Chinese-LLaMA-Alpaca
Open Chinese LLaMA and instruction-tuned Alpaca language models that extend the original LLaMA vocabulary for Chinese, further pretrained and fine-tuned on Chinese text and instructions, for local CPU or GPU deployment.
This project releases Chinese LLaMA models and instruction-tuned Chinese Alpaca models to support open research on large language models in the Chinese NLP community. The models extend the original LLaMA vocabulary with Chinese tokens and are further pretrained on Chinese text, improving Chinese semantic understanding, and the Alpaca models are additionally fine-tuned on Chinese instruction data.
Pretraining and instruction fine-tuning scripts are released so users can further train the models themselves, and the models can be quantized and run locally on a personal computer's CPU or GPU. They work with ecosystems including Transformers, llama.cpp, text-generation-webui, LlamaChat, LangChain and privateGPT, and are available in 7B, 13B and 33B sizes, each with base, Plus and Pro variants.
The project has continued with a newer Chinese-LLaMA-Alpaca-3 release based on the Llama 3 models, and earlier users are encouraged to upgrade.