Skip to content

Dataset · VillanovaAI

Multilingual synthetic document VQA dataset

VillanovaAI/Multi-CoSyn-400K

100K to 1M question-answer pairs with reasoning over text-dense images across 5 European languages.

What's inside

This is a multilingual extension of the CoSyn-400K dataset (a successor to PixMo-Docs), translated and adapted for English, Italian, German, Spanish, and French. It contains synthetic text-rich images generated via code rendering, along with questions and detailed answers featuring chain-of-thought reasoning from text LLMs. The corpus size ranges from 100,000 to 1 million examples.

What you can do with it

  • Training multimodal models (VLMs) in document reading and comprehension across 5 languages
  • Fine-tuning VQA systems to generate step-by-step reasoning over charts and tables
  • Improving OCR capabilities of vision-language models in European languages
Tasks
Visual question answering
Languages
Italian, English, German, Spanish, French
Size
100K–1M rows
Downloads this month
28.7K
Likes
0
Access
on request on Hugging Face

How to load it

from datasets import load_dataset

ds = load_dataset("VillanovaAI/Multi-CoSyn-400K")