Reproduce “InfraRSI: Harnessing Recursive Self-Improvement for Training Infrastructure” 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

InfraRSI Team, Yuchuan Tian, Haocheng Mei, Ye Yuan, Chao Xu, Boxun Li, Guohao Dai, Kai Han and 3 more

InfraRSI searches distributed-training launch settings and code together while keeping the training task fixed and checking that promoted results are faithful and repeatably faster. It updates estimates during each search and carries verified results into later searches. Across four dense and mixture-of-experts workloads, it reports a 1.83× geometric-mean speedup; this result is limited to the evaluated setups.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.