metadata
language:
- ha
- ig
- pcm
- yo
- sw
- am
- so
- xh
- sn
- lg
- ln
license: apache-2.0
tags:
- africa
- llama
- fine-tuned
- multilingual
🌳 BaobabAI v0.1
Africa's first continental AI model, fine-tuned on 11 African languages.
Languages
🇳🇬 Hausa | Igbo | Yoruba | Nigerian Pidgin 🇰🇪 Swahili | 🇪🇹 Amharic | 🇸🇴 Somali 🇿🇦 Xhosa | 🇿🇼 Shona | 🇺🇬 Luganda | 🇨🇩 Lingala
What It Can Do
- Summarize African news articles in local languages
- Classify African content by category
- Understand Nigerian Pidgin English natively
Training Details
| Parameter | Value |
|---|---|
| Base Model | Llama 3.2 3B Instruct |
| Training Pairs | 27,178 |
| Training Steps | 100 |
| Final Loss | 2.81 |
| Method | QLoRA (r=16) via Unsloth |
Note
This is v0.1 — the proof of concept. For the improved version with 15 languages and 86,604 training pairs, see BaobabAI-v0.2
Built By
Pascal Okafor Ogbonna | SabiFlow Technologies Limited 🌍 baobabai.dev · GitHub
Vision
1.4 billion Africans are underrepresented in every major AI model. BaobabAI is changing that — one language at a time.
License
Apache 2.0 — free to use, modify and build on.
The tree is planted. The continent will feel its shade. 🌳