goldfish-models/sot_latn_full

license: apache-2.0 language:

• —sot datasets:
• —cis-lmu/Glot500
• —allenai/c4
• —legacy-datasets/wikipedia
• —allenai/MADLAD-400
• —allenai/nllb libraryname: transformers pipelinetag: text-generation tags:
• —goldfish
• —arxiv:2408.10441 ---

sotlatnfull

Goldfish is a suite of monolingual language models trained for 350 languages. This model is the <b>Southern Sotho</b> (Latin script) model trained on 332MB of data (all our data in the language), after accounting for an estimated byte premium of 1.17; content-matched text in Southern Sotho takes on average 1.17x 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: sotlatn is an [individual language](https://iso639-3.sil.org/codetables/639/data) code. It is not contained in any macrolanguage codes contained in Goldfish (for script latn).

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): 388.18MB
• —Training text data (byte premium scaled): 332.915MB
• —Training tokens: 94144000 (x10 epochs)
• —Vocabulary size: 50000
• —Compute cost: 4.80449939963904e+17 FLOPs or ~45.4 NVIDIA A6000 GPU hours

Training datasets (percentages prior to deduplication):

• —34.03982%: Glot500, including AfroMAFT, CORP.NCHLT, Wortschatz Leipzig Data, Mburisano_Covid, MC4, Wikipedia Hugging Face
• —33.19565%: MADLAD-400 (CommonCrawl)
• —32.65729%: NLLB (CommonCrawl and ParaCrawl)
• —0.10724%: Wikipedia 2023/08

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},
}