Sparse Autoencoding on GPT-2 activations
98KL ScoreStandard SAE
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Standard SAETraining Token Budget=300M2026.06 | 98 | 98 | 99.1 | 60 | |
| SASATraining Token Budget=150M2026.06 | 97 | 97 | 98.8 | 60 |
| Method | Links | ||||
|---|---|---|---|---|---|
| Standard SAETraining Token Budget=300M2026.06 | 98 | 98 | 99.1 | 60 | |
| SASATraining Token Budget=150M2026.06 | 97 | 97 | 98.8 | 60 |