This project extends the Mixtral-8x7B model with a Chinese-optimized vocabulary to improve encoding and decoding efficiency for Chinese text. It provides incremental pretraining code on large-scale open-source corpora and releases both LoRA adapter weights and fully merged model weights for download. The extended vocabulary significantly boosts the model’s Chinese language generation and comprehension capabilities.