Math Word Problems on GSM8K (ACC and Length)
91.4AccuracyADaPTthink
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ADaPTthinkBackbone=Qwen2.5-7B2026.06 | 91.4 | 803 | |
| SFT+GRPOBackbone=Qwen2.5-7B2026.06 | 91.1 | 801 | |
| ADaPTBackbone=Qwen2.5-7B2026.06 | 90.8 | 348 | |
| ARMBackbone=Qwen2.5-7B2026.06 | 89.2 | 297 | |
| SFT+GRPOBackbone=Qwen2.5-3B2026.06 | 88.9 | 873 | |
| R-4BBackbone=Qwen2.5-7B2026.06 | 88.6 | 310 | |
| ADaPTthinkBackbone=Qwen2.5-3B2026.06 | 88.6 | 816 | |
| TLMREBackbone=Qwen2.5-7B2026.06 | 88.3 | 584 | |
| ADaPTBackbone=Qwen2.5-3B2026.06 | 86.3 | 317 | |
| TLMREBackbone=Qwen2.5-3B2026.06 | 85.8 | 674 | |
| ADaPTanswerBackbone=Qwen2.5-7B2026.06 | 84.7 | 152 | |
| ARMBackbone=Qwen2.5-3B2026.06 | 84.2 | 286 | |
| R-4BBackbone=Qwen2.5-3B2026.06 | 83.6 | 427 | |
| BaseBackbone=Qwen2.5-7B2026.06 | 81.8 | 332 | |
| ADaPTanswerBackbone=Qwen2.5-3B2026.06 | 80.1 | 177 | |
| SFTBackbone=Qwen2.5-7B2026.06 | 68.3 | 541 | |
| BaseBackbone=Qwen2.5-3B2026.06 | 64.1 | 258 | |
| SFTBackbone=Qwen2.5-3B2026.06 | 48.3 | 194 |