Prefix Teach, Suffix Fade: Local Teachability Collapse in Strong-to-Weak On-Policy Distillation

Documenti analoghi

• Multi-Objective and Mixed-Reward Reinforcement Learning via Reward-Decorrelated Policy Optimization
  di: Bai, Yang, et al.
  Pubblicazione: (2026)
• Where Did This Sentence Come From? Tracing Provenance in LLM Reasoning Distillation
  di: Liu, Kaiyuan, et al.
  Pubblicazione: (2025)
• Distribution-Aligned Sequence Distillation for Superior Long-CoT Reasoning
  di: Yan, Shaotian, et al.
  Pubblicazione: (2026)
• Self-Control of LLM Behaviors by Compressing Suffix Gradient into Prefix Controller
  di: Cai, Min, et al.
  Pubblicazione: (2024)
• Maximizing Local Entropy Where It Matters: Prefix-Aware Localized LLM Unlearning
  di: Zhai, Naixin, et al.
  Pubblicazione: (2026)