Reproduce “ActionOPD: Distilling Decisions, Not Tokens” 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
Huining Yuan, Xiangmin Yi, Zelai Xu, Zixu Wang, Xinlei Chen, Xiao-Ping Zhang, Yu Wang, Chao Yu
ActionOPD distills tool-using agents by matching semantic teacher and student actions at student-encountered states, rather than matching wording or token probabilities. It samples teacher responses and rewards student actions expressing the same meaning. After 100 steps on 6,697 tasks, all four tested configurations improve an equally weighted four-benchmark average by 3.3–6.9 percentage points, without gains on every metric.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.