Chinese-LLaMA-Alpaca-3

Why we included this project

These are Chinese-adapted Llama-3 models, covering both base and instruction-tuned versions, built by continuing pretraining on large-scale Chinese data and then fine-tuning with curated instruction sets. Along with the weights, the project releases the pretraining and fine-tuning scripts plus the instruction data, so you can reproduce the training or keep going on your own corpus. The models run in the standard Llama-3 tooling, from llama.cpp on a laptop CPU to vLLM for production serving. Teams building Chinese-language applications, or looking for a strong multilingual base for domain fine-tuning, get working releases and documentation instead of having to assemble the whole stack themselves.