LLM from Scratch

A 37-chapter public course explaining how large language models work internally, from tokenization and embeddings through attention, training loops, quantization, distributed training, LoRA, DPO, GRPO, distillation, RAG, tool use, AI agents, and safety. Includes a small runnable decoder-only Transformer in PyTorch with tests, plus guides for building datasets from Hugging Face, Kaggle, and Common Crawl. Explicitly framed as educational material, not a production LLM framework.