Share your thoughts, 1 month free Claude Pro on us
See more
Home
/
Benchmarks
Machine Unlearning on WMDP-Cyber (Pre/Post Accuracy)
Loading...
41.5
Accuracy (Pre-Unlearning)
Original model
23.924
28.487
33.05
37.613
Apr 15, 2026
Accuracy (Pre-Unlearning)
Accuracy (Post-Unlearning)
Updated 3d ago
Evaluation Results
Method
Method
Links
Accuracy (Pre-Unlearning)
Accuracy (Post-Unlearning)
Original model
2026.04
41.5
-
SimNPO
retain free=false, Tim...
2026.04
40.8
41.4
RMU
retain free=false, Tim...
2026.04
40.2
39.8
NPO
retain free=false, Tim...
2026.04
39.6
41
MC-WIN-U
retain free=true, Time...
2026.04
37.6
39.5
GradDiff
retain free=false, Tim...
2026.04
27.4
37.4
GradAscent
retain free=true, Time...
2026.04
24.6
26.5
Feedback
Search any
task
Search any
task