Mathematical Reasoning on AIME 2024 (ACC, Length)
15.1Accuracy (ACC)ADaPTthink
Evaluation results
| Method | Links | ||
|---|---|---|---|
| ADaPTthinkBackbone=Qwen2.5-7B2026.06 | 15.1 | 3,772 | |
| SFT+GRPOBackbone=Qwen2.5-7B2026.06 | 14.4 | 3,857 | |
| TLMREBackbone=Qwen2.5-7B2026.06 | 14.2 | 3,491 | |
| ADaPTBackbone=Qwen2.5-7B2026.06 | 13.7 | 3,044 | |
| ARMBackbone=Qwen2.5-7B2026.06 | 13.1 | 3,374 | |
| R-4BBackbone=Qwen2.5-7B2026.06 | 11.7 | 2,883 | |
| ADaPTanswerBackbone=Qwen2.5-7B2026.06 | 6.3 | 1,201 | |
| BaseBackbone=Qwen2.5-7B2026.06 | 4.6 | 923 | |
| ADaPTthinkBackbone=Qwen2.5-3B2026.06 | 4.6 | 3,404 | |
| SFT+GRPOBackbone=Qwen2.5-3B2026.06 | 4.3 | 3,433 | |
| ARMBackbone=Qwen2.5-3B2026.06 | 3.9 | 3,094 | |
| ADaPTBackbone=Qwen2.5-3B2026.06 | 3.8 | 2,872 | |
| TLMREBackbone=Qwen2.5-3B2026.06 | 3.5 | 3,293 | |
| R-4BBackbone=Qwen2.5-3B2026.06 | 2.5 | 2,794 | |
| ADaPTanswerBackbone=Qwen2.5-3B2026.06 | 2 | 1,349 | |
| BaseBackbone=Qwen2.5-3B2026.06 | 1.7 | 1,103 | |
| SFTBackbone=Qwen2.5-7B2026.06 | 1.3 | 1,484 | |
| SFTBackbone=Qwen2.5-3B2026.06 | 0 | 982 |