Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Backdoor Mitigation on SFT-based Poisoning Long trigger

94.8Clean Accuracy (CACC)

Base

89.3490.757592.17593.5925Oct 11, 2025
Updated 20d ago

Evaluation Results

MethodLinks
2025.10
94.8100
2025.10
94.898
2025.10
94.71.25
2025.10
94.624.2
2025.10
94.694.5
2025.10
94.692
2025.10
94.644
2025.10
94.628
2025.10
94.692.25
2025.10
94.345
2025.10
9452.5
2025.10
93.848
2025.10
93.87.8
2025.10
93.686
2025.10
93.218.75
2025.10
92.68100
2025.10
90.5100
2025.10
89.55100