Instruction-following on Just-Eval
4.25HelpfulnessICD
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| ICDModel=Vicuna2024.02 | 4.25 | 4.892 | 4.48 | 3.821 | 4.509 | 4.39 | |
| No DefenseModel=Vicuna2024.02 | 4.247 | 4.778 | 4.34 | 3.922 | 4.435 | 4.344 | |
| Self-ExaminationModel=Vicuna2024.02 | 4.207 | 4.758 | 4.322 | 3.877 | 4.395 | 4.312 | |
| No DefenseModel=Llama22024.02 | 4.146 | 4.892 | 4.424 | 3.974 | 4.791 | 4.445 | |
| SafeDecodingModel=Vicuna, m=2, c=5, alpha=32024.02 | 4.072 | 4.842 | 4.402 | 3.714 | 4.452 | 4.296 | |
| ParaphraseModel=Vicuna2024.02 | 3.981 | 4.702 | 4.174 | 3.742 | 4.324 | 4.185 | |
| SafeDecodingModel=Llama2, m=2, c=5, alpha=32024.02 | 3.926 | 4.824 | 4.343 | 3.825 | 4.66 | 4.32 | |
| ParaphraseModel=Llama22024.02 | 3.909 | 4.794 | 4.238 | 3.809 | 4.67 | 4.284 | |
| ICDModel=Llama22024.02 | 3.524 | 4.527 | 3.934 | 3.516 | 4.269 | 3.954 | |
| Self-ExaminationModel=Llama22024.02 | 1.504 | 3.025 | 2.348 | 1.482 | 1.77 | 2.206 |