Mathematical Reasoning on GSM8K (Pass@1, #Tokens)
94.1Pass@1 AccuracyDRP
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| DRPBackbone=R1-Distill-Qwen-7B, Evaluation Setting=zero-shot2025.05 | 94.1 | 328 | |
| BaseBackbone=R1-Distill-Qwen-7B, Evaluation Setting=zero-shot2025.05 | 91.7 | 917 | |
| TALEBackbone=R1-Distill-Qwen-7B, Evaluation Setting=zero-shot2025.05 | 91 | 522 | |
| Cot ValveBackbone=R1-Distill-Qwen-7B, Evaluation Setting=zero-shot2025.05 | 90.8 | 364 | |
| DRPBackbone=R1-Distill-Qwen-1.5B, Evaluation Setting=zero-shot2025.05 | 83.4 | 721 | |
| ThinkpruneBackbone=R1-Distill-Qwen-1.5B, Evaluation Setting=zero-shot2025.05 | 80 | 712 | |
| BaseBackbone=R1-Distill-Qwen-1.5B, Evaluation Setting=zero-shot2025.05 | 70.7 | 1,443 | |
| Cot ValveBackbone=R1-Distill-Qwen-1.5B, Evaluation Setting=zero-shot2025.05 | 70.4 | 805 | |
| TALEBackbone=R1-Distill-Qwen-1.5B, Evaluation Setting=zero-shot2025.05 | 70.1 | 1,170 |