Question Answering on HotpotQA (AUROC, AUPRC)
63.1AUROCP(True) + CoT-UQ
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| P(True) + CoT-UQBackbone=Llama 3.1-8B, CoT-UQ=true2025.02 | 63.1 | 37.29 | |
| P(True)Backbone=Llama 3.1-8B, CoT-UQ=false2025.02 | 62.39 | 35.68 | |
| Probas-mean + CoT-UQBackbone=Llama 3.1-8B, CoT-UQ=true2025.02 | 62.01 | 32.25 | |
| TOKENSAR + CoT-UQBackbone=Llama 3.1-8B, CoT-UQ=true2025.02 | 61.07 | 31.27 | |
| Probas-meanBackbone=Llama 3.1-8B, CoT-UQ=false2025.02 | 53.73 | 29.14 | |
| TOKENSARBackbone=Llama 3.1-8B, CoT-UQ=false2025.02 | 53.57 | 28.41 |