Truthfulness on TruthfulQA 4 options (test)
80.1Accuracy (ACC)KP
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| KPModel=Qwen 2.5 7B2025.09 | 80.1 | 100 | 0 | |
| KPModel=Llama 3.1 8B2025.09 | 76.3 | 100 | 0 | |
| KAPPAModel=Llama 3.1 8B, Intervention layers=62025.09 | 73.5 | 77.6 | 0.46 | |
| KAPPAModel=Llama 3.1 8B, Intervention layers=12025.09 | 67.8 | 78.8 | 0.6 | |
| KAPPAModel=Llama 3.1 8B, Intervention layers=32025.09 | 65.6 | 72.9 | 0.48 | |
| KAPPAModel=Qwen 2.5 7B, Intervention layers=62025.09 | 64.1 | 67.3 | 0.95 | |
| CAAModel=Llama 3.1 8B2025.09 | 62.3 | 67.2 | 0.6 | |
| KAPPAModel=Qwen 2.5 7B, Intervention layers=32025.09 | 61.9 | 65.1 | 0.97 | |
| CAAModel=Qwen 2.5 7B2025.09 | 61.1 | 63.5 | 0.98 | |
| KAPPAModel=Qwen 2.5 7B, Intervention layers=12025.09 | 60.6 | 64 | 0.99 | |
| BaseModel=Qwen 2.5 7B2025.09 | 58.8 | 61.8 | 1.01 | |
| DoLAModel=Qwen 2.5 7B2025.09 | 58.5 | 61.3 | 1.02 | |
| BaseModel=Llama 3.1 8B2025.09 | 56.7 | 62.1 | 0.63 | |
| DoLAModel=Llama 3.1 8B2025.09 | 55.6 | 61.6 | 0.76 |