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

Safety Alignment on BeaverTails V

93.37Safety Score

SaFeR-ToolKit (+ SFT+GRPO) [3B]

28.52645.360562.19579.0295Mar 3, 2026Mar 18, 2026Apr 2, 2026Apr 17, 2026May 2, 2026May 17, 2026Jun 2, 2026
Updated 1mo ago

Evaluation Results

MethodLinks
2026.03
93.3779.4292.69
2026.03
93.2289.6694.92
2026.03
92.8882.289.49
2026.06
90.4993.32-
2026.03
87.4658.8178.64
2026.03
87.4470.4639.05
2026.03
84.5853.2250
2026.06
84.2175.27-
2026.06
84.0460.75-
2026.06
83.5392.91-
2026.03
83.2270.1783.22
2026.03
82.8875.0882.37
2026.03
79.668.8117.8
2026.06
77.7681.29-
2026.06
76.2382.15-
2026.06
74.5383.94-
2026.06
65.8725.43-
2026.03
63.974.2426.61
2026.06
63.8450.7-
2026.06
62.1462.65-
2026.06
59.5956.45-
2026.06
52.2946.85-
2026.06
5050-
2026.06
5050-
2026.03
37.865.766.78
2026.03
36.6156.276.27
2026.03
31.0239.664.75