Preprint
Aug 2026
PCSD: Persistent Consistency for Self-Distillation in Agentic Reinforcement Learning
Persistent Consistency Self-Distillation (PCSD) is proposed, which derives token-level distillation weights from the local persistence of teacher-favoring signals, and combines adaptive windows with exponentially decayed aggregation to capture persistent relative teacher support.
Chunji Lv, Yangguang Wei, Junlin Liu et al.
· 0 citations