Mathematical Reasoning on MATH500 (Pass@8, Pass@16, Pass@32)
93Pass@8w/ Clip
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| w/ ClipBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.12 | 93 | 93 | 93.8 | |
| SENTBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.12 | 93 | 93.8 | 95.2 | |
| w/ MaskBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.12 | 92.8 | 93.6 | 94.4 | |
| w/ CovBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.12 | 92.6 | 93.8 | 94.8 | |
| GRPOBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.12 | 91.6 | 94 | 95 | |
| w/ EnBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.12 | 91.6 | 93.2 | 94.6 | |
| w/ High EnBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.12 | 91 | 92.8 | 94.2 | |
| Base modelBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.12 | 90 | 92.4 | 93.2 | |
| w/ AdvBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.12 | 89.6 | 91.6 | 94 |