | 5120 | Per-token timing unavailable (not a request failure) |
| hf-027 | — | — | — | 5120 | Per-token timing unavailable (not a request failure) |
| hf-023 | — | — | — | 3458 | candidate output below tolerance |
| hf-003 | — | — | — | 5120 | Per-token timing unavailable (not a request failure) |
| hf-019 | — | — | — | 4670 | Per-token timing unavailable (not a request failure) |
| hf-015 | — | — | — | 5120 | Per-token timing unavailable (not a request failure) |
| hf-012 | — | — | — | 5120 | Per-token timing unavailable (not a request failure) |
| hf-025 | — | — | — | 4162 | Per-token timing unavailable (not a request failure) |
| hf-021 | — | — | — | 5120 | Per-token timing unavailable (not a request failure) |
| hf-030 | — | — | — | 5120 | Per-token timing unavailable (not a request failure) |
| hf-013 | — | — | — | 4491 | Per-token timing unavailable (not a request failure) |
| hf-007 | — | — | — | 3326 | candidate output below tolerance |
| 80.537s |
| 5120 |
| 38.54ms |
| 2840 gaps for 5120 tokens |
| hf-005 | 692.88ms | 77.887s | 5120 | 38.55ms | 2666 gaps for 5120 tokens |
| hf-006 | 696.79ms | 78.500s | 4697 | 38.20ms | 2341 gaps for 4697 tokens |
| hf-007 | 690.62ms | 62.793s | 3326 | 38.24ms | 1837 gaps for 3326 tokens |
| hf-008 | 693.12ms | 82.301s | 4750 | 38.53ms | 2974 gaps for 4750 tokens |
| hf-009 | 698.63ms | 76.521s | 5120 | 38.54ms | 2586 gaps for 5120 tokens |
| hf-010 | 703.92ms | 61.745s | 2979 | 37.85ms | 1828 gaps for 2979 tokens |
| hf-011 | 703.19ms | 43.648s | 2611 | 37.86ms | 1262 gaps for 2611 tokens |
| hf-012 | 705.23ms | 84.094s | 5120 | 38.84ms | 2764 gaps for 5120 tokens |
| hf-013 | 698.33ms | 86.157s | 4945 | 37.97ms | 2949 gaps for 4945 tokens |
| hf-014 | 701.88ms | 56.736s | 3582 | 37.82ms | 1668 gaps for 3582 tokens |
| hf-015 | 692.68ms | 79.446s | 5120 | 38.53ms | 2766 gaps for 5120 tokens |
| hf-016 | 697.70ms | 84.559s | 5120 | 37.76ms | 2802 gaps for 5120 tokens |
| hf-018 | 690.28ms | 60.437s | 4071 | 38.60ms | 1832 gaps for 4071 tokens |
| hf-019 | 688.18ms | 76.012s | 4670 | 38.49ms | 2559 gaps for 4670 tokens |
| hf-020 | 683.69ms | 79.263s | 5120 | 37.81ms | 2367 gaps for 5120 tokens |
| hf-021 | 689.63ms | 82.890s | 5120 | 38.54ms | 3026 gaps for 5120 tokens |
| hf-022 | 684.89ms | 81.441s | 5120 | 38.48ms | 2907 gaps for 5120 tokens |
| hf-023 | 688.03ms | 60.929s | 3458 | 38.49ms | 1849 gaps for 3458 tokens |
| hf-024 | 683.08ms | 54.766s | 3211 | 37.97ms | 1588 gaps for 3211 tokens |
| hf-025 | 693.77ms | 88.619s | 5120 | 37.82ms | 3213 gaps for 5120 tokens |
| hf-026 | 697.24ms | 81.618s | 4564 | 38.03ms | 2592 gaps for 4564 tokens |
| hf-027 | 694.55ms | 84.849s | 5120 | 38.13ms | 2828 gaps for 5120 tokens |
| hf-029 | 692.76ms | 53.783s | 2950 | 37.84ms | 1575 gaps for 2950 tokens |
| hf-030 | 695.38ms | 87.301s | 5120 | 37.80ms | 3064 gaps for 5120 tokens |
| hf-031 | 697.73ms | 32.049s | 1539 | 38.25ms | 919 gaps for 1539 tokens |
| Baseline char-16 |
|---|
| Drop |
|---|
| Limit |
|---|
| Result |
|---|
| hf-001 | 0.9797 | 0.9203 | -0.0594 | 0.1000 | Passed |
| hf-002 | 0.9562 | 0.9763 | 0.0201 | 0.1000 | Passed |
| hf-003 | 0.9189 | 0.8336 | -0.0853 | 0.1000 | Passed |
| hf-005 | 0.9543 | 0.9704 | 0.0161 | 0.1000 | Passed |
| hf-006 | 0.9530 | 0.9593 | 0.0063 | 0.1000 | Passed |
| hf-007 | 0.9787 | 0.7025 | -0.2762 | 0.1000 | Passed |
| hf-008 | 0.9612 | 0.7725 | -0.1888 | 0.1000 | Passed |
| hf-009 | 0.9502 | 0.9725 | 0.0223 | 0.1000 | Passed |
| hf-010 | 0.9991 | 0.9876 | -0.0114 | 0.1000 | Passed |
| hf-011 | 0.8934 | 0.9432 | 0.0498 | 0.1000 | Passed |
| hf-012 | 0.9538 | 0.9591 | 0.0053 | 0.1000 | Passed |
| hf-013 | 0.9529 | 0.9490 | -0.0039 | 0.1000 | Passed |
| hf-014 | 0.9669 | 0.9603 | -0.0065 | 0.1000 | Passed |
| hf-015 | 0.9574 | 0.9432 | -0.0142 | 0.1000 | Passed |
| hf-016 | 0.9802 | 0.9732 | -0.0070 | 0.1000 | Passed |
| hf-018 | 0.7423 | 0.5820 | -0.1603 | 0.1000 | Passed |
| hf-019 | 0.9633 | 0.9726 | 0.0093 | 0.1000 | Passed |
| hf-020 | 0.6792 | 0.9137 | 0.2345 | 0.1000 | Failed distinct 16-gram ratio 0.679 is more than 0.10 below baseline 0.914 over 7924 chars |
| hf-021 | 0.9607 | 0.9389 | -0.0219 | 0.1000 | Passed |
| hf-022 | 0.9590 | 0.9048 | -0.0542 | 0.1000 | Passed |
| hf-023 | 0.9817 | 0.8942 | -0.0876 | 0.1000 | Passed |
| hf-024 | 0.9860 | 0.9312 | -0.0548 | 0.1000 | Passed |
| hf-025 | 0.9147 | 0.9246 | 0.0100 | 0.1000 | Passed |
| hf-026 | 0.7420 | 0.9627 | 0.2207 | 0.1000 | Failed distinct 16-gram ratio 0.742 is more than 0.10 below baseline 0.963 over 25161 chars |
| hf-027 | 0.9538 | 0.8809 | -0.0729 | 0.1000 | Passed |
| hf-029 | 0.9796 | 0.9088 | -0.0708 | 0.1000 | Passed |
| hf-030 | 0.9685 | 0.9677 | -0.0008 | 0.1000 | Passed |
| hf-031 | 0.9715 | 0.8850 | -0.0865 | 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-011 | 1 | 0.8957 | 1 | 0 |
| hf-007 | 1 | 0.714609 | 1 | 0 |
| hf-031 | 1 | 0.776567 | 1 | 0 |
| hf-002 | 1 | 0.711992 | 1 | 0 |
| hf-027 | 1 | 0.911383 | 1 | 0 |
| hf-010 | 1 | 0.938026 | 1 | 0 |
| hf-024 | 1 | 0.995315 | 1 | 0 |
| hf-029 | 1 | 0.752413 | 1 | 0 |
| hf-013 | 1 | 0.234207 | 1 | 0 |
| hf-018 | 1 | 0.737151 | 1 | 0 |
| hf-023 | 1 | 0.502734 | 1 | 0 |
| hf-006 | 1 | 0.852864 | 1 | 0 |
| hf-019 | 1 | 0.450884 | 1 | 0 |
| hf-020 | 1 | 0.854241 | 1 | 0 |
| hf-009 | 1 | 0.970461 | 1 | 0 |
| hf-001 | 1 | 0.732727 | 1 | 0 |
| hf-005 | 1 | 0.663059 | 1 | 0 |
| hf-014 | 1 | 0.690995 | 1 | 0 |
| hf-016 | 1 | 0.996831 | 1 | 0 |
| hf-015 | 1 | 0.468658 | 1 | 0 |
| hf-003 | 1 | 0.506793 | 1 | 0 |
| hf-022 | 1 | 0.407771 | 1 | 0 |
| hf-012 | 1 | 0.444494 | 1 | 0 |
| hf-008 | 1 | 0.391391 | 1 | 0 |
| hf-021 | 1 | 0.295024 | 1 | 0 |
| hf-026 | 1 | 0.258588 | 1 | 0 |
| hf-030 | 1 | 0.282037 | 1 | 0 |
| hf-025 | 1 | 0.119657 | 1 | 0 |
| hf-001 | 2 | 0.732727 | 1 | 0 |
| hf-026 | 2 | 0.258588 | 1 | 0 |
| hf-018 | 2 | 0.737151 | 1 | 0 |
| hf-014 | 2 | 0.690995 | 1 | 0 |
| hf-024 | 2 | 0.995315 | 1 | 0 |
| hf-011 | 2 | 0.8957 | 1 | 0 |
| hf-007 | 2 | 0.714609 | 1 | 0 |
| hf-029 | 2 | 0.752413 | 1 | 0 |
| hf-010 | 2 | 0.938026 | 1 | 0 |
| hf-006 | 2 | 0.852864 | 1 | 0 |
| hf-020 | 2 | 0.854241 | 1 | 0 |
| hf-023 | 2 | 0.502734 | 1 | 0 |
| hf-012 | 2 | 0.444494 | 1 | 0 |
| hf-009 | 2 | 0.970461 | 1 | 0 |
| hf-003 | 2 | 0.506793 | 1 | 0 |
| hf-002 | 2 | 0.711992 | 1 | 0 |
| hf-019 | 2 | 0.450884 | 1 | 0 |
| hf-021 | 2 | 0.295024 | 1 | 0 |
| hf-015 | 2 | 0.468658 | 1 | 0 |
| hf-016 | 2 | 0.996831 | 1 | 0 |
| hf-027 | 2 | 0.911383 | 1 | 0 |
| hf-031 | 2 | 0.776567 | 1 | 0 |
| hf-022 | 2 | 0.407771 | 1 | 0 |
| hf-013 | 2 | 0.234207 | 1 | 0 |
| hf-008 | 2 | 0.391391 | 1 | 0 |
| hf-030 | 2 | 0.282037 | 1 | 0 |
| hf-025 | 2 | 0.119657 | 1 | 0 |
| hf-005 | 2 | 0.663059 | 1 | 0 |
| hf-009 | 3 | 0.970461 | 1 | 0 |
| hf-031 | 3 | 0.776567 | 1 | 0 |
| hf-023 | 3 | 0.502734 | 1 | 0 |
| hf-011 | 3 | 0.8957 | 1 | 0 |
| hf-003 | 3 | 0.506793 | 1 | 0 |
| hf-029 | 3 | 0.752413 | 1 | 0 |
| hf-014 | 3 | 0.690995 | 1 | 0 |
| hf-010 | 3 | 0.938026 | 1 | 0 |
| hf-018 | 3 | 0.737151 | 1 | 0 |
| hf-005 | 3 | 0.663059 | 1 | 0 |
| hf-019 | 3 | 0.450884 | 1 | 0 |
| hf-015 | 3 | 0.468658 | 1 | 0 |
| hf-021 | 3 | 0.295024 | 1 | 0 |
| hf-007 | 3 | 0.714609 | 1 | 0 |
| hf-024 | 3 | 0.995315 | 1 | 0 |
| hf-008 | 3 | 0.391391 | 1 | 0 |
| hf-022 | 3 | 0.407771 | 1 | 0 |
| hf-026 | 3 | 0.258588 | 1 | 0 |
| hf-002 | 3 | 0.711992 | 1 | 0 |
| hf-006 | 3 | 0.852864 | 1 | 0 |
| hf-001 | 3 | 0.732727 | 1 | 0 |
| hf-020 | 3 | 0.854241 | 1 | 0 |
| hf-012 | 3 | 0.444494 | 1 | 0 |
| hf-016 | 3 | 0.996831 | 1 | 0 |
| hf-027 | 3 | 0.911383 | 1 | 0 |
| hf-013 | 3 | 0.234207 | 1 | 0 |
| hf-030 | 3 | 0.282037 | 1 | 0 |
| hf-025 | 3 | 0.119657 | 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