LG’s EXAONE Tabular ranked first for classification on TabArena by Elo score, a user reports
The user describes classification and regression models with about 21 million parameters each. They note a licensing distinction: released weights are limited to non-commercial research and education, while inference code permits commercial use.
TLDR
In a September 11 post, a user reported that LG AI Research’s EXAONE Tabular ranked first for classification and second for regression on TabArena by Elo score, and shared its technical report. They described models with about 21 million parameters each that handle missing values directly, without filling them in first, and use built-in feature selection for tables exceeding the feature limit. The user also noted that the released weights are limited to non-commercial research and education, though the inference code permits commercial use.
Combined views
—
1 Source, first seen ago
