Rust-ML: From Zero to GPT-2

A deep learning framework built entirely from scratch in Pure Rust.

rust-ml-1.mov

Project Mission

My main objective with this repository is to learn and practice Rust by implementing the main deep learning concepts from scratch - using no libraries other than the standard Rust library.

Additionally, because this is a learning project, I will absolutely not use AI for code generation. All code in this repository was written by me.

The Roadmap

Phase 1: Foundations (Current)

• Basic Matrix Engine.
• MNIST Digit Classification (Single Layer).
• Deep Neural Network (MLP) with modular layers.
• Next: Framework polish (Optimizers, Checkpointing, and Testing).

Phase 2: Computer Vision

• Convolutional Neural Networks (CNNs).
• Data Augmentation and Batch Normalization.
• CIFAR-10 / CIFAR-100 Classification.

Phase 3: Generative Models

• Generative Adversarial Networks (GANs).
• Latent Diffusion Models.

Phase 4: Natural Language Processing

• Recurrent Neural Networks (RNNs) & LSTMs.
• Attention Mechanisms & Transformers.
• GPT-2: Training a small-scale language model from scratch.

Project Structure

Under construction.