Build and train your own small language model from scratch

FK

Made by Fareed Khan

View profile on GitHub (opens in a new tab)

This is a hands-on tutorial that walks through building a language model entirely from scratch in PyTorch, with no shortcuts through libraries like transformers, trl or peft that normally do the heavy lifting.

It starts from the Transformer architecture in the original "Attention is All You Need" paper and goes all the way through modern post-training steps — supervised fine-tuning, a reward model, and reinforcement learning methods like DPO, PPO and GRPO — so the reader ends with a small model that can hold a conversation, trainable on a single GPU. Every code block is preceded by a plain explanation and followed by the output you should expect.