Mathematical Reasoning on AIME 25 (avg@16)
25.3Avg@16Scaf-GRPO
Evaluation Results
| Method | Links | |
|---|---|---|
| Scaf-GRPOBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.10 | 25.3 | |
| Vanilla GRPOBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.10 | 22.9 | |
| Scaf-GRPOBackbone=Qwen2.5-Math-7B2025.10 | 14.6 | |
| LUFFYBackbone=Qwen2.5-Math-7B2025.10 | 12 | |
| Oat-ZeroBackbone=Qwen2.5-Math-7B2025.10 | 11.5 | |
| SimpleRL-ZeroBackbone=Qwen2.5-Math-7B2025.10 | 11 | |
| Scaf-GRPOBackbone=Qwen2.5-7B2025.10 | 11 | |
| Scaf-GRPOBackbone=Qwen2.5-Math-1.5B2025.10 | 10.8 | |
| Vanilla GRPOBackbone=Qwen2.5-Math-7B2025.10 | 10.8 | |
| Vanilla GRPOBackbone=Qwen2.5-7B2025.10 | 9.5 | |
| Vanilla GRPOBackbone=Qwen2.5-Math-1.5B2025.10 | 8.2 | |
| Scaf-GRPOBackbone=Llama-3.2-3B-Instruct2025.10 | 0.3 | |
| Vanilla GRPOBackbone=Llama-3.2-3B-Instruct2025.10 | 0 |