arXiv:2512.001 BitNet Architecture
Intelligence
at the Edge.
We are moving beyond the era of massive GPU clusters. Carthigan Research is pioneering 1-Bit LLMs and quantized models designed to run natively on African hardware.
Bit Models
Current AI models are too bloated for edge devices and consumer hardware. We are researching 1.58-bit quantization techniques that reduce model size by 90% while retaining reasoning capabilities.
- • Inference on 2GB RAM devices
- • Zero-latency offline reasoning
- • Energy-efficient matrix multiplication
Data Sovereignty
An AI is only as good as its training data. We are building the largest structured dataset of Bantu languages and African cultural contexts, ensuring our models understand "Ubuntu" not just as a word, but as a concept.
- • RLHF with local linguistic experts
- • Cultural alignment safety layers
- • Preserving oral histories and cultural heritage
The Carthigan Lab
We publish our findings. Our goal is to advance the field of efficient computing for the entire Global South.