Question Answering on Out-of-domain QA (ARC-C, GPQA, MMLU-Pro) (test)
84.7ARC-C ScoreExGRPO
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| ExGRPOSource domain=mathematical reasoning, re-tuning=none, transfer_mode=direct2026.05 | 84.7 | 37.4 | 52.9 | 58.3 | |
| DoTS (ExGRPO + ReLIFT)Source domain=mathematical reasoning, re-tuning=none, transfer_mode=direct2026.05 | 84.6 | 42.9 | 53.9 | 60.5 | |
| ReLIFTSource domain=mathematical reasoning, re-tuning=none, transfer_mode=direct2026.05 | 82.8 | 36.9 | 53.9 | 57.9 | |
| LUFFYSource domain=mathematical reasoning, re-tuning=none, transfer_mode=direct2026.05 | 80.5 | 39.9 | 53 | 57.8 |