Concept Identifiability on BINARY (2, 2)
0.8896MCCSSAE
Evaluation results
| Method | Links | |
|---|---|---|
| SSAEModel Backbone=Gemma-2-2B2025.02 | 0.8896 | |
| Linear ProbeModel Backbone=Gemma-2-2B2025.02 | 0.8562 | |
| GemmaScopeModel Backbone=Gemma-2-2B2025.02 | 0.8387 | |
| TopK-SAEModel Backbone=Gemma-2-2B2025.02 | 0.8269 | |
| ReLU-SAEModel Backbone=Gemma-2-2B2025.02 | 0.8128 | |
| JumpReLU SAEModel Backbone=Gemma-2-2B2025.02 | 0.8015 | |
| SSAEEvaluation Protocol=Recovered decoder directions evaluated on paired observations (f(x), f(x˜))2025.02 | 0.7198 | |
| PythiaSAEEvaluation Protocol=Recovered decoder directions evaluated on paired observations (f(x), f(x˜))2025.02 | 0.7087 | |
| Linear ProbeEvaluation Protocol=Recovered decoder directions evaluated on paired observations (f(x), f(x˜))2025.02 | 0.6883 | |
| TopK-SAEEvaluation Protocol=Recovered decoder directions evaluated on paired observations (f(x), f(x˜))2025.02 | 0.6719 | |
| ReLU-SAEEvaluation Protocol=Recovered decoder directions evaluated on paired observations (f(x), f(x˜))2025.02 | 0.6594 | |
| JumpReLU SAEEvaluation Protocol=Recovered decoder directions evaluated on paired observations (f(x), f(x˜))2025.02 | 0.6512 |