limix-ldm-ai/LimiX

Leverages a transformer architecture with dual-dimension attention across samples and features to unify classification, regression, missing-value imputation, feature selection, and causal inference in a single model. Includes optional sample retrieval-based ensemble inference for enhanced performance, with configurable speed/memory tradeoffs via separate inference configs. Distributed across two model variants (LimiX-16M and LimiX-2M) optimized for different hardware constraints, integrating with PyTorch and supporting mixed-precision inference on NVIDIA GPUs.

3,340 stars.