Datasets:

ArXiv:
License:
File size: 1,501 Bytes
5af0e81
9d06c9b
5af0e81
 
 
 
 
 
 
 
 
 
 
 
 
1298212
5af0e81
1298212
3b699c1
1298212
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
a1c5c16
1298212
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
9d06c9b
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
---
license: odc-by
language:
- ar
- cy
- en
- es
- fr
- de
- id
- it
- ru
- sw
---

# TransWebEdu

**TransWebEdu** is a machine-translated, multi-way parallel, multilingual dataset at pretrain scale, supporting ten languages: Arabic, Welsh, German, English, Spanish, French, Indonesian, Italian, Russian, and Swahili.  
It is used to pretrain the [**TransWebLLM**](https://huggingface.co/britllm/TransWebLLM) model from scratch, with a focus on multilingual web-based education content.

For more information, see the paper:  
[Multilingual Language Model Pretraining using Machine-translated Data](https://arxiv.org/abs/2502.13252)

## Languages Supported

- Arabic (`ar`)
- Welsh (`cy`)
- English (`en`)
- Spanish (`es`)
- French (`fr`)
- German (`de`)
- Indonesian (`id`)
- Italian (`it`)
- Russian (`ru`)
- Swahili (`sw`)

## Dataset Notes

- The dataset files are chunked due to large file sizes.
- Each language's data files use a specific prefix to indicate language and script.

### Data File Prefix Mapping

| Language     | Identifier  |
|--------------|-------------|
| Arabic       | `arb_Arab`  |
| English      | `eng_Latn`  |
| French       | `fra_Latn`  |
| German       | `deu_Latn`  |
| Indonesian   | `ind_Latn`  |
| Italian      | `ita_Latn`  |
| Russian      | `rus_Cyrl`  |
| Spanish      | `spa_Latn`  |
| Swahili      | `swh_Latn`  |
| Welsh        | `cym_Latn`  |

## Contact

For questions or feedback, please contact:  
[[email protected]](mailto:[email protected])