Awal LM ⵣ

A lightweight, standalone Neural Network & Tifinagh Language Engine built completely from scratch in pure C++ without any external machine learning libraries.

Developed by Rachid Nichan on Arch Linux.

Live Demo: awallm.com

Features

Pure C++ Implementation: No PyTorch, no TensorFlow, no external dependencies.
Custom Tifinagh Tokenizer: Native multi-byte UTF-8 handling for the official 33 IRCAM Amazigh characters.
Fast Sparse Deep Learning: Custom multi-layer neural network with Backpropagation, ReLU activations, and Softmax.
Gradient Safety: Built-in gradient clipping and early-stopping to prevent overshooting.

Quick Start on Arch Linux

1. Compile

g++ -O3 src/main.cpp -o my_ai

2. Run

./my_ai

Dataset

Trained on a curated Tamazight dialogue dataset supporting native Tifinagh interactions.

📦 Get the Full Dataset: The clean, production-ready Tamazight NLP dataset (dataset.jsonl with 900+ curated conversations in Tifinagh) is available for researchers, linguists, and developers for only $10:

👉 Download Tamazight NLP Dataset Here

Downloads last month: -; Downloads are not tracked for this model. How to track