Round 1 breakdown

scoredbaseline

Round score

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

Workload

Temperature range

0.1 to 1.01 (per prompt)

Request concurrency

C16

Tier scheduling

2k + 4k together, then 8k + 16k

Output policy

Natural EOS · Minimum 90% of baseline tokens per request

Thinking

Disabled

Context limit

1,048,576 tokens

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.

Request trace

13 requests

Request	TTFT	Total	Tokens	Slowest gap	Stream
hf-000	610.70ms	76.263s	4642	35.57ms	2827 gaps for 4642 tokens
hf-001	608.89ms	78.006s	4577	35.34ms	2728 gaps for 4577 tokens
hf-002	681.75ms	66.546s

Observed request concurrency

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	C16	16	16	12.02
2	0.000s	C16	16	16	13.44
3	0.000s	C16	16	16	12.38

Replay sampling

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

Round 1 breakdown

Request

TTFT

Total

Tokens

Slowest gap

Stream

hf-000

610.70ms

76.263s

4642

35.57ms

2827 gaps for 4642 tokens

hf-001

608.89ms

78.006s

4577

35.34ms

2728 gaps for 4577 tokens

hf-002

681.75ms

66.546s

Repetition

Group start

Requested

Effective cap

Peak

Mean

0.000s

C16

12.02

0.000s

C16

13.44

0.000s

C16

12.38

hf-005	1	0.510856	1
hf-015	1	0.683513	1
hf-012	1	0.440298	1
hf-002	1	0.927754	1
hf-004	1	0.800229	1
hf-006	1	0.943683	1
hf-007	1	0.984061	1
hf-010	1	0.868359	1
hf-013	1	0.684286	1
hf-008	1	0.428242	1
hf-003	1	0.581732	1
hf-014	1	0.704503	1
hf-000	1	0.285333	1
hf-001	1	0.241279	1
hf-009	1	0.210266	1
hf-011	1	0.101751	1
hf-010	2	0.868359	1
hf-013	2	0.684286	1
hf-014	2	0.704503	1
hf-004	2	0.800229	1
hf-002	2	0.927754	1
hf-007	2	0.984061	1
hf-006	2	0.943683	1
hf-015	2	0.683513	1
hf-003	2	0.581732	1
hf-005	2	0.510856	1
hf-012	2	0.440298	1
hf-008	2	0.428242	1
hf-001	2	0.241279	1
hf-000	2	0.285333	1
hf-011	2	0.101751	1
hf-009	2	0.210266	1
hf-012	3	0.440298	1
hf-014	3	0.704503	1
hf-000	3	0.285333	1
hf-004	3	0.800229	1
hf-015	3	0.683513	1
hf-013	3	0.684286	1
hf-002	3	0.927754	1
hf-003	3	0.581732	1
hf-007	3	0.984061	1
hf-008	3	0.428242	1
hf-005	3	0.510856	1
hf-010	3	0.868359	1
hf-001	3	0.241279	1
hf-009	3	0.210266	1
hf-011	3	0.101751	1
hf-006	3	0.943683	1