Hallucination Detection on HaloQuest
0.834ROCLogistic
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| LogisticBackbone=LLaVA-8B2026.01 | 0.834 | 0.84 | 74.6 | 85.7 | 1,874.895 | 0.114 | |
| FaithSCANBackbone=Qwen3-VL-8B2026.01 | 0.829 | 0.941 | 53.2 | 97.5 | 1,741.652 | 1.191 | |
| FaithSCANBackbone=LLaVA-8B2026.01 | 0.827 | 0.848 | 76.2 | 90.2 | 1,898.95 | 0.741 | |
| LogisticBackbone=Qwen3-VL-8B2026.01 | 0.795 | 0.887 | 44.6 | 91 | 1,689.458 | 0.102 | |
| SelfCheckGPTBackbone=Qwen3-VL-8B2026.01 | 0.783 | 0.878 | 50.1 | 90.3 | 2,165.013 | 367.522 | |
| LogisticBackbone=IB-T5-XL2026.01 | 0.735 | 0.833 | 85.5 | 80.7 | 1,570.349 | 0.087 | |
| P(True)Backbone=LLaVA-8B2026.01 | 0.716 | 0.761 | 55.3 | 73 | 1,178.084 | 19.511 | |
| FaithSCANBackbone=IB-T5-XL2026.01 | 0.713 | 0.854 | 87 | 88.5 | 1,560.237 | 0.605 | |
| SEUBackbone=Qwen3-VL-8B2026.01 | 0.678 | 0.868 | 50 | 91.7 | 2,199.427 | 1.821 | |
| SEUBackbone=IB-T5-XL2026.01 | 0.66 | 0.498 | 77.6 | 51.5 | 1,685.975 | 1.845 | |
| EmbVarBackbone=Qwen3-VL-8B2026.01 | 0.641 | 0.875 | 44.4 | 90.2 | 2,194.586 | 0.107 | |
| P(True)Backbone=Qwen3-VL-8B2026.01 | 0.631 | 0.776 | 39.4 | 83.2 | 1,143.72 | 24.299 | |
| SEBackbone=Qwen3-VL-8B2026.01 | 0.606 | 0.873 | 31.6 | 87.3 | 2,019.877 | 13.812 | |
| EmbVarBackbone=IB-T5-XL2026.01 | 0.605 | 0.521 | 79 | 57.8 | 1,430.884 | 0.06 | |
| SelfCheckGPTBackbone=LLaVA-8B2026.01 | 0.603 | 0.576 | 51 | 53.4 | 2,588.092 | 340.683 | |
| SEBackbone=LLaVA-8B2026.01 | 0.595 | 0.28 | 55.7 | 27.5 | 2,632.833 | 11.212 | |
| PEBackbone=IB-T5-XL2026.01 | 0.578 | 0.526 | 80.5 | 53.4 | 1,326.308 | 0.002 | |
| PEBackbone=Qwen3-VL-8B2026.01 | 0.573 | 0.726 | 22.9 | 78.9 | 1,328.434 | 0.002 | |
| T-NLLBackbone=IB-T5-XL2026.01 | 0.571 | 0.55 | 73 | 54.9 | 1,257.512 | 0.001 | |
| EmbVarBackbone=LLaVA-8B2026.01 | 0.546 | 0.355 | 51.5 | 28.4 | 2,504.584 | 0.175 | |
| T-NLLBackbone=Qwen3-VL-8B2026.01 | 0.542 | 0.733 | 25.7 | 78.4 | 1,330.996 | 0.001 | |
| SEBackbone=IB-T5-XL2026.01 | 0.537 | 0.595 | 80.5 | 60.8 | 1,381.929 | 4.913 | |
| SEUBackbone=LLaVA-8B2026.01 | 0.536 | 0.324 | 43.4 | 32.8 | 2,443.89 | 1.769 | |
| P(True)Backbone=IB-T5-XL2026.01 | 0.515 | 0.609 | 77 | 64.2 | 1,127.961 | 14.938 | |
| SelfCheckGPTBackbone=IB-T5-XL2026.01 | 0.474 | 0.623 | 79.7 | 63.7 | 1,602.887 | 78.613 | |
| T-NLLBackbone=LLaVA-8B2026.01 | 0.447 | 0.372 | 44.6 | 37.3 | 1,524.74 | 0.001 | |
| PEBackbone=LLaVA-8B2026.01 | 0.44 | 0.368 | 51.5 | 34.8 | 1,491.448 | 0.002 |