Reproduce “RT-J: Large-Scale Pretraining of Relational Transformers for Context-Efficient Predictions” 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
Rishabh Ranjan, Vignesh Kothapalli, Harshvardhan Agarwal, Charilaos Kanatsoulis, Roshan Upendra, Tom Palczewski, Carlos Guestrin, Jure Leskovec
RT-J is an 86-million-parameter relational model for few-example prediction from linked databases. It is pretrained on synthetic PluRel data and 639 databases, using random walks to select relevant rows. On RelBench regression, it matches the strongest pipeline’s error with about 10 times fewer labels; evaluation covers 21 tasks across seven databases, with contexts limited to 8,192 cells.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.