Out-of-Domain (OOD) Bias Evaluation on StereoSet
67.2AccuracyC2PO
Evaluation Results
| Method | Links | |
|---|---|---|
| C2POBackbone=LLaMA-2-13B-Chat2025.12 | 67.2 | |
| CPOBackbone=LLaMA-2-13B-Chat2025.12 | 64 | |
| GRPOBackbone=LLaMA-2-13B-Chat2025.12 | 60.4 | |
| CPOBackbone=DeepSeek-R1-Distill-Qwen-14B2025.12 | 60.1 | |
| C2POBackbone=DeepSeek-R1-Distill-Qwen-14B2025.12 | 60.1 | |
| GRPOBackbone=DeepSeek-R1-Distill-Qwen-14B2025.12 | 60 | |
| FRBackbone=DeepSeek-R1-Distill-Qwen-14B2025.12 | 58.1 | |
| IPOBackbone=LLaMA-2-13B-Chat2025.12 | 53.8 | |
| BCOBackbone=LLaMA-2-13B-Chat2025.12 | 53 | |
| IPOBackbone=DeepSeek-R1-Distill-Qwen-14B2025.12 | 44.9 | |
| FRBackbone=LLaMA-2-13B-Chat2025.12 | 42.8 | |
| DPOBackbone=LLaMA-2-13B-Chat2025.12 | 41.8 | |
| DPOBackbone=DeepSeek-R1-Distill-Qwen-14B2025.12 | 33.4 | |
| BCOBackbone=DeepSeek-R1-Distill-Qwen-14B2025.12 | 33.3 |