Hallucination Detection on TriviaQA (val)
0.8297AUROC (Token Entropy)UA-CLM
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| UA-CLMModel=Llama-2-13B2024.12 | 0.8297 | 0.8352 | 0.8033 | 0.8447 | |
| UA-CLMModel=Llama-2-7B2024.12 | 0.8293 | 0.8393 | 0.8197 | 0.8423 | |
| CLMModel=Llama-2-13B2024.12 | 0.8264 | 0.8333 | 0.7971 | 0.8407 | |
| ULTModel=Llama-2-13B2024.12 | 0.824 | 0.8485 | 0.8245 | 0.8456 | |
| CLMModel=Llama-2-7B2024.12 | 0.8135 | 0.8192 | 0.8108 | 0.8371 | |
| UA-CLMModel=Gemma-2B2024.12 | 0.8085 | 0.8211 | 0.796 | 0.8228 | |
| CLMModel=Gemma-2B2024.12 | 0.803 | 0.8138 | 0.7989 | 0.8018 | |
| Pre-trainedModel=Llama-2-13B2024.12 | 0.7984 | 0.8123 | 0.7801 | 0.8365 | |
| ULTModel=Gemma-2B2024.12 | 0.7935 | 0.8134 | 0.7912 | 0.8035 | |
| CTModel=Llama-2-7B2024.12 | 0.7714 | 0.8211 | 0.8037 | 0.8233 | |
| Pre-trainedModel=Llama-2-7B2024.12 | 0.7687 | 0.822 | 0.8191 | 0.8315 | |
| ULTModel=Llama-2-7B2024.12 | 0.7676 | 0.8003 | 0.8004 | 0.8276 | |
| Pre-trainedModel=Gemma-2B2024.12 | 0.7633 | 0.7719 | 0.792 | 0.8127 | |
| CTModel=Llama-2-13B2024.12 | 0.7338 | 0.7897 | 0.7991 | 0.8222 |