Bias Evaluation on RaceEthnicity
15Mean ImprovementiPASa
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| iPASaModel=Llama-3.1-8B-Instruct, Steering Target=residual2025.09 | 15 | 14 | 0 | |
| iPASaModel=DeepSeek-R1-Distill-8B, Steering Target=residual2025.09 | 14 | 13 | 0 | |
| iPASwoModel=Llama-3.1-8B-Instruct, Steering Target=residual2025.09 | 14 | 13 | 0 | |
| iPASwoModel=DeepSeek-R1-Distill-8B, Steering Target=residual2025.09 | 13 | 12 | 0 | |
| iPASwoModel=Nous-Hermes-2-Mistral-7B-DPO, Steering Target=residual2025.09 | 12 | 11 | 0 | |
| PASfModel=Nous-Hermes-2-Mistral-7B-DPO, Steering Target=residual2025.09 | 10 | 9 | 0 | |
| iPASaModel=Nous-Hermes-2-Mistral-7B-DPO, Steering Target=residual2025.09 | 8 | 7 | 0 | |
| PASfModel=Llama-3.1-8B-Instruct, Steering Target=residual2025.09 | 3 | 2 | 0 | |
| PASfModel=DeepSeek-R1-Distill-8B, Steering Target=residual2025.09 | 0 | -1 | 0.85 |