Skip to content

mni-ml/mni-ml.github.io

Repository files navigation

Logo

mni-ml

Understanding machine learning by building it from scratch.

mni-ml.github.io

About

mni-ml is a curriculum-style blog that teaches machine learning from the ground up. Rather than treating models as black boxes, each article walks through the math, intuition, and implementation — building everything from scratch using the mni-ml/framework, a TypeScript ML framework with Rust native backends.

What it covers

The curriculum progresses from fundamentals to full model training:

  • Intro to ML — What machine learning is, how models learn from data, and the core loop of prediction, loss, and optimization.
  • Gradient Descent — How gradients drive learning, computation graphs, the forward and backward pass, and automatic differentiation.
  • CUDA & GPU Computing — Why GPUs matter for ML, how CUDA parallelism works, and how the framework leverages native GPU acceleration.
  • Transformers — Attention mechanisms, the transformer architecture, and training a language model on TinyStories from scratch.

Interactive demos

  • Transformer Token Explorer — A 12M-parameter GPT trained on TinyStories, running inference entirely in the browser using the framework's pure TypeScript web backend. See next-token predictions and their probability distributions in real time.

Development

pnpm install
pnpm dev

Articles are authored in Notion and pulled at build time via a custom Astro content loader.

Built with

  • Astro — Static site generator
  • @mni-ml/framework — TypeScript ML framework with Rust native backends
  • React — Interactive demo components
  • Notion — Content management

About

Technical writing

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors