Confidence Estimation on Global-MMLU Japanese
0.72AUROCLast Answer
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Last AnswerProbe hidden state=last answer token, Training Language=French, Test Language=Japanese, Zero-shot Transfer=true2026.05 | 0.72 | 0.43 | 0.28 | 0.29 | |
| Last QueryProbe hidden state=last query token, Training Language=French, Test Language=Japanese, Zero-shot Transfer=true2026.05 | 0.69 | 0.32 | 0.2 | 0.18 |