Safety Alignment on Harmful Dataset (test)
81Harmful ScoreNon-Aligned
Evaluation Results
| Method | Links | |
|---|---|---|
| Non-AlignedHarmful ratio (p)=0.05, Sample number (n)=10002024.02 | 81 | |
| Non-AlignedHarmful ratio (p)=0.2, Sample number (n)=10002024.02 | 79.2 | |
| Non-AlignedHarmful ratio (p)=0.1, Sample number (n)=10002024.02 | 77.6 | |
| Non-AlignedHarmful ratio (p)=0.01, Sample number (n)=10002024.02 | 65.6 | |
| SFTHarmful ratio (p)=0.2, Sample number (n)=10002024.02 | 60 | |
| KLHarmful ratio (p)=0.2, Sample number (n)=10002024.02 | 56.6 | |
| VaccineHarmful ratio (p)=0.2, Sample number (n)=10002024.02 | 56.6 | |
| SFTHarmful ratio (p)=0.1, Sample number (n)=10002024.02 | 55.2 | |
| KLHarmful ratio (p)=0.05, Sample number (n)=10002024.02 | 55.2 | |
| VlguardHarmful ratio (p)=0.1, Sample number (n)=10002024.02 | 54.4 | |
| KLHarmful ratio (p)=clean, Sample number (n)=10002024.02 | 54.4 | |
| VlguardHarmful ratio (p)=0.05, Sample number (n)=10002024.02 | 54 | |
| KLHarmful ratio (p)=0.1, Sample number (n)=10002024.02 | 54 | |
| VlguardHarmful ratio (p)=0.2, Sample number (n)=10002024.02 | 53.6 | |
| KLHarmful ratio (p)=0.01, Sample number (n)=10002024.02 | 53.6 | |
| SFTHarmful ratio (p)=0.05, Sample number (n)=10002024.02 | 52.6 | |
| EWCHarmful ratio (p)=clean, Sample number (n)=10002024.02 | 50.6 | |
| EWCHarmful ratio (p)=0.01, Sample number (n)=10002024.02 | 50.6 | |
| EWCHarmful ratio (p)=0.05, Sample number (n)=10002024.02 | 50.6 | |
| EWCHarmful ratio (p)=0.1, Sample number (n)=10002024.02 | 50.6 | |
| EWCHarmful ratio (p)=0.2, Sample number (n)=10002024.02 | 50.6 | |
| VlguardHarmful ratio (p)=0.01, Sample number (n)=10002024.02 | 50 | |
| SFTHarmful ratio (p)=0.01, Sample number (n)=10002024.02 | 49.8 | |
| VlguardHarmful ratio (p)=clean, Sample number (n)=10002024.02 | 49.4 | |
| SFTHarmful ratio (p)=clean, Sample number (n)=10002024.02 | 48.6 | |
| VaccineHarmful ratio (p)=0.1, Sample number (n)=10002024.02 | 48.2 | |
| VaccineHarmful ratio (p)=0.05, Sample number (n)=10002024.02 | 42.8 | |
| VaccineHarmful ratio (p)=clean, Sample number (n)=10002024.02 | 42.4 | |
| VaccineHarmful ratio (p)=0.01, Sample number (n)=10002024.02 | 42.2 | |
| Non-AlignedHarmful ratio (p)=clean, Sample number (n)=10002024.02 | 34.2 |