Selective Generation on TruthfulQA (PRR AlignScore)
35.3PRR (AlignScore)SATRMD+MSP
Evaluation results
| Method | Links | |
|---|---|---|
| SATRMD+MSPModel=Llama 8b v3.12025.02 | 35.3 | |
| HUQ-SATRMDModel=Llama 8b v3.12025.02 | 30.8 | |
| Maximum Sequence ProbabilityModel=Llama 8b v3.12025.02 | 27.7 | |
| SentenceSARModel=Llama 8b v3.12025.02 | 18.5 | |
| PerplexityModel=Llama 8b v3.12025.02 | 17.8 | |
| Semantic EntropyModel=Llama 8b v3.12025.02 | 17.1 | |
| DegMat NLI Score Entail.Model=Llama 8b v3.12025.02 | 15.6 | |
| EigValLaplacian NLI Score Entail.Model=Llama 8b v3.12025.02 | 15.2 | |
| Eccentricity NLI Score Entail.Model=Llama 8b v3.12025.02 | 12.2 | |
| SAPLMAModel=Llama 8b v3.12025.02 | 11.2 | |
| SARModel=Llama 8b v3.12025.02 | 10.5 | |
| EigenScoreModel=Llama 8b v3.12025.02 | 2.3 | |
| FactoscopeModel=Llama 8b v3.12025.02 | 1.7 | |
| Lexical Similarity ROUGE-LModel=Llama 8b v3.12025.02 | 0.8 |