Transformer from Scratch

A decoder-only transformer language model implemented from first principles — no high-level model libraries.

Built the attention mechanism, positional encodings, and training loop by hand to deeply understand the architecture behind modern LLMs.