SFT-based Poisoning
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
SFT-based Poisoning Word trigger
96.7Clean Accuracy (CACC)
18
SFT-based Poisoning Phrase trigger
95.7Clean Accuracy (CACC)
18
SFT-based Poisoning Long trigger
94.8Clean Accuracy (CACC)
18