Mathematical Reasoning on MIN
54.41Accuracy (maj@8)GRPO
Evaluation results
| Method | Links | |
|---|---|---|
| GRPOBackbone=Qwen3-8B-Base2026.04 | 54.41 | |
| DAPOBackbone=Qwen3-8B-Base2026.04 | 53.68 | |
| CDEBackbone=Qwen3-8B-Base2026.04 | 53.68 | |
| DiPOBackbone=Qwen3-8B-Base2026.04 | 52.25 | |
| DAPO w/ ELBackbone=Qwen3-8B-Base2026.04 | 52.21 | |
| DiPOBackbone=Qwen3-4B-Base2026.04 | 50.7 | |
| DAPO w/ ELBackbone=Qwen3-4B-Base2026.04 | 50.4 | |
| GRPOBackbone=Qwen3-4B-Base2026.04 | 50.37 | |
| CDEBackbone=Qwen3-4B-Base2026.04 | 49.63 | |
| DAPOBackbone=Qwen3-4B-Base2026.04 | 49.26 | |
| DAPO w/ ELBackbone=Qwen2.5-7B2026.04 | 43.01 | |
| DAPOBackbone=Qwen2.5-7B2026.04 | 41.91 | |
| GRPOBackbone=Qwen2.5-7B2026.04 | 41.18 | |
| CDEBackbone=Qwen2.5-7B2026.04 | 41.18 | |
| DiPOBackbone=Qwen2.5-7B2026.04 | 41.18 | |
| Base modelBackbone=Qwen3-8B-Base2026.04 | 36.76 | |
| Base modelBackbone=Qwen3-4B-Base2026.04 | 30.15 | |
| Base modelBackbone=Qwen2.5-7B2026.04 | 20.59 |