Datasets:
Tasks:
Text Generation
Modalities:
Text
Formats:
parquet
Sub-tasks:
language-modeling
Size:
100K - 1M
License:
metadata
language:
- zul
- sot
- xho
- nde
- nbl
- tso
- tsn
- nso
- ssw
- ven
license: apache-2.0
annotations_creators:
- no-annotation
language_creators:
- found
size_categories:
- 100K<n<1M
source_datasets:
- original
task_categories:
- text-generation
task_ids:
- language-modeling
pretty_name: South African Languages Dataset
tags:
- south-african-languages
- african-languages
- indigenous-languages
datasets:
- HuggingFaceFW/fineweb-2
South African Languages Dataset
Dataset Overview
Language | Training Documents | Training GPT2 Tokens | Avg Tokens/Doc | Max Tokens | Test Documents | Test GPT2 Tokens | Test Avg Tokens/Doc | Test Max Tokens |
---|---|---|---|---|---|---|---|---|
isiZulu | 116,693 | 192,622,799 | 1,650.68 | 335,530 | 687 | 1,080,961 | 1,573.45 | 15,691 |
Sesotho | 83,329 | 144,337,938 | 1,732.15 | 98,542 | 841 | 1,393,086 | 1,656.46 | 14,071 |
isiXhosa | 99,567 | 141,484,241 | 1,421.00 | 113,710 | 788 | 1,161,296 | 1,473.73 | 17,220 |
isiNdebele | 21,922 | 17,533,799 | 799.83 | 42,701 | 222 | 170,111 | 766.27 | 6,615 |
Xitsonga | 5,591 | 12,561,979 | 2,246.82 | 57,826 | 58 | 114,059 | 1,966.53 | 8,142 |
Setswana | 5,530 | 10,405,062 | 1,881.57 | 154,365 | 56 | 99,303 | 1,773.27 | 6,422 |
Sepedi | 5,180 | 9,947,520 | 1,920.37 | 25,017 | 45 | 98,001 | 2,177.80 | 8,602 |
siSwati | 1,668 | 3,148,007 | 1,887.29 | 24,129 | 17 | 29,753 | 1,750.18 | 6,682 |
Tshivenda | 3,994 | 25,756,055 | 6,448.69 | 243,593 | 35 | 60,390 | 1,725.43 | 8,272 |
This dataset covers 9 of South Africa's 11 official languages (excluding English and Afrikaans)
Dataset Structure
Data Instances
Each instance contains:
text
: string feature containing the document textlanguage
: string feature indicating the language name
Data Fields
text
: The text content of the documentlanguage
: The language of the text (one of: isiZulu, Sesotho, isiXhosa, isiNdebele, Xitsonga, Setswana, Sepedi, siSwati, Tshivenda)
Data Splits
The dataset is divided into training and test splits as shown in the table above.
Data Source
This dataset was collected from the FineWeb-2 dataset: https://huggingface.co/datasets/HuggingFaceFW/fineweb-2
Additional Information
- License: Apache 2.0
- Version: 1.0.0
- Collection Date: January 2025