Mathematics on AMC 23 (avg@32 accuracy)
57.9Avg@32 AccuracyIBTPO
Evaluation Results
| Method | Links | |
|---|---|---|
| IBTPOBase Model=Qwen3-8B-Base2026.05 | 57.9 | |
| TreeRLBase Model=Qwen3-8B-Base2026.05 | 56.1 | |
| TreePOBase Model=Qwen3-8B-Base2026.05 | 55.8 | |
| IBROBase Model=Qwen3-8B-Base2026.05 | 55.3 | |
| GRPO w/ Entropy Reg†Base Model=Qwen3-8B-Base2026.05 | 54.3 | |
| Vanilla GRPOBase Model=Qwen3-8B-Base2026.05 | 53.5 | |
| GRPO w/ Clip-higherBase Model=Qwen3-8B-Base2026.05 | 53.3 | |
| Initial ModelBase Model=Qwen3-8B-Base2026.05 | 41.7 | |
| IBTPOBase Model=Qwen3-1.7B-Base2026.05 | 38.7 | |
| TreeRLBase Model=Qwen3-1.7B-Base2026.05 | 37 | |
| TreePOBase Model=Qwen3-1.7B-Base2026.05 | 36.5 | |
| Vanilla GRPOBase Model=Qwen3-1.7B-Base2026.05 | 35.6 | |
| GRPO w/ Clip-higherBase Model=Qwen3-1.7B-Base2026.05 | 35 | |
| IBROBase Model=Qwen3-1.7B-Base2026.05 | 34.6 | |
| GRPO w/ Entropy Reg†Base Model=Qwen3-1.7B-Base2026.05 | 33.6 | |
| Initial ModelBase Model=Qwen3-1.7B-Base2026.05 | 24.5 |