Xiaomi-TabLDM

Xiaomi's tabular foundation model, an in-context learner for classification and regression pretrained exclusively on synthetic data from structural causal models, with dual-stream feature grouping, a sparse mixture-of-experts backbone, and test-time compute scaling. The report ranks it first on OpenML-CTR23 and second on regression across TALENT, TabArena, and BCCO, and claims 82% less training time than TabFM at matched quality. Apache 2.0; the parameter count is not stated. One of three Chinese tabular foundation models in a fortnight alongside LimiX-2 and Amazon's Mitra-v2.