Classification on SST-2
4.4ASR ErrorPROTOPURIFY
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| PROTOPURIFYAttack=CBA, Backbone=Llama3-8B2026.02 | 4.4 | 91.3 | |
| PROTOPURIFYAttack=BadEdit, Backbone=Llama3-8B2026.02 | 9.3 | 92.5 | |
| PROTOPURIFYAttack=BadNet, Backbone=Llama3-8B2026.02 | 9.3 | 92.5 | |
| PROTOPURIFYAttack=BadEdit, Backbone=Mistral-7B2026.02 | 9.6 | 85.5 | |
| PROTOPURIFYAttack=BadNet, Backbone=Mistral-7B2026.02 | 10.7 | 87 | |
| PROTOPURIFYAttack=VPI, Backbone=Llama3-8B2026.02 | 16.3 | 93.8 | |
| PROTOPURIFYAttack=CBA, Backbone=Mistral-7B2026.02 | 16.5 | 89.3 | |
| PROTOPURIFYAttack=VPI, Backbone=Mistral-7B2026.02 | 22 | 92.9 |