---
license: apache-2.0
language:
- mai
datasets:
- allenai/nllb
- allenai/MADLAD-400
- cis-lmu/Glot500
- sil-ai/bloom-lm
- legacy-datasets/wikipedia
- oscar-corpus/OSCAR-2109
library_name: transformers
pipeline_tag: text-generation
tags:
- goldfish
- arxiv:2408.10441
---

# mai_deva_full

Goldfish is a suite of monolingual language models trained for 350 languages.
This model is the <b>Maithili</b> (Devanagari script) model trained on 41MB of data (all our data in the language), after accounting for an estimated byte premium of 2.39; content-matched text in Maithili takes on average 2.39x 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: mai_deva 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 deva).

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: 124770816
* Maximum sequence length: 512 tokens
* Training text data (raw): 99.73MB
* Training text data (byte premium scaled): 41.735MB
* Training tokens: 11159040 (x10 epochs)
* Vocabulary size: 50000
* Compute cost: 5.6935272480768e+16 FLOPs or ~5.4 NVIDIA A6000 GPU hours

Training datasets (percentages prior to deduplication):
* 67.00767%: [NLLB (CommonCrawl and ParaCrawl)](https://huggingface.co/datasets/allenai/nllb)
* 12.01673%: [MADLAD-400 (CommonCrawl)](https://huggingface.co/datasets/allenai/MADLAD-400)
* 10.36867%: [Glot500](https://huggingface.co/datasets/cis-lmu/Glot500), including [BLOOM](https://huggingface.co/datasets/sil-ai/bloom-lm), [Wortschatz Leipzig Data](https://wortschatz.uni-leipzig.de/en/download), [OSCAR](https://oscar-project.org/), [Wikipedia Hugging Face](https://huggingface.co/datasets/legacy-datasets/wikipedia)
* 8.75652%: [Wikipedia 2023/08](https://dumps.wikimedia.org/)
* 1.83462%: [eBible](https://ebible.org/find/)
* 0.01578%: [OSCAR 2021/09](https://huggingface.co/datasets/oscar-corpus/OSCAR-2109)


## 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},
}
```