LatinNLP/mt5-small-en-la-translation: Checkpoints & Model Storage

Overview

This repository stores the trained checkpoints and tokenizer files for the mt5-small-en-la-translation model. It is part of the Latin NLP project, focusing on English-Latin translation using mt5-small. The trained model supports translation between English and Latin, fine-tuned without LoRA for 6 epochs.

Project Structure

Files & Directories

• .gitattributes - Configures Git LFS to track large files.
• .gitignore - Specifies ignored files to keep the repository clean.
• install_large_models.sh - A script to download models from Hugging Face.
• README.md - This documentation file.
• added_tokens.json - Custom token mappings for specialized tasks.
• special_tokens_map.json - Maps special tokens to tokenizer components.
• spiece.model - SentencePiece tokenizer model used in mt5-small.
• tokenizer.json - Serialized tokenizer configuration.
• tokenizer_config.json - Additional tokenizer settings.
• Model Checkpoints (Git LFS-tracked):
  • mt5-small-en-la-translation-epoch0.pt (1.2 GB)
  • mt5-small-en-la-translation-epoch1.pt (1.2 GB)
  • mt5-small-en-la-translation-epoch2.pt (1.2 GB)
  • mt5-small-en-la-translation-epoch3.pt (1.2 GB)
  • mt5-small-en-la-translation-epoch4.pt (1.2 GB)
  • mt5-small-en-la-translation-epoch5.pt (1.2 GB)

Installation & Setup

To download and set up the models, use the provided script:

bash install_large_models.sh

This script:

1. Installs huggingface-cli if not already present.
2. Creates directories for the models.
3. Downloads Mistral-7B-Instruct-v0.3 and mt5-small from Hugging Face.

License

The model and training data are released under Creative Commons Attribution 4.0 (CC-BY-4.0).

Citation

@misc{LatinNLP-mt5-small-en-la,
  author = {Axel Delaval, Elsa Lubek},
  title = {English-Latin Translation Model (mt5-small)},
  year = {2025},
  url = {https://huggingface.co/LatinNLP/mt5-small-en-la-translation}
}