Reproduce “CoReMem: Cost-Internalizing Memory for Successive Coding Tasks with Prospective Valuation” 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
Peiwen Li, Harshit Verma, Sophia Zhou, Haiwen Wang, Ngoc Bui, Heng Chang, Fei Fei, David van Dijk and 1 more
CoReMem evaluates which information from one coding task helps an unknown next task on the inherited repository, then trains an actor to use useful repository locations and executable checks. It values handoff details by measuring next-task steps and tokens. On ChainSWE with Qwen3.8-27B-FP8, the full post-trained system raises resolution from 21.0% to 25.7% and reduces steps and tokens versus an untrained no-memory actor; memory alone does not reduce tokens.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.