LLMs-Zero-to-Hero

Why we included this project

This Chinese-language course is for people who want to see what goes on inside a large language model rather than just treating it as a black box. It starts with a nanoGPT written from scratch, then moves to Mixture-of-Experts and DeepSeek's MLA attention, before walking through pre-training, SFT, DPO, and RLHF fine-tuning. Deployment and quantization are on the roadmap. Notebooks run directly, and video explanations fill in the dense parts. Anyone with a single 3090 or 4090 can follow the training demos, so the whole path stays practical even on modest hardware. If you prefer building things yourself over importing transformers, this is a coherent study route.