TabBench: Tabular Embedding Benchmark
A Comprehensive Evaluation Suite for Tabular Embedding Models
Overview
TabBench is a comprehensive benchmark designed to evaluate the tabular understanding capability of embedding models. It assesses two critical dimensions of tabular representation: linear separability (via classification) and semantic alignment (via retrieval).
TabBench aggregates diverse datasets from four authoritative repositories and provides a… See the full description on the dataset page: https://huggingface.co/datasets/qiangminjie27/TabBench.
