Logo Lanfrica
  • Accueil
  • Atlas
  • Analyses
  • Documentation
  • Sign in

© 2026 Lanfrica. Tous droits réservés. Tous les droits d'auteur des ressources affichées sur le site Web Lanfrica appartiennent aux détenteurs de droits d'auteur d'origine, sauf indication contraire explicite.

malouka2003/Gemma-4-Darija-Making-AI-Accessible-to-300-Million-North-Africans

Domaine:

natural language processing

Type de record:

model
Créateur:
mal
Hôte:
WHAT: Fine-tuned Google's Gemma 4 2B model to understand and generate Algerian Darija (Derja) - a dialect spoken by 30+ million people in Algeria alone. 🌍 WHY: 300+ million North Africans are excluded from the AI revolution because mainstream LLMs don't support their native dialect. This project bridges that gap. # 🇩🇿 Gemma 4 Darija - Making AI Accessible in Algerian Dialect > Fine-tuning Gemma 4 for Algerian Darija (Derja) - Making AI accessible to **300 million North Africans** ## 🎯 Overview This project fine-tunes Google's **Gemma 4 2B** model to understand and generate Algerian Darija (Derja), a dialect spoken by 30+ million people in Algeria alone and 300+ million across North Africa. ### The Problem 300 million people in North Africa are excluded from the AI revolution because mainstream Large Language Models don't support their native dialect. Every time an Algerian tries to use AI in their language, they face incomprehension. ### The Solution Using **Unsloth** and **LoRA adapters**, we fine-tuned Gemma 4 to understand Darija with minimal computational resources - completely reproducible on free kaggle hardware. --- ## 🏆 Hackathon Submission This project is submitted for the **Gemma 4 Good Hackathon** by Google DeepMind. - **Track**: Impact Track - Digital Equity & Inclusivity - **Competition**: Gemma 4 Good Hackathon --- ## 📊 Results | Metric | Base Gemma 4 | Gemma 4 Darija | Improvement | |--------|-------------|-----------------|-------------| | Darija Understanding | 35% | 85% | **+50 pts** | | Code-switching Support | 28% | 78% | **+50 pts** | | Cultural Relevance | 30% | 82% | **+52 pts** | | **Overall Score** | 31% | **82%** | **+51 pts** | --- ## 🚀 Quick Start ### 1. Run on kaggle (Free) [ ]) 📈 Impact & Vision Immediate Impact ✅ Proof that Gemma 4 can be adapted for Darija ✅ Open-source weights and dataset for community use ✅ Complete documentation for reproducibility Long-term Vision 🎓 Educational Tools: AI tutors for Algerian students 🏥 Healthcare Assistants: Symptom checkers in Darija 🏛️ Government Services: Chatbots for public services 🌾 Agricultural Advisory: Weather and farming tips 🌎 Cross-dialect Model: Extending to Moroccan, Tunisian, Libyan dialects SDG Alignment SDG 4 (Quality Education): Personalized learning in native langu …

Visit

github.com

Languages

Arabic, Algerian SpokenArabic, Tunisian Spoken

Similaires

younesAbaaoud/gemma-4-E4B-it-northern-darija-sftYamemaru/gemma-3-4b-DarijaJoe254h/gemma-4-africa-reasoningglothebeatz/gemma-4-isizulu-script-writerYamemaru/gemma-3-4b-Darija-GRPO-16bitMaking Virtual Learning Environments Accessible to People with Disabilities in Universities in Uganda

younesAbaaoud/gemma-4-E4B-it-northern-darija-sft

Yamemaru/gemma-3-4b-Darija

Joe254h/gemma-4-africa-reasoning

Reproducible multilingual reasoning research for African languages with Gemma 4 and HPC-ready workfl

glothebeatz/gemma-4-isizulu-script-writer

Yamemaru/gemma-3-4b-Darija-GRPO-16bit

Making Virtual Learning Environments Accessible to People with Disabilities in Universities in Uganda

Public and private universities in Uganda have been using Virtual Learning Environments (VLEs) since