Concept Steering on AXBENCH D_L20^G9B
1.079Steering ScorePrompt
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Promptsource=ours2026.05 | 1.079 | — | |
| Promptsource=Wu et al. (2025a)2026.05 | 1.075 | — | |
| PrOSVObjective=SimPO2026.05 | 0.905 | — | |
| FSSVObjective=SimPO, source=Wu et al. (2025b)2026.05 | 0.892 | — | |
| FSSV + Joint trainingObjective=SimPO2026.05 | 0.886 | — | |
| FSSVObjective=SimPO, source=ours2026.05 | 0.877 | 1.022 | |
| PrOSVObjective=Lang.2026.05 | 0.859 | — | |
| FSSV + Joint trainingObjective=Lang.2026.05 | 0.821 | — | |
| FSSVObjective=Lang.2026.05 | 0.788 | — | |
| LoReFT2026.05 | 0.777 | — | |
| SAE2026.05 | 0.546 | — | |
| DiffMean2026.05 | 0.322 | — |