General Reasoning on EvoLM-4B OOD Benchmarks
13.2GPQA ScoreModSFT
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| ModSFTRL=Yes2026.06 | 13.2 | 22.9 | 6.2 | 14.1 | 17 | |
| RejuvenationRL=Yes2026.06 | 12.9 | 26.2 | 6.9 | 15.3 | 17.5 | |
| OverSFTRL=Yes2026.06 | 9.9 | 21.3 | 5.6 | 12.2 | 14.9 | |
| DFTRL=Yes2026.06 | 8.3 | 22.3 | 5.7 | 12.1 | 15.2 | |
| OverSFTRL=No2026.06 | 7.8 | 13.7 | 3.3 | 8.3 | 10 | |
| DFTRL=No2026.06 | 5.8 | 7.2 | 3.3 | 5.4 | 8.5 | |
| ModSFTRL=No2026.06 | 5.3 | 13.4 | 3.4 | 7.4 | 9.1 | |
| RejuvenationRL=No2026.06 | 3.6 | 2.6 | 2.1 | 2.8 | 5.4 |