Uncertainty Estimation on MuSiQue
65.6AUROCPredictive Entropy
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Predictive EntropyModel=Phi-3.8B2025.06 | 65.6 | 29.6 | — | |
| P(YES) (naive avg)Model=Phi-3.8B2025.06 | 65.5 | 29.5 | 63.2 | |
| Internal ConfidenceModel=Phi-3.8B2025.06 | 65.5 | 30.2 | 9.3 | |
| P(YES) (top right)Model=Phi-3.8B2025.06 | 65.4 | 29.8 | 9.5 | |
| Internal ConfidenceModel=Llama-8B2025.06 | 64.3 | 29.8 | 74.4 | |
| P(YES) (top right)Model=Llama-8B2025.06 | 64.1 | 27.3 | 74.5 | |
| P(YES) (naive avg)Model=Llama-8B2025.06 | 63.2 | 31.9 | 18.5 | |
| Min-K EntropyModel=Phi-3.8B2025.06 | 59.9 | 21.9 | — | |
| P(YES) (naive avg)Model=Qwen-14B2025.06 | 58.4 | 16.9 | 6.5 | |
| P(YES) (top right)Model=Qwen-14B2025.06 | 57.7 | 14.4 | 22.4 | |
| Internal ConfidenceModel=Qwen-14B2025.06 | 57.6 | 14.6 | 5.4 | |
| Attentional EntropyModel=Llama-8B2025.06 | 57.4 | 17.7 | — | |
| Predictive EntropyModel=Llama-8B2025.06 | 56.4 | 13.1 | — | |
| Min-K EntropyModel=Llama-8B2025.06 | 56 | 15.2 | — | |
| PerplexityModel=Phi-3.8B2025.06 | 55.2 | 7.1 | — | |
| Attentional EntropyModel=Qwen-14B2025.06 | 54.9 | 13.2 | — | |
| Attentional EntropyModel=Phi-3.8B2025.06 | 54.7 | 11.6 | — | |
| PerplexityModel=Llama-8B2025.06 | 54.2 | 5.9 | — | |
| Min-K EntropyModel=Qwen-14B2025.06 | 54.1 | 2.3 | — | |
| Predictive EntropyModel=Qwen-14B2025.06 | 53.8 | 9.4 | — | |
| PerplexityModel=Qwen-14B2025.06 | 53.7 | — | — | |
| Max(− log p)Model=Phi-3.8B2025.06 | 53.2 | 4.7 | — | |
| Max(− log p)Model=Llama-8B2025.06 | 53.2 | 6.3 | — | |
| Max(− log p)Model=Qwen-14B2025.06 | 52.5 | 6.8 | — |