Hallucination Detection on UMWP
89.6ROC-AUCSinkProbe
Evaluation Results
| Method | Links | |
|---|---|---|
| SinkProbeLLM=Phi3.52026.04 | 89.6 | |
| SinkProbeLLM=Llama3.1-8B2026.04 | 87.9 | |
| LookbackLensLLM=Phi3.52026.04 | 87.7 | |
| SinkProbeLLM=Mistral-Nemo2026.04 | 87.6 | |
| MTopDivLLM=Phi3.52026.04 | 87.2 | |
| LookbackLensLLM=Mistral-Nemo2026.04 | 86.6 | |
| AttnLogDetLLM=Phi3.52026.04 | 86.4 | |
| AttnEigvalLLM=Phi3.52026.04 | 86.4 | |
| LapEigvalLLM=Phi3.52026.04 | 86.2 | |
| AttnLogDetLLM=Mistral-Nemo2026.04 | 84.7 | |
| LapEigvalLLM=Mistral-Nemo2026.04 | 84.7 | |
| MTopDivLLM=Mistral-Nemo2026.04 | 84.1 | |
| LookbackLensLLM=Llama3.1-8B2026.04 | 83.8 | |
| LapEigvalLLM=Llama3.1-8B2026.04 | 83.4 | |
| AttnEigvalLLM=Mistral-Nemo2026.04 | 82.6 | |
| AttnLogDetLLM=Llama3.1-8B2026.04 | 82.5 | |
| AttnEigvalLLM=Llama3.1-8B2026.04 | 81.8 | |
| AttnLogDetLLM=Llama3.2-3B2026.04 | 80.9 | |
| MTopDivLLM=Llama3.1-8B2026.04 | 80.9 | |
| AttnScoreLLM=Mistral-Nemo2026.04 | 77.3 | |
| SinkProbeLLM=Llama3.2-3B2026.04 | 75.7 | |
| AttnScoreLLM=Phi3.52026.04 | 75.7 | |
| LookbackLensLLM=Llama3.2-3B2026.04 | 74.9 | |
| MTopDivLLM=Llama3.2-3B2026.04 | 73.5 | |
| AttnScoreLLM=Llama3.1-8B2026.04 | 72.2 | |
| LapEigvalLLM=Llama3.2-3B2026.04 | 72 | |
| AttnEigvalLLM=Llama3.2-3B2026.04 | 70.4 | |
| AttnScoreLLM=Llama3.2-3B2026.04 | 70.2 |