Mini-LLM. This project aims to replicate mainstream open-source model architectures with limited computational resources, implementing mini models with 100-200M parameters.