ruvimx/UkrLM-wiki

---
license: cc-by-sa-4.0
pretty_name: UkrLM — Wikipedia Corpus
configs:
- config_name: wikipedia
data_files:
- split: train
path: wikipedia/train-*
tags:
- ukrainian
- pretraining
- wikipedia
size_categories:
- 100M<n<1B
dataset_info:
- config_name: wikipedia
features:
- name: text
dtype: large_string
- name: title
dtype: large_string
- name: source
dtype: large_string
- name: license
dtype: large_string
- name: date
dtype: timestamp[us]
splits:
- name: train
num_bytes: 5647322338
num_examples: 1134416
download_size: 2631278461
dataset_size: 5647322338
language:
- uk
---

# UkrLM — Wikipedia Corpus

A large-scale Ukrainian Wikipedia corpus for language model pretraining.
Includes cleaned Ukrainian Wikipedia (uk.wikipedia.org).

## Subsets

| Config | Description | Records |
|--------|-------------|---------|
| `wikipedia` | Articles from uk.wikipedia.org, cleaned from wikimarkup | 1,134,432 |

## Usage

```python
from datasets import load_dataset

ds = load_dataset("ruvimx/UkrLM-wiki", "wikipedia")
```

## Fields

| Field | Type | Description |
|-------|------|-------------|
| `text` | string | Article text |
| `title` | string | Article title |
| `source` | string | Source identifier (`wikipedia_uk`) |
| `license` | string | Content license (`CC-BY-SA-4.0`) |
| `date` | string | Collection date |

## Known Limitations

Some articles contain recurring templated sections that are structurally similar but not identical across entries.

## License

Content is licensed under **CC-BY-SA 4.0** in accordance with Wikipedia's licensing terms.
See [Creative Commons](https://creativecommons.org/licenses/by-sa/4.0/).

## Citation

```bibtex
@dataset{savytskyi2026ukrlm,
author = {Savytskyi, Ruvim},
title = {UkrLM: Ukrainian Wikipedia Corpus},
year = {2026},
publisher = {Hugging Face},
url = {https://huggingface.co/datasets/ruvimx/UkrLM-wiki}
}
```