A lightweight framework for Neural Networks and AI

1. 📖 About

AIFrame aims to be a lightweight framework written in pure Python and accelerated by OpenCL and Numpy. This repository is a clean and more well-written version than the actual prototype that isn't available for public use.

Currently, the project is under development and will need some time to be actually used!

2. 🔧 Features (planned)

• ✨ Simplicity for fast prototyping
• 🏎️ Speed and performance for low-end hardware (CPU & GPU support)
• 💻 Training pipelines for easy training
• 📄 Buildscripts for automated use
• 🔌 Low-level access for custom logic
• 💾 Custom file format
• 🗺️ Adaption for other file formats

3. Demo Setup

Current demo (demo-01.06.2025.py) needs following requirements: gzip, requests, numpy (no specific version, newest is enough).

Run the demo.

python demo-01.06.2025.py

This demo downloads the MNIST-Dataset from https://storage.googleapis.com/cvdf-datasets/mnist/. After an example training of all 60k samples, it will evaluate the 10k test samples. Results may very in speed and result, but training accuracy should be around 94-98%. Tested on a AMD Ryzen 5 3600 (6 Cores); training took 12.5 seconds to complete.