Reproduce “Communicative On-Policy Distillation: Distilling Black-Box Teacher Capabilities through Communication for Long-Horizon Tasks” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Lirui Luo, Kelong Mao, Xinwei Yang, Rongqing Li, Luyu Chen, Kieran Wong, Yudong Guo, Xinrui Wang and 4 more
COPD lets a language-model student consult a text-only adviser during tasks, then trains it on fresh adviser-free attempts using successful consultations as privileged context. It combines on-policy self-distillation with task-reward training. On CAMG-II, COPD-4B scores 59.8 versus 51.2 for ActionOPD on an equal-weight average of five domain metrics. However, other 4B methods lead in Shop and General.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.