Classification on COLA
0.175ASR ScorePROTOPURIFY
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| PROTOPURIFYAttack=BadNet, Backbone=Mistral-7B2026.02 | 0.175 | 0.771 | |
| PROTOPURIFYAttack=VPI, Backbone=Mistral-7B2026.02 | 0.157 | 0.878 | |
| PROTOPURIFYAttack=CBA, Backbone=Mistral-7B2026.02 | 0.154 | 0.863 | |
| PROTOPURIFYAttack=VPI, Backbone=Llama3-8B2026.02 | 0.15 | 0.825 | |
| PROTOPURIFYAttack=CBA, Backbone=Llama3-8B2026.02 | 0.097 | 0.875 | |
| PROTOPURIFYAttack=BadEdit, Backbone=Mistral-7B2026.02 | 0.089 | 0.719 | |
| PROTOPURIFYAttack=BadEdit, Backbone=Llama3-8B2026.02 | 0.081 | 0.742 | |
| PROTOPURIFYAttack=BadNet, Backbone=Llama3-8B2026.02 | 0.027 | 0.813 |