Reproduce “ReuseGR: Offline--Online Decoupled Generative Retrieval via Adaptive Prefix Reuse” 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
Xujing Wang, Yufeng Chen, Boxuan Zhang, Chao Wei, Cai Xu, Jie Zhao, Weiru Zhang, Wei Zhao and 1 more
ReuseGR addresses repeated computation in search advertising by caching query-specific prefixes of four-token Semantic IDs, which an online 0.5B model completes using each request’s history. A 7B offline planner prepares the prefixes. On a 3.38-million-item catalog, prepared-prefix ReuseGR achieves 85.82% HR@100, comparable to 7B, with 57.7% lower measured online latency; cold misses use 0.5B end-to-end.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.