llms-from-scratch-cn

Why we included this project

For anyone who has used an LLM through an API and wondered what is actually happening inside, this Chinese-language tutorial is a good place to find out. It assumes only basic Python and a little PyTorch, then walks the full pipeline, from tokenization and pretraining to fine-tuning, using runnable Jupyter notebooks. Alongside the GPT-style core it adds step-by-step implementations of GLM4, Llama3, and RWKV6. Engineers can read the notebooks to see how the models they already work with behave under the hood, and educators can lift them straight into course material, which matters because most tutorials jump to fine-tuning and deployment without covering the architecture underneath. It is a learning resource rather than something you deploy, so treat it as the reference you read before attempting a serious from-scratch project.