Chinese-LLaMA-Alpaca is a landmark open-source project that adapted Meta’s LLaMA models for the Chinese language. It expanded the original tokenizer with a large Chinese vocabulary, continued pre-training on Chinese text, and released instruction-tuned Alpaca variants, together with the full training and deployment toolchain. Launched in 2023, it became one of the most influential Chinese LLM projects and has earned about 18,900 stars on GitHub, with later generations continuing in sibling repositories.
The project releases LoRA weights for Chinese LLaMA base models and Chinese Alpaca instruction models in 7B, 13B and 33B sizes, in Base, Plus and Pro editions trained on progressively more data. Because the original LLaMA license did not allow direct redistribution, the weights ship as LoRA adapters that you merge with the base model using the provided scripts. Everything needed is open: the vocabulary-expansion code, pre-training and instruction-tuning scripts, quantization recipes, and deployment guides for running the models locally on a personal computer’s CPU or GPU through llama.cpp, Hugging Face Transformers, text-generation-webui and other tools.
The code and LoRA weights are free and open source under the Apache-2.0 license. Note that use also remains subject to Meta’s original LLaMA license terms for the underlying base models.
Chinese-LLaMA-Alpaca is for Chinese NLP researchers and developers who want to study, fine-tune or locally deploy Chinese-adapted LLaMA models and learn from a complete, documented training recipe.