04:00
2026-08-18
arxiv.org
machine-learning
SMOPD: Selective Token-Entropy Masking for Dirty-History Multi-Turn On-Policy Self-Distillation
Researchers introduced SMOPD (Selective Masking for On-Policy Distillation), a loss-only stabilization method for multi-turn on-policy self-distillation (OPSD) that removes the lowest-entropy 20% of t…