This project is a hands-on tutorial for training large language models from scratch. It doesn't just call transformers to run an example, but implements the complete process from the ground up using PyTorch, including Transformer architecture, pre-training, supervised fine-tuning, reward modeling, and evaluation.