Hugging Face's logo

Datasets:
metuKKhud
/
bashqort-raw

Modalities:
Text
Formats:
parquet
Languages:
Bashkir
Size:
100K - 1M
ArXiv:
Tags:
bashkir
low-resource-language
continual-training
news
llm-adaptation
Libraries:
Datasets
pandas
Polars
License:
Dataset card Data Studio Files
xet
 Community
1
bashqort-raw
File size: 2,756 Bytes
4b4a0d0
02b0365
 
 
 
 
 
 
 
 
 
 
4b4a0d0
02b0365
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
---
language: ba
license: mit
tags:
- bashkir
- low-resource-language
- continual-training
- news
- llm-adaptation
pretty_name: Bashqort Raw Corpus
size_categories:
- 10M<n<100M tokens
---

# Bashqort Raw Corpus

## Description

This dataset contains raw Bashkir text collected for **continual training** of large language models (LLMs). It is part of the project "Adapting Open-Source LLMs for the Bashkir Language", which aims to evaluate adaptation methods proposed by LlamaTurk (Toraman, 2024) and Persian adaptation (Mahdizadeh Sani et al., 2024).

The corpus is assembled from multiple sources to provide a diverse linguistic foundation for language modeling.

## Sources

| Source | Type | # articles / tokens |
|--------|------|----------------------|
| [bashgazet.ru](https://bashgazet.ru/articles) | daily socio-political newspaper | ~1,400 articles |
| [neftcity.ru](https://neftcity.ru) | regional news | ~548 articles |
| [bash.news](https://bash.news) | news & analytics | ~68,708 articles |
| [bashkir-corpus](https://github.com/nevmenandr/bashkir-corpus) | mixed (public domain + shuffled) | ~20.9M tokens |

Total tokens: see dataset statistics on Hugging Face.

## Preprocessing

- Deduplication (document and sentence level)
- Removal of sentences with <5 words
- Removal of HTML artifacts, ads, and metadata
- No train/val split (the whole corpus is for self-supervised learning)

## Format

Each example is a JSON object with the following fields:

```json
{
  "text": "Bashkir sentence or paragraph",
  "source": "bashgazet.ru | neftcity.ru | bash.news | bash-corpus | etc",
  "is_shuffled": true/false
}
```

For plain text mode, the dataset also provides a text column with raw sentences.
Intended Use

    Continual pre‑training / further pretraining of LLMs (e.g., Llama‑2, Llama‑3)

    Next‑token prediction (causal language modeling)

    Any research aiming to improve Bashkir language representation in NLP

Licensing

This dataset is released under the MIT License.
Citation

If you use this dataset, please cite the project repository:
```bibtex

@misc{khudiakova2025bashqortraw,
  author       = {Khudiakova, Kseniia},
  title        = {Bashqort Raw Corpus: Bashkir Text Collection for Continual Training},
  year         = {2025},
  howpublished = {Hugging Face Datasets},
  url          = {https://huggingface.co/datasets/metuKKhud/bashqort-raw}
}
```

Additionally, consider citing the original data sources and the adaptation framework:

    Toraman, C. (2024). LlamaTurk: Adapting open-source generative large language models for low-resource language. arXiv:2405.07745.

    Mahdizadeh Sani, S., et al. (2024). Extending LLMs to new languages: A case study of Llama and Persian adaptation. arXiv:2412.13375.