Reproduce “AGR: Agentic Generative Retrieval with Tool-Augmented Search” 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
Ming Yang, Yaqi Yin, Zhibo Wang, Xiaoze Jiang, Jingwei Zhuo, Chao Zhang, Quanzhi Zhu, Weiguo Zheng and 7 more
AGR makes generative retrieval adaptive by requesting external features during semantic ID decoding and using returned observations to guide later codewords. Its tool-call positions are trained with rule-constructed trajectories, not an optimal learned policy. On Kuaishou’s Industry search dataset, AGR reaches Recall@50 of 0.3264 versus 0.2863 for TIGER, a 14.01% relative gain; this does not establish equal gains across settings.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.