TabICL vv2
Open Source
Infrastructure
LLMs
TabICL is a tabular foundation model designed for in-context learning on large datasets, focusing on classification tasks. It processes tabular data through three sequential stages: column-wise embedding, row-wise interaction, and dataset-wise in-context learning.
Column-wise embedding for distribution-aware feature representations
Row-wise interaction capturing feature interactions within each row
Dataset-wise in-context learning leveraging labeled examples for predictions