---
language:
- de
task_categories:
- text-generation
size_categories:
- 100k<n<1M
task_ids:
- language-modeling
- masked-language-modeling
---
German validation dataset from WECHSEL () to evaluate LLM perplexity.

JSON-line files (on JSON object per line):

- `valid.json.gz`: Gzipped validation set as generated by the paper (163,698 docs)
- `valid.random_1636.json.gz`: Random 1% (1636 docs) of the validation set