avishake/neural-networks-from-scratch

Neural Networks From ScratchNeural Networks From Scratch.

From derivatives to a GPT, one notebook at a timeFrom derivatives to a GPT, one notebook at a time

model card · description

A lecture series of Jupyter notebooks building neural networks from first principles: derivative-based fundamentals, the NameWeave bigram → MLP progression, activations/gradients/batch-norm, manual backpropagation, a WaveNet-style CNN, and finally a decoder-only GPT with self-attention.