Head Recovery on Greater-Than task GPT-2 (N=200)
83.2Accuracy @5MS-HVP
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| MS-HVPK=52026.06 | 83.2 | 59.4 | 53.8 | |
| IGS=102026.06 | 73 | 52.4 | 49.4 | |
| Std HVPK=12026.06 | 72.5 | 52.4 | 49.1 | |
| Attribution patching2026.06 | 70.1 | 51.7 | 49.1 | |
| GIM2026.06 | 68.2 | 49.7 | 46.8 |