GPU Linux虚拟主机GN7型安装配置文档定稿

sockstack / 95 / 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 referrerpolicy="no-referrer" 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 referrerpolicy="no-referrer" 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 referrerpolicy="no-referrer" 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 referrerpolicy="no-referrer" 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 referrerpolicy="no-referrer" 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 referrerpolicy="no-referrer" 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 referrerpolicy="no-referrer" 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 referrerpolicy="no-referrer" 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 referrerpolicy="no-referrer" 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 referrerpolicy="no-referrer" 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 referrerpolicy="no-referrer" 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 referrerpolicy="no-referrer" 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 referrerpolicy="no-referrer" 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 referrerpolicy="no-referrer" 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 referrerpolicy="no-referrer" 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 referrerpolicy="no-referrer" 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 referrerpolicy="no-referrer" 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 referrerpolicy="no-referrer" 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 referrerpolicy="no-referrer" 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

2024-07-25

上一篇：软件：常用 Linux 软件汇总，值得收藏下一篇：Llama2开源大模型的新篇章以及在阿里云的实践

尚未登录

GPU Linux虚拟主机GN7型安装配置文档定稿

文章分类

博客重构之路

Spring Boot简单入门

k8s 入门教程

MySQL 知识

NSQ 消息队列

ThinkPHP5 源码分析

使用 Docker 从零开始搭建私人代码仓库

日常开发汇总

标签列表

springboot

hyperf

swoole

webman

php

多线程

数据结构

docker

k8s

thinkphp

mysql

tailwindcss

flowbite

css

前端

GPU Linux虚拟主机GN7型安装配置文档 定稿

文章分类

博客重构之路

Spring Boot简单入门

k8s 入门教程

MySQL 知识

NSQ 消息队列

ThinkPHP5 源码分析

使用 Docker 从零开始搭建私人代码仓库

日常开发汇总

标签列表

springboot

hyperf

swoole

webman

php

多线程

数据结构

docker

k8s

thinkphp

mysql

tailwindcss

flowbite

css

前端

GPU Linux虚拟主机GN7型安装配置文档定稿