Can Native Safety Travel? Reusing Safety Pretraining Across Models via On-Policy Distillation

Key points

• Safety pretraining has emerged as a promising paradigm for embedding safety deeply into a model during pretraining, yielding robustness to subsequent fine-tuning that conventional safety post-training often lacks.1
• We refer to this safety acquired during model pretraining as native safety.1
• We show that native safety can "travel" through on-policy distillation.1
• Unlike conventional large-to-small knowledge distillation, our framework enables selective, small-to-large safety transfer: it reuses the robust safety expertise of a smaller pretrained teacher without inheriting its utility limitations.1

Sources (1)

• [1]Can Native Safety Travel? Reusing Safety Pretraining Across Models via On-Policy Distillation
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 20, 03:13 AM

  In this work, we rethink the practical value of safety-pretrained models and introduce, for the first time, the native safety reuse problem: Can native safety be transferred from a specialized, potentially smaller, safety-pretrained "teacher" to standard-pretrained "students," enabling them to inherit its robustness while preserving their general utility?

  Safety pretraining has emerged as a promising paradigm for embedding safety deeply into a model during pretraining, yielding robustness to subsequent fine-tuning that conventional safety post-training often lacks.

Extractive summary: sentences quoted from the sources.