On-Policy Self-Distillation
Definition
A method in which a base model is trained to match the token-level predictions of a teacher with the full session context accumulated, distilling session learning into weights using dense per-token signals without requiring an external verifiable reward.
Cast within
No topic-region cast yet — this appears once On-Policy Self-Distillation's compiled claims are aligned into a topic region's argument tree.
Claims mentioning On-Policy Self-Distillation (6)
Your notes
Loading notes...