Mathematical Reasoning on GSM8k Base-7 (test)
49.9AccuracySFT+AdaBack
Evaluation Results
| Method | Links | |
|---|---|---|
| SFT+AdaBackModel Size=3B2025.06 | 49.9 | |
| SFT+RLModel Size=3B2025.06 | 45.4 | |
| AdaBackModel Size=3B2025.06 | 43.9 | |
| SFT+AdaBackModel Size=1B2025.06 | 24.5 | |
| AdaBackModel Size=1B2025.06 | 18.4 | |
| SFT+RLModel Size=1B2025.06 | 14.4 | |
| Base+RLModel Size=3B2025.06 | 4.9 | |
| Base+RLModel Size=1B2025.06 | 4.8 |