Media

A description of the project, figures and the logo for use in publications. They can be used without asking.

In one sentence

Wieszcz is an independent research project that builds language models trained only on Polish from 1800 to 1918.

In one paragraph

Wieszcz is an independent research project run by Szymon Kocur. It builds language models trained from scratch on Polish texts published between 1800 and 1918. The first release, Wieszcz-XIX, consists of a corpus of 3.1 billion words in 294,369 documents and three models with 47, 107 and 349 million parameters. The corpus, the code and the model weights are published.

Figures and facts

Corpus
3.1 billion words in 294,369 documents
Period
texts published from 1800 to 1918
Sources
the Internet Archive and Wolne Lektury
Models
47, 107 and 349 million parameters, trained from scratch
Paper
arXiv:2610.10592
Author
Szymon Kocur, independent researcher, Warsaw
Licences
code and weights Apache 2.0, corpus CC0 1.0 on the compilation layer

Worth knowing before you publish

  • The published models continue a given text. A conversational version is being worked on.
  • The models know only what was in texts published up to 1918.
  • The models reproduce the prejudices of the period, including antisemitic ones. Please say so when quoting generated text.

The name

“Wieszcz” is an old Polish word for a poet-prophet, the title given to the great Romantic poets. It is pronounced roughly VYESH-ch. “Wieszcz” names the project and “Wieszcz-XIX” its first release.

How to cite

@misc{kocur2026wieszcz,
  title  = {Wieszcz-XIX: A 3.1-Billion-Word Corpus of Pre-1918 Polish
            and Temporally Bounded Language Models Trained From Scratch},
  author = {Kocur, Szymon},
  year   = {2026},
  eprint = {2610.10592},
  archivePrefix = {arXiv},
  primaryClass  = {cs.CL},
  url    = {https://arxiv.org/abs/2610.10592},
}

Written about the project

Media contact

We answer in Polish and in English.

[email protected]