Math and Reasoning Accuracy on GSM8K
83.85AccuracyBFPO
Evaluation Results
| Method | Links | |
|---|---|---|
| BFPOBackbone=Qwen2.5-7B-Instruct2025.12 | 83.85 | |
| PeCANBackbone=Qwen2.5-7B-Instruct2025.12 | 82.87 | |
| NSPOBackbone=Qwen2.5-7B-Instruct2025.12 | 81.96 | |
| Qwen2.5-7B-Instruct2025.12 | 81.65 | |
| MoCANBackbone=Qwen2.5-7B-Instruct2025.12 | 81.65 | |
| DPO-MixBackbone=Qwen2.5-7B-Instruct2025.12 | 81.35 | |
| DPO-SBackbone=Qwen2.5-7B-Instruct2025.12 | 80.67 | |
| DPO-HBackbone=Qwen2.5-7B-Instruct2025.12 | 80.21 | |
| BFPOBackbone=Llama3-8B-Instruct2025.12 | 78.32 | |
| NSPOBackbone=Llama3-8B-Instruct2025.12 | 76.42 | |
| Llama3-8B-Instruct2025.12 | 75.44 | |
| SafeRLHFBackbone=Qwen2.5-7B-Instruct2025.12 | 75.28 | |
| MoCANBackbone=Llama3-8B-Instruct2025.12 | 74.98 | |
| W-DOORBackbone=Llama3-8B-Instruct2025.12 | 71.11 | |
| DPO-MixBackbone=Llama3-8B-Instruct2025.12 | 63.76 | |
| DPO-SBackbone=Llama3-8B-Instruct2025.12 | 62.55 | |
| DPO-HBackbone=Llama3-8B-Instruct2025.12 | 59.97 | |
| PeCANBackbone=Llama3-8B-Instruct2025.12 | 59.36 | |
| W-DOORBackbone=Qwen2.5-7B-Instruct2025.12 | 57.85 | |
| SafeRLHFBackbone=Llama3-8B-Instruct2025.12 | 51.71 |