This project open-sources Chinese LLaMA and instruction-tuned Alpaca models built on top of Meta’s LLaMA. The models extend the original LLaMA vocabulary with additional Chinese tokens to improve encoding efficiency and are further pretrained on Chinese corpora. The Alpaca models additionally undergo instruction tuning on Chinese datasets to better understand and follow user instructions. The project provides training scripts, quantization tools, and local deployment guides for both CPU and GPU inference.