Mathematical Reasoning on EvoLM-4B In-Distribution Math Benchmarks
51.6GSM8K ScoreModSFT
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| ModSFTRL=Yes2026.06 | 51.6 | 25.7 | 6.4 | 11.6 | 4.7 | 20 | |
| RejuvenationRL=Yes2026.06 | 49.1 | 26.2 | 8 | 10.8 | 4.2 | 19.7 | |
| OverSFTRL=Yes2026.06 | 45.2 | 22 | 6.6 | 9.9 | 4.1 | 17.5 | |
| DFTRL=Yes2026.06 | 43.6 | 24.4 | 7.5 | 11.9 | 3.9 | 18.3 | |
| OverSFTRL=No2026.06 | 31.7 | 15.8 | 3.8 | 5.1 | 2.4 | 11.8 | |
| DFTRL=No2026.06 | 30.9 | 14.5 | 5.3 | 5.2 | 2.3 | 11.7 | |
| ModSFTRL=No2026.06 | 28.1 | 15.2 | 4.3 | 4 | 2.4 | 10.8 | |
| RejuvenationRL=No2026.06 | 18.5 | 12 | 3 | 4.3 | 2.5 | 8.1 |