--- license: apache-2.0 language: - npi - nep datasets: - allenai/MADLAD-400 - oscar-corpus/OSCAR-2109 - allenai/nllb - cis-lmu/Glot500 library_name: transformers pipeline_tag: text-generation tags: - goldfish - arxiv:2408.10441 --- # nep_deva_10mb Goldfish is a suite of monolingual language models trained for 350 languages. This model is the <b>Nepali</b> (Devanagari script) model trained on 10MB of data, after accounting for an estimated byte premium of 2.63; content-matched text in Nepali takes on average 2.63x 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: nep_deva is a [macrolanguage](https://iso639-3.sil.org/code_tables/639/data) code. None of its contained individual languages are included 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: 39087104 * Maximum sequence length: 512 tokens * Training text data (raw): 26.30MB * Training text data (byte premium scaled): 10.005MB * Training tokens: 2264064 (x10 epochs) * Vocabulary size: 50000 * Compute cost: 1712214622863360.0 FLOPs or ~0.2 NVIDIA A6000 GPU hours Training datasets (percentages prior to deduplication): * 53.66687%: [MADLAD-400 (CommonCrawl)](https://huggingface.co/datasets/allenai/MADLAD-400) * 23.99086%: [OSCAR 2021/09](https://huggingface.co/datasets/oscar-corpus/OSCAR-2109) * 16.12038%: [NLLB (CommonCrawl and ParaCrawl)](https://huggingface.co/datasets/allenai/nllb) * 5.97982%: [Glot500](https://huggingface.co/datasets/cis-lmu/Glot500), including [CCNet](https://github.com/facebookresearch/cc_net), [Earthlings](https://publicdata.canterbury.ac.nz/Research/Geocorpus/CCGLU_v5.0/), [Tatoeba](https://tatoeba.org/en/), [TICO](https://tico-19.github.io/), [W2C](https://lindat.mff.cuni.cz/repository/xmlui/handle/11858/00-097C-0000-0022-6133-9), [WikiMatrix](https://github.com/facebookresearch/LASER/tree/main/tasks/WikiMatrix) * 0.24207%: [eBible](https://ebible.org/find/) ## 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}, } ```