LimiX · Large Data Models for Structured Data

洞悉数据背后的因果关联 —— 一模型，多任务，零训练，零调参

🌐 Website · 🚀 Live Demo · 💻 GitHub · 🤗 LimiX-2

🧬 What is LimiX?

LimiX is a family of foundation models for structured (tabular) data — Large Data Models (LDMs) that read tables the way LLMs read text.

• 🧠 In-context learning — show the model a handful of labeled rows, like writing a prompt. No training, no tuning: predictions emerge in a single forward pass.
• 🧬 Causality-aware — the CMN architecture encodes direct feature-to-feature causal relations in its attention, making predictions interpretable.
• 🏆 State of the art — Elo #1 on all three public benchmarks (TabArena · TALENT · BCCO), surpassing AutoGluon 1.6 and other tabular foundation models.

Try it now: upload a CSV at www.limix.ai/demo — classification, regression and missing-value imputation in a single forward pass.

🇨🇳 中文简介

LimiX 是面向结构化（表格）数据的基础模型系列（Large Data Models, LDMs）。 像大模型读文本一样读表格：把带标签的行作为上下文提示，一次前向传播即可完成 分类 · 回归 · 缺失值填充，零训练、零调参。CMN 架构天然编码特征间的直接因果关系， 在 TabArena / TALENT / BCCO 三大权威基准中 Elo 评分全部第一。

_{Stable AI · www.limix.ai}