Mathematical Reasoning on MATH500 (Accuracy (Pass@1) and Token Usage)
96Accuracy (Pass@1)RFTshortest
Evaluation results
| Method | Links | ||
|---|---|---|---|
| RFTshortestBackbone=Qwen3-4B-Thinking, Evaluation Protocol=Pass@12026.02 | 96 | 6,062 | |
| ConPressBackbone=Qwen3-4B-Thinking, Evaluation Protocol=Pass@12026.02 | 96 | 2,661 | |
| OriginalBackbone=Qwen3-4B-Thinking, Evaluation Protocol=Pass@12026.02 | 95.6 | 6,634 | |
| DPOshortestBackbone=Qwen3-4B-Thinking, Evaluation Protocol=Pass@12026.02 | 95.2 | 5,616 | |
| RFTshortestBackbone=R1-Distill-Qwen-7B, Evaluation Protocol=Pass@12026.02 | 92.8 | 3,148 | |
| ConPressBackbone=R1-Distill-Qwen-7B, Evaluation Protocol=Pass@12026.02 | 92.4 | 1,720 | |
| AdaptThinkBackbone=R1-Distill-Qwen-7B, Evaluation Protocol=Pass@12026.02 | 91.8 | 1,550 | |
| OriginalBackbone=R1-Distill-Qwen-7B, Evaluation Protocol=Pass@12026.02 | 91.6 | 3,136 | |
| DPOshortestBackbone=R1-Distill-Qwen-7B, Evaluation Protocol=Pass@12026.02 | 90.4 | 1,918 | |
| LC-R1Backbone=R1-Distill-Qwen-7B, Evaluation Protocol=Pass@12026.02 | 87.6 | 1,487 | |
| ThinkPruneBackbone=R1-Distill-Qwen-1.5B, Evaluation Protocol=Pass@12026.02 | 83 | 2,587 | |
| RFTshortestBackbone=R1-Distill-Qwen-1.5B, Evaluation Protocol=Pass@12026.02 | 82.4 | 4,625 | |
| DPOshortestBackbone=R1-Distill-Qwen-1.5B, Evaluation Protocol=Pass@12026.02 | 82 | 3,133 | |
| OriginalBackbone=R1-Distill-Qwen-1.5B, Evaluation Protocol=Pass@12026.02 | 81.2 | 4,622 | |
| ConPressBackbone=R1-Distill-Qwen-1.5B, Evaluation Protocol=Pass@12026.02 | 80.8 | 2,255 | |
| AdaptThinkBackbone=R1-Distill-Qwen-1.5B, Evaluation Protocol=Pass@12026.02 | 80.4 | 1,945 | |
| LC-R1Backbone=R1-Distill-Qwen-1.5B, Evaluation Protocol=Pass@12026.02 | 80.2 | 2,313 |