ZdravkoGrad

Archived learning note: ZdravkoGrad is my hands-on implementation of Andrej Karpathy's micrograd lecture and micrograd repository.

I worked through the project in a blank Jupyter notebook to understand what automatic differentiation and backpropagation do beneath libraries such as PyTorch.

What it contains

• A scalar Value object that records operations in a computation graph.
• Reverse-mode automatic differentiation and backpropagation through that graph.
• Small Neuron, Layer, and multilayer-perceptron abstractions built on top of the engine.
• A PyTorch-like API used to train small neural networks.

The implementation is available in the ZdravkoGrad repository. The previous version of this page closely followed the lecture step by step, so I have replaced it with this short source-led record instead.