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

Violation Detection on PolicyGuardBench

88.83Safety F1

LLAMA-3.3-70B-INSTRUCT

-3.386820.554144.49568.4359Oct 3, 2025Nov 9, 2025Dec 17, 2025Jan 24, 2026Mar 2, 2026Apr 9, 2026May 17, 2026
Updated 14d ago

Evaluation Results

MethodLinks
2025.10
88.8390.54305
2026.05
87.77--
2025.10
87.7389.6451.3
2025.10
87.5990.1422.5
2025.10
86.988.693,640
2025.10
86.7889.831,238
2025.10
86.0788.25205
2025.10
85.288.573.6
2025.10
85.0287.13596.1
2025.10
84.0786.133,270
2025.10
81.9884.57265
2026.05
77.85--
2026.05
77.64--
2025.10
77.6478.7670.8
2026.05
77.02--
2025.10
67.261.83250
2025.10
64.0764.08115.8
2025.10
59.5742.5687.5
2025.10
59.5442.39175.3
2026.05
59.52--
2025.10
59.5242.46164.8
2026.05
53.48--
2025.10
53.4868.9725.6
2025.10
42.2266.4785
2026.05
34.72--
2025.10
34.7254.5739.8
2025.10
33.1757.3532.6
2025.10
27.6760.6744.3
2025.10
18.3455.5545
2025.10
0.1657.5340