nllb 200 distilled 600 M

Introduction

The NLLB-200-Distilled-600M is a machine translation model developed by Meta, designed to support translation across 200 languages. It is particularly focused on low-resource languages and aims to advance research in machine translation. This model is not intended for production use but serves researchers and the machine translation community.

Architecture

NLLB-200 is a distilled model variant containing 600 million parameters. It utilizes the Flores-200 dataset, providing comprehensive language coverage. The model is designed to translate single sentences and is evaluated using metrics such as BLEU, spBLEU, and chrF++.

Training

The model was trained using parallel multilingual data from diverse sources, supplemented by monolingual data from Common Crawl. The training algorithm, data handling, and strategies are detailed in the paper "No Language Left Behind: Scaling Human-Centered Machine Translation." Ethical considerations include prioritized human user safety, potential misinformation risks, and data privacy issues.

Guide: Running Locally

1. Install Dependencies: Ensure you have PyTorch and the transformers library installed.
2. Download the Model: Access the model via the Hugging Face Model Hub.
3. Load the Model: Use the transformers library to load the model for translation tasks.
4. Run Translation: Input sentences for translation and evaluate outputs.

To enhance performance, consider running the model on cloud GPUs such as those offered by AWS, GCP, or Azure.

License

The NLLB-200-Distilled-600M model is released under the CC-BY-NC-4.0 license, which allows use for non-commercial purposes with appropriate attribution. For further inquiries, visit the GitHub issues page of the Fairseq repository.