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

Unlearning on WMDP retain

55.6Retain

GU+SimNPO

24.19232.34640.548.654May 14, 2026
Updated 2mo ago

Evaluation Results

MethodLinks
2026.05
55.6
2026.05
54.7
2026.05
54.6
2026.05
51.1
2026.05
47.6
2026.05
46.8
2026.05
46
2026.05
45.2
2026.05
45.1
2026.05
44.3
2026.05
43.1
2026.05
43.1
2026.05
42.9
2026.05
42.1
2026.05
39.1
2026.05
38
2026.05
35.7
2026.05
34.5
2026.05
34.2
2026.05
34.1
2026.05
31.7
2026.05
25.4