基本硬件:
V100-32G,python3.10
1.下载项目(我是本地下载再映射上去,映射路径文件夹起名为Chinese-Llama,注意一下)
https://github.com/ymcui/Chinese-LLaMA-Alpaca-3
也可以走git那一套流程
git clone https://github.com/ymcui/Chinese-LLaMA-Alpaca-3.git
2.创建conda环境
conda create -n llava python=3.10
conda activate llava
cd Chinese-Llama
4.下载包
pip install -r requirements.txt
5.下载基座权重
https://huggingface.co/hfl/llama-3-chinese-8b-instruct-v2/tree/main
6.下载Lora权重(为了省钱免训练)
https://huggingface.co/hfl/llama-3-chinese-8b-instruct-v2-lora/tree/main
7.上传到服务器(绝对路径加粗部分根据自己的服务器改)(注意权重巨多,需要最少60G空间存储数据,·)
路径为:/root/siton-tmp/Chinese-Llama/llama-3-chinese-8b-instruct-v2(基座)
路径为:/root/siton-tmp/Chinese-Llama/llama3-Lora(Lora)