A coding assistant needs the files relevant to a requested change, without filling its input with unrelated code. Relace describes a two-stage approach for large repositories. First, precomputed numerical representations of code help find a shortlist resembling the request. A second model, the reranker, reads the request alongside each shortlisted candidate and scores its relevance more directly. This concentrates the more expensive comparison on fewer files. The retrieval article emphasizes preserving relevant files during filtering: excluding a necessary file can leave the coding model without information it needs to make the change.
Terms in this piece · Glossary
embedding — A list of numbers representing a piece of text's meaning, so that similar meanings end up numerically close and can be searched.
token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
articleVITAL-RAG: Invariance Race for Context Allocation in Coding AgentsZijian Lu, Yonghua Lu, Mingcai Chen, Yiping Zuo, Xin He, Weijun Wang, Weibei Fan
articleExecRetrieval: Measuring the Functional-Correctness Gap in Code-Embedding RetrievalAaryan Kapoor, Md Abdullah Al Hafiz Khan
articleBetter Call Grep: Evaluating and Improving Grep-Like Lexical Retrieval for Repository-Level Code CompletionBaoyi Wang, Xingliang Wang, Guochang Li, Chen Zhi, Junxiao Han, Xinkui Zhao, Nan Wang, Shuiguang Deng, Jianwei Yin