Report · resources

Rolling chat

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

Measured results / Rolling chat / Run 933bd14af8e24e5690f7c9d6

Measured 2026-10-07 21:07 UTC · Record date 2026-10-07

Failed. 50/50 turns

Answer checks

50/50 turns

Rolling-chat turn checks
All rolling-chat turns
Person · turnFilled inputFirst wordRecent · pinned · oldResultRun
1 · 1
20208
21.33 s
true · true · recalled
Passed
1 · 2
40334
25.96 s
true · true · recalled
Passed
1 · 3
60462
48.91 s
true · true · recalled
Passed
1 · 4
40512
74.31 s
true · true · recalled
Passed
1 · 5
60640
46.11 s
true · true · recalled
Passed
1 · 6
40504
140.01 s
true · true · recalled
Passed
1 · 7
60633
127.06 s
true · true · recalled
Passed
1 · 8
40505
174.34 s
true · true · recalled
Passed
1 · 9
60635
136.46 s
true · true · recalled
Passed
1 · 10
40515
174.25 s
true · true · recalled
Passed
2 · 1
20214
48.47 s
true · true · recalled
Passed
2 · 2
40338
46.27 s
true · true · recalled
Passed
2 · 3
60465
38.51 s
true · true · recalled
Passed
2 · 4
40510
136.95 s
true · true · recalled
Passed
2 · 5
60639
129.84 s
true · true · recalled
Passed
2 · 6
40509
174.01 s
true · true · recalled
Passed
2 · 7
60642
136.44 s
true · true · recalled
Passed
2 · 8
40512
174.10 s
true · true · recalled
Passed
2 · 9
60646
112.45 s
true · true · recalled
Passed
2 · 10
40514
130.24 s
true · true · recalled
Passed
3 · 1
20209
28.17 s
true · true · recalled
Passed
3 · 2
40339
35.05 s
true · true · recalled
Passed
3 · 3
60471
103.27 s
true · true · recalled
Passed
3 · 4
40507
152.32 s
true · true · recalled
Passed
3 · 5
60642
123.65 s
true · true · recalled
Passed
3 · 6
40513
189.65 s
true · true · recalled
Passed
3 · 7
60645
131.76 s
true · true · recalled
Passed
3 · 8
40507
159.56 s
true · true · recalled
Passed
3 · 9
60637
109.01 s
true · true · recalled
Passed
3 · 10
40513
134.30 s
true · true · recalled
Passed
4 · 1
20212
89.27 s
true · true · recalled
Passed
4 · 2
40341
129.31 s
true · true · recalled
Passed
4 · 3
60478
124.47 s
true · true · recalled
Passed
4 · 4
40510
189.38 s
true · true · recalled
Passed
4 · 5
60637
127.74 s
true · true · recalled
Passed
4 · 6
40505
157.65 s
true · true · recalled
Passed
4 · 7
60637
109.04 s
true · true · recalled
Passed
4 · 8
40518
133.96 s
true · true · recalled
Passed
4 · 9
60647
55.89 s
true · true · recalled
Passed
4 · 10
40518
89.42 s
true · true · recalled
Passed
5 · 1
20214
43.70 s
true · true · recalled
Passed
5 · 2
40342
128.91 s
true · true · recalled
Passed
5 · 3
60472
124.89 s
true · true · recalled
Passed
5 · 4
40514
190.31 s
true · true · recalled
Passed
5 · 5
60641
134.05 s
true · true · recalled
Passed
5 · 6
40513
158.35 s
true · true · recalled
Passed
5 · 7
60647
110.68 s
true · true · recalled
Passed
5 · 8
40511
133.19 s
true · true · recalled
Passed
5 · 9
60641
56.41 s
true · true · recalled
Passed
5 · 10
40517
88.89 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

Bounded tuning pilot; incomplete window growth cannot pass the full workload grade. 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 did not work

  • Bounded tuning pilot; incomplete window growth cannot pass the full workload grade.

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
190.31 s
per person
max
not recorded
roll_ttft_max_seconds
requests ok
50.00 count
machine total
single
not recorded
requests_ok
roll ttft p50 seconds
154.99 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
136.46 s
per person
max
not recorded
normal_ttft_max_seconds
rolled off correct
35.00 count
machine total
single
not recorded
rolled_off_correct
normal ttft p50 seconds
109.02 s
per person
median
not recorded
normal_ttft_p50_seconds
rolled off total
35.00 count
machine total
single
not recorded
rolled_off_total
requests total
50.00 count
machine total
single
not recorded
requests_total

Data and citation

Skilak Keep evidence, run 933bd14af8e24e5690f7c9d6, measured 2026-10-07T21:07:12.591006+00:00.
https://skilak.ai/keep/resources/evidence/r/933bd14af8e24e5690f7c9d6

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.