Total runs:
534.8K
Run Growth:
-254.4K
Growth Rate:
-47.39%
Sponsored by
AdsCreator.com
FINAL-Bench / Armoring-Qwen3.6-35B-A3B
Model's Last Updated: August 16 2026
text-generation
Introduction of Armoring-Qwen3.6-35B-A3B
Model Details of Armoring-Qwen3.6-35B-A3B
Armoring-Qwen3.6-35B-A3B
This private repository contains a model-card draft for a VIDRAFT Attention Armoring study on
Qwen/Qwen3.6-35B-A3B
.
The repository is intentionally card-first. No full model weights, implementation artifact, internal transformation procedure, training command, internal mapping, or infrastructure path is included.
One-Screen Result
This is the strongest recovery narrative in the current armoring series. The candidate started with the largest measured quality damage after transformation, then recovered most of that damage while preserving local format behavior.
| What changed | Why it matters |
|---|---|
| Immediate quality ratio started at 1.3435x | This was the hardest recovery case among the three draft cards. |
| Final quality ratio recovered to 1.0214x | The candidate recovered close to the reference behavior in the local validation lane. |
| Best observed ratio reached 1.0207x | The best observed measurement stayed close to the final value. |
| Damage recovery measured 32.21 percentage points | About 93.8% of the initial measured damage was recovered. |
| Format probe passed 15 / 15 | Basic response-format behavior was preserved after recovery. |
These figures should be read as recovery and behavior-preservation evidence, not as a broad capability claim.
Before / After Architecture View
Standard Attention vs. Attention Armoring
In a conventional Transformer-style attention block, the model projects the current hidden state into query, key, and value streams. During long-context generation, key and value states are retained in the KV cache so later tokens can attend back to earlier context. This is useful and standard, but it creates a direct memory bottleneck: the longer the context and the more concurrent sessions, the more cache state the serving system must keep alive.
VIDRAFT Attention Armoring keeps the public model interface intact while changing the long-context attention-state path inside the eligible attention block. The public point is not the private transformation procedure; the public point is the measured operational effect. In this 35B-A3B study, the important signal is the recovery curve: the candidate absorbed a large initial transformation shock and recovered most of the measured damage while keeping local format behavior intact.
Public Transformation Summary
| Layer of the system | Before | After Armoring |
|---|---|---|
| Base model family | Qwen3.6-35B-A3B-class model | Same base-family interface |
| Public architecture view | Original attention-state path | Attention subsystem is armored at a high level |
| Main validation signal | Baseline behavior before transformation | Large recovery after transformation shock |
| Operational value | Tests armoring on a larger sparse-activation target | Shows recovery feasibility under a harder condition |
| Release posture | Standard model-card summary | Private card; implementation details retained privately |
Measured Summary
| Measurement | Result |
|---|---|
| Immediate quality ratio after transformation | 1.3435x |
| Final quality ratio | 1.0214x |
| Best observed ratio | 1.0207x |
| Recovered damage | 32.21 percentage points |
| Share of initial measured damage recovered | 93.8% |
| Format probe | 15 / 15 |
Operating Notes
This card is primarily a recovery and feasibility card. It should not be used to claim that this candidate is faster, deployment-ready, or broadly better than the base model.
The armoring family is intended for long-context efficiency, but this 35B-A3B card should not borrow the 27B serving economics as if they were directly measured on this model. Any long-context capacity or serving claim for this model needs its own paired measurement.
Important Limitations
This repository does not claim:
- broad model-capability improvement over the base model;
- better per-user decode speed than the base model;
- production readiness;
- a completed long-context serving capacity result for this model;
- full safety, reasoning, coding, multilingual, tool-use, or multimodal coverage;
- disclosure of the proprietary armoring procedure.
Quality ratios from narrow local measurements should not be used to rank armoring variants at the 1 percentage-point level. The recovery magnitude is the useful signal here; small differences between final ratios are not.
License
The base model
Qwen/Qwen3.6-35B-A3B
is listed on Hugging Face with the Apache-2.0 license. This card follows the same license metadata while retaining VIDRAFT-specific implementation details privately.
Citation
VIDRAFT / FINAL-Bench. "Armoring-Qwen3.6-35B-A3B." Private Hugging Face Model Repository, 2026.
Runs of FINAL-Bench Armoring-Qwen3.6-35B-A3B on huggingface.co
0
Total runs
0
24-hour runs
0
3-day runs
0
7-day runs
0
30-day runs
More Information About Armoring-Qwen3.6-35B-A3B huggingface.co Model
Url of Armoring-Qwen3.6-35B-A3B
Armoring-Qwen3.6-35B-A3B huggingface.co Url
Provider of Armoring-Qwen3.6-35B-A3B huggingface.co
Other API from FINAL-Bench
Updated:September 25 2026
Total runs:
271.4K
Run Growth:
-61.7K
Growth Rate:
-22.74%
Total runs:
52.3K
Run Growth:
51.6K
Growth Rate:
100.00%
Total runs:
3.5K
Run Growth:
658
Growth Rate:
90.26%
Total runs:
2.5K
Run Growth:
2.5K
Growth Rate:
100.00%
Total runs:
2.2K
Run Growth:
1.7K
Growth Rate:
85.64%
Total runs:
1.5K
Run Growth:
1.4K
Growth Rate:
97.81%
Total runs:
1.3K
Run Growth:
1.2K
Growth Rate:
100.00%
Total runs:
995
Run Growth:
605
Growth Rate:
60.80%
Total runs:
984
Run Growth:
8
Growth Rate:
0.76%
Total runs:
976
Run Growth:
955
Growth Rate:
100.00%
Total runs:
960
Run Growth:
960
Growth Rate:
100.00%
Total runs:
912
Run Growth:
497
Growth Rate:
54.50%
Total runs:
806
Run Growth:
535
Growth Rate:
66.38%
Total runs:
600
Run Growth:
-122
Growth Rate:
-20.33%
Total runs:
491
Run Growth:
322
Growth Rate:
65.58%
Total runs:
471
Run Growth:
465
Growth Rate:
98.73%
Total runs:
432
Run Growth:
432
Growth Rate:
100.00%
Total runs:
401
Run Growth:
239
Growth Rate:
59.60%
Total runs:
385
Run Growth:
49
Growth Rate:
12.73%
Total runs:
376
Run Growth:
351
Growth Rate:
100.00%
Total runs:
346
Run Growth:
260
Growth Rate:
75.14%
Total runs:
336
Run Growth:
301
Growth Rate:
93.19%
Total runs:
330
Run Growth:
178
Growth Rate:
50.14%
Total runs:
305
Run Growth:
60
Growth Rate:
19.67%
Total runs:
194
Run Growth:
183
Growth Rate:
100.00%
Total runs:
193
Run Growth:
23
Growth Rate:
11.44%
Total runs:
191
Run Growth:
191
Growth Rate:
100.00%
Total runs:
181
Run Growth:
-7
Growth Rate:
-3.65%
Total runs:
178
Run Growth:
-17
Growth Rate:
-1.77%
Total runs:
162
Run Growth:
89
Growth Rate:
54.94%
Total runs:
153
Run Growth:
-82
Growth Rate:
-53.59%
Total runs:
127
Run Growth:
114
Growth Rate:
89.76%
Total runs:
119
Run Growth:
72
Growth Rate:
63.16%
Total runs:
96
Run Growth:
86
Growth Rate:
100.00%
Total runs:
72
Run Growth:
-180
Growth Rate:
-250.00%
Total runs:
63
Run Growth:
37
Growth Rate:
58.73%
Total runs:
61
Run Growth:
61
Growth Rate:
100.00%
Total runs:
60
Run Growth:
60
Growth Rate:
100.00%
Total runs:
60
Run Growth:
17
Growth Rate:
28.33%
Total runs:
52
Run Growth:
52
Growth Rate:
100.00%
Total runs:
52
Run Growth:
52
Growth Rate:
100.00%
Total runs:
51
Run Growth:
51
Growth Rate:
100.00%
Total runs:
50
Run Growth:
50
Growth Rate:
100.00%
Total runs:
49
Run Growth:
25
Growth Rate:
65.79%
Total runs:
48
Run Growth:
48
Growth Rate:
100.00%
Total runs:
47
Run Growth:
16
Growth Rate:
34.04%
Total runs:
46
Run Growth:
46
Growth Rate:
100.00%
Total runs:
46
Run Growth:
46
Growth Rate:
100.00%
Total runs:
43
Run Growth:
43
Growth Rate:
100.00%
Total runs:
40
Run Growth:
40
Growth Rate:
100.00%
Total runs:
31
Run Growth:
22
Growth Rate:
70.97%
Total runs:
13
Run Growth:
6
Growth Rate:
46.15%
Total runs:
0
Run Growth:
0
Growth Rate:
0.00%
Total runs:
0
Run Growth:
0
Growth Rate:
0.00%
Total runs:
0
Run Growth:
0
Growth Rate:
0.00%
Total runs:
0
Run Growth:
0
Growth Rate:
0.00%
Total runs:
0
Run Growth:
0
Growth Rate:
0.00%