Compositional Steering on 3-behavior combinations (seen)
87.9AccuracyHybrid: Composition (<and>) + Instr.
Evaluation results
| Method | Links | |
|---|---|---|
| Hybrid: Composition (<and>) + Instr.Backbone=Qwen-8B2026.01 | 87.9 | |
| Instruction SteeringBackbone=Qwen-8B2026.01 | 83.7 | |
| Composition (<and>)Backbone=Qwen-8B2026.01 | 83.1 | |
| ConcatenationBackbone=Qwen-8B2026.01 | 59.6 | |
| LoRA DAREBackbone=Qwen-8B2026.01 | 58.4 | |
| LM-SteerBackbone=Qwen-8B2026.01 | 2.2 |