Mathematical Reasoning on AIME 2024 (Avg@k Sampling)
27.92Avg Recall@8SENT
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| SENTBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.12 | 27.92 | 25.21 | 26.56 | |
| w/ MaskBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.12 | 23.33 | 24.79 | 24.79 | |
| w/ ClipBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.12 | 23.33 | 22.5 | 24.27 | |
| w/ EnBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.12 | 22.5 | 20.62 | 22.29 | |
| w/ AdvBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.12 | 21.67 | 19.58 | 19.58 | |
| w/ High EnBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.12 | 19.58 | 21.25 | 18.96 | |
| w/ CovBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.12 | 19.17 | 21.67 | 20.52 | |
| Base modelBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.12 | 18.33 | 18.96 | 18.54 | |
| GRPOBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.12 | 14.17 | 16.46 | 15.63 |