Weak-to-Strong Generalization via Direct On-Policy Distillation

Weak-to-Strong Generalization via Direct On-Policy Distillation

Direct On-Policy Distillation transfers reinforcement learning improvements from smaller to larger models by using the policy shift induced by RL as an implicit reward signal, enab…

Hugging Face · Daily Papers ·Shiyuan Feng, Huan-ang Gao · ·▲ 92 upvotes

Este artigo está em destaque na seleção diária de papers do Hugging Face, curada pela comunidade de pesquisa em IA.

Autores: Shiyuan Feng, Huan-ang Gao, Haohan Chi, Hanlin Wu, Zhilong Zhang, Zheng Jiang

  • 92 upvotes da comunidade
  • Temas: reinforcement learning, verifiable rewards, weak-to-strong transfer, direct on-policy distillation, policy shift, implicit reward

Resumo

Resumo original (em inglês), extraído do paper:

Direct On-Policy Distillation transfers reinforcement learning improvements from smaller to larger models by using the policy shift induced by RL as an implicit reward signal, enabling efficient scaling of training without re-running expensive RL on the target model.

Onde ler

compartilhar: