Reproduce “MARS Technical Report” 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

Tong Liu, Chaoqun Hou, Longxiang Xiong, Junhong Lou, Jiaqi Zheng, Qixin Li, Zixin Zhao, Yong Duan and 9 more

MARS personalizes recommendation orchestration by choosing pipeline settings for each request, not just which items to show. A small language model uses behavior and scenario context to generate and cache legal configurations without delaying live serving. In a two-week Baiyibutie A/B test, MARS achieved relative lifts of 5.22% V2P and 7.34% GMV over production; these end-to-end results do not identify which settings caused the gains.

Summary by alphaXiv.

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