File size: 2,704 Bytes
41d3691
 
 
 
 
 
 
 
 
 
 
50663fc
41d3691
 
 
 
 
 
 
 
 
 
50663fc
41d3691
 
 
 
 
 
 
2cbbd25
41d3691
50663fc
41d3691
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
50663fc
41d3691
 
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

---
license: apache-2.0
language:
- mdf
datasets:
- allenai/MADLAD-400
library_name: transformers
pipeline_tag: text-generation
tags:
- goldfish
- arxiv:2408.10441
---

# mdf_cyrl_5mb

Goldfish is a suite of monolingual language models trained for 350 languages.
This model is the <b>Moksha</b> (Cyrillic script) model trained on 5MB of data, after accounting for an estimated byte premium of 1.71; content-matched text in Moksha takes on average 1.71x as many UTF-8 bytes to encode as English.
The Goldfish models are trained primarily for comparability across languages and for low-resource languages; Goldfish performance for high-resource languages is not designed to be comparable with modern large language models (LLMs).

Note: mdf_cyrl is an [individual language](https://iso639-3.sil.org/code_tables/639/data) code. It is not contained in any macrolanguage codes contained in Goldfish (for script cyrl).

All training and hyperparameter details are in our paper, [Goldfish: Monolingual Language Models for 350 Languages (Chang et al., 2024)](https://www.arxiv.org/abs/2408.10441).

Training code and sample usage: https://github.com/tylerachang/goldfish

Sample usage also in this Google Colab: [link](https://colab.research.google.com/drive/1rHFpnQsyXJ32ONwCosWZ7frjOYjbGCXG?usp=sharing)

## Model details:

To access all Goldfish model details programmatically, see https://github.com/tylerachang/goldfish/blob/main/model_details.json.
All models are trained with a [CLS] (same as [BOS]) token prepended, and a [SEP] (same as [EOS]) token separating sequences.
For best results, make sure that [CLS] is prepended to your input sequence (see sample usage linked above)!
Details for this model specifically:

* Architecture: gpt2
* Parameters: 39087104
* Maximum sequence length: 512 tokens
* Training text data (raw): 8.56MB
* Training text data (byte premium scaled): 5.005MB
* Training tokens: 1070592 (x10 epochs)
* Vocabulary size: 50000
* Compute cost: 809619743047680.0 FLOPs or ~0.1 NVIDIA A6000 GPU hours

Training datasets (percentages prior to deduplication):
* 79.42795%: [MADLAD-400 (CommonCrawl)](https://huggingface.co/datasets/allenai/MADLAD-400)
* 18.39365%: [Wikipedia 2023/08](https://dumps.wikimedia.org/)
* 2.17561%: [Languages of Russia](http://web-corpora.net/wsgi3/minorlangs/download)
* 0.00279%: [Tatoeba](https://tatoeba.org/en/)


## Citation

If you use this model, please cite:

```
@article{chang-etal-2024-goldfish,
  title={Goldfish: Monolingual Language Models for 350 Languages},
  author={Chang, Tyler A. and Arnett, Catherine and Tu, Zhuowen and Bergen, Benjamin K.},
  journal={Preprint},
  year={2024},
  url={https://www.arxiv.org/abs/2408.10441},
}
```