# Training data sources and attribution

The model repository does not redistribute source documents or processed
training datasets. The model was trained on sampled text from:

- [FineWeb2](https://huggingface.co/datasets/HuggingFaceFW/fineweb-2),
  configuration `deu_Latn`, released under ODC-By 1.0.
- [FineWiki](https://huggingface.co/datasets/HuggingFaceFW/finewiki),
  configuration `de`, derived from German Wikipedia text and identified by its
  dataset card as CC BY-SA 4.0 and GFDL.
- [FineWeb](https://huggingface.co/datasets/HuggingFaceFW/fineweb),
  configuration `default`, released under ODC-By 1.0 and additionally subject
  to Common Crawl's Terms of Use.

FineWiki attributes its text to Wikipedia contributors. FineWeb and FineWeb2
should be cited according to their dataset cards. Users who reproduce,
continue, or redistribute training data are responsible for complying with
the source licenses and terms.

The repository's MIT license covers HanseLM's release code and model weights.
It does not relicense third-party source datasets or grant rights to training
examples that may be reproduced by the model.