micrograd

What problem does it solve?

Train tiny neural networks and visualize backpropagation steps.

Core Features & Use Cases

• Value: scalar autograd node with full backpropagation
• Neuron / Layer / MLP: build tiny neural networks from scratch
• Training loop: SGD with MSE, hinge, and binary cross-entropy loss functions
• Interactive walkthrough: 10-step guided tour from "what is a Value?" to "train a moon classifier"
• Canvas visualization: decision boundary grids, parameter snapshots, and training curves for the Canvas UI

Quick Start

Start by creating a session, define a tiny network architecture, and run a short training to observe backpropagation in action.