Mathematical Reasoning on GSM8K (Acc, Len, L-Acc)
96.6AccuracyPPO
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| PPOBackbone=Qwen2.5-Math-7B-Instruct2025.06 | 96.6 | 305 | 91.5 | |
| Bingo-ABackbone=Qwen2.5-Math-7B-Instruct2025.06 | 96.1 | 283 | 91.5 | |
| Bingo-EBackbone=Qwen2.5-Math-7B-Instruct2025.06 | 96 | 241 | 92 | |
| BaseBackbone=Qwen2.5-Math-7B-Instruct2025.06 | 95.8 | 331 | 90.3 | |
| PPOBackbone=DeepSeek-R1-Distill-Qwen-7B2025.06 | 92.9 | 918 | 87.5 | |
| Bingo-ABackbone=DeepSeek-R1-Distill-Qwen-7B2025.06 | 92.3 | 371 | 90.2 | |
| Bingo-EBackbone=DeepSeek-R1-Distill-Qwen-7B2025.06 | 91.8 | 366 | 89.7 | |
| Bingo-ABackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.06 | 87 | 563 | 83.9 | |
| Bingo-EBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.06 | 86.7 | 345 | 84.9 | |
| BaseBackbone=DeepSeek-R1-Distill-Qwen-7B2025.06 | 85.7 | 1,001 | 80.3 | |
| PPOBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.06 | 85.4 | 1,310 | 78.2 | |
| BaseBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.06 | 73.2 | 2,025 | 63.5 |