- First word afterttft_median_s · s
- 115.22 sper person ·
- Slowest first wordttft_max_s · s
- 115.22 sper person ·
- Machine totalaggregate_decode_tps · tok/s
- 30.59 tok/smachine total · single · sample count not recorded
- Machine total including first-word waitaggregate_e2e_tps · tok/s
- 0.81 tok/smachine total · single · sample count not recorded
- Everyone answered inwall_seconds · s
- 118.33 smachine total · single · sample count not recorded
- Peak memory, whole hostpeak_host_unavailable_gib · GiB
- 34.96 GiBmachine total · max · sample count not recorded
Runs on this setup
Earlier runs keep their own engine versions and outcomes. Workload shapes stay separate.
History and measured sweep
| Date | Shape | Each person · median | Machine total | First word · median | Result | Run |
|---|---|---|---|---|---|---|
2026-10-07 | 5 people · shared · warm | not recorded | 1.09 tok/s | 356.62 s | Failed | |
2026-10-07 | 5 people · shared · cold | not recorded | 1.07 tok/s | 373.11 s | Failed | |
2026-10-07 | 5 people · distinct · cold | not recorded | 0.88 tok/s | 375.66 s | Failed | |
2026-10-07 | 1 people · distinct · cold | not recorded | 31.17 tok/s | 118.04 s | Passed | |
2026-10-06 | 5 people · distinct · cold | not recorded | 0.95 tok/s | 357.17 s | Failed | |
2026-10-06 | 1 people · distinct · cold | not recorded | 30.59 tok/s | 115.22 s | Failed |
Each person in this run
First-word wait against writing speed after that word, including pauses. Earlier streams can pause while other people’s prompts are read. The table preserves filled input, cached-token counts and each person’s checks.
Per-person table · all measured request rows
| Person | Filled input | Cached tokens | Output | First word | Writing after first word | Total time | Checks | Run |
|---|---|---|---|---|---|---|---|---|
1 | 250,016 | 0 | 96 | 115.22 s | 30.59 tok/sone person · includes waits | 118.32 s | 0/3 Hidden facts found · completed · checks missed |
Per-person table · all measured request rows
| Person | Filled input | Cached tokens | Output | First word | Writing after first word | Total time | Checks | Run |
|---|---|---|---|---|---|---|---|---|
1 | 250,016 | 0 | 96 | 115.22 s | 30.59 tok/sone person · includes waits | 118.32 s | 0/3 Hidden facts found · completed · checks missed |
Answer checks
0/3 Hidden facts found0/1 request checks
Method
Synchronized long public-domain prompts, three exact hidden facts per user. Decode rate uses completion tokens minus one over first output through stream completion; effective prefill includes scheduling and first-token latency. Aggregate decode uses the union window from first output to final completion; end-to-end rate includes prefill.
Limits
Lab observation, not customer acceptance. Prompt content length excludes the server chat template; actual per-request usage is recorded. Synthesis is requested and retained but semantic accuracy is unassessed. All requested users must overlap in observed decode to pass.
What did not work
- One or more requests did not meet completion, exact retrieval, synthesis-presence, reasoning-off or token-accounting requirements.
What ran
Exact identity
- Original publisher model
- NVIDIA-Nemotron-3.5-Lightning-30B-A3B
- Publisher revision
a9904d24bcc1d289a1950fa9d2b978c47cf903b9- Variant · format · precision
- Q4_K_M GGUF variant · GGUF · Q4_K_M
- Weights SHA-256
- not recorded
- Engine · version · feature
- llama.cpp CUDA ·
version not recorded· plain - Hardware
- NVIDIA GB10, 128 GB unified memory
- Harness / workload
keep-long-context-concurrency-v1 · 1- Workload suite · version
not recorded- Harness SHA-256
not recorded- Fixture SHA-256
04cf656671acc024178598a26462c2de38b6f97137ea5abf44a09b83b8047c5e
Workload shape
- cache state
- cold
- People in this run
- 1
- Configured window (tokens)
- 262144
- Filled content target (tokens)
- 250000
- fixture variant
- distinct-v1
- KV cache precision
- q8_0
- Output token ceiling
- 512
- scenario
- distinct
- Thinking
- off
Full reproduction pack available to customers and evaluators on request
All recorded numbers
Metric definitions, scope and statistics
| Metric | Value | Scope | Statistic | Sample count | Basis / source field | Run |
|---|---|---|---|---|---|---|
wall seconds | 118.33 s | machine total | single | not recorded | wall_seconds | |
ttft median s | 115.22 s | per person | median | 1 | ttft_median_s | |
hidden facts total | 3.00 count | machine total | single | not recorded | needle_total | |
ttft max s | 115.22 s | per person | max | 1 | ttft_max_s | |
requests ok | 0.00 count | machine total | single | not recorded | requests_ok | |
hidden facts correct | 0.00 count | machine total | single | not recorded | needle_correct | |
latency bar seconds | 300.00 s | machine total | single | not recorded | latency_bar_seconds | |
decode overlap | 1.00 count | machine total | single | not recorded | decode_overlap | |
aggregate concurrency | 1.00 count | machine total | single | not recorded | aggregate_concurrency | |
requests total | 1.00 count | machine total | single | not recorded | requests_total | |
aggregate decode tps | 30.59 tok/s | machine total | single | not recorded | aggregate_tps | |
aggregate e2e tps | 0.81 tok/s | machine total | single | not recorded | aggregate_end_to_end_tps | |
peak host unavailable gib | 34.96 GiB | machine total | max | not recorded | peak_host_unavailable_gib |
Data and citation
Skilak Keep evidence, run 1698e65ee74543aea36ae56d, measured 2026-10-06T16:09:24.389242+00:00. https://skilak.ai/keep/resources/evidence/r/1698e65ee74543aea36ae56d
