Report · resources

Rolling chat

Qwen3.6-35B-A3B · NVFP4 variant · NVIDIA NGC vLLM 0.29

Measured results / Rolling chat / Run 7cde518077a3402d96c45d36

Measured 2026-10-07 19:26 UTC · Record date 2026-10-07

Failed. 49/50 turns

Answer checks

49/50 turns

Rolling-chat turn checks
All rolling-chat turns
Person · turnFilled inputFirst wordRecent · pinned · oldResultRun
1 · 1
20208
3.97 s
true · true · recalled
Passed
1 · 2
40330
5.73 s
true · true · recalled
Passed
1 · 3
60454
7.00 s
true · true · recalled
Passed
1 · 4
80583
8.79 s
true · true · recalled
Passed
1 · 5
100707
10.51 s
true · true · recalled
Passed
1 · 6
120833
11.05 s
true · true · recalled
Passed
1 · 7
80753
27.63 s
true · true · recalled
Passed
1 · 8
100875
10.67 s
true · true · recalled
Passed
1 · 9
121001
16.26 s
true · true · recalled
Passed
1 · 10
80759
32.22 s
true · true · recalled
Passed
2 · 1
20214
8.94 s
true · true · recalled
Passed
2 · 2
40334
6.66 s
true · true · recalled
Passed
2 · 3
60457
15.63 s
true · true · recalled
Passed
2 · 4
80585
11.85 s
true · true · recalled
Passed
2 · 5
100712
10.51 s
true · true · recalled
Passed
2 · 6
120832
11.06 s
true · true · recalled
Passed
2 · 7
80760
27.87 s
true · true · recalled
Passed
2 · 8
100884
10.60 s
true · true · recalled
Passed
2 · 9
121014
17.19 s
true · true · recalled
Passed
2 · 10
80765
33.59 s
true · true · recalled
Passed
3 · 1
20209
8.31 s
true · true · recalled
Passed
3 · 2
40335
6.02 s
true · true · recalled
Passed
3 · 3
60463
7.34 s
true · true · recalled
Passed
3 · 4
80585
17.83 s
true · true · recalled
Passed
3 · 5
100714
13.32 s
true · true · recalled
Passed
3 · 6
120840
11.46 s
true · true · recalled
Passed
3 · 7
80763
24.56 s
true · true · recalled
Passed
3 · 8
100887
10.17 s
true · true · recalled
Passed
3 · 9
121013
11.18 s
true · true · recalled
Passed
3 · 10
80758
27.71 s
true · true · recalled
Passed
4 · 1
20212
10.22 s
true · true · recalled
Passed
4 · 2
40337
5.81 s
true · true · recalled
Passed
4 · 3
60470
7.02 s
true · true · recalled
Passed
4 · 4
80594
8.85 s
true · true · recalled
Passed
4 · 5
100719
10.10 s
true · true · recalled
Passed
4 · 6
120840
12.66 s
true · true · recalled
Passed
4 · 7
80755
29.36 s
true · true · recalled
Passed
4 · 8
100888
10.62 s
true · true · recalled
Passed
4 · 9
121013
18.54 s
true · true · recalled
Passed
4 · 10
80770
34.95 s
true · true · recalled
Passed
5 · 1
20214
10.35 s
true · true · recalled
Passed
5 · 2
40338
9.02 s
true · true · incorrect
Failed
5 · 3
60464
7.32 s
true · true · recalled
Passed
5 · 4
80590
9.24 s
true · true · recalled
Passed
5 · 5
100716
19.11 s
true · true · recalled
Passed
5 · 6
120844
14.17 s
true · true · recalled
Passed
5 · 7
80765
24.99 s
true · true · recalled
Passed
5 · 8
100891
10.18 s
true · true · recalled
Passed
5 · 9
121017
11.18 s
true · true · recalled
Passed
5 · 10
80766
24.52 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

  • 49/50 requests passed.

What ran

Exact identity

Original publisher model
Qwen3.6-35B-A3B
Publisher revision
995ad96eacd98c81ed38be0c5b274b04031597b0
Variant · format · precision
NVFP4 variant · safetensors · NVFP4
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
34.95 s
per person
max
not recorded
roll_ttft_max_seconds
requests ok
49.00 count
machine total
single
not recorded
requests_ok
roll ttft p50 seconds
27.79 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
19.11 s
per person
max
not recorded
normal_ttft_max_seconds
rolled off correct
20.00 count
machine total
single
not recorded
rolled_off_correct
normal ttft p50 seconds
10.43 s
per person
median
not recorded
normal_ttft_p50_seconds
rolled off total
20.00 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 7cde518077a3402d96c45d36, measured 2026-10-07T19:26:02.104296+00:00.
https://skilak.ai/keep/resources/evidence/r/7cde518077a3402d96c45d36

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.