Report · resources

Rolling chat

Qwen3.8-27B · INT4 variant · NVIDIA NGC vLLM 0.29

Measured results / Rolling chat / Run 861767d1989c495e851fd421

Measured 2026-10-07 22:28 UTC · Record date 2026-10-07

Passed. 105/105 turns

Runs on this setup

Earlier runs keep their own engine versions and outcomes. Workload shapes stay separate.

History and measured sweep
History of this setup, newest first
DateShapeEach person · medianMachine totalFirst word · medianResultRun
2026-10-07
5 people
not recorded
not recorded
not recorded
Passed
2026-10-07
5 people
not recorded
not recorded
not recorded
Failed
2026-10-07
1 people
not recorded
not recorded
not recorded
Failed

Answer checks

105/105 turns

Rolling-chat turn checks
All rolling-chat turns
Person · turnFilled inputFirst wordRecent · pinned · oldResultRun
1 · 1
20208
21.30 s
true · true · recalled
Passed
1 · 2
20380
21.61 s
true · true · recalled
Passed
1 · 3
20381
34.67 s
true · true · recalled
Passed
1 · 4
20380
27.18 s
true · true · recalled
Passed
1 · 5
20381
36.00 s
true · true · recalled
Passed
1 · 6
20375
38.84 s
true · true · recalled
Passed
1 · 7
20382
40.54 s
true · true · recalled
Passed
1 · 8
20381
61.33 s
true · true · recalled
Passed
1 · 9
20382
49.72 s
true · true · recalled
Passed
1 · 10
20383
50.01 s
true · true · recalled
Passed
1 · 11
20380
51.15 s
true · true · recalled
Passed
1 · 12
20383
66.67 s
true · true · recalled
Passed
1 · 13
20379
70.47 s
true · true · recalled
Passed
1 · 14
20384
79.82 s
true · true · recalled
Passed
1 · 15
20380
67.96 s
true · true · recalled
Passed
1 · 16
20383
63.04 s
true · true · recalled
Passed
1 · 17
20381
52.81 s
true · true · recalled
Passed
1 · 18
20379
51.72 s
true · true · recalled
Passed
1 · 19
20381
57.44 s
true · true · recalled
Passed
1 · 20
20380
58.21 s
true · true · recalled
Passed
1 · 21
20378
52.64 s
true · true · recalled
Passed
2 · 1
20215
36.15 s
true · true · recalled
Passed
2 · 2
20383
28.93 s
true · true · recalled
Passed
2 · 3
20384
45.36 s
true · true · recalled
Passed
2 · 4
20382
40.25 s
true · true · recalled
Passed
2 · 5
20378
41.78 s
true · true · recalled
Passed
2 · 6
20384
61.36 s
true · true · recalled
Passed
2 · 7
20382
51.83 s
true · true · recalled
Passed
2 · 8
20383
48.16 s
true · true · recalled
Passed
2 · 9
20384
52.92 s
true · true · recalled
Passed
2 · 10
20380
69.10 s
true · true · recalled
Passed
2 · 11
20383
70.50 s
true · true · recalled
Passed
2 · 12
20387
83.86 s
true · true · recalled
Passed
2 · 13
20380
65.89 s
true · true · recalled
Passed
2 · 14
20382
22.15 s
true · true · recalled
Passed
2 · 15
20383
25.95 s
true · true · recalled
Passed
2 · 16
20382
32.49 s
true · true · recalled
Passed
2 · 17
20383
22.20 s
true · true · recalled
Passed
2 · 18
20383
23.04 s
true · true · recalled
Passed
2 · 19
20385
30.96 s
true · true · recalled
Passed
2 · 20
20383
42.50 s
true · true · recalled
Passed
2 · 21
20384
42.54 s
true · true · recalled
Passed
3 · 1
20213
32.83 s
true · true · recalled
Passed
3 · 2
20381
21.98 s
true · true · recalled
Passed
3 · 3
20378
22.00 s
true · true · recalled
Passed
3 · 4
20381
42.37 s
true · true · recalled
Passed
3 · 5
20382
25.12 s
true · true · recalled
Passed
3 · 6
20383
34.90 s
true · true · recalled
Passed
3 · 7
20381
29.66 s
true · true · recalled
Passed
3 · 8
20383
41.85 s
true · true · recalled
Passed
3 · 9
20383
46.81 s
true · true · recalled
Passed
3 · 10
20378
79.62 s
true · true · recalled
Passed
3 · 11
20379
71.31 s
true · true · recalled
Passed
3 · 12
20379
63.08 s
true · true · recalled
Passed
3 · 13
20383
54.75 s
true · true · recalled
Passed
3 · 14
20382
51.76 s
true · true · recalled
Passed
3 · 15
20380
59.44 s
true · true · recalled
Passed
3 · 16
20383
55.95 s
true · true · recalled
Passed
3 · 17
20381
54.56 s
true · true · recalled
Passed
3 · 18
20380
42.43 s
true · true · recalled
Passed
3 · 19
20382
42.41 s
true · true · recalled
Passed
3 · 20
20377
42.52 s
true · true · recalled
Passed
3 · 21
20382
40.64 s
true · true · recalled
Passed
4 · 1
20214
58.59 s
true · true · recalled
Passed
4 · 2
20385
43.78 s
true · true · recalled
Passed
4 · 3
20382
46.72 s
true · true · recalled
Passed
4 · 4
20378
45.32 s
true · true · recalled
Passed
4 · 5
20380
68.71 s
true · true · recalled
Passed
4 · 6
20384
68.26 s
true · true · recalled
Passed
4 · 7
20379
82.24 s
true · true · recalled
Passed
4 · 8
20386
75.42 s
true · true · recalled
Passed
4 · 9
20385
63.47 s
true · true · recalled
Passed
4 · 10
20387
51.69 s
true · true · recalled
Passed
4 · 11
20382
51.46 s
true · true · recalled
Passed
4 · 12
20385
56.93 s
true · true · recalled
Passed
4 · 13
20386
53.95 s
true · true · recalled
Passed
4 · 14
20383
47.53 s
true · true · recalled
Passed
4 · 15
20384
42.73 s
true · true · recalled
Passed
4 · 16
20382
42.76 s
true · true · recalled
Passed
4 · 17
20383
43.26 s
true · true · recalled
Passed
4 · 18
20384
41.75 s
true · true · recalled
Passed
4 · 19
20383
22.06 s
true · true · recalled
Passed
4 · 20
20381
21.99 s
true · true · recalled
Passed
4 · 21
20387
22.09 s
true · true · recalled
Passed
5 · 1
20213
42.40 s
true · true · recalled
Passed
5 · 2
20383
24.21 s
true · true · recalled
Passed
5 · 3
20383
25.91 s
true · true · recalled
Passed
5 · 4
20386
29.32 s
true · true · recalled
Passed
5 · 5
20382
22.15 s
true · true · recalled
Passed
5 · 6
20385
22.26 s
true · true · recalled
Passed
5 · 7
20381
22.18 s
true · true · recalled
Passed
5 · 8
20386
22.29 s
true · true · recalled
Passed
5 · 9
20383
22.22 s
true · true · recalled
Passed
5 · 10
20382
22.15 s
true · true · recalled
Passed
5 · 11
20381
22.16 s
true · true · recalled
Passed
5 · 12
20382
22.22 s
true · true · recalled
Passed
5 · 13
20383
24.17 s
true · true · recalled
Passed
5 · 14
20382
30.92 s
true · true · recalled
Passed
5 · 15
20385
42.89 s
true · true · recalled
Passed
5 · 16
20388
43.08 s
true · true · recalled
Passed
5 · 17
20387
22.21 s
true · true · recalled
Passed
5 · 18
20386
22.21 s
true · true · recalled
Passed
5 · 19
20384
22.02 s
true · true · recalled
Passed
5 · 20
20383
21.93 s
true · true · recalled
Passed
5 · 21
20383
22.02 s
true · true · recalled
Passed

Method

Staggered multi-turn fictional archive, exact engine-tokenized rolling gateway; independent recent, pinned and old probes each turn.

Limits

Lab retrieval and rolling-history evidence only. Bounded literal user-source excerpts, with verified source hashes; details outside them remain unavailable. Shared engine cache intervals include overlapping users and are not per-user hits. Gateway path, not authenticated document ingestion or customer acceptance.

What ran

Exact identity

Original publisher model
Qwen3.8-27B
Publisher revision
1d4bf0f2ff6012fd82039f2fa52739d0dd7c60c0
Variant · format · precision
INT4 variant · safetensors · INT4
Weights SHA-256
not recorded
Engine · version · feature
NVIDIA NGC vLLM · 0.29.0+5013de39.dev · plain
Hardware
NVIDIA GB10, 128 GB unified memory
Harness / workload
keep-rolling-context-v1 · 1
Workload suite · version
not recorded
Harness SHA-256
not recorded
Fixture SHA-256
not recorded

Workload shape

People in this run
5
Configured window (tokens)
262144
Configured window (tokens)
262144
Output token ceiling
256

Full reproduction pack available to customers and evaluators on request

All recorded numbers

Metric definitions, scope and statistics
All aggregate numeric result fields in this run
MetricValueScopeStatisticSample countBasis / source fieldRun
roll ttft max seconds
83.86 s
per person
max
not recorded
roll_ttft_max_seconds
requests ok
105.00 count
machine total
single
not recorded
requests_ok
roll ttft p50 seconds
42.64 s
per person
median
not recorded
roll_ttft_p50_seconds
rolled off abstained
0.00 count
machine total
single
not recorded
rolled_off_abstained
rolled off invalid
0.00 count
machine total
single
not recorded
rolled_off_invalid
rolled off wrong
0.00 count
machine total
single
not recorded
rolled_off_wrong
normal ttft max seconds
58.59 s
per person
max
not recorded
normal_ttft_max_seconds
rolled off correct
100.00 count
machine total
single
not recorded
rolled_off_correct
normal ttft p50 seconds
36.15 s
per person
median
not recorded
normal_ttft_p50_seconds
rolled off total
100.00 count
machine total
single
not recorded
rolled_off_total
requests total
105.00 count
machine total
single
not recorded
requests_total

Data and citation

Skilak Keep evidence, run 861767d1989c495e851fd421, measured 2026-10-07T22:28:13.809032+00:00.
https://skilak.ai/keep/resources/evidence/r/861767d1989c495e851fd421

Ready to scope it?

Start with one workload.

Tell us who uses it, what it can read, and what must stay inside. We will map the build.