Structural Bias Evaluation on HANS
99.6AccuracyC2PO
Evaluation Results
| Method | Links | |
|---|---|---|
| C2POBackbone=DeepSeek-R1-Distill-Qwen-14B2025.12 | 99.6 | |
| GRPOBackbone=DeepSeek-R1-Distill-Qwen-14B2025.12 | 99 | |
| IPOBackbone=DeepSeek-R1-Distill-Qwen-14B2025.12 | 97.7 | |
| CPOBackbone=DeepSeek-R1-Distill-Qwen-14B2025.12 | 96.9 | |
| C2POBackbone=LLaMA-2-13B-Chat2025.12 | 95.9 | |
| BCOBackbone=DeepSeek-R1-Distill-Qwen-14B2025.12 | 86.8 | |
| BCOBackbone=LLaMA-2-13B-Chat2025.12 | 86.1 | |
| CPOBackbone=LLaMA-2-13B-Chat2025.12 | 85.7 | |
| IPOBackbone=LLaMA-2-13B-Chat2025.12 | 77.9 | |
| GRPOBackbone=LLaMA-2-13B-Chat2025.12 | 66.6 | |
| DPOBackbone=LLaMA-2-13B-Chat2025.12 | 57.9 | |
| FRBackbone=LLaMA-2-13B-Chat2025.12 | 54.2 | |
| FRBackbone=DeepSeek-R1-Distill-Qwen-14B2025.12 | 53.6 | |
| DPOBackbone=DeepSeek-R1-Distill-Qwen-14B2025.12 | 51.2 |