metadata

license: apache-2.0
language:
  - aym
datasets:
  - breakend/nllb-multi-domain
  - allenai/nllb
  - allenai/MADLAD-400
  - cis-lmu/Glot500
  - legacy-datasets/wikipedia
library_name: transformers
pipeline_tag: text-generation
tags:
  - goldfish
  - arxiv:2408.10441

aym_latn_full

Goldfish is a suite of monolingual language models trained for 350 languages. This model is the Aymara (Latin script) model trained on 30MB of data (all our data in the language), after accounting for an estimated byte premium of 1.21; content-matched text in Aymara takes on average 1.21x 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: aym_latn is a macrolanguage code. Individual language code ayr_latn (Central Aymara) is included in Goldfish, although with less data.

All training and hyperparameter details are in our paper, Goldfish: Monolingual Language Models for 350 Languages (Chang et al., 2024).

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

Sample usage also in this Google Colab: link

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): 37.31MB
Training text data (byte premium scaled): 30.745MB
Training tokens: 9201152 (x10 epochs)
Vocabulary size: 50000
Compute cost: 4.6943465177088e+16 FLOPs or ~4.4 NVIDIA A6000 GPU hours

Training datasets (percentages prior to deduplication):

61.19759%: NLLB (CommonCrawl and ParaCrawl) and NLLB Multi-Domain
24.44245%: MADLAD-400 (CommonCrawl)
7.18196%: Glot500, including Wortschatz Leipzig Data, Wikipedia Hugging Face
3.90026%: Wikipedia 2023/08
2.73814%: AmericasNLP (excluding AmericasNLI)
0.53960%: Wortschatz Leipzig Data

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},
}