TurnOPD: Making On-Policy Distillation Turn-Aware for Efficient Long-Horizon Agent Training

TurnOPD: Making On-Policy Distillation Turn-Aware for Efficient Long-Horizon Agent Training

Turn-level budgeting strategy for efficient on-policy distillation in long-horizon agent training addresses inefficiencies in full-horizon rollouts and shallow token concentration.

Hugging Face · Daily Papers ·Yuhang Zhou, Kai Zheng · ·▲ 12 upvotes

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

Autores: Yuhang Zhou, Kai Zheng, Haoling Li, Dengyun Peng, Can Xu, Jingjing Chen

  • 12 upvotes da comunidade
  • Temas: on-policy distillation, teacher-student framework, trajectory-level KL objectives, turn-level budgeting, adaptive rollout-depth budgeting, progressive turn-normalized loss budgeting

Resumo

Resumo original (em inglês), extraído do paper:

Turn-level budgeting strategy for efficient on-policy distillation in long-horizon agent training addresses inefficiencies in full-horizon rollouts and shallow token concentration.

Onde ler

compartilhar: