Mathematical Reasoning on AMC23 (Avg@32)
90Avg@32DSRL
Evaluation Results
| Method | Links | |
|---|---|---|
| DSRLBackbone=Qwen3-8B2026.04 | 90 | |
| PPOBackbone=Qwen3-8B2026.04 | 89.69 | |
| DSRLBackbone=Qwen3-4B2026.04 | 89.22 | |
| DAPOBackbone=Qwen3-8B2026.04 | 88.52 | |
| DAPOBackbone=Qwen3-4B2026.04 | 88.2 | |
| GRPOBackbone=Qwen3-8B2026.04 | 88.05 | |
| PPOBackbone=Qwen3-4B2026.04 | 87.89 | |
| GRPOBackbone=Qwen3-4B2026.04 | 87.81 | |
| Dr.GRPOBackbone=Qwen3-8B2026.04 | 87.66 | |
| Dr.GRPOBackbone=Qwen3-4B2026.04 | 87.58 | |
| RLOOBackbone=Qwen3-8B2026.04 | 87.27 | |
| RLOOBackbone=Qwen3-4B2026.04 | 85 | |
| Reinforce++Backbone=Qwen3-8B2026.04 | 84.14 | |
| Reinforce++Backbone=Qwen3-4B2026.04 | 79.37 | |
| VanillaBackbone=Qwen3-8B2026.04 | 69.53 | |
| VanillaBackbone=Qwen3-4B2026.04 | 68.28 |