WDL-OPD co-trains an anchor and an auxiliary policy through a shared geometric-mixture reverse-KL loss to a frozen teacher, yielding stronger Qwen3 math and code checkpoints and avoiding entropy-collapse failures seen in single-policy on-policy distillation.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
WDL-OPD: Weak-Driven On-Policy Distillation via Mixture-Constrained Co-Training
WDL-OPD co-trains an anchor and an auxiliary policy through a shared geometric-mixture reverse-KL loss to a frozen teacher, yielding stronger Qwen3 math and code checkpoints and avoiding entropy-collapse failures seen in single-policy on-policy distillation.