Skip to content

Dataset · LiquidAI

Structured JSON extraction from documents in 16 languages

LiquidAI/structured-extraction

Multilingual set of synthetic documents paired with JSON Schema and gold extracted objects for SFT.

What's inside

The LiquidAI dataset includes 6,000 training examples and 500 test examples, covering 35 types of fictional documents (letters, receipts, bookings, forms, announcements) in 16 languages. Each text includes a target JSON Schema and a correct JSON object; all names, companies, and identifiers are generated and contain no real personal data.

What you can do with it

  • SFT fine-tuning of LLMs for reliable parsing of unstructured text into strict JSON by a given schema
  • Training automated processing systems for receipts, requests, letters, and forms in multiple languages
  • Benchmarking structured output quality in language models
Tasks
Text generation
Languages
English, Arabic, Chinese, French, German, Italian, Japanese, Korean and 4 more
Size
1K–10K rows
Downloads this month
821
Likes
45

How to load it

from datasets import load_dataset

ds = load_dataset("LiquidAI/structured-extraction")