Automated Short Answer Scoring on Private-Physics (OOD)
45.2ACCEDIT-SFT
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| EDIT-SFTStrategy=EDIT, Training=SFT2026.06 | 45.2 | 81.3 | 0.784 | 0.472 | |
| EDIT-RLStrategy=EDIT, Training=RL2026.06 | 44.5 | 81.7 | 0.772 | 0.487 | |
| DGPOStrategy=RL-only2026.06 | 43.5 | 79.3 | 0.82 | 0.449 | |
| InTStrategy=InT, Training=RL2026.06 | 43.3 | 81.5 | 0.84 | 0.409 | |
| GRPOStrategy=RL-only2026.06 | 43.1 | 80.8 | 0.805 | 0.465 | |
| BaseStrategy=Base2026.06 | 42.9 | 79.2 | 0.829 | 0.456 | |
| InTStrategy=InT, Training=SFT2026.06 | 37.4 | 78.4 | 0.95 | 0.393 |