Xiaomi Ships a Tabular Foundation Model Pretrained Only on Synthetic Structural Causal Model Data

arXiv 2609.03880·medium signalXBlueskyLinkedInCopy link

Xiaomi-TabLDM does classification and regression through in-context learning with no task-specific fine-tuning, pretrained exclusively on synthetic data generated from structural causal models. It ranks 1st on OpenML-CTR23 and 2nd on regression across TALENT, TabArena and BCCO, and the efficiency trade is the notable part: on TabArena regression it takes the second-highest Elo while using 82% less training time and 68% less prediction time than top-ranked TabFM. The architecture combines a three-stage training strategy with dual-stream feature grouping, a lightweight Attention Residual and a sparse mixture of experts.