Share your thoughts, 1 month free Claude Pro on us
See more
Home
/
Benchmarks
SAE Interpretability and Faithfulness Evaluation on ConvNeXt Base activations V2
Loading...
6
LLM Rank
ConvNeXtV2-Base (backbone only)
0.8
2.15
3.5
4.85
May 29, 2026
LLM Rank
MS-Score
Accuracy
R2
Updated 2d ago
Evaluation Results
Method
Method
Links
LLM Rank
MS-Score
Accuracy
R2
ConvNeXtV2-Base (backbone only)
Backbone=ConvNeXtV2-Base
2026.05
6
8.74
86.74
100
Archetypal SAE
Backbone=ConvNeXtV2-Base
2026.05
5
17.35
86.66
73.85
ELUDe
Backbone=ConvNeXtV2-Base
2026.05
4
32.66
86.74
100
Matryoshka SAE
Backbone=ConvNeXtV2-Base
2026.05
3
23.44
86.67
84.42
Matching Pursuit
Backbone=ConvNeXtV2-Base
2026.05
2
30.43
86.69
84.03
BatchTopK-SAE
Backbone=ConvNeXtV2-Base
2026.05
1
25.65
86.7
84.77
Feedback
Search any
task
Search any
task