参考链接：https://zhuanlan.zhihu.com/p/15194336245
注意：/sbin/ldconfig.real: /usr/local/cuda-11.8/targets/x86_64-linux/lib/libcudnn_cnn_infer.so.8 is not a symbolic link这里不是报错的意思，只是警告，这里其实已经安装成功了，用nvidia-ctk --version命令检查若输出版本号就成功安装了。
二、离线安装NVIDIA Container Toolkit
1. 在NVIDIA的GitHub主页找到Ubuntu系统对应的NVIDIA Container Toolkit安装包
该页面的安装包较多，搜索关键词“1.14.1”，下载所有含有“1.14.1”的安装包，安装包的说明如下：
libnvidia-container1_1.14.1-1_amd64.deb # 基础库包，提供了最基本的功能，其他包都依赖于它
libnvidia-container-tools_1.14.1-1_amd64.deb # 基础工具包，依赖于 libnvidia-container1
nvidia-container-toolkit-base_1.14.1-1_amd64.deb # 基础组件包，依赖于前面的包
nvidia-container-toolkit_1.14.1-1_amd64.deb # 主要的工具包，依赖于以上所有包
libnvidia-container1-dbg_1.14.1-1_amd64.deb # 调试符号包，只在调试问题时使用
libnvidia-container-dev_1.14.1-1_amd64.deb # 开发包，只在进行开发时使用其中最后两个安装包可以选择不下载和不安装
2. 执行下列命令安装NVIDIA Container Toolkit：
安装上面6个deb文件
sudo dpkg *.deb
3. 查看NVIDIA Container Toolkit的版本以验证是否安装成功
nvidia-ctk --version
4. 设置Docker默认使用NVIDIA runtime
sudo nvidia-ctk runtime configure --runtime=docker
复现结果：
(base) gpu@gpu01:~/dockerdir/nvidia-toolkit/install1.14$ sudo dpkg -i *.deb
(Reading database ... 220709 files and directories currently installed.)
Preparing to unpack libnvidia-container1_1.14.1-1_amd64.deb ...
Unpacking libnvidia-container1:amd64 (1.14.1-1) over (1.12.0-1) ...
Selecting previously unselected package libnvidia-container1-dbg:amd64.
Preparing to unpack libnvidia-container1-dbg_1.14.1-1_amd64.deb ...
Unpacking libnvidia-container1-dbg:amd64 (1.14.1-1) ...
Selecting previously unselected package libnvidia-container-dev:amd64.
Preparing to unpack libnvidia-container-dev_1.14.1-1_amd64.deb ...
Unpacking libnvidia-container-dev:amd64 (1.14.1-1) ...
Preparing to unpack libnvidia-container-tools_1.14.1-1_amd64.deb ...
Unpacking libnvidia-container-tools (1.14.1-1) over (1.12.0-1) ...
dpkg: warning: downgrading nvidia-container-toolkit from 1.17.8-1 to 1.14.1-1
Preparing to unpack nvidia-container-toolkit_1.14.1-1_amd64.deb ...
Unpacking nvidia-container-toolkit (1.14.1-1) over (1.17.8-1) ...
Preparing to unpack nvidia-container-toolkit-base_1.14.1-1_amd64.deb ...
Unpacking nvidia-container-toolkit-base (1.14.1-1) over (1.12.0-1) ...
dpkg: warning: unable to delete old directory '/etc/nvidia-container-runtime': Directory not empty
Setting up libnvidia-container1:amd64 (1.14.1-1) ...
Setting up libnvidia-container1-dbg:amd64 (1.14.1-1) ...
Setting up libnvidia-container-dev:amd64 (1.14.1-1) ...
Setting up libnvidia-container-tools (1.14.1-1) ...
Setting up nvidia-container-toolkit-base (1.14.1-1) ...
Setting up nvidia-container-toolkit (1.14.1-1) ...
Processing triggers for libc-bin (2.35-0ubuntu3.9) ...
/sbin/ldconfig.real: /usr/local/cuda-11.8/targets/x86_64-linux/lib/libcudnn_cnn_infer.so.8 is not a symbolic link
/sbin/ldconfig.real: /usr/local/cuda-11.8/targets/x86_64-linux/lib/libcudnn_adv_train.so.8 is not a symbolic link
/sbin/ldconfig.real: /usr/local/cuda-11.8/targets/x86_64-linux/lib/libcudnn.so.8 is not a symbolic link
/sbin/ldconfig.real: /usr/local/cuda-11.8/targets/x86_64-linux/lib/libcudnn_ops_infer.so.8 is not a symbolic link
/sbin/ldconfig.real: /usr/local/cuda-11.8/targets/x86_64-linux/lib/libcudnn_cnn_train.so.8 is not a symbolic link
/sbin/ldconfig.real: /usr/local/cuda-11.8/targets/x86_64-linux/lib/libcudnn_ops_train.so.8 is not a symbolic link
/sbin/ldconfig.real: /usr/local/cuda-11.8/targets/x86_64-linux/lib/libcudnn_adv_infer.so.8 is not a symbolic link
(base) gpu@gpu01:~/dockerdir/nvidia-toolkit/install1.14$ nvidia-ctk --version
NVIDIA Container Toolkit CLI version 1.14.1
commit: 6094effd58d88becdfb7900ef5df7fa274686620
(base) gpu@gpu01:~/dockerdir/nvidia-toolkit/install1.14$ sudo nvidia-ctk runtime configure --runtime=docker
INFO[0000] Config file does not exist; using empty config
INFO[0000] Wrote updated config to /etc/docker/daemon.json
INFO[0000] It is recommended that docker daemon be restarted. 注意：/sbin/ldconfig.real: /usr/local/cuda-11.8/targets/x86_64-linux/lib/libcudnn_cnn_infer.so.8 is not a symbolic link这里不是报错的意思，只是警告，这里其实已经安装成功了，用nvidia-ctk --version命令检查若输出版本号就成功安装了。