ymcui/chinese-llama-alpaca

This project provides Chinese-language versions of the LLaMA and Alpaca large language models, AI systems capable of understanding and generating text. The core problem it solves: the original LLaMA model was primarily trained on English text, so its Chinese language ability was limited. This project takes LLaMA as a starting point, expands its vocabulary with Chinese characters, and then re-trains it on Chinese text data to dramatically improve its Chinese comprehension. Two model variants are offered. Chinese LLaMA is the base language model good at text completion, give it the start of a sentence and it generates the rest. Chinese Alpaca goes a step further by training with instruction-following data, making it behave more like a chat assistant (similar to ChatGPT) that can answer questions, write content, and follow directions in Chinese. A key practical feature is local deployment: you can run these models on a personal laptop using just the CPU or a consumer GPU, without sending data to any cloud service. The models are distributed as LoRA weights, a compact "patch" file that you merge with the original LLaMA model weights to get the full model. Supported tools include llama.cpp, transformers, text-generation-webui, LangChain, and privateGPT. Available in 7B, 13B, and 33B parameter sizes, written in Python.