Demos
Runnable demonstrations of MbazaNLP models. Each demo includes a command-line script, a Jupyter notebook, and full install instructions.
English ↔ Kinyarwanda Translation
Fine-tuned NLLB-200 models for general, education, and tourism domains. Works offline once models are downloaded.
Models
Nllb_finetuned_general_en_kin
Everyday text and mixed-domain content.
Nllb_finetuned_education_en_kin
Academic, instructional, and curriculum text.
Nllb_finetuned_tourism_en_kin
Travel, hospitality, and cultural content.
Requirements
- Python 3.9+
- ~3 GB disk space per model (downloaded automatically on first run and cached)
- GPU optional — CPU inference works but is slower
Quick Start
# 1. Clone the repository
git clone https://github.com/ronn13/aimbaza.org.git
cd aimbaza.org/demos/translation
# 2. Install dependencies
pip install -r requirements.txt
# 3. Translate
python translate.py "Hello, how are you?"
Models are downloaded from HuggingFace automatically on first use
and cached in ~/.cache/huggingface/.
CLI Usage
# English → Kinyarwanda (default)
python translate.py "The student passed the exam."
# Kinyarwanda → English
python translate.py "Muraho, amakuru?" --src kin_Latn --tgt eng_Latn
# Choose a domain model
python translate.py "Book a room near the national park." --domain tourism
python translate.py "Submit your assignment before the deadline." --domain education| Flag | Default | Options |
|---|---|---|
--src |
eng_Latn |
eng_Latn (English) · kin_Latn (Kinyarwanda) |
--tgt |
kin_Latn |
eng_Latn (English) · kin_Latn (Kinyarwanda) |
--domain |
general |
general · education · tourism |
Jupyter Notebook
The notebook translation_demo.ipynb walks through English→Kinyarwanda,
Kinyarwanda→English, domain comparison across all three models, batch translation,
and an interactive translation cell.
pip install jupyter
jupyter notebook demos/translation/translation_demo.ipynbComing after the June 2026 sprint
A second demonstrator will be published following the community sprint on 14–15 June 2026. Topic will be determined by sprint participants. Join Slack to take part.