Hallucination Detection on SQuAD v2
0.798ROC-AUCSinkProbe
Evaluation Results
| Method | Links | |
|---|---|---|
| SinkProbeLLM=Llama3.1-8B2026.04 | 0.798 | |
| SinkProbeLLM=Mistral-Nemo2026.04 | 0.79 | |
| MTopDivLLM=Llama3.1-8B2026.04 | 0.787 | |
| LapEigvalLLM=Llama3.1-8B2026.04 | 0.785 | |
| LookbackLensLLM=Mistral-Nemo2026.04 | 0.783 | |
| MTopDivLLM=Phi3.52026.04 | 0.78 | |
| LapEigvalLLM=Mistral-Nemo2026.04 | 0.777 | |
| LookbackLensLLM=Llama3.1-8B2026.04 | 0.776 | |
| MTopDivLLM=Mistral-Nemo2026.04 | 0.77 | |
| SinkProbeLLM=Llama3.2-3B2026.04 | 0.769 | |
| LapEigvalLLM=Phi3.52026.04 | 0.767 | |
| LookbackLensLLM=Phi3.52026.04 | 0.767 | |
| LookbackLensLLM=Llama3.2-3B2026.04 | 0.765 | |
| MTopDivLLM=Llama3.2-3B2026.04 | 0.764 | |
| SinkProbeLLM=Phi3.52026.04 | 0.764 | |
| AttnEigvalLLM=Llama3.1-8B2026.04 | 0.761 | |
| AttnLogDetLLM=Phi3.52026.04 | 0.76 | |
| LapEigvalLLM=Llama3.2-3B2026.04 | 0.753 | |
| AttnLogDetLLM=Llama3.1-8B2026.04 | 0.752 | |
| AttnEigvalLLM=Phi3.52026.04 | 0.75 | |
| AttnLogDetLLM=Mistral-Nemo2026.04 | 0.749 | |
| AttnEigvalLLM=Mistral-Nemo2026.04 | 0.745 | |
| AttnLogDetLLM=Llama3.2-3B2026.04 | 0.739 | |
| AttnEigvalLLM=Llama3.2-3B2026.04 | 0.738 | |
| AttnScoreLLM=Phi3.52026.04 | 0.678 | |
| AttnScoreLLM=Llama3.1-8B2026.04 | 0.656 | |
| AttnScoreLLM=Mistral-Nemo2026.04 | 0.641 | |
| AttnScoreLLM=Llama3.2-3B2026.04 | 0.637 |