Mathematical Reasoning on Math500 (pass@8)
82.6Pass@8VI-CuRL
Evaluation results
| Method | Links | |
|---|---|---|
| VI-CuRLBackbone=Qwen2.5-Math-7B, Reward Setting=Oracle (w. Verifier)2026.02 | 82.6 | |
| No CurriculumBackbone=Qwen2.5-Math-7B, Reward Setting=Oracle (w. Verifier)2026.02 | 82.4 | |
| VI-CuRLBackbone=DeepSeek-R1-Distill-Qwen-1.5B, Intrinsic Reward Signal=Majority Vote2026.02 | 82 | |
| VCRLBackbone=Qwen2.5-Math-7B, Reward Setting=Oracle (w. Verifier)2026.02 | 79.8 | |
| AdaRFTBackbone=DeepSeek-R1-Distill-Qwen-1.5B, Reward Setting=Oracle (w. Verifier)2026.02 | 78.6 | |
| AdaRFTBackbone=Qwen2.5-Math-7B, Reward Setting=Oracle (w. Verifier)2026.02 | 78 | |
| VI-CuRLBackbone=DeepSeek-R1-Distill-Qwen-1.5B, Reward Setting=Oracle (w. Verifier)2026.02 | 75.7 | |
| VI-CuRLBackbone=Qwen2.5-Math-1.5B, Intrinsic Reward Signal=Majority Vote2026.02 | 73.9 | |
| VCRLBackbone=DeepSeek-R1-Distill-Qwen-1.5B, Reward Setting=Oracle (w. Verifier)2026.02 | 72.9 | |
| AdaRFTBackbone=Qwen2.5-Math-1.5B, Reward Setting=Oracle (w. Verifier)2026.02 | 72.4 | |
| VI-CuRLBackbone=Qwen2.5-Math-1.5B, Reward Setting=Oracle (w. Verifier)2026.02 | 71.3 | |
| No CurriculumBackbone=DeepSeek-R1-Distill-Qwen-1.5B, Reward Setting=Oracle (w. Verifier)2026.02 | 71.2 | |
| No CurriculumBackbone=Qwen2.5-Math-1.5B, Reward Setting=Oracle (w. Verifier)2026.02 | 70.9 | |
| VCRLBackbone=Qwen2.5-Math-1.5B, Reward Setting=Oracle (w. Verifier)2026.02 | 70.2 | |
| BaseBackbone=DeepSeek-R1-Distill-Qwen-1.5B, Reward Setting=No RL2026.02 | 67.6 | |
| BaseBackbone=Qwen2.5-Math-7B, Reward Setting=No RL2026.02 | 66.6 | |
| BaseBackbone=Qwen2.5-Math-1.5B, Reward Setting=No RL2026.02 | 62.8 | |
| VI-CuRLBackbone=Llama3.2-3B-Instruct, Reward Setting=Oracle (w. Verifier)2026.02 | 49.6 | |
| AdaRFTBackbone=Llama3.2-3B-Instruct, Reward Setting=Oracle (w. Verifier)2026.02 | 45.6 | |
| No CurriculumBackbone=Llama3.2-3B-Instruct, Reward Setting=Oracle (w. Verifier)2026.02 | 42.5 | |
| BaseBackbone=Llama3.2-3B-Instruct, Reward Setting=No RL2026.02 | 40.8 | |
| VCRLBackbone=Llama3.2-3B-Instruct, Reward Setting=Oracle (w. Verifier)2026.02 | 40.8 |