Distribution Alignment on UK Dataset Real Setting (test)
0.078Alignment ErrorSteering Token Calibration
Evaluation Results
| Method | Links | |
|---|---|---|
| Steering Token CalibrationModel=Qwen2.5-7B2026.04 | 0.078 | |
| Steering Token CalibrationModel=Llama-3.1-8B2026.04 | 0.084 | |
| Steering Token CalibrationModel=Llama-3.2-1B2026.04 | 0.088 | |
| Steering Token CalibrationModel=Qwen2.5-1.5B2026.04 | 0.095 | |
| IFTModel=Llama-3.1-8B2026.04 | 0.109 | |
| IFTModel=Llama-3.2-1B2026.04 | 0.147 | |
| IFTModel=Qwen2.5-7B2026.04 | 0.158 | |
| DPOModel=Llama-3.2-1B2026.04 | 0.178 | |
| IFTModel=Qwen2.5-1.5B2026.04 | 0.187 | |
| DPOModel=Qwen2.5-7B2026.04 | 0.201 | |
| DPOModel=Llama-3.1-8B2026.04 | 0.205 | |
| DPOModel=Qwen2.5-1.5B2026.04 | 0.265 |