Sentiment Analysis on SST2 POLISHED Attack (test)
99ASRUnpatched
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| UnpatchedModel=Llama2026.06 | 99 | 95 | |
| FinePruningModel=Llama2026.06 | 98 | 94 | |
| FT_l2Model=Llama2026.06 | 97 | 94 | |
| ROMEModel=Llama2026.06 | 97 | 92 | |
| FT_linfModel=Llama2026.06 | 96 | 94 | |
| SPPModel=Llama2026.06 | 96 | 91 | |
| FTModel=Llama2026.06 | 95 | 95 | |
| BAERASERModel=Llama2026.06 | 95 | 90 | |
| OneShotModel=Llama2026.06 | 95 | 90 | |
| MENDModel=Llama2026.06 | 94 | 93 | |
| UnpatchedBackbone=Qwen2.5-32B2026.06 | 92 | 94 | |
| MudjackingModel=Llama2026.06 | 82 | 90 | |
| PatcherModel=Llama2026.06 | 1 | 92 | |
| PatcherBackbone=Qwen2.5-32B2026.06 | 1 | 93 |