Skip to content

Dataset · dell-research-harvard

Structured US historical newspapers

dell-research-harvard/AmericanStories

Structured historical English newspaper texts for LLM pre-training and RAG.

What's inside

American Stories provides high-quality structured data from historical newspapers. According to the card, it is suitable for pre-training large language models and improves understanding of historical English and world knowledge. The data can also be connected to retrieval-augmented models and used for multimodal layout analysis; language is English, size is 100M<n<1B.

What you can do with it

  • Pre-training LLMs on historical English
  • Populating historical newspaper RAG databases
  • Training and evaluating newspaper layout analysis models
Tasks
Text classification, Text generation, Text retrieval, Summarization, Question answering
Languages
English
Size
100M–1B rows
Downloads this month
18K
Likes
184

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("dell-research-harvard/AmericanStories")