Tiny GPT in Go. Optimised for Understanding. Trained on Jules Verne Books
SMRTR summary
A pure Go implementation of GPT, trained on Jules Verne novels, built for learning rather than performance. It skips batching to keep matrix math intuitive, removes gonum for radical simplicity, and pairs with Karpathy's Neural Networks: Zero to Hero course. Git tags let you trace the model's evolution from naive bigram to full transformer with self-attention.
SMRTR provides this summary for quick context. The original article belongs to Hacker News.
Read the original article