KL for a KL: On-Policy Distillation with Control Variate Baseline

Documenti analoghi

• Your Language Model is Its Own Critic: Reinforcement Learning with Value Estimation from Actor's Internal States
  di: Choi, Yunho, et al.
  Pubblicazione: (2026)
• Future Policy Approximation for Offline Reinforcement Learning Improves Mathematical Reasoning
  di: Oh, Minjae, et al.
  Pubblicazione: (2025)
• Context-Robust Knowledge Editing for Language Models
  di: Park, Haewon, et al.
  Pubblicazione: (2025)
• On the Design of KL-Regularized Policy Gradient Algorithms for LLM Reasoning
  di: Zhang, Yifan, et al.
  Pubblicazione: (2025)
• Decoupling KL and Trajectories: A Unified Perspective for SFT, DAgger, Offline RL, and OPD in LLM Distillation
  di: Zhao, Anhao, et al.
  Pubblicazione: (2026)