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