Self-Verification on MATH
88.3Pass@1FABSVer-7B
Evaluation Results
| Method | Links | |
|---|---|---|
| FABSVer-7BBackbone=Qwen2.5-7B, Training=FABSVer2026.05 | 88.3 | |
| FABSVer-3BBackbone=Qwen2.5-3B, Training=FABSVer2026.05 | 85.2 | |
| RISE-7BBackbone=Qwen2.5-7B, Training=RISE2026.05 | 83.8 | |
| GPT-4oModel=GPT-4o2026.05 | 83.4 | |
| FABSVer-1.5BBackbone=Qwen2.5-1.5B, Training=FABSVer2026.05 | 81 | |
| RISE-3BBackbone=Qwen2.5-3B, Training=RISE2026.05 | 81 | |
| Qwen2.5-7B InstructBackbone=Qwen2.5-7B, Training=Instruct2026.05 | 77.2 | |
| RISE-1.5BBackbone=Qwen2.5-1.5B, Training=RISE2026.05 | 75.9 | |
| Qwen2.5-7B Zero-RLBackbone=Qwen2.5-7B, Training=Zero-RL2026.05 | 75.9 | |
| Qwen2.5-3B InstructBackbone=Qwen2.5-3B, Training=Instruct2026.05 | 65.6 | |
| Qwen2.5-3B Zero-RLBackbone=Qwen2.5-3B, Training=Zero-RL2026.05 | 64.9 | |
| Qwen2.5-7B BaseBackbone=Qwen2.5-7B, Training=Base2026.05 | 58.4 | |
| Qwen2.5-1.5B Zero-RLBackbone=Qwen2.5-1.5B, Training=Zero-RL2026.05 | 54.1 | |
| Qwen2.5-1.5B InstructBackbone=Qwen2.5-1.5B, Training=Instruct2026.05 | 48.8 | |
| Qwen2.5-7B SFTBackbone=Qwen2.5-7B, Training=SFT2026.05 | 40.5 | |
| Qwen2.5-3B BaseBackbone=Qwen2.5-3B, Training=Base2026.05 | 39.5 | |
| Qwen2.5-3B SFTBackbone=Qwen2.5-3B, Training=SFT2026.05 | 21.5 | |
| Qwen2.5-1.5B BaseBackbone=Qwen2.5-1.5B, Training=Base2026.05 | 19.4 | |
| Qwen2.5-1.5B SFTBackbone=Qwen2.5-1.5B, Training=SFT2026.05 | 19 |