Indirect Object Identification on Wang-style IOI templates
4.49Lift (routing)Manifestation Units
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Manifestation UnitsBehaviour=IO-token routing, Intervention=mean ablation, k=102026.06 | 4.49 | — | — | |
| Manifestation UnitsBehaviour=IOI logit-difference, Intervention=path patching, k=302026.06 | — | 1.75 | — |