List your company

Safety alignment on SORRY-Bench (Score, log C)

85.7Scoreπref
19.24436.49753.7571.003Jul 3, 2026

Evaluation results

MethodLinks
πrefBackbone=zephyr-7b, N=12026.0785.75.44
πrefBackbone=zephyr-7b, N=162026.0773.9—
URIALBackbone=zephyr-7b, N=12026.07633.15
URIALBackbone=zephyr-7b, N=162026.0752.7—
BoBNBackbone=zephyr-7b, N=12026.0743.9—
BoBNBackbone=zephyr-7b, N=42026.0741.22.87
BoBNBackbone=zephyr-7b, N=162026.0736.1—
πrefBackbone=tulu3-8b, N=12026.0728.61.59
πrefBackbone=tulu3-8b, N=162026.0726.8—
URIALBackbone=tulu3-8b, N=12026.0725.71.6
BoBNBackbone=tulu3-8b, N=12026.0724.3—
BoBNBackbone=tulu3-8b, N=42026.0723.61.59
URIALBackbone=tulu3-8b, N=162026.0722.6—
BoBNBackbone=tulu3-8b, N=162026.0721.8—