Math Reasoning on Math Reasoning Tasks Task-average (test)
41.5Pass@1OrderGrad (Top2@4)
Evaluation Results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| OrderGrad (Top2@4)Backbone=Qwen3-4B-Base, Selection Strategy=Top2@42026.06 | 41.5 | 49.4 | 55.5 | 60.7 | 65.6 | 70.2 | 74.2 | 77.8 | 81 | |
| GRPOBackbone=Qwen3-4B-Base2026.06 | 41.1 | 46.6 | 50.9 | 54.4 | 57.8 | 61.3 | 64.6 | 68 | 71.9 | |
| MaxPOBackbone=Qwen3-4B-Base, K=22026.06 | 40.3 | 47.4 | 53.1 | 58.1 | 62.8 | 67.4 | 71.4 | 74.9 | 78.1 | |
| OrderGrad (Top1@2)Backbone=Qwen3-4B-Base, Selection Strategy=Top1@22026.06 | 38.9 | 46 | 52 | 57.5 | 62.7 | 67.5 | 71.5 | 75.1 | 78.5 | |
| OrderGrad (Top3@6)Backbone=Qwen3-4B-Base, Selection Strategy=Top3@62026.06 | 37.8 | 47.5 | 54.5 | 60.1 | 65.2 | 69.8 | 73.9 | 77.2 | 80.2 | |
| OrderGrad (Top2@6)Backbone=Qwen3-4B-Base, Selection Strategy=Top2@62026.06 | 37 | 46.1 | 53.1 | 58.9 | 64.2 | 69.2 | 73.8 | 77.7 | 81.2 | |
| MaxPOBackbone=Qwen3-4B-Base, K=42026.06 | 35.9 | 45.3 | 52.2 | 57.5 | 62.1 | 66.5 | 71 | 75 | 78.5 | |
| MaxPOBackbone=Qwen3-4B-Base, K=62026.06 | 31.9 | 42.2 | 50 | 56 | 61 | 65.5 | 69.6 | 73.4 | 76.8 | |
| baseBackbone=Qwen3-4B-Base2026.06 | 31 | 39.9 | 47.5 | 54 | 59.4 | 64.3 | 68.8 | 72.9 | 76.4 |