Skip to content

Dataset · llamaindex

ParseBench benchmark for document parsing

llamaindex/ParseBench

A benchmark for evaluating real enterprise document parsing systems across five dimensions.

What's inside

ParseBench is a benchmark for testing document extraction and parsing systems on real enterprise documents. Evaluation is divided into five dimensions: tables, charts, content accuracy, semantic formatting, and visual grounding. Each dimension has its own metrics; the data includes English documents, images, and text.

What you can do with it

  • Evaluate OCR/parsing systems on tables and charts
  • Test document data extraction pipelines
  • Compare document understanding models
Languages
English
Size
100K–1M rows
Downloads this month
45.1K
Likes
173

How to load it

from datasets import load_dataset

ds = load_dataset("llamaindex/ParseBench")