大模型部署

23 阅读1分钟

下载cuda
wget developer.download.nvidia.com/compute/cud…

安装必要依赖
yum -y install kernel-headers-$(uname -r)

yum -y install kernel-devel-$(uname -r)
yum update -y
yum -y install pkgconfig libglvnd-devel
yum -y install gcc gcc-c++ make tar

安装cuda
chmod 755 cuda_12.4.1_550.54.15_linux.run
./cuda_12.4.1_550.54.15_linux.run

安装验证
nvidia-smi image.png

安装 NVIDIA 容器工具包
curl -s -L nvidia.github.io/libnvidia-c… | \sudo tee /etc/yum.repos.d/nvidia-container-toolkit.repo
启用镜像源
yum-config-manager --enable nvidia-container-toolkit-experimental
安装
yum install -y nvidia-container-toolkit
重启docker
systemctl  restart docker

部署模型
拉取推理框架
docker pull registry.cn-hangzhou.aliyuncs.com/xprobe_xinf… 运行推理框架
docker run -d
-v /data/xinference/data:/data
-v /data/models:/data/models
-e XINFERENCE_MODEL_SRC=modelscope
-e XINFERENCE_HOME=/data
--net=host
--privileged
--restart=always
--gpus all
--shm-size=100G
--name xinference-latest
registry.cn-hangzhou.aliyuncs.com/xprobe_xinference/xinference:latest
xinference-local -H 0.0.0.0

拉取模型
git lfs install
git lfs clone www.modelscope.cn/Qwen/Qwen2.…