Question Answering on HotpotQA (Acc, AUROC, Brier, ECE)
63AccuracyRLVR
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| RLVR2025.07 | 63 | 0.5 | 0.37 | 0.37 | |
| RLVR + BCE Classifier2025.07 | 63 | 0.66 | 0.22 | 0.07 | |
| RLVR + Brier Classifier2025.07 | 63 | 0.65 | 0.22 | 0.09 | |
| RLVR + Probe2025.07 | 63 | 0.55 | 0.24 | 0.1 | |
| Answer Probability2025.07 | 63 | 0.72 | 0.36 | 0.36 | |
| RLCR2025.07 | 62.1 | 0.69 | 0.21 | 0.03 | |
| Base2025.07 | 39.7 | 0.54 | 0.53 | 0.53 |