List your company

Post-training Safety and Utility Alignment on Mistral-7B

0Unsafe Rate (%)CB
-0.9125.24411.417.556Apr 17, 2026

Evaluation results

MethodLinks
CBBackbone=Mistral-7B2026.0401.041002.838.161870
GoalBackbone=Mistral-7B2026.0409.9559.78.047.33—231.5
BeamBackbone=Mistral-7B, Pruning Framework=Beam2026.041.177.13467.737.131869.9
GreedyBackbone=Mistral-7B, Pruning Framework=Greedy2026.041.677.12467.556.961870.3
One-PassBackbone=Mistral-7B, Pruning Framework=One-Pass2026.0411.26.1736.32.972.71869
DPOBackbone=Mistral-7B2026.04177.4723.77.887.953669.5
BaseBackbone=Mistral-7B2026.0422.86.3822.77.718.07—70.4