GPT-2 FROM SCRATCH IN MLX

Summary from the TLDR newsletter

This repository contains a Python script that defines and trains GPT-2 from scratch. The model is trained on around 1 million characters from Shakespeare. It can be trained in around 10 minutes on a MacBook to produce coherent Shakespeare-like text. The script only uses MLX and NumPy as dependencies.