Round score
0
weighted tier speedup after failure cap and deduction
Eligible requests
—
trusted baseline exclusions removed
Failed eligible requests
—
each request counted once
Raw weighted speedup
—
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.
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.
Reference engine
The baseline is what candidates are measured against, so it has no speedup of its own. Its SLA replay is below.
16 requests
| Request | TTFT | Total | Tokens | Slowest gap | Stream |
|---|---|---|---|---|---|
| hf-000 | 97.08ms | 66.015s | 5120 | 37.57ms | 1868 gaps for 5120 tokens |
| hf-001 | 89.06ms | 76.354s | 5120 | 78.73ms | 2063 gaps for 5120 tokens |
| hf-002 | 96.72ms | 76.683s |
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 | C4 | 4 | 4 | 3.86 |
| 1 | 142.497s | C4 | 4 | 4 | 3.80 |
| 2 | 0.000s | C4 | 4 | 4 | 3.56 |
| 2 | 152.275s | C4 | 4 | 4 | 3.82 |
| 3 | 0.000s | C4 | 4 | 4 | 3.73 |
| 3 | 151.271s | C4 | 4 | 4 | 3.79 |
| 5120 |
| 58.19ms |
| 2166 gaps for 5120 tokens |
| hf-003 | 128.53ms | 73.097s | 5065 | 59.29ms | 1973 gaps for 5065 tokens |
| hf-004 | 126.15ms | 67.904s | 5120 | 59.47ms | 1834 gaps for 5120 tokens |
| hf-005 | 98.17ms | 74.544s | 5120 | 58.04ms | 2110 gaps for 5120 tokens |
| hf-006 | 95.53ms | 77.222s | 5120 | 58.33ms | 2185 gaps for 5120 tokens |
| hf-007 | 100.15ms | 58.297s | 4556 | 56.58ms | 1649 gaps for 4556 tokens |
| hf-008 | 97.13ms | 67.633s | 5120 | 56.56ms | 1927 gaps for 5120 tokens |
| hf-009 | 133.36ms | 66.981s | 4637 | 40.03ms | 1810 gaps for 4637 tokens |
| hf-010 | 98.13ms | 74.584s | 5120 | 55.76ms | 2158 gaps for 5120 tokens |
| hf-011 | 138.37ms | 72.484s | 4855 | 59.42ms | 1960 gaps for 4855 tokens |
| hf-012 | 136.04ms | 76.437s | 5120 | 59.19ms | 2070 gaps for 5120 tokens |
| hf-013 | 101.87ms | 83.064s | 5120 | 58.62ms | 2293 gaps for 5120 tokens |
| hf-014 | 99.05ms | 72.093s | 5101 | 40.93ms | 1961 gaps for 5101 tokens |
| hf-015 | 96.50ms | 65.421s | 5120 | 38.54ms | 1874 gaps for 5120 tokens |
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-000 | 1 | 0.313855 | 1 | 0 |
| hf-005 | 1 | 0.707523 | 1 | 0 |
| hf-006 | 1 | 0.799884 | 1 | 0 |
| hf-002 | 1 | 0.902649 | 1 | 0 |
| hf-007 | 1 | 0.227869 | 1 | 0 |
| hf-010 | 1 | 0.896142 | 1 | 0 |
| hf-015 | 1 | 0.408634 | 1 | 0 |
| hf-008 | 1 | 0.137534 | 1 | 0 |
| hf-009 | 1 | 0.663781 | 1 | 0 |
| hf-004 | 1 | 0.233423 | 1 | 0 |
| hf-003 | 1 | 0.320146 | 1 | 0 |
| hf-001 | 1 | 0.681613 | 1 | 0 |
| hf-011 | 1 | 0.425865 | 1 | 0 |
| hf-012 | 1 | 0.725123 | 1 | 0 |
| hf-014 | 1 | 0.37922 | 1 | 0 |
| hf-013 | 1 | 0.929007 | 1 | 0 |
| hf-000 | 2 | 0.313855 | 1 | 0 |
| hf-005 | 2 | 0.707523 | 1 | 0 |
| hf-006 | 2 | 0.799884 | 1 | 0 |
| hf-002 | 2 | 0.902649 | 1 | 0 |
| hf-007 | 2 | 0.227869 | 1 | 0 |
| hf-015 | 2 | 0.408634 | 1 | 0 |
| hf-008 | 2 | 0.137534 | 1 | 0 |
| hf-010 | 2 | 0.896142 | 1 | 0 |
| hf-009 | 2 | 0.663781 | 1 | 0 |
| hf-004 | 2 | 0.233423 | 1 | 0 |
| hf-003 | 2 | 0.320146 | 1 | 0 |
| hf-001 | 2 | 0.681613 | 1 | 0 |
| hf-011 | 2 | 0.425865 | 1 | 0 |
| hf-012 | 2 | 0.725123 | 1 | 0 |
| hf-014 | 2 | 0.37922 | 1 | 0 |
| hf-013 | 2 | 0.929007 | 1 | 0 |
| hf-000 | 3 | 0.313855 | 1 | 0 |
| hf-005 | 3 | 0.707523 | 1 | 0 |
| hf-002 | 3 | 0.902649 | 1 | 0 |
| hf-006 | 3 | 0.799884 | 1 | 0 |
| hf-007 | 3 | 0.227869 | 1 | 0 |
| hf-008 | 3 | 0.137534 | 1 | 0 |
| hf-015 | 3 | 0.408634 | 1 | 0 |
| hf-010 | 3 | 0.896142 | 1 | 0 |
| hf-004 | 3 | 0.233423 | 1 | 0 |
| hf-009 | 3 | 0.663781 | 1 | 0 |
| hf-003 | 3 | 0.320146 | 1 | 0 |
| hf-001 | 3 | 0.681613 | 1 | 0 |
| hf-011 | 3 | 0.425865 | 1 | 0 |
| hf-012 | 3 | 0.725123 | 1 | 0 |
| hf-014 | 3 | 0.37922 | 1 | 0 |
| hf-013 | 3 | 0.929007 | 1 | 0 |
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