OOD Bias Evaluation on Winobias
0.507AccuracyDPO
Evaluation Results
| Method | Links | |
|---|---|---|
| DPOBackbone=LLaMA-2-13B-Chat2025.12 | 0.507 | |
| FRBackbone=LLaMA-2-13B-Chat2025.12 | 0.506 | |
| IPOBackbone=DeepSeek-R1-Distill-Qwen-14B2025.12 | 0.501 | |
| C2POBackbone=DeepSeek-R1-Distill-Qwen-14B2025.12 | 0.501 | |
| CPOBackbone=DeepSeek-R1-Distill-Qwen-14B2025.12 | 0.5 | |
| BCOBackbone=DeepSeek-R1-Distill-Qwen-14B2025.12 | 0.5 | |
| GRPOBackbone=DeepSeek-R1-Distill-Qwen-14B2025.12 | 0.5 | |
| CPOBackbone=LLaMA-2-13B-Chat2025.12 | 0.499 | |
| FRBackbone=DeepSeek-R1-Distill-Qwen-14B2025.12 | 0.499 | |
| C2POBackbone=LLaMA-2-13B-Chat2025.12 | 0.496 | |
| GRPOBackbone=LLaMA-2-13B-Chat2025.12 | 0.494 | |
| DPOBackbone=DeepSeek-R1-Distill-Qwen-14B2025.12 | 0.492 | |
| IPOBackbone=LLaMA-2-13B-Chat2025.12 | 0.491 | |
| BCOBackbone=LLaMA-2-13B-Chat2025.12 | 0.484 |