Compositional Steering on 2-behavior combinations (seen)
0.922AccuracyHybrid: Composition (<and>) + Instr.
Evaluation Results
| Method | Links | |
|---|---|---|
| Hybrid: Composition (<and>) + Instr.Backbone=Qwen-8B2026.01 | 0.922 | |
| Composition (<and>)Backbone=Qwen-8B2026.01 | 0.909 | |
| Instruction SteeringBackbone=Qwen-8B2026.01 | 0.907 | |
| LoRA DAREBackbone=Qwen-8B2026.01 | 0.815 | |
| ConcatenationBackbone=Qwen-8B2026.01 | 0.813 | |
| LM-SteerBackbone=Qwen-8B2026.01 | 0.181 |