Reproduce “DecisionRank: On-Policy Distillation for Decision-Model Rerankers” 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
Yanran Wu, Dian Yu, Chongyang Bai, Jieyi Zhang, Rui Liu, Tianyi Chen, Shike Mei, Hang Cui and 5 more
DecisionRank trains a one-pass reranker to use conditional feedback from a generative teacher. It sorts the student’s first-token candidate scores, conditions the teacher on that ranking, and aligns choice distributions over candidate identities; at inference, the student scores once and sorts. On TREC DL19–23, its 1.7B student surpasses the 27B teacher, under the reported evaluation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.