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

RLHF Alignment Evaluation on HHA

76.1Harmlessness Win Rate (Base, A)

DPO

17.96433.05748.1563.243Feb 16, 2025
Updated 26d ago

Evaluation Results

MethodLinks
2025.02
76.120.543.3776.7714.818.4249.8344.16.0662.9725.5911.4566.4263.19
2025.02
63.326.2610.4489.93.76.461.9632.665.3979.1312.128.7573.5776.99
2025.02
55.5634.3410.172.7316.510.7748.4842.429.0969.3619.8610.7761.5363.53
2025.02
51.5233.6714.8188.558.083.3758.5934.347.0785.8610.443.771.1377.67
2025.02
42.4242.4215.1568.6919.1912.1233.3354.2112.4653.231.6515.1549.4151.74
2025.02
20.259.620.249.4932.3218.1812.4671.3816.1634.6847.4717.8429.2132.21