metadata

license: apache-2.0
language:
  - quz
  - que
datasets:
  - cis-lmu/Glot500
  - sil-ai/bloom-lm
library_name: transformers
pipeline_tag: text-generation
tags:
  - goldfish

quz_latn_5mb

Goldfish is a suite of monolingual language models trained for 350 languages. This model is the Cusco Quechua (Latin script) model trained on 5MB of data, after accounting for an estimated byte premium of 1.30; content-matched text in Cusco Quechua takes on average 1.30x 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: quz_latn is an individual language code. Macrolanguage code que_latn (Quechua) is included in Goldfish. Consider using that model depending on your use case.

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/model_details.json. All models are trained with a [CLS] (same as [BOS]) token prepended, and a [SEP] (same as [EOS]) token separating sequences. Details for this model specifically:

Architecture: gpt2
Parameters: 38038528
Maximum sequence length: 512 tokens
Training text data (raw): 6.49MB
Training text data (byte premium scaled): 5.005MB
Training tokens: 1143808 (x10 epochs)
Vocabulary size: 48160
Compute cost: 864746251223040.0 FLOPs or ~0.1 NVIDIA A6000 GPU hours

Training datasets (percentages prior to deduplication):

99.99007%: AmericasNLP (excluding AmericasNLI)
0.00993%: Glot500, including BLOOM

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},
}