GPU Linux虚拟主机GN7型安装配置文档定稿

sockstack / 12 / 2023-11-14 11:48:55

ChatGPT 可用网址，仅供交流学习使用，如对您有所帮助，请收藏并推荐给需要的朋友。 <a href="https://ckai.xyz/?sockstack&section=detail" target="__blank">https://ckai.xyz</a> <article class="_2rhmJa">  我现在配的虚拟主机缺了颗GPU，一些使用GPU的算法无法在线演示，有点美中不足。网上搜了一圈，腾讯云现在有个推广活动，花点小钱就可以配一个实验了，比较便宜，其它一些厂的个人担负不起，所以在腾讯云上买了一个实例，试用一个月，以完成配置测试的实验。
  完全没有用过Ubuntu，大概需要重装很多次才能搞定。进度很难估计，先用一个月看看。所以要写下每一步的详细文档，以便随时重装。我将安装tensorflow2.6、pytorch1.11.0与HanLP2.1，它们的版本不会冲突。对应的是CUDA11.2与cuDNN8.5，cuDNN8.5适配CUDA11.X（最后改回cuDNN8.1了），Python 3.9。然后会在Rstudio中通过reticulate包、tensorflow包与keras包调用它们。如果有时间，也会测试一下R语言实现的<a href="https://links.jianshu.com/go?to=https%3A%2F%2Fcran.r-project.org%2Fweb%2Fpackages%2Ftorch%2Findex.html" target="_blank">torch包</a>，它提供了类似PyTorch的功能，直接调用<a href="https://links.jianshu.com/go?to=https%3A%2F%2Fgithub.com%2Fpytorch%2Fpytorch%2Fblob%2Fmaster%2Fdocs%2Flibtorch.rst" target="_blank">libtorch</a>。
  腾讯云GPU计算型虚拟主机 GN7，搭载 NVIDIA T4 GPU，8核CPU+32G RAM+100G SSD+1颗T4，带宽5M，￥80/试用一个月，试用计划<a href="https://links.jianshu.com/go?to=https%3A%2F%2Fcloud.tencent.com%2Fact%2Fpro%2Fgpu-study%3FfromSource%3Dgwzcw.6866631.6866631.6866631%26utm_medium%3Dcpc%26utm_id%3Dgwzcw.6866631.6866631.6866631%26bd_vid%3D11503804075512292365" target="_blank">GPU实验室</a>，<a href="https://links.jianshu.com/go?to=https%3A%2F%2Fcloud.tencent.com%2Fdeveloper%2Farticle%2F2073793" target="_blank">入门教程</a>。
一、从镜像安装操作系统。
  不同的GPU驱动版本，可选的CUDA版本不同，要选460.106.00版。
公共镜像：Ubuntu Server 18.04.1 LTS64位
后台自动安装GPU驱动
GPU 驱动版本：460.106.00
CUDA版本: 11.2.2
cuDNN版本： 8.2.1
用户名: ubuntu
网址 ：172.16.XX.XX（内）106.52.XX.XX（公）
  安装完成，用SecureCRT或PuTTY连接，它的SSH服务器启用了更新的密钥交换算法，SecureCRT要升级到9.0版以上。
1、登录机器后，先启用root账户，<a href="https://links.jianshu.com/go?to=https%3A%2F%2Fblog.csdn.net%2Fweixin_42677788%2Farticle%2Fdetails%2F120488072" target="_blank">参阅资料</a>。设置root账户密码：
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-bash"><code class=" language-bash">$sudo passwd root
</code></pre>
</div>
账户切换：
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-bash"><code class=" language-bash">$su root
#su ubuntu
</code></pre>
</div>
如果要允许root在SSH登录，参阅<a href="https://links.jianshu.com/go?to=https%3A%2F%2Fwp.gxnas.com%2F10746.html" target="_blank">资料1</a>与<a href="https://links.jianshu.com/go?to=https%3A%2F%2Fblog.csdn.net%2Faiwowuzui%2Farticle%2Fdetails%2F126930388" target="_blank">资料2</a>。
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-bash"><code class=" language-bash"># vi /etc/ssh/sshd_config
</code></pre>
</div>
找到这一段：
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-bash"><code class=" language-bash"># Authentication:
#LoginGraceTime 2m
#PermitRootLogin prohibit-password
#StrictModes yes
#MaxAuthTries 6
#MaxSessions 10
</code></pre>
</div>
改成这样：
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-bash"><code class=" language-bash"># Authentication:
#LoginGraceTime 2m
#PermitRootLogin prohibit-password
PermitRootLogin yes
StrictModes yes
#MaxAuthTries 6
#MaxSessions 10
</code></pre>
</div>
重启SSH服务器：
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-bash"><code class=" language-bash"># systemctl restart sshd.service
</code></pre>
</div>
为了方便后面安装软件，关闭sudo命令的PATH限制，<a href="https://links.jianshu.com/go?to=https%3A%2F%2Fsuperuser.com%2Fquestions%2F927512%2Fhow-to-set-path-for-sudo-commands" target="_blank">参阅资料</a>，用wq!存盘：
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-bash"><code class=" language-bash"># vi /etc/sudoers
</code></pre>
</div>
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-objectivec"><code class=" language-objectivec">Defaults env_reset
Defaults mail_badpass
# Defaults secure_path="/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin:/snap/bin"
</code></pre>
</div>
然后用 -E选项运行sudo命令可以继承当前用户的环境变量设置，这样安装软件也可以不用登录到root用户，比如后面用conda命令安装Python软件包：
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-ruby"><code class=" language-ruby">(gpu) ubuntu@VM-0-14-ubuntu:~$ sudo -E conda list hanlp
# packages in environment at /usr/local/anaconda3/envs/gpu:
#
# Name Version Build Channel
hanlp 2.1.0b42 pypi_0 pypi
hanlp-common 0.0.18 pypi_0 pypi
hanlp-downloader 0.0.25 pypi_0 pypi
hanlp-trie 0.0.5 pypi_0 pypi
</code></pre>
</div>
2、大约需要10～15分钟进行安装，可以用以下命令查看当前安装进程：
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-cpp"><code class=" language-cpp">root@VM-0-14-ubuntu:~# ps aux | grep -i install
root 8158 0.0 0.0 13776 1156 pts/0 S+ 08:50 0:00 grep --color=auto -i install
</code></pre>
</div>
如上面所示，里面没有nv_driver_install.sh及nv_cuda_install.sh，则表示驱动安装已经完成。
3、验证GPU驱动安装成功。
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-ruby"><code class=" language-ruby">root@VM-0-14-ubuntu:~# nvidia-smi
Sat Oct 29 08:52:11 2022 
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 460.106.00 Driver Version: 460.106.00 CUDA Version: 11.2 |
|-------------------------------+----------------------+----------------------+
| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|===============================+======================+======================|
| 0 Tesla T4 On | 00000000:00:08.0 Off | 0 |
| N/A 28C P8 8W / 70W | 0MiB / 15109MiB | 0% Default |
| | | N/A |
+-------------------------------+----------------------+----------------------+
 
+-----------------------------------------------------------------------------+
| Processes: |
| GPU GI CI PID Type Process name GPU Memory |
| ID ID Usage |
|=============================================================================|
| No running processes found |
+-----------------------------------------------------------------------------+
</code></pre>
</div>
4、验证CUDA 安装成功。上面入门教程写的不适用于这个配置组合，/usr/local/cuda是到/usr/local/cuda-11.2的链接。
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-ruby"><code class=" language-ruby">root@VM-0-14-ubuntu:~# cat /usr/local/cuda/version.txt
cat: /usr/local/cuda/version.txt: No such file or directory
root@VM-0-14-ubuntu:~# find / -name cuda
/usr/local/cuda-11.2/targets/x86_64-linux/include/cuda
/usr/local/cuda-11.2/targets/x86_64-linux/include/thrust/system/cuda
/usr/local/cuda
root@VM-0-14-ubuntu:~# cd /usr/local/cuda
root@VM-0-14-ubuntu:/usr/local/cuda# ls
bin DOCS extras lib64 nsight-compute-2020.3.1 nsight-systems-2020.4.3 nvvm README share targets version.json
compute-sanitizer EULA.txt include libnvvp nsightee_plugins nvml nvvm-prev samples src tools
root@VM-0-14-ubuntu:/usr/local/cuda# cd bin
root@VM-0-14-ubuntu:/usr/local/cuda/bin# ./nvcc -V
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2021 NVIDIA Corporation
Built on Sun_Feb_14_21:12:58_PST_2021
Cuda compilation tools, release 11.2, V11.2.152
Build cuda_11.2.r11.2/compiler.29618528_0
</code></pre>
</div>
5、验证cuDNN安装，上面入门教程写的同样不适用，系统从镜像安装cuDNN没有成功。
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-ruby"><code class=" language-ruby">root@VM-0-14-ubuntu:/usr/local/cuda/bin# cat /usr/include/cudnn_version.h | grep CUDNN_MAJOR -A 2
cat: /usr/include/cudnn_version.h: No such file or directory
</code></pre>
</div>
二、手工安装cuDNN，<a href="https://links.jianshu.com/go?to=https%3A%2F%2Fblog.csdn.net%2Ftangjiahao10%2Farticle%2Fdetails%2F125227005" target="_blank">参阅资料</a>。
cuDNN下载要登录Nvidia的网站，所以用下面的命令是不行的：
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-ruby"><code class=" language-ruby">wget https://developer.nvidia.com/compute/cudnn/secure/8.5.0/local_installers/11.7/cudnn-linux-x86_64-8.5.0.96_cuda11-archive.tar.xz
</code></pre>
</div>
1、在笔记本上下载好，再用SecureFX从SSH端口传到服务器上，解压安装。Linux上验证过的CUDA与cuDNN等的匹配关系<a href="https://links.jianshu.com/go?to=https%3A%2F%2Ftensorflow.google.cn%2Finstall%2Fsource%23linux" target="_blank">参阅该资料</a>。
<div class="image-package">
<div class="image-container" style="max-width: 700px; max-height: 341px;">
<div class="image-container-fill" style="padding-bottom: 31.369999999999997%;"></div>
<div class="image-view" data-width="1087" data-height="341"><img data-original-src="//upload-images.jianshu.io/upload_images/28576403-d154a2fc6aae162b.PNG" data-original-width="1087" data-original-height="341" data-original-format="image/png" data-original-filesize="30538" data-image-index="0" style="cursor: zoom-in;" class="image-loading"></div>
</div>
<div class="image-caption">tensorflow-cuda-cudnn-python版本对照表</div>
</div>
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-css"><code class=" language-css"># tar -xvf cudnn-linux-x86_64-8.5.0.96_cuda11-archive.tar.xz
# cd cudnn-linux-x86_64-8.5.0.96_cuda11-archive
# cp lib/* /usr/local/cuda/lib64/
# cp include/* /usr/local/cuda/include/
# chmod a+r /usr/local/cuda/lib64/*
# chmod a+r /usr/local/cuda/include/*
</code></pre>
</div>
2、将CUDA目录加入全局环境变量：
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-bash"><code class=" language-bash"># vi /etc/profile
</code></pre>
</div>
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-bash"><code class=" language-bash">export PATH=/usr/local/cuda-11.2/bin:$PATH
export LD_LIBRARY_PATH=/usr/local/cuda-11.2/lib64:$LD_LIBRARY_PATH
export CUDA_HOME=/usr/local/cuda-11.2
</code></pre>
</div>
3、source /etc/profile使它生效，或者logout再login，验证cuDNN安装：
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-bash"><code class=" language-bash">root@VM-0-14-ubuntu:/usr/local/cuda/bin# source /etc/profile
root@VM-0-14-ubuntu:/usr/local/cuda/bin# echo $PATH
/usr/local/cuda-11.2/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin:/usr/games:/usr/local/games:/snap/bin
root@VM-0-14-ubuntu:/usr/local/cuda/bin# nvcc -V
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2021 NVIDIA Corporation
Built on Sun_Feb_14_21:12:58_PST_2021
Cuda compilation tools, release 11.2, V11.2.152
Build cuda_11.2.r11.2/compiler.29618528_0
root@VM-0-14-ubuntu:/usr/local/cuda/bin# cat /usr/local/cuda/include/cudnn_version.h | grep CUDNN_MAJOR -A 2
#define CUDNN_MAJOR 8
#define CUDNN_MINOR 5
#define CUDNN_PATCHLEVEL 0
--
#define CUDNN_VERSION (CUDNN_MAJOR * 1000 + CUDNN_MINOR * 100 + CUDNN_PATCHLEVEL)

#endif /* CUDNN_VERSION_H */
</code></pre>
</div>
三、安装Anaconda
1、下载安装Anaconda，装在/usr/local/anaconda3目录。
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-ruby"><code class=" language-ruby">$ wget https://mirrors.tuna.tsinghua.edu.cn/anaconda/archive/Anaconda3-2022.10-Linux-x86_64.sh
$ sudo bash Anaconda3-2022.10-Linux-x86_64.sh
</code></pre>
</div>
安装完成，选择运行 conda init：
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-bash"><code class=" language-bash">done
installation finished.
Do you wish the installer to initialize Anaconda3
by running conda init? [yes|no]
[no] &gt;&gt;&gt; yes
modified /usr/local/anaconda3/condabin/conda
modified /usr/local/anaconda3/bin/conda
modified /usr/local/anaconda3/bin/conda-env
no change /usr/local/anaconda3/bin/activate
no change /usr/local/anaconda3/bin/deactivate
no change /usr/local/anaconda3/etc/profile.d/conda.sh
no change /usr/local/anaconda3/etc/fish/conf.d/conda.fish
no change /usr/local/anaconda3/shell/condabin/Conda.psm1
no change /usr/local/anaconda3/shell/condabin/conda-hook.ps1
no change /usr/local/anaconda3/lib/python3.9/site-packages/xontrib/conda.xsh
no change /usr/local/anaconda3/etc/profile.d/conda.csh
modified /root/.bashrc

==&gt; For changes to take effect, close and re-open your current shell. &lt;==

If you'd prefer that conda's base environment not be activated on startup, 
   set the auto_activate_base parameter to false:

conda config --set auto_activate_base false

Thank you for installing Anaconda3!

===========================================================================

Working with Python and Jupyter is a breeze in DataSpell. It is an IDE
designed for exploratory data analysis and ML. Get better data insights
with DataSpell.

DataSpell for Anaconda is available at: https://www.anaconda.com/dataspell
</code></pre>
</div>
编辑全局变量脚本，把设置conda环境的脚本加到最后，以便所有用户都可用。
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-bash"><code class=" language-bash"># vi /etc/profile
</code></pre>
</div>
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-bash"><code class=" language-bash"># &gt;&gt;&gt; conda initialize &gt;&gt;&gt;
# !! Contents within this block are managed by 'conda init' !!
__conda_setup="$('/usr/local/anaconda3/bin/conda' 'shell.bash' 'hook' 2&gt; /dev/null)"
if [ $? -eq 0 ]; then
 eval "$__conda_setup"
else
 if [ -f "/usr/local/anaconda3/etc/profile.d/conda.sh" ]; then
 . "/usr/local/anaconda3/etc/profile.d/conda.sh"
 else
 export PATH="/usr/local/anaconda3/bin:$PATH"
 fi
fi
unset __conda_setup
# &lt;&lt;&lt; conda initialize &lt;&lt;&lt;
</code></pre>
</div>
运行~/.bashrc使conda base环境生效，或者logout再login。
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-bash"><code class=" language-bash"># source ~/.bashrc
</code></pre>
</div>
2、root安装tensorflow-gpu 2.6。
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-bash"><code class=" language-bash"># conda create --name gpu python=3.9
# pip install ipykernel
# python -m ipykernel install --user --name gpu
# conda activate gpu
# pip install tensorflow-gpu==2.6
</code></pre>
</div>
3、ubuntu用户测试安装。
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-ruby"><code class=" language-ruby">(base) ubuntu@VM-0-14-ubuntu:~$ conda activate gpu
(gpu) ubuntu@VM-0-14-ubuntu:~$ python
Python 3.9.13 (main, Oct 13 2022, 21:15:33) 
[GCC 11.2.0] :: Anaconda, Inc. on linux
Type "help", "copyright", "credits" or "license" for more information.
&gt;&gt;&gt; import tensorflow as tf
&gt;&gt;&gt; tf.test.is_built_with_cuda() 
True
&gt;&gt;&gt; a = tf.constant(1.)
2022-10-29 18:14:29.577429: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2022-10-29 18:14:29.585025: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2022-10-29 18:14:29.585898: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2022-10-29 18:14:29.587034: I tensorflow/core/platform/cpu_feature_guard.cc:142] This TensorFlow binary is optimized with oneAPI Deep Neural Network Library (oneDNN) to use the following CPU instructions in performance-critical operations: AVX2 AVX512F FMA
To enable them in other operations, rebuild TensorFlow with the appropriate compiler flags.
2022-10-29 18:14:29.587744: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2022-10-29 18:14:29.588624: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2022-10-29 18:14:29.589442: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2022-10-29 18:14:30.245462: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2022-10-29 18:14:30.246301: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2022-10-29 18:14:30.247122: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2022-10-29 18:14:30.247901: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1510] Created device /job:localhost/replica:0/task:0/device:GPU:0 with 13803 MB memory: -&gt; device: 0, name: Tesla T4, pci bus id: 0000:00:08.0, compute capability: 7.5
&gt;&gt;&gt; b = tf.constant(2.)
&gt;&gt;&gt; print(a+b)
tf.Tensor(3.0, shape=(), dtype=float32)
&gt;&gt;&gt; 
</code></pre>
</div>
四、配置Jupyter Notebook
  Jupyter Notebook的安装配置要简单一点，先配起它来验证GPU环境的安装，<a href="https://links.jianshu.com/go?to=https%3A%2F%2Fblog.csdn.net%2Fm0_46459047%2Farticle%2Fdetails%2F121567309" target="_blank">参阅资料</a>。
1、安装Anaconda3时base环境已经安装了Jupyter Notebook，但上面建立的虚拟环境"gpu"里面没有安装，要安装一下，先用conda activate激活环境再装。
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-bash"><code class=" language-bash">(base) root@VM-0-14-ubuntu:~# conda activate gpu
(gpu) root@VM-0-14-ubuntu:~# conda list jupyter
# packages in environment at /usr/local/anaconda3/envs/gpu:
#
# Name Version Build Channel
(gpu) root@VM-0-14-ubuntu:~# conda install jupyter notebook
Collecting package metadata (current_repodata.json): done
Solving environment: done

## Package Plan ##

environment location: /usr/local/anaconda3/envs/gpu

added / updated specs:
    - jupyter
    - notebook

The following packages will be downloaded:

package                    |            build
    ---------------------------|-----------------
    asttokens-2.0.5            |     pyhd3eb1b0_0          20 KB
......
Proceed ([y]/n)? y

Downloading and Extracting Packages
soupsieve-2.3.2.post | 65 KB | ################################################################################################################################################## | 100% 
......
asttokens-2.0.5 | 20 KB | ################################################################################################################################################## | 100% 
Preparing transaction: done
Verifying transaction: done
Executing transaction: done
Retrieving notices: ...working... done
 
</code></pre>
</div>
2、为用户ubuntu配置Jupyter Notebook。
1）产生配置文件。
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-tsx"><code class=" language-tsx">(base) ubuntu@VM-0-14-ubuntu:~$ jupyter notebook --generate-config
Writing default config to: /home/ubuntu/.jupyter/jupyter_notebook_config.py
</code></pre>
</div>
2）产生登录口令的Hash。
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-ruby"><code class=" language-ruby">(base) ubuntu@VM-0-14-ubuntu:~$ python
Python 3.9.13 (main, Aug 25 2022, 23:26:10) 
[GCC 11.2.0] :: Anaconda, Inc. on linux
Type "help", "copyright", "credits" or "license" for more information.
&gt;&gt;&gt; from notebook.auth import passwd
&gt;&gt;&gt; passwd()
Enter password: 
Verify password: 
'argon2:$argon2id$v=19$m=10240,t=10,p=xxxxxxxxxxxxxxxxxxx'
&gt;&gt;&gt; 
</code></pre>
</div>
3、编辑配置文件，拷贝上面登录口令的Hash到配置文件。
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-ruby"><code class=" language-ruby">$ vi ~/.jupyter/jupyter_notebook_config.py
</code></pre>
</div>
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-php"><code class=" language-php">c.NotebookApp.ip='*' # 就是设置所有ip皆可访问 
c.NotebookApp.password = 'argon2:$argon2id$v=19$m=10240,t=10,p=xxxxxxxxxxxxxxxxxxx' # 上面复制的那个sha密文' 
c.NotebookApp.open_browser = False # 禁止自动打开浏览器 
c.NotebookApp.port =8888 # 端口
c.NotebookApp.notebook_dir = '/home/ubuntu/jupyternotebook' #设置Notebook启动进入的目录
</code></pre>
</div>
4、启动Jupyter Notebook，注意要先激活使用"gpu"环境，用的是它。
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-ruby"><code class=" language-ruby">(base) ubuntu@VM-0-14-ubuntu:~$ conda activate gpu
(gpu) ubuntu@VM-0-14-ubuntu:~$ conda list jupyter
# packages in environment at /usr/local/anaconda3/envs/gpu:
#
# Name Version Build Channel
jupyter 1.0.0 py39h06a4308_8 
jupyter_client 7.3.5 py39h06a4308_0 
jupyter_console 6.4.3 pyhd3eb1b0_0 
jupyter_core 4.11.1 py39h06a4308_0 
jupyter_server 1.18.1 py39h06a4308_0 
jupyterlab 3.4.4 py39h06a4308_0 
jupyterlab_pygments 0.1.2 py_0 
jupyterlab_server 2.15.2 py39h06a4308_0 
jupyterlab_widgets 1.0.0 pyhd3eb1b0_1 
(gpu) ubuntu@VM-0-14-ubuntu:~$ jupyter notebook &amp;
[1] 16510
(gpu) ubuntu@VM-0-14-ubuntu:~$ [W 07:53:21.094 NotebookApp] WARNING: The notebook server is listening on all IP addresses and not using encryption. This is not recommended.
[W 2022-10-30 07:53:21.326 LabApp] 'ip' has moved from NotebookApp to ServerApp. This config will be passed to ServerApp. Be sure to update your config before our next release.
[W 2022-10-30 07:53:21.326 LabApp] 'password' has moved from NotebookApp to ServerApp. This config will be passed to ServerApp. Be sure to update your config before our next release.
[W 2022-10-30 07:53:21.326 LabApp] 'password' has moved from NotebookApp to ServerApp. This config will be passed to ServerApp. Be sure to update your config before our next release.
[W 2022-10-30 07:53:21.326 LabApp] 'port' has moved from NotebookApp to ServerApp. This config will be passed to ServerApp. Be sure to update your config before our next release.
[W 2022-10-30 07:53:21.326 LabApp] 'notebook_dir' has moved from NotebookApp to ServerApp. This config will be passed to ServerApp. Be sure to update your config before our next release.
[W 2022-10-30 07:53:21.326 LabApp] 'notebook_dir' has moved from NotebookApp to ServerApp. This config will be passed to ServerApp. Be sure to update your config before our next release.
[I 2022-10-30 07:53:21.333 LabApp] JupyterLab extension loaded from /usr/local/anaconda3/envs/gpu/lib/python3.9/site-packages/jupyterlab
[I 2022-10-30 07:53:21.333 LabApp] JupyterLab application directory is /usr/local/anaconda3/envs/gpu/share/jupyter/lab
[I 07:53:21.337 NotebookApp] Serving notebooks from local directory: /home/ubuntu/jupyternotebook
[I 07:53:21.337 NotebookApp] Jupyter Notebook 6.4.12 is running at:
[I 07:53:21.337 NotebookApp] http://VM-0-14-ubuntu:8888/
[I 07:53:21.337 NotebookApp] Use Control-C to stop this server and shut down all kernels (twice to skip confirmation).
</code></pre>
</div>
5、浏览器访问，输入上面设置的密码登录，然后新建一个测试的notebook测试GPU环境的安装。
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-dart"><code class=" language-dart">import tensorflow as tf
tf.test.is_built_with_cuda() 
a = tf.constant(1.)
b = tf.constant(2.)
print(a+b)
</code></pre>
</div>
<div class="image-package">
<div class="image-container" style="max-width: 700px; max-height: 376px;">
<div class="image-container-fill" style="padding-bottom: 53.65%;"></div>
<div class="image-view" data-width="1920" data-height="1030"><img data-original-src="//upload-images.jianshu.io/upload_images/28576403-7c228fb43658a541.PNG" data-original-width="1920" data-original-height="1030" data-original-format="image/png" data-original-filesize="162808" data-image-index="1" style="cursor: zoom-in;" class="image-loading"></div>
</div>
<div class="image-caption">JupyterNotebook测试tensorflow-gpu安装</div>
</div>
 
6、新建一个测试的notebook测试keras与cuDNN。

# 一、数据集处理
 
# 构建模型
(x_train_raw, y_train_raw),(x_test_raw,y_test_raw) = datasets.mnist.load_data()
print(y_train_raw[0]) # 5
print(x_train_raw.shape, y_train_raw.shape) # (60000,28,28)6万张训练集
print(x_test_raw.shape, y_test_raw.shape) # (10000,28,28)1万张测试集
 
num_classes = 10
y_train= keras.utils.to_categorical(y_train_raw,num_classes) # 将分类标签变为独热码(onehot)
y_test = keras.utils.to_categorical(y_test_raw,num_classes)
print(y_train[0]) # [0. 0. 0. 0. 0. 1. 0. 0. 0. 0.]
 
# 数据可视化，看看测试的数据
plt.figure()
for i in range(9):
 plt.subplot(3,3,i+1)
 plt.imshow(x_train_raw[i])
 plt.axis('off')
plt.show()

# 二、构建并编译全连接神经网络
 
# 编译全连接层
x_train = x_train_raw.reshape(60000,784) # 将28*28的图像展开成784*1的向量
x_test = x_test_raw.reshape(10000,784) # 将图像像素值归一化0~1
x_train= x_train.astype('float32')/255
x_test = x_test.astype('float32')/255 
 
model = keras.Sequential([ # 创建模型。模型包括3个全连接层和两个RELU激活函数
 layers.Dense(512,activation='relu', input_dim = 784), # 降维处理
 layers.Dense(256,activation='relu'),
 layers.Dense(124,activation='relu'),
 layers.Dense(num_classes,activation='softmax')
])

# 三、训练网络
 
Optimizer = optimizers.Adam(0.001)
model.compile(loss=keras.losses.categorical_crossentropy,
 optimizer=Optimizer, # Adam优化器 
 metrics=['accuracy']
)
model.fit(x_train,y_train, # 训练集数据标签
 batch_size=128, # 批大小 
 epochs=10, # 训练的轮次
 verbose=1) # 输出日志

# 四、测试模型
 
score = model.evaluate(x_test,y_test,verbose=0)
print('Test loss:', score[0]) # 损失函数: 0.0853068439
print('Test accuracy:', score[1]) # 精确度: 0.9767
 
test_loss,test_acc = model.evaluate(x=x_test,y=y_test)
print("Test Accuracy %.2f"%test_acc) # 精确度: 0.9

# 五、保存模型
 
model.save('./final_DNN_mode1.h5') # 保存DNN模型

# 六、加载保存的模型
new_model = load_model('./final_DNN_mode1.h5')
new_model.summary()

# 七、CNN 模型测试 -----------------------------------------------------------------------------------------------------

# 将数据扩充维度，以适应CNN模型
X_train=x_train.reshape(60000,28,28,1)
X_test=x_test.reshape(10000,28,28,1)

# 定义卷积神经网络模型
model=keras.Sequential([ # 创建网络序列
 layers.Conv2D(filters=32,kernel_size = 5,strides = (1,1), padding ='same',activation = tf.nn.relu,input_shape = (28,28,1)),
 # 添加第一层卷积层和池化层
 layers.MaxPool2D(pool_size=(2,2),strides = (2,2),padding = 'valid'),
 # 添加第二层卷积层和泄化层
 layers.Conv2D(filters=64, kernel_size = 3, strides=(1, 1),padding='same', activation = tf.nn.relu),
 layers.MaxPool2D(pool_size=(2,2),strides = (2,2),padding = 'valid'),
 # 添加dropout层 以减少过拟合
 layers.Dropout(0.25), # 随机丢弃神经元的比例 
 layers.Flatten(),
 # 添加两层全连接层
 layers.Dense(units=128,activation = tf.nn.relu),
 layers.Dropout(0.5),
 layers.Dense(units=10,activation = tf.nn.softmax)
])

# 编译并训练模型
Optimizer = optimizers.Adam(0.001)
model.compile(Optimizer,loss="categorical_crossentropy",metrics=['accuracy'])
model.fit(x=X_train,y=y_train,epochs=5,batch_size=128) # 轮次为5

# 保存CNN模型
model.save('./final_CNN_model.h5') 
# 加载保存的模型
new_model = load_model('./final_CNN_model.h5')

# 八、测试数据进行可视化测试
 
# @matplotlib.inline
def res_Visual(n):
 # 参阅 https://blog.csdn.net/yiyihuazi/article/details/122323349
 # keras 2.6删除了predict_classes()函数
 # final_opt_a=new_model.predict_classes(X_test[0:n]) # 通过模型预测测试集
 # 用下面的语句代替
 predicts = new_model.predict(X_test[0:n])
 final_opt_a = np.argmax(predicts, axis=1)
 
 fig, ax = plt.subplots(nrows=int(n/5), ncols=5)
 ax = ax.flatten()
 print('前{}张图片预测结果为:'.format(n))
 for i in range(n): 
 print(final_opt_a[i],end='.')
 if int((i+1)%5)==0:
 print('\t')
 
 # 图片可视化展示
 img = X_test[i].reshape((28,28)) # 读取每行数据，格式为Ndarry
 plt.axis("off")
 ax[i].imshow(img,cmap='Greys',interpolation='nearest') # 可视化
 ax[i].axis("off")
 print('测试集前{}张图片为:'.format(n))
 
 
res_Visual(20) 
</code></pre>
</div>
keras要降低版本到2.6.0，否则出错，<a href="https://links.jianshu.com/go?to=https%3A%2F%2Fblog.csdn.net%2Fqq_43440490%2Farticle%2Fdetails%2F125458488" target="_blank">参阅资料</a>。
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-jsx"><code class=" language-jsx">ImportError: cannot import name 'dtensor' from 'tensorflow.compat.v2.experimental' 
</code></pre>
</div>
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-ruby"><code class=" language-ruby">
(gpu) root@VM-0-14-ubuntu:~# conda list keras
# packages in environment at /usr/local/anaconda3/envs/gpu:
#
# Name Version Build Channel
keras 2.10.0 pypi_0 pypi
keras-preprocessing 1.1.2 pypi_0 pypi
(gpu) root@VM-0-14-ubuntu:~# pip install keras==2.6
</code></pre>
</div>
  测试程序前面DNN全连接神经网络的部分通过了，后面使用cuDNN的CNN卷积神经网络部分没有通过，cuDNN8.5的版本可能过高，<a href="https://links.jianshu.com/go?to=https%3A%2F%2Fgithub.com%2Ftensorflow%2Ftensorflow%2Fissues%2F45044" target="_blank">参阅资料</a>。需要降回经过测试确认的8.1版。报错：
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-css"><code class=" language-css">OP_REQUIRES failed at conv_ops.cc:1276 : Not found: No algorithm worked!
</code></pre>
</div>
7、降低cuDNN版本到8.1。笔记本上下载并用SecureFX通过SSH传到服务器上，拷贝并替换cuDNN8.5的文件。
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-bash"><code class=" language-bash"># tar -xvf cudnn-11.2-linux-x64-v8.1.1.33.tgz
# cd cuda
# cp -f lib64/* /usr/local/cuda/lib64/
# cp -f include/* /usr/local/cuda/include/
# chmod a+r /usr/local/cuda/lib64/*
# chmod a+r /usr/local/cuda/include/*
</code></pre>
</div>
  在全局环境变量中加入下面的设置，否则跑CNN测试时可能会报错说申请的内存过大，导致算法运行失败：
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-bash"><code class=" language-bash"># vi /etc/profile
</code></pre>
</div>
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-undefined"><code class=" language-undefined">TF_GPU_ALLOCATOR=cuda_malloc_async
</code></pre>
</div>
  更新动态链接库的Cache，否则链接不对，重启系统:
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-bash"><code class=" language-bash"># ldconfig -X
# reboot now
</code></pre>
</div>
  用ubuntu用户登录，激活"gpu"环境并启动Jupyter Notebook：
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-ruby"><code class=" language-ruby">$ conda activate gpu
$ jupyter notebook &amp;
</code></pre>
</div>
8、重新运行刚才的notebook测试GPU环境的安装，通过。
<div class="image-package">
<div class="image-container" style="max-width: 700px; max-height: 376px;">
<div class="image-container-fill" style="padding-bottom: 53.65%;"></div>
<div class="image-view" data-width="1920" data-height="1030"><img data-original-src="//upload-images.jianshu.io/upload_images/28576403-153f298830bd13ca.PNG" data-original-width="1920" data-original-height="1030" data-original-format="image/png" data-original-filesize="110246" data-image-index="2" style="cursor: zoom-in;" class="image-loading"></div>
</div>
<div class="image-caption">1、加载tensorflow识别手写数字例子数据集</div>
</div>
<div class="image-package">
<div class="image-container" style="max-width: 700px; max-height: 376px;">
<div class="image-container-fill" style="padding-bottom: 53.65%;"></div>
<div class="image-view" data-width="1920" data-height="1030"><img data-original-src="//upload-images.jianshu.io/upload_images/28576403-a83eab0eb5a33c2a.PNG" data-original-width="1920" data-original-height="1030" data-original-format="image/png" data-original-filesize="186918" data-image-index="3" style="cursor: zoom-in;" class="image-loading"></div>
</div>
<div class="image-caption">2、构建并编译DNN神经网络</div>
</div>
<div class="image-package">
<div class="image-container" style="max-width: 700px; max-height: 376px;">
<div class="image-container-fill" style="padding-bottom: 53.65%;"></div>
<div class="image-view" data-width="1920" data-height="1030"><img data-original-src="//upload-images.jianshu.io/upload_images/28576403-46e421ea1e6c88fe.PNG" data-original-width="1920" data-original-height="1030" data-original-format="image/png" data-original-filesize="124233" data-image-index="4" style="cursor: zoom-in;" class="image-loading"></div>
</div>
<div class="image-caption">3、训练网络</div>
</div>
<div class="image-package">
<div class="image-container" style="max-width: 700px; max-height: 376px;">
<div class="image-container-fill" style="padding-bottom: 53.65%;"></div>
<div class="image-view" data-width="1920" data-height="1030"><img data-original-src="//upload-images.jianshu.io/upload_images/28576403-bf6df69caaac3ad3.PNG" data-original-width="1920" data-original-height="1030" data-original-format="image/png" data-original-filesize="112141" data-image-index="5" style="cursor: zoom-in;" class="image-loading"></div>
</div>
<div class="image-caption">4、测试模型</div>
</div>
<div class="image-package">
<div class="image-container" style="max-width: 700px; max-height: 376px;">
<div class="image-container-fill" style="padding-bottom: 53.65%;"></div>
<div class="image-view" data-width="1920" data-height="1030"><img data-original-src="//upload-images.jianshu.io/upload_images/28576403-cfc5645f409210b2.PNG" data-original-width="1920" data-original-height="1030" data-original-format="image/png" data-original-filesize="137679" data-image-index="6" style="cursor: zoom-in;" class="image-loading"></div>
</div>
<div class="image-caption">5、CNN 模型测试</div>
</div>
<div class="image-package">
<div class="image-container" style="max-width: 700px; max-height: 376px;">
<div class="image-container-fill" style="padding-bottom: 53.65%;"></div>
<div class="image-view" data-width="1920" data-height="1030"><img data-original-src="//upload-images.jianshu.io/upload_images/28576403-51725c7e1844a943.PNG" data-original-width="1920" data-original-height="1030" data-original-format="image/png" data-original-filesize="125099" data-image-index="7" style="cursor: zoom-in;" class="image-loading"></div>
</div>
<div class="image-caption">6、测试数据进行可视化</div>
</div>
五、安装Pytorch与HanLP
  我在同一个虚拟环境"gpu"中安装Tensorflow、Pytorch与HanLP，因为要跑HanLP2.1，它同时支持两个后端。
1、安装Pytorch。
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-bash"><code class=" language-bash">(gpu) root@VM-0-14-ubuntu:~# conda install pytorch==1.11.0 torchvision==0.12.0 torchaudio==0.11.0 cudatoolkit=11.3 -c pytorch
Collecting package metadata (current_repodata.json): done
Solving environment: failed with initial frozen solve. Retrying with flexible solve.
Collecting package metadata (repodata.json): done
Solving environment: done

## Package Plan ##

environment location: /usr/local/anaconda3/envs/gpu

added / updated specs:
    - cudatoolkit=11.3
    - pytorch==1.11.0
    - torchaudio==0.11.0
    - torchvision==0.12.0

The following packages will be downloaded:

package                    |            build
    ---------------------------|-----------------
    cudatoolkit-11.3.1         |       h2bc3f7f_2       549.3 MB
......
  torchvision        pytorch/linux-64::torchvision-0.12.0-py39_cu113 None

Proceed ([y]/n)? y

Downloading and Extracting Packages
lame-3.100           | 323 KB    | ################################################################################################################################################## | 100% 
......
######################################################################## | 100% 
Preparing transaction: done
Verifying transaction: done
Executing transaction: | By downloading and using the CUDA Toolkit conda packages, you accept the terms and conditions of the CUDA End User License Agreement (EULA): https://docs.nvidia.com/cuda/eula/index.html

done
Retrieving notices: ...working... done
</code></pre>
</div>
2、安装HanLP。
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-csharp"><code class=" language-csharp">(gpu) root@VM-0-14-ubuntu:~# pip install hanlp
Looking in indexes: http://mirrors.tencentyun.com/pypi/simple
Collecting hanlp
......
Successfully built hanlp-common hanlp-trie hanlp-downloader phrasetree
Installing collected packages: toposort, tokenizers, phrasetree, tqdm, regex, pyyaml, pynvml, hanlp-common, filelock, huggingface-hub, hanlp-trie, hanlp-downloader, transformers, hanlp
Successfully installed filelock-3.8.0 hanlp-2.1.0b42 hanlp-common-0.0.18 hanlp-downloader-0.0.25 hanlp-trie-0.0.5 huggingface-hub-0.10.1 phrasetree-0.0.8 pynvml-11.4.1 pyyaml-6.0 regex-2022.9.13 tokenizers-0.11.6 toposort-1.5 tqdm-4.64.1 transformers-4.23.1
WARNING: Running pip as the 'root' user can result in broken permissions and conflicting behaviour with the system package manager. It is recommended to use a virtual environment instead: https://pip.pypa.io/warnings/venv
</code></pre>
</div>
安装fasttext，这是HanLP一些Tensorflow预训练模型要用的：
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-csharp"><code class=" language-csharp">(gpu) root@VM-0-14-ubuntu:~# pip install fasttext
Looking in indexes: http://mirrors.tencentyun.com/pypi/simple
Collecting fasttext
 Downloading http://mirrors.tencentyun.com/pypi/packages/f8/85/e2b368ab6d3528827b147fdb814f8189acc981a4bc2f99ab894650e05c40/fasttext-0.9.2.tar.gz (68 kB)
 ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 68.8/68.8 kB 332.3 kB/s eta 0:00:00
 Preparing metadata (setup.py) ... done
Collecting pybind11&gt;=2.2
 Using cached http://mirrors.tencentyun.com/pypi/packages/1d/53/e6b27f3596278f9dd1d28ef1ddb344fd0cd5db98ef2179d69a2044e11897/pybind11-2.10.1-py3-none-any.whl (216 kB)
Requirement already satisfied: setuptools&gt;=0.7.0 in /usr/local/anaconda3/envs/gpu/lib/python3.9/site-packages (from fasttext) (65.5.0)
Requirement already satisfied: numpy in /usr/local/anaconda3/envs/gpu/lib/python3.9/site-packages (from fasttext) (1.23.3)
Building wheels for collected packages: fasttext
 Building wheel for fasttext (setup.py) ... done
 Created wheel for fasttext: filename=fasttext-0.9.2-cp39-cp39-linux_x86_64.whl size=299146 sha256=4dee6f6dc5fb53404fb5cbb69c2cc3a2faef7f3af0500567ad49dc01f26d89d7
 Stored in directory: /root/.cache/pip/wheels/ca/08/ee/d0dd871c6c089c4c3971722067bd577f8827c9b4d5d6f2477a
Successfully built fasttext
Installing collected packages: pybind11, fasttext
</code></pre>
</div>
3、测试PyTorch及HanLP。
  先简单测试下，后面会继续测试。
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-swift"><code class=" language-swift">import torch

print(torch.__version__)
print(torch.cuda.is_available())
</code></pre>
</div>
<div class="image-package">
<div class="image-container" style="max-width: 700px; max-height: 376px;">
<div class="image-container-fill" style="padding-bottom: 53.65%;"></div>
<div class="image-view" data-width="1920" data-height="1030"><img data-original-src="//upload-images.jianshu.io/upload_images/28576403-3ee8f3f284424de8.PNG" data-original-width="1920" data-original-height="1030" data-original-format="image/png" data-original-filesize="82396" data-image-index="8" style="cursor: zoom-in;" class="image-loading"></div>
</div>
<div class="image-caption">pytorch识别出GPU</div>
</div>
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-go"><code class=" language-go"># 先运行Tensorflow模型再运行PyTorch模型就成功，如果前面先运行过PyTorch模型，这里就会失败。
import hanlp
tokenizer = hanlp.load(hanlp.pretrained.tok.LARGE_ALBERT_BASE)
text = 'NLP统计模型没有加规则，聪明人知道自己加。英文、数字、自定义词典统统都是规则。'
print(tokenizer(text))

# 后面的测试不受运行顺序的影响

import hanlp
HanLP = hanlp.pipeline() \
 .append(hanlp.utils.rules.split_sentence, output_key='sentences') \
 .append(hanlp.load('FINE_ELECTRA_SMALL_ZH'), output_key='tok') \
 .append(hanlp.load('CTB9_POS_ELECTRA_SMALL'), output_key='pos') \
 .append(hanlp.load('MSRA_NER_ELECTRA_SMALL_ZH'), output_key='ner', input_key='tok') \
 .append(hanlp.load('CTB9_DEP_ELECTRA_SMALL', conll=0), output_key='dep', input_key='tok')\
 .append(hanlp.load('CTB9_CON_ELECTRA_SMALL'), output_key='con', input_key='tok')
HanLP('2021年HanLPv2.1为生产环境带来次世代最先进的多语种NLP技术。阿婆主来到北京立方庭参观自然语义科技公司。')

HanLP('2021年HanLPv2.1为生产环境带来次世代最先进的多语种NLP技术。').pretty_print()

import hanlp

tok = hanlp.load(hanlp.pretrained.tok.COARSE_ELECTRA_SMALL_ZH)
tok(['商品和服务。', '阿婆主来到北京立方庭参观自然语义科技公司。'])

tok_fine = hanlp.load(hanlp.pretrained.tok.FINE_ELECTRA_SMALL_ZH)
tok_fine('阿婆主来到北京立方庭参观自然语义科技公司')

pos = hanlp.load(hanlp.pretrained.pos.CTB9_POS_ELECTRA_SMALL)
pos(["我", "的", "希望", "是", "希望", "张晚霞", "的", "背影", "被", "晚霞", "映红", "。"])
</code></pre>
</div>
<div class="image-package">
<div class="image-container" style="max-width: 700px; max-height: 376px;">
<div class="image-container-fill" style="padding-bottom: 53.65%;"></div>
<div class="image-view" data-width="1920" data-height="1030"><img data-original-src="//upload-images.jianshu.io/upload_images/28576403-f83d01b114c0fb84.PNG" data-original-width="1920" data-original-height="1030" data-original-format="image/png" data-original-filesize="129921" data-image-index="9" style="cursor: zoom-in;" class="image-loading"></div>
</div>
<div class="image-caption">分词与词性标注</div>
</div>
<div class="image-package">
<div class="image-container" style="max-width: 700px; max-height: 376px;">
<div class="image-container-fill" style="padding-bottom: 53.65%;"></div>
<div class="image-view" data-width="1920" data-height="1030"><img data-original-src="//upload-images.jianshu.io/upload_images/28576403-93308dfea49a1938.PNG" data-original-width="1920" data-original-height="1030" data-original-format="image/png" data-original-filesize="119564" data-image-index="10" style="cursor: zoom-in;" class="image-loading"></div>
</div>
<div class="image-caption">管道操作</div>
</div>
<div class="image-package">
<div class="image-container" style="max-width: 700px; max-height: 376px;">
<div class="image-container-fill" style="padding-bottom: 53.65%;"></div>
<div class="image-view" data-width="1920" data-height="1030"><img data-original-src="//upload-images.jianshu.io/upload_images/28576403-d4799d7861662ebf.PNG" data-original-width="1920" data-original-height="1030" data-original-format="image/png" data-original-filesize="128877" data-image-index="11" style="cursor: zoom-in;" class="image-loading"></div>
</div>
<div class="image-caption">打印语法树</div>
</div>
<div class="image-package">
<div class="image-container" style="max-width: 700px; max-height: 376px;">
<div class="image-container-fill" style="padding-bottom: 53.65%;"></div>
<div class="image-view" data-width="1920" data-height="1030"><img data-original-src="//upload-images.jianshu.io/upload_images/28576403-8c9d89a3337aa19c.PNG" data-original-width="1920" data-original-height="1030" data-original-format="image/png" data-original-filesize="127723" data-image-index="12" style="cursor: zoom-in;" class="image-loading"></div>
</div>
<div class="image-caption">各种预训练模型模型分词</div>
</div>
六、安装配置JupyterHub
  Linux GPU虚拟主机作为科研、开发、测试或生产环境，多用户是很自然的，Jupyter Notebook是单用户的，<a href="https://links.jianshu.com/go?to=https%3A%2F%2Fjupyterhub.readthedocs.io%2Fen%2Fstable%2Findex.html" target="_blank">JupyterHub</a>则提供了一层多用户的代理，让大家可以通过它登录系统，使用各自的Jupyter Notebook或Jupyter Lab，后者是前者的下一代版本。

<div class="image-package">
<div class="image-container" style="max-width: 700px; max-height: 525px;">
<div class="image-container-fill" style="padding-bottom: 75.0%;"></div>
<div class="image-view" data-width="1024" data-height="768"><img data-original-src="//upload-images.jianshu.io/upload_images/28576403-9b941db52827e473.png" data-original-width="1024" data-original-height="768" data-original-format="image/png" data-original-filesize="126076" data-image-index="13" style="cursor: zoom-in;" class="image-loading"></div>
</div>
<div class="image-caption">Jupyterhub统一代理各用户的Jupyterlab，从而实现多用户服务</div>
</div>
 
  根据<a href="https://links.jianshu.com/go?to=https%3A%2F%2Fstackoverflow.com%2Fquestions%2F47605853%2F500-error-in-jupyterhub" target="_blank">该帖子</a>，如果曾经运行过Jupyter Notebook，那么它在$HOME/.jupyter下的配置文件会与Jupyterhub要启动的用户Jupyter Lab或Jupyter Notebook Server冲突，导致服务进程不能启动，代理转发失败，这是个BUG？所以如果曾经运行过Jupyter Notebook，像前面那样，要先删除那个目录。这个问题搞了两天，几乎要崩溃，还是stackoverflow给力。
  <a href="https://www.jianshu.com/p/3585d51eb577" target="_blank">参阅资料1</a>，<a href="https://links.jianshu.com/go?to=https%3A%2F%2Fcloud.tencent.com%2Fdeveloper%2Farticle%2F1816547" target="_blank">参阅资料2</a>，<a href="https://links.jianshu.com/go?to=https%3A%2F%2Fzhuanlan.zhihu.com%2Fp%2F363793325" target="_blank">参阅资料3</a>，<a href="https://links.jianshu.com/go?to=https%3A%2F%2Fjupyterhub.readthedocs.io%2Fen%2Fstable%2Fquickstart.html%23installation" target="_blank">参阅资料4</a>。
1、安装并升级node.js与npm。
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-bash"><code class=" language-bash"># #从软件源获取最新软件列表，更新系统软件
# apt-get update 
# apt-get upgrade
# #安装依赖
# apt install -y npm nodejs
</code></pre>
</div>
升级node.js，不要安装最新的18版，兼容性有问题，会报错，<a href="https://links.jianshu.com/go?to=https%3A%2F%2Fstackoverflow.com%2Fquestions%2F72921215%2Fgetting-glibc-2-28-not-found" target="_blank">参阅资料</a>，JupyterHub要求版本10以上，而Ubuntu18安装的是版本8。
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-bash"><code class=" language-bash">##----- 先清除 npm cache
# npm cache clean -f 
##----- 安装 n 模块
# npm install -g n
</code></pre>
</div>
升级node.js：
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-tsx"><code class=" language-tsx">root@VM-0-14-ubuntu:~# n 16.18.0 # 指定版本16.18.0
 installing : node-v16.18.0
 mkdir : /usr/local/n/versions/node/16.18.0
 fetch : https://nodejs.org/dist/v16.18.0/node-v16.18.0-linux-x64.tar.xz
 copying : node/16.18.0
 installed : v16.18.0 (with npm 8.19.2)

Note: the node command changed location and the old location may be remembered in your current shell.
 old : /usr/bin/node
 new : /usr/local/bin/node
If "node --version" shows the old version then start a new shell, or reset the location hash with:
hash -r (for bash, zsh, ash, dash, and ksh)
rehash (for csh and tcsh)

root@VM-0-14-ubuntu:~# hash -r
root@VM-0-14-ubuntu:~# node -v
v16.18.0
root@VM-0-14-ubuntu:~# npm -v
8.19.2
</code></pre>
</div>
2、安装configurable-http-proxy。
可以用npm装：
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-undefined"><code class=" language-undefined">npm install -g configurable-http-proxy
</code></pre>
</div>
不过推荐用conda装，会把其它依赖包一起装上，它也会安装一个node.js版本11，也可以用，注意要切换并安装到相应的虚拟环境中，这里是"gpu"。
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-ruby"><code class=" language-ruby">(gpu) root@VM-0-14-ubuntu:~# conda install configurable-http-proxy
(gpu) root@VM-0-14-ubuntu:~# conda list configurable-http-proxy
# packages in environment at /usr/local/anaconda3/envs/gpu:
#
# Name Version Build Channel
configurable-http-proxy 4.0.1 node6_0 
(gpu) root@VM-0-14-ubuntu:~# configurable-http-proxy -V
4.0.1
(gpu) root@VM-0-14-ubuntu:~# 
</code></pre>
</div>
3、在虚拟环境中安装JupyterHub等。
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-ruby"><code class=" language-ruby">(gpu) root@VM-0-14-ubuntu:~# conda install jupyter jupyterlab jupyterhub
(gpu) root@VM-0-14-ubuntu:~# conda list jupyter
# packages in environment at /usr/local/anaconda3/envs/gpu:
#
# Name Version Build Channel
jupyter 1.0.0 py39h06a4308_8 
jupyter_client 7.3.5 py39h06a4308_0 
jupyter_console 6.4.3 pyhd3eb1b0_0 
jupyter_core 4.11.1 py39h06a4308_0 
jupyter_server 1.18.1 py39h06a4308_0 
jupyter_telemetry 0.1.0 py_0 
jupyterhub 2.0.0 pyhd3eb1b0_0 
jupyterlab 3.4.4 py39h06a4308_0 
jupyterlab_pygments 0.1.2 py_0 
jupyterlab_server 2.15.2 py39h06a4308_0 
jupyterlab_widgets 1.0.0 pyhd3eb1b0_1 
</code></pre>
</div>
4、配置JupyterHub。
新建目录/etc/jupyterhub，在该目录下新建一个配置文件，编辑文件。
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-ruby"><code class=" language-ruby">(gpu) root@VM-0-14-ubuntu:~# mkdir /etc/jupyterhub
(gpu) root@VM-0-14-ubuntu:~# cd /etc/jupyterhub
(gpu) root@VM-0-14-ubuntu:/etc/jupyterhub# jupyterhub --generate-config
Writing default config to: jupyterhub_config.py
(gpu) root@VM-0-14-ubuntu:/etc/jupyterhub# vi jupyterhub_config.py
</code></pre>
</div>
内容如下：
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-bash"><code class=" language-bash"># Added by Jean 2022/10/31
c.Authenticator.whitelist = {'ubuntu'} # 允许使用Jupyterhub的用户列表，逗号分隔。
c.Authenticator.admin_users = {'ubuntu'} #Jupyterhub的管理员用户列表
c.Spawner.notebook_dir = '/home/{username}' #浏览器登录后进入用户的主目录
c.Spawner.default_url = '/lab' # 使用Jupyterlab而不是Notebook
c.JupyterHub.extra_log_file = '/var/log/jupyterhub.log'
</code></pre>
</div>
5、用root用户后台启动JupyterHub。
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-ruby"><code class=" language-ruby">(gpu) root@VM-0-14-ubuntu:/etc/jupyterhub# jupyterhub -f /etc/jupyterhub/jupyterhub_config.py &amp;
</code></pre>
</div>
6、在浏览器中访问，输入的是Linux系统中已有的用户名，网址是<a href="https://links.jianshu.com/go?to=http%3A%2F%2Fip%3A8000" target="_blank">http://ip:8000</a>，后面再配SSL加密。 

<div class="image-package">
<div class="image-container" style="max-width: 700px; max-height: 376px;">
<div class="image-container-fill" style="padding-bottom: 53.65%;"></div>
<div class="image-view" data-width="1920" data-height="1030"><img data-original-src="//upload-images.jianshu.io/upload_images/28576403-e9a1ef224701a9b6.PNG" data-original-width="1920" data-original-height="1030" data-original-format="image/png" data-original-filesize="110567" data-image-index="14" style="cursor: zoom-in;" class="image-loading"></div>
</div>
<div class="image-caption">JupyterHub中运行Jupyter Lab</div>
</div>
 
JupyterHub里可以打开终端窗口，执行各种操作，用户的身份就是登录的用户。如果SSH端口被屏蔽，这样就可以通过HTTP端口建立隧道。执行su命令就可以root。
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-bash"><code class=" language-bash">(base) ubuntu@VM-0-14-ubuntu:~$ su --help
Usage: su [options] [LOGIN]

Options:
  -c, --command COMMAND         pass COMMAND to the invoked shell
  -h, --help                    display this help message and exit
  -, -l, --login                make the shell a login shell
  -m, -p,
  --preserve-environment        do not reset environment variables, and
                                keep the same shell
  -s, --shell SHELL             use SHELL instead of the default in passwd

(base) ubuntu@VM-0-14-ubuntu:~$ su --preserve-environment
Password: 
(base) root@VM-0-14-ubuntu:~# 
</code></pre>
</div>
<div class="image-package">
<div class="image-container" style="max-width: 700px; max-height: 376px;">
<div class="image-container-fill" style="padding-bottom: 53.65%;"></div>
<div class="image-view" data-width="1920" data-height="1030"><img data-original-src="//upload-images.jianshu.io/upload_images/28576403-a9340f980403d025.png" data-original-width="1920" data-original-height="1030" data-original-format="image/png" data-original-filesize="109262" data-image-index="15" style="cursor: zoom-in;" class="image-loading"></div>
</div>
<div class="image-caption">JupyterHub中打开终端窗口</div>
</div>
7、配置SSL加密。
  这是配好后SSL加密连接登录的截图，可以打开网址前面的锁图标看证书链的内容，前面的截图可见，如果是非加密连接，网址前面显示的是“不安全”提示。此处自签的数字证书是签给IP，因为这个虚拟主机还没有申请域名。
<div class="image-package">
<div class="image-container" style="max-width: 700px; max-height: 394px;">
<div class="image-container-fill" style="padding-bottom: 56.25%;"></div>
<div class="image-view" data-width="1920" data-height="1080"><img data-original-src="//upload-images.jianshu.io/upload_images/28576403-b430487c92f55e4d.PNG" data-original-width="1920" data-original-height="1080" data-original-format="image/png" data-original-filesize="210245" data-image-index="16" style="cursor: zoom-in;" class="image-loading"></div>
</div>
<div class="image-caption">用自签证书给JupyterHub建立SSL加密通道</div>
</div>
 
1）先讲讲JupyterHub配置。在配置文件中增加两行指出使用的服务器密钥文件和证书文件即可，后面再讲用openssl自建CA及签发该数字证书。因为是root用户，server.key没有指定访问密码。

<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-bash"><code class=" language-bash"># Added by Jean for SSL 2022/03/19
c.JupyterHub.ssl_key = '/root/cert/server.key'
c.JupyterHub.ssl_cert = '/root/cert/server.crt'
</code></pre>
</div>
重启JupyterHub后，把自建CA的根证书拷出并导入浏览器（后面讲）,用<a href="https://links.jianshu.com/go?to=https%3A%2F%2Fip%3A8000" target="_blank">https://ip:8000</a>访问即可，如上图所示。
2）自建CA签发自签服务器证书。
<a href="https://links.jianshu.com/go?to=https%3A%2F%2Fwww.cnblogs.com%2Fpcxie%2Fp%2F12900666.html" target="_blank">参阅资料</a>。
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-kotlin"><code class=" language-kotlin">(gpu) root@VM-0-14-ubuntu:~# cd /root
(gpu) root@VM-0-14-ubuntu:~# mkdir cert
(gpu) root@VM-0-14-ubuntu:~# cd cert
(gpu) root@VM-0-14-ubuntu:~/cert# mkdir demoCA &amp;&amp; cd demoCA
(gpu) root@VM-0-14-ubuntu:~/cert/demoCA# mkdir private newcerts
(gpu) root@VM-0-14-ubuntu:~/cert/demoCA# touch index.txt
(gpu) root@VM-0-14-ubuntu:~/cert/demoCA# echo '01' &gt; serial
(gpu) root@VM-0-14-ubuntu:~/cert/demoCA# cd private
(gpu) root@VM-0-14-ubuntu:~/cert/demoCA/private# openssl genrsa -out cakey.pem 2048
Generating RSA private key, 2048 bit long modulus (2 primes)
...............................................................................+++++
....................+++++
e is 65537 (0x010001)
(gpu) root@VM-0-14-ubuntu:~/cert/demoCA/private# openssl req -sha256 -new -x509 -days 3650 -key cakey.pem -out cacert.pem \
&gt; -subj "/C=CN/ST=GD/L=ZhuHai/O=Jean/OU=Study/CN=RootCA"
(gpu) root@VM-0-14-ubuntu:~/cert/demoCA/private# ls
cacert.pem cakey.pem
(gpu) root@VM-0-14-ubuntu:~/cert/demoCA/private# cd .. &amp;&amp; mv ./private/cacert.pem ./
(gpu) root@VM-0-14-ubuntu:~/cert/demoCA# ls
cacert.pem index.txt newcerts private serial
</code></pre>
</div>
上面的命令执行了一系列的操作：
A、在root用户的HOME目录/root下新建了/root/cert目录。
B、然后在其下建立了自建CA的目录结构./demoCA，因为openssl默认的配置文件中，建在当前目录的./demoCA目录下。
C、然后产生了CA的密钥cakey.pem。
D、签发了CA的自签数字证书cacert.pem，然后移动到./demoCA目录下。后面自建CA签发服务器证书时会到那里找CA根证书，这是openssl默认的配置。
E、最后列出了demoCA的目录结构。
可以找出openssl默认的配置文件看一下，自建CA在当前目录的./demoCA目录下：
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-bash"><code class=" language-bash">(gpu) root@VM-0-14-ubuntu:~# find / -name openssl.cnf
/usr/lib/ssl/openssl.cnf
/usr/local/anaconda3/pkgs/openssl-1.1.1q-h7f8727e_0/ssl/openssl.cnf
/usr/local/anaconda3/ssl/openssl.cnf
/usr/local/anaconda3/envs/gpu/ssl/openssl.cnf
/usr/local/anaconda3/envs/hub/ssl/openssl.cnf
/etc/ssl/openssl.cnf
(gpu) root@VM-0-14-ubuntu:~# vi /usr/lib/ssl/openssl.cnf
</code></pre>
</div>
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-php"><code class=" language-php">####################################################################
[ ca ]
default_ca = CA_default # The default ca section

####################################################################
[ CA_default ]

dir = ./demoCA # Where everything is kept
certs = $dir/certs # Where the issued certs are kept
crl_dir = $dir/crl # Where the issued crl are kept
database = $dir/index.txt # database index file.
#unique_subject = no # Set to 'no' to allow creation of
 # several certs with same subject.
new_certs_dir = $dir/newcerts # default place for new certs.

certificate = $dir/cacert.pem # The CA certificate
serial = $dir/serial # The current serial number
crlnumber = $dir/crlnumber # the current crl number
 # must be commented out to leave a V1 CRL
crl = $dir/crl.pem # The current CRL
private_key = $dir/private/cakey.pem# The private key
RANDFILE = $dir/private/.rand # private random number file

x509_extensions = usr_cert # The extensions to add to the cert
</code></pre>
</div>
F、生成服务器证书的密钥与证书请求。
参考<a href="https://links.jianshu.com/go?to=https%3A%2F%2Fstackoverflow.com%2Fquestions%2F63893662%2Fcant-load-root-rnd-into-rng" target="_blank">帖子1</a>与<a href="https://links.jianshu.com/go?to=https%3A%2F%2Fgithub.com%2Fnode-opcua%2Fnode-opcua-pki%2Fissues%2F7" target="_blank">帖子2</a>，要先执行下面的命令产生/root/.rnd文件，否则产生服务器密钥的命令会出错。
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-csharp"><code class=" language-csharp">openssl rand -out /root/.rnd -hex 256
</code></pre>
</div>
  切换到./demoCA的父目录/root/cert，然后执行下面的命令产生服务器证书的密钥与证书请求，产生证书请求用配置文件/usr/lib/ssl/openssl.cnf，额外增加了认证的主体别名，Chrome浏览器使用主体别名来检查证书的主体别名与网址是否一致。因为用<a href="https://links.jianshu.com/go?to=https%3A%2F%2Fip" target="_blank">https://ip</a>访问，这里的主体别名为IP.1:106.52.33.185，表示是该证书认证的第一个IP，还可以有IP.2等等。如果是认证域名，可以是DNS.1 = <a href="https://links.jianshu.com/go?to=http%3A%2F%2Fjeanye.cn" target="_blank">jeanye.cn</a>等等，如此类推。产生证书请求文件server.csr。
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-kotlin"><code class=" language-kotlin">(gpu) root@VM-0-14-ubuntu:~/cert# openssl genrsa -out server.key 2048
(gpu) root@VM-0-14-ubuntu:~/cert# openssl req -new \
&gt; -sha256 \
&gt; -key server.key \
&gt; -subj "/C=CN/ST=GD/L=ZhuHai/O=Jean/OU=Study/CN=106.52.33.185" \
&gt; -reqexts SAN \
&gt; -config &lt;(cat /usr/lib/ssl/openssl.cnf \
&gt; &lt;(printf "[SAN]\nsubjectAltName=IP.1:106.52.33.185")) \
&gt; -out server.csr
</code></pre>
</div>
G、签署服务器证书。
  openssl会在默认子目录./demoCA中找到cakey.pem与cacert.pem，按照证书请求文件server.csr的请求，使用配置文件/usr/lib/ssl/openssl.cnf，以及与请求一样的证书扩展（主体别名）签署证书，输出成server.crt。
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-csharp"><code class=" language-csharp">(gpu) root@VM-0-14-ubuntu:~/cert# openssl ca -in server.csr \
&gt; -md sha256 \
&gt; -extensions SAN \
&gt; -config &lt;(cat /usr/lib/ssl/openssl.cnf \
&gt; &lt;(printf "[SAN]\nsubjectAltName=IP.1:106.52.33.185")) \
&gt; -out server.crt
Using configuration from /dev/fd/63
Check that the request matches the signature
Signature ok
Certificate Details:
 Serial Number: 1 (0x1)
 Validity
 Not Before: Nov 2 09:47:58 2022 GMT
 Not After : Nov 2 09:47:58 2023 GMT
 Subject:
 countryName = CN
 stateOrProvinceName = GD
 organizationName = Jean
 organizationalUnitName = Study
 commonName = 106.52.33.185
 X509v3 extensions:
 X509v3 Subject Alternative Name: 
 IP Address:106.52.33.185
Certificate is to be certified until Nov 2 09:47:58 2023 GMT (365 days)
Sign the certificate? [y/n]:y

1 out of 1 certificate requests certified, commit? [y/n]y
Write out database with 1 new entries
Data Base Updated
(gpu) root@VM-0-14-ubuntu:~/cert# ls
demoCA server.crt server.csr server.key
</code></pre>
</div>
H、自建CA根证书导入浏览器。
  把自建CA的根证书/root/cert/demoCA/cacert.pem下载到客户端（比如Win10），在浏览器（比如Chrome）中导入到受信任根证书颁证机构中。
Google浏览器:
  设置-&gt;隐私设置和安全性-&gt;安全-&gt;高级-&gt;管理证书-&gt;受信任根证书颁证机构-&gt;导入-&gt;下一步-&gt;浏览-&gt;所有文件(*.*)
<div class="image-package">
<div class="image-container" style="max-width: 700px; max-height: 394px;">
<div class="image-container-fill" style="padding-bottom: 56.25%;"></div>
<div class="image-view" data-width="1920" data-height="1080"><img data-original-src="//upload-images.jianshu.io/upload_images/28576403-34bdb975893b50ea.PNG" data-original-width="1920" data-original-height="1080" data-original-format="image/png" data-original-filesize="264024" data-image-index="17" style="cursor: zoom-in;" class="image-loading"></div>
</div>
<div class="image-caption">导入自建CA根证书到浏览器受信任根证书颁发机构列表</div>
</div>
 
I、浏览器中输入网址https://106.52.33.185:8000访问，输入用户名/密码登录。

<div class="image-package">
<div class="image-container" style="max-width: 700px; max-height: 394px;">
<div class="image-container-fill" style="padding-bottom: 56.25%;"></div>
<div class="image-view" data-width="1920" data-height="1080"><img data-original-src="//upload-images.jianshu.io/upload_images/28576403-dd0b3eaed1b5905c.png" data-original-width="1920" data-original-height="1080" data-original-format="image/png" data-original-filesize="153563" data-image-index="18" style="cursor: zoom-in;" class="image-loading"></div>
</div>
<div class="image-caption">输入用户名/密码登录，启动自己的Jupyter Lab实例</div>
</div>
8、配置JupyterHub为开机自启动服务。
1）建立服务配置文件。
先看看conda虚拟环境"gpu"的PATH设置：
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-tsx"><code class=" language-tsx">(gpu) root@VM-0-14-ubuntu:~# echo $PATH
/usr/local/anaconda3/envs/gpu/bin:/usr/local/anaconda3/condabin:/usr/local/cuda-11.2/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin:/usr/games:/usr/local/games:/snap/bin
(gpu) root@VM-0-14-ubuntu:~# 
</code></pre>
</div>
然后新建一个系统守护进程的配置文件：
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-ruby"><code class=" language-ruby">(gpu) root@VM-0-14-ubuntu:~# vi /etc/systemd/system/jupyterhub.service
</code></pre>
</div>
内容如下，几个要点。
A、以root运行。
B、设定PATH路径，因为开机启动进程没有登录的过程，不会执行/etc/profile等设置环境变量，把上面的PATH拷进去。
C、用全路径引用执行jupyterhub。
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-ruby"><code class=" language-ruby">[Unit]
Description=Jupyterhub service
After=syslog.target network.target

[Service]
User=root
Environment="PATH=/usr/local/anaconda3/envs/gpu/bin:/usr/local/anaconda3/condabin:/usr/local/cuda-11.2/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin"
ExecStart=/usr/local/anaconda3/envs/gpu/bin/jupyterhub -f /etc/jupyterhub/config.py

[Install]
WantedBy=multi-user.target
</code></pre>
</div>
然后让服务配置文件生效：
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-ruby"><code class=" language-ruby">(gpu) root@VM-0-14-ubuntu:~# systemctl enable jupyterhub.service
</code></pre>
</div>
然后可以用下面几个命令来管理服务：
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-bash"><code class=" language-bash"># systemctl status jupyterhub.service
# systemctl start jupyterhub.service
# systemctl stop jupyterhub.service
</code></pre>
</div>
用下面的命令来查看服务的日志：
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-ruby"><code class=" language-ruby">(gpu) root@VM-0-14-ubuntu:~# journalctl -u jupyterhub.service -f
</code></pre>
</div>
上面Jupyterhub的配置文件中，日志也另外输出到以下的文件：
<div class="_2Uzcx_">
<button class="VJbwyy" type="button" aria-label="复制代码"><svg viewbox="64 64 896 896" focusable="false" class="" data-icon="copy" width="1em" height="1em" fill="currentColor" aria-hidden="true"><path d="M832 64H296c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h496v688c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V96c0-17.7-14.3-32-32-32zM704 192H192c-17.7 0-32 14.3-32 32v530.7c0 8.5 3.4 16.6 9.4 22.6l173.3 173.3c2.2 2.2 4.7 4 7.4 5.5v1.9h4.2c3.5 1.3 7.2 2 11 2H704c17.7 0 32-14.3 32-32V224c0-17.7-14.3-32-32-32zM350 856.2L263.9 770H350v86.2zM664 888H414V746c0-22.1-17.9-40-40-40H232V264h432v624z"></path></svg></button><pre class="line-numbers language-swift"><code class=" language-swift">c.JupyterHub.extra_log_file = '/var/log/jupyterhub.log'
</code></pre>
</div>
所以也可以打开日志文件来看。
这样，每次服务器重启，Jupyterhub都会自动启动了。
本篇到此结束，Linux GPU虚拟主机与GPU、Python深度学习运行与开发环境相关的部分就配好了，Rstudio、Shiny等其它部分另起文章。
</article>

GPU Linux虚拟主机GN7型安装配置文档定稿

作者

sockstack

许可协议

CC BY 4.0

发布于

2023-11-14

修改于

2023-12-04

上一篇：【AI聊天丨 ChatGPT应用案例一】— 仅用30分钟，ChatGPT帮你完成专利交底书！下一篇：Llama2开源大模型的新篇章以及在阿里云的实践