Mathematical Reasoning on AIME24 (Score)
19.7AIME24 ScoreSGPO
Evaluation results
| Method | Links | |
|---|---|---|
| SGPOBackbone=Qwen2.5-7B-Instruct2026.06 | 19.7 | |
| SFT+GRPOBackbone=Qwen2.5-7B-Instruct2026.06 | 18 | |
| HPTBackbone=Qwen2.5-7B-Instruct2026.06 | 17 | |
| LUFFYBackbone=Qwen2.5-7B-Instruct2026.06 | 16.4 | |
| SFTBackbone=Qwen2.5-7B-Instruct2026.06 | 12.4 | |
| Qwen2.5-7B-InstructBackbone=Qwen2.5-7B-Instruct2026.06 | 11.8 | |
| SGPOBackbone=Llama-3.2-8B-Instruct2026.06 | 10 | |
| HPTBackbone=Llama-3.2-8B-Instruct2026.06 | 9.8 | |
| LUFFYBackbone=Llama-3.2-8B-Instruct2026.06 | 9.4 | |
| SGPOBackbone=Qwen2.5-1.5B-Instruct2026.06 | 9 | |
| LUFFYBackbone=Qwen2.5-1.5B-Instruct2026.06 | 8.4 | |
| SFT+GRPOBackbone=Llama-3.2-8B-Instruct2026.06 | 8 | |
| HPTBackbone=Qwen2.5-1.5B-Instruct2026.06 | 7.7 | |
| SFTBackbone=Llama-3.2-8B-Instruct2026.06 | 6 | |
| SFT+GRPOBackbone=Qwen2.5-1.5B-Instruct2026.06 | 5.5 | |
| SFTBackbone=Qwen2.5-1.5B-Instruct2026.06 | 4.7 | |
| Qwen2.5-1.5B-InstructBackbone=Qwen2.5-1.5B-Instruct2026.06 | 3 | |
| Llama-3.2-8B-InstructBackbone=Llama-3.2-8B-Instruct2026.06 | 3 |