Chinese LLaMA-Alpaca 3

Chinese LLaMA-Alpaca 3 is the third generation of the well-known Chinese LLaMA and Alpaca model family from Harbin Institute of Technology researcher Yiming Cui, this time built on Meta’s Llama 3 generation. The project extends and instruction-tunes Llama 3 for Chinese, releasing model weights, training recipes and inference tooling so the community can run strong bilingual large language models locally. It is an open-source project on GitHub with around 1,900 stars, and its codebase saw fresh updates in 2026.

Overview

Earlier generations of the series were among the most downloaded Chinese open LLMs, and the third phase continues the formula: expanded Chinese vocabulary, large-scale Chinese pre-training on top of Llama 3, and instruction tuning that produces both base and chat-ready Alpaca variants. The repository ships the models in multiple sizes, together with scripts for fine-tuning, merging vocabulary extensions, quantization and deployment with popular inference stacks. Documentation is provided in Chinese and English, with a wiki and an online arena for comparing model outputs.

Key Features

• Chinese-extended Llama 3 base and instruction-tuned Alpaca models
• Expanded Chinese tokenizer vocabulary for more efficient Chinese text encoding
• Training and fine-tuning scripts for full-parameter and parameter-efficient tuning
• Support for quantization and deployment with mainstream inference frameworks
• Bilingual documentation, wiki guides and a community model arena
• Apache-2.0 licensed code with openly released model checkpoints

💳 Pricing

Free. The code is open source under the Apache-2.0 license and the model weights are released for the community to download and run on their own hardware, subject to Meta’s Llama 3 license terms.

Who Is It For

Chinese-language NLP researchers, developers building bilingual chatbots or content tools, and anyone who wants a proven, locally runnable Chinese LLM family with mature community tooling around it.