Refusal on Llama-3-8B n≈200
0ASRFine-tune
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Fine-tuneAttack Type=BadNet2026.04 | 0 | 9 | — | |
| CleanGenAttack Type=BadNet2026.04 | 0 | 0.5 | — | |
| CleanGenAttack Type=VPI2026.04 | 0 | 1 | — | |
| LCFAttack Type=VPI2026.04 | 0 | — | 10.5 | |
| CROWAttack Type=Sleeper2026.04 | 0 | 28.4 | — | |
| CleanGenAttack Type=StyleBkd2026.04 | 0 | 0.5 | — | |
| PruningAttack Type=CTBA2026.04 | 0 | 85 | — | |
| CleanGenAttack Type=MTBA2026.04 | 0 | 1 | — | |
| LCFAttack Type=MTBA2026.04 | 0 | — | 8.5 | |
| PruningAttack Type=BadNet2026.04 | 0.5 | 78.5 | — | |
| PruningAttack Type=Sleeper2026.04 | 0.5 | 82.5 | — | |
| CleanGenAttack Type=Sleeper2026.04 | 0.5 | 1.5 | — | |
| CleanGenAttack Type=SynBkd2026.04 | 0.5 | 1.5 | — | |
| LCFAttack Type=SynBkd2026.04 | 0.5 | — | 11 | |
| Fine-tuneAttack Type=CTBA2026.04 | 0.5 | 12 | — | |
| CleanGenAttack Type=CTBA2026.04 | 0.5 | 1.5 | — | |
| Fine-tuneAttack Type=Sleeper2026.04 | 1 | 12.5 | — | |
| PruningAttack Type=SynBkd2026.04 | 1 | 69.5 | — | |
| LCFAttack Type=StyleBkd2026.04 | 1.1 | — | 9.5 | |
| PruningAttack Type=StyleBkd2026.04 | 1.6 | 68 | — | |
| CROWAttack Type=BadNet2026.04 | 2 | 26.4 | — | |
| CROWAttack Type=SynBkd2026.04 | 3.5 | 14.9 | — | |
| PruningAttack Type=VPI2026.04 | 4.5 | 82 | — | |
| DecodingAttack Type=SynBkd2026.04 | 5 | 3 | — | |
| LCFAttack Type=Sleeper2026.04 | 5.5 | — | 12.5 | |
| PruningAttack Type=MTBA2026.04 | 5.5 | 81.5 | — | |
| DecodingAttack Type=Sleeper2026.04 | 6 | 1 | — | |
| Fine-tuneAttack Type=SynBkd2026.04 | 6.5 | 7 | — | |
| CROWAttack Type=StyleBkd2026.04 | 6.5 | 23.9 | — | |
| DecodingAttack Type=BadNet2026.04 | 9 | 0.5 | — | |
| CROWAttack Type=CTBA2026.04 | 11.5 | 24.4 | — | |
| LCFAttack Type=BadNet2026.04 | 12 | — | 13 | |
| Fine-tuneAttack Type=MTBA2026.04 | 13.5 | 8.5 | — | |
| CROWAttack Type=MTBA2026.04 | 13.5 | 21.9 | — | |
| Fine-tuneAttack Type=StyleBkd2026.04 | 14 | 5.5 | — | |
| Fine-tuneAttack Type=VPI2026.04 | 19.5 | 13.5 | — | |
| DecodingAttack Type=CTBA2026.04 | 28 | 1 | — | |
| LCFAttack Type=CTBA2026.04 | 30 | — | 10 | |
| DecodingAttack Type=VPI2026.04 | 30.5 | 3.5 | — | |
| CROWAttack Type=VPI2026.04 | 33 | 26.4 | — | |
| DecodingAttack Type=StyleBkd2026.04 | 45.2 | 1 | — | |
| DecodingAttack Type=MTBA2026.04 | 58 | 3.5 | — |