Round score
0.5985
weighted tier speedup after failure cap and deduction
Eligible requests
16
trusted baseline exclusions removed
Failed eligible requests
1
each request counted once
Raw weighted speedup
+60.48%
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(0.604793, 0) = 0.604793. 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.
16 prompts
| Prompt | Diagnostic speedup | Baseline | Candidate | Aligned output | Diagnostic / failure |
|---|---|---|---|---|---|
| hf-002 | — | — | — | 4970 | Per-token timing unavailable (not a request failure) |
| hf-001 | — | — | — | 4721 | Per-token timing unavailable (not a request failure) |
| hf-005 | — | — | — |
16 requests
| Request | TTFT | Total | Tokens | Slowest gap | Stream |
|---|---|---|---|---|---|
| hf-000 | 82.21ms | 29.674s | 5120 | 42.00ms | 1875 gaps for 5120 tokens |
| hf-001 | 85.95ms | 29.127s | 4721 | 41.96ms | 1842 gaps for 4721 tokens |
| hf-002 | 85.26ms | 26.247s |
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 | 3.67 |
| 1 | 61.209s | C4 | 4 | 4 | 3.65 |
| 2 | 0.000s | C4 | 4 | 4 | 3.78 |
| 2 | 59.114s | C4 | 4 | 4 | 3.78 |
| 3 | 0.000s | C4 | 4 | 4 | 3.71 |
| 3 | 56.860s | C4 | 4 | 4 | 3.91 |
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 |
|---|---|---|---|---|---|---|
| 8k | 8 | 50% | 142.176s | 59.113s | +58.42% | +29.21% |
| 16k | 8 | 50% | 164.590s | 61.663s | +62.54% | +31.27% |
| 4599 |
| Per-token timing unavailable (not a request failure) |
| hf-000 | — | — | — | 5120 | Per-token timing unavailable (not a request failure) |
| hf-006 | — | — | — | 4266 | candidate output below tolerance |
| hf-013 | — | — | — | 4082 | Per-token timing unavailable (not a request failure) |
| hf-015 | — | — | — | 4907 | Per-token timing unavailable (not a request failure) |
| hf-014 | — | — | — | 5120 | Per-token timing unavailable (not a request failure) |
| hf-008 | — | — | — | 4801 | Per-token timing unavailable (not a request failure) |
| hf-007 | — | — | — | 5120 | Per-token timing unavailable (not a request failure) |
| hf-004 | — | — | — | 5033 | Per-token timing unavailable (not a request failure) |
| hf-003 | — | — | — | 4956 | Per-token timing unavailable (not a request failure) |
| hf-009 | — | — | — | 4930 | Per-token timing unavailable (not a request failure) |
| hf-010 | — | — | — | 4801 | Per-token timing unavailable (not a request failure) |
| hf-012 | — | — | — | 4822 | Per-token timing unavailable (not a request failure) |
| hf-011 | — | — | — | 5120 | Per-token timing unavailable (not a request failure) |
| 5120 |
| 17.81ms |
| 1663 gaps for 5120 tokens |
| hf-003 | 112.27ms | 28.978s | 4956 | 57.22ms | 1735 gaps for 4956 tokens |
| hf-004 | 125.45ms | 28.701s | 5033 | 43.70ms | 1720 gaps for 5033 tokens |
| hf-005 | 85.13ms | 26.958s | 5120 | 41.96ms | 1706 gaps for 5120 tokens |
| hf-006 | 74.36ms | 26.264s | 4266 | 40.76ms | 1660 gaps for 4266 tokens |
| hf-007 | 149.33ms | 29.516s | 5120 | 41.36ms | 1766 gaps for 5120 tokens |
| hf-008 | 122.48ms | 27.436s | 4801 | 44.69ms | 1647 gaps for 4801 tokens |
| hf-009 | 93.44ms | 31.739s | 4993 | 46.23ms | 1898 gaps for 4993 tokens |
| hf-010 | 75.18ms | 29.074s | 4801 | 42.66ms | 1748 gaps for 4801 tokens |
| hf-011 | 91.00ms | 31.184s | 5120 | 41.74ms | 1883 gaps for 5120 tokens |
| hf-012 | 91.34ms | 31.120s | 4822 | 19.11ms | 1881 gaps for 4822 tokens |
| hf-013 | 72.83ms | 25.153s | 4286 | 41.44ms | 1592 gaps for 4286 tokens |
| hf-014 | 72.77ms | 29.983s | 5120 | 39.48ms | 1927 gaps for 5120 tokens |
| hf-015 | 72.78ms | 26.791s | 4907 | 18.06ms | 1720 gaps for 4907 tokens |
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 | Baseline char-16 | Drop | Limit | Result |
|---|---|---|---|---|---|
| hf-000 | 0.9414 | 0.9676 | 0.0262 | 0.1000 | Passed |
| hf-001 | 0.9455 | 0.9245 | -0.0211 | 0.1000 | Passed |
| hf-002 | 0.5767 | 0.8694 | 0.2927 | 0.1000 | Failed longest repeated span ratio 0.270 at or above harness ceiling over 24820 chars |
| hf-003 | 0.9484 | 0.9721 | 0.0237 | 0.1000 | Passed |
| hf-004 | 0.9145 | 0.9758 | 0.0613 | 0.1000 | Passed |
| hf-005 | 0.9279 | 0.8465 | -0.0815 | 0.1000 | Passed |
| hf-006 | 0.9491 | 0.8428 | -0.1063 | 0.1000 | Passed |
| hf-007 | 0.9125 | 0.9345 | 0.0220 | 0.1000 | Passed |
| hf-008 | 0.8938 | 0.8052 | -0.0886 | 0.1000 | Passed |
| hf-009 | 0.9590 | 0.8808 | -0.0782 | 0.1000 | Passed |
| hf-010 | 0.9694 | 0.9677 | -0.0017 | 0.1000 | Passed |
| hf-011 | 0.9802 | 0.9814 | 0.0012 | 0.1000 | Passed |
| hf-012 | 0.9401 | 0.7489 | -0.1912 | 0.1000 | Passed |
| hf-013 | 0.9610 | 0.9546 | -0.0064 | 0.1000 | Passed |
| hf-014 | 0.9509 | 0.9718 | 0.0209 | 0.1000 | Passed |
| hf-015 | 0.9217 | 0.9487 | 0.0270 | 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 | 0.667054 | 1 | 0 |
| hf-000 | 1 | 0.789264 | 1 | 0 |
| hf-001 | 1 | 0.260726 | 1 | 0 |
| hf-002 | 1 | 0.42793 | 1 | 0 |
| hf-015 | 1 | 0.504619 | 1 | 0 |
| hf-013 | 1 | 0.572124 | 1 | 0 |
| hf-006 | 1 | 0.232062 | 1 | 0 |
| hf-014 | 1 | 0.9079 | 1 | 0 |
| hf-008 | 1 | 0.503348 | 1 | 0 |
| hf-003 | 1 | 0.737657 | 1 | 0 |
| hf-004 | 1 | 0.823636 | 1 | 0 |
| hf-007 | 1 | 0.857577 | 1 | 0 |
| hf-011 | 1 | 1.000972 | 1 | 0 |
| hf-010 | 1 | 0.897985 | 1 | 0 |
| hf-009 | 1 | 0.293209 | 1 | 0 |
| hf-012 | 1 | 0.298366 | 1 | 0 |
| hf-002 | 2 | 0.42793 | 1 | 0 |
| hf-005 | 2 | 0.667054 | 1 | 0 |
| hf-001 | 2 | 0.260726 | 1 | 0 |
| hf-000 | 2 | 0.789264 | 1 | 0 |
| hf-006 | 2 | 0.232062 | 1 | 0 |
| hf-013 | 2 | 0.572124 | 1 | 0 |
| hf-015 | 2 | 0.504619 | 1 | 0 |
| hf-014 | 2 | 0.9079 | 1 | 0 |
| hf-008 | 2 | 0.503348 | 1 | 0 |
| hf-004 | 2 | 0.823636 | 1 | 0 |
| hf-003 | 2 | 0.737657 | 1 | 0 |
| hf-007 | 2 | 0.857577 | 1 | 0 |
| hf-010 | 2 | 0.897985 | 1 | 0 |
| hf-009 | 2 | 0.293209 | 1 | 0 |
| hf-011 | 2 | 1.000972 | 1 | 0 |
| hf-012 | 2 | 0.298366 | 1 | 0 |
| hf-002 | 3 | 0.42793 | 1 | 0 |
| hf-005 | 3 | 0.667054 | 1 | 0 |
| hf-001 | 3 | 0.260726 | 1 | 0 |
| hf-000 | 3 | 0.789264 | 1 | 0 |
| hf-006 | 3 | 0.232062 | 1 | 0 |
| hf-013 | 3 | 0.572124 | 1 | 0 |
| hf-014 | 3 | 0.9079 | 1 | 0 |
| hf-015 | 3 | 0.504619 | 1 | 0 |
| hf-008 | 3 | 0.503348 | 1 | 0 |
| hf-007 | 3 | 0.857577 | 1 | 0 |
| hf-003 | 3 | 0.737657 | 1 | 0 |
| hf-004 | 3 | 0.823636 | 1 | 0 |
| hf-010 | 3 | 0.897985 | 1 | 0 |
| hf-009 | 3 | 0.293209 | 1 | 0 |
| hf-011 | 3 | 1.000972 | 1 | 0 |
| hf-012 | 3 | 0.298366 | 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