This project aims to fine-tune the NLLB-200 (No Language Left Behind model from Meta AI for language translations. Yoruba, Igbo, Hausa, Wolof, Swahili, and Zulu being a low-resource African languages, lacks robust machine translation systems, and this work contributes toward addressing that gap.
# 🧠 Fine-Tuning NLLB for Language Translation
This project fine-tunes Meta AI’s NLLB (No Language Left Behind) model to translate from **English** to **selected language (LaTn)**, especially low-resource languages. The training pipeline leverages Hugging Face’s ecosystem to tokenize data, fine-tune multilingual transformers, and evaluate translation quality using standard metrics.
---
## 📌 Project Highlights
- 🔤 Language Pair: `English → ....`
- 🧰 Model Base: facebook/nllb-200-distilled-1.3B
- ⚙️ Tools: Hugging Face Transformers, Datasets, Evaluate, and Colab (or local GPU)
- 📄 Format: Python code implemented in a Jupyter Notebook
---
## 🚀 Example Usage
### ✅ Install Required Packages
```bash
pip install transformers datasets sacrebleu evaluate sentencepiece ...
```
## 🧪 Evaluation
After training, the model is evaluated using:
BLEU Score – Measures n-gram overlap between candidate and reference translations.
## 🙏 Acknowledgements
- Meta AI – for releasing the NLLB-200 model
- Hugging Face – for transformers, datasets, and evaluation libraries
- JW300 / MENYO-20k / ROGENDO – multilingual parallel corpora for low-resource African languages
- Google Colab – free GPU access for prototyping and training
## Take-Away
Efficient translation models capable of an appreciable degree of accurate English-to-African-language translation. These notebooks show the implementation steps for a Seq2Seq fine-tuning process.