LG-AI-Research/exaone-tabular-demo

EXAONE Tabular

A Gradio demo for `LG-AI-Research/EXAONE-Tabular`, a ~21M-parameter in-context foundation model for tabular classification and regression.

Upload a CSV, choose the column to predict, and the model produces predictions in a single forward pass — no gradient updates, no per-dataset training. Rows whose target cell is blank are treated as the query set; if nothing is blank, a slice of the table is held out so the demo can report held-out scores (and compare against a scikit-learn gradient-boosting baseline).

Powered by the `exaonetabular` runtime on ZeroGPU.

Example data

The bundled examples are the datasets from the model card's own quickstart, exported to CSV from scikit-learn (BSD-3-Clause):

• —examples/breast_cancer.csv — UCI Breast Cancer Wisconsin (Diagnostic), CC BY 4.0
• —examples/wine.csv / examples/wine_unlabeled.csv — UCI Wine recognition, CC BY 4.0
• —examples/diabetes.csv — Efron et al. (2004) diabetes dataset, distributed with scikit-learn

License

The exaonetabular runtime code is BSD-3-Clause-LG AI Research. The model weights are licensed under the EXAONE AI Model License Agreement 1.2 – NC and may be used for non-commercial research and education only.