Activation Steering on AxBench Gemma-2-9B layer 20
1.12Steering ScoreA-PSRMSE
Evaluation Results
| Method | Links | |
|---|---|---|
| A-PSRMSEMethod Type=Multi-rank and/or multi-layer2026.05 | 1.12 | |
| Hyper-SteerMethod Type=Multi-rank and/or multi-layer2026.05 | 1.091 | |
| promptMethod Type=Multi-rank and/or multi-layer2026.05 | 1.075 | |
| A-ConstMSEMethod Type=Multi-rank and/or multi-layer2026.05 | 1.053 | |
| S-ConstMSERank=1, Intervention type=single-layer2026.05 | 0.903 | |
| S-PSRMSERank=1, Intervention type=single-layer2026.05 | 0.9 | |
| ΦSV,r=1Rank=1, Intervention type=single-layer2026.05 | 0.892 | |
| A-PSRLLMethod Type=Multi-rank and/or multi-layer2026.05 | 0.827 | |
| LoReFT-RePSMethod Type=Multi-rank and/or multi-layer2026.05 | 0.757 | |
| A-ConstLLMethod Type=Multi-rank and/or multi-layer2026.05 | 0.757 | |
| S-PSRLLRank=1, Intervention type=single-layer2026.05 | 0.667 | |
| S-ConstLLRank=1, Intervention type=single-layer2026.05 | 0.633 | |
| LoRA-RePSMethod Type=Multi-rank and/or multi-layer2026.05 | 0.631 | |
| ReFT-r1Rank=1, Intervention type=single-layer2026.05 | 0.63 | |
| LoRAMethod Type=Multi-rank and/or multi-layer2026.05 | 0.602 | |
| S-ConstDiMRank=1, Intervention type=single-layer2026.05 | 0.322 | |
| SAERank=1, Intervention type=single-layer2026.05 | 0.191 |