Binary Classification on SAEBench Pythia-160m (full)
95.3Bias in BIOS Set 1 AccRSAE
Evaluation Results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| RSAEDictionary usage=Full-dictionary probe, Sparsity constraint=No k-sparsity, Teacher SAE family=ReLU2026.06 | 95.3 | 93.04 | 91.22 | 88 | 91.4 | 96.5 | 94.25 | 99.96 | 93.71 | |
| ReLU teacher (trainer 3)Dictionary usage=Full-dictionary probe, Sparsity constraint=No k-sparsity, Teacher SAE family=ReLU2026.06 | 95.28 | 93.12 | 90.94 | 87.98 | 91.4 | 96.64 | 94.18 | 99.96 | 93.69 |