Confidence Estimation on MKQA Russian / ru (test)
80AUROCCross-lingual Probe
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Cross-lingual ProbeBackbone=Qwen 3 8B, Evaluation Protocol=Zero-shot2026.05 | 80 | 43 | 15 | 16 | |
| Seq. LikelihoodBackbone=Qwen 3 8B, Evaluation Protocol=Zero-shot2026.05 | 79 | 35 | 49 | 62 | |
| Mass-Mean ProbeBackbone=Qwen 3 8B, Evaluation Protocol=Zero-shot2026.05 | 73 | 51 | 40 | 12 | |
| Verbalized Conf.Backbone=Qwen 3 8B, Evaluation Protocol=Zero-shot2026.05 | 72 | 36 | 55 | 56 | |
| P(True)Backbone=Qwen 3 8B, Evaluation Protocol=Zero-shot2026.05 | 43 | 12 | 12 | 12 |