openpd-gpt2
A GPT-2-sized language model (124M parameters) trained from scratch only on public-domain text, mostly books from before 1930. It runs entirely in your browser: nothing you type leaves your computer.
Try a prompt
More settings
The model's continuation appears here.
This is a small base model with no instruction tuning and no safety filtering. It continues text; it does not answer questions. It knows little about the world after about 1930, and it reflects the attitudes of its sources, including their prejudices. Expect confident nonsense.
Training data: Project Gutenberg, pre-1929 books, Library of Congress books, case law, US government documents, CC0 science papers, and Ubuntu IRC chat (with pseudonymized nicknames). Weights released under CC0. Details, evaluations, and known issues: model card.