Bias-Penalized Accuracy Evaluation on English
98.78Bias-Penalized Accuracy (BPA)SFT
Evaluation Results
| Method | Links | |
|---|---|---|
| SFTModel=Gemma-2-9B2026.02 | 98.78 | |
| SFTModel=Qwen-2.5-7B2026.02 | 98.57 | |
| SteeringModel=Gemma-2-9B2026.02 | 98.25 | |
| SteeringModel=Qwen-2.5-7B2026.02 | 94.99 | |
| SteeringModel=Mistral-7B2026.02 | 94.77 | |
| SFTModel=Mistral-7B2026.02 | 88.09 | |
| CoTModel=Qwen-2.5-7B2026.02 | 81.65 | |
| CoTModel=Gemma-2-9B2026.02 | 69.92 | |
| CoTModel=Mistral-7B2026.02 | 34.25 |