Bias Evaluation on BOLD
1.037Bias ScoreCDA
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| CDABackbone=LLaMA-3-8B2026.03 | 1.037 | 0.281 | — | — | — | — | — | |
| KLAADBackbone=LLaMA-3-8B2026.03 | 1.267 | 0.611 | — | — | — | — | — | |
| UGIDBackbone=LLaMA-3-8B2026.03 | 1.998 | 1 | — | — | — | — | — | |
| ORIGINALBackbone=LLaMA-3-8B2026.03 | 11.343 | 1 | — | — | — | — | — | |
| BaseModel=BB3-175B2026.05 | — | — | 19.3 | — | — | — | — | |
| Base2026.06 | — | — | — | 2.93 | 1.16 | 0.6 | 7.03 | |
| DebiasRAGModel=BB3-175B2026.05 | — | — | 17.4 | — | — | — | — | |
| DPO2026.06 | — | — | — | 2.22 | 1 | 0.74 | 4.91 | |
| GRPO2026.06 | — | — | — | 1.4 | 0.78 | 0.49 | 2.95 | |
| PPO2026.06 | — | — | — | 2.68 | 1.21 | 0.68 | 6.16 | |
| PromptingModel=BB3-175B2026.05 | — | — | 17.7 | — | — | — | — | |
| Self-DebiasModel=BB3-175B2026.05 | — | — | 17.9 | — | — | — | — |