不会在网速上受苦，脑溢血玩家的最好选择。

常规版 数据集版 老司机版

0文件处理

0.1下载cookies并上传到share文件夹

cookie.txt

<aside> 💡 我会拿小号开cpu机一直开着不下线。1000豆用完大概要80天。

</aside>

0.2文件下载

（你可以再开个命令栏开始设置环境安装cuda并且一步一步往下了，理论上来说下的比装的快）

cd ~/share
wget --no-check-certificate --referer=https://jiutian.10086.cn --cookies=on --load-cookies=cookie.txt <https://jiutian.10086.cn/edu/ingress/cc4b3bc9-6862-4009-96c9-c59b50483903-dl-221116221939m68-pod/files/share/1.zip>
wget --no-check-certificate --referer=https://jiutian.10086.cn --cookies=on --load-cookies=cookie.txt <https://jiutian.10086.cn/edu/ingress/cc4b3bc9-6862-4009-96c9-c59b50483903-dl-221116221939m68-pod/files/share/2.zip>
wget --no-check-certificate --referer=https://jiutian.10086.cn --cookies=on --load-cookies=cookie.txt －-keep-session-cookies <https://jiutian.10086.cn/edu/ingress/cc4b3bc9-6862-4009-96c9-c59b50483903-dl-221116221939m68-pod/files/share/3.zip>
wget --no-check-certificate --referer=https://jiutian.10086.cn --cookies=on --load-cookies=cookie.txt －-keep-session-cookies <https://jiutian.10086.cn/edu/ingress/cc4b3bc9-6862-4009-96c9-c59b50483903-dl-221116221939m68-pod/files/share/4.zip>
wget --no-check-certificate --referer=https://jiutian.10086.cn --cookies=on --load-cookies=cookie.txt －-keep-session-cookies <https://jiutian.10086.cn/edu/ingress/cc4b3bc9-6862-4009-96c9-c59b50483903-dl-221116221939m68-pod/files/share/5.zip>
wget --no-check-certificate --referer=https://jiutian.10086.cn --cookies=on --load-cookies=cookie.txt －-keep-session-cookies <https://jiutian.10086.cn/edu/ingress/cc4b3bc9-6862-4009-96c9-c59b50483903-dl-221116221939m68-pod/files/share/6.zip>
wget --no-check-certificate --referer=https://jiutian.10086.cn --cookies=on --load-cookies=cookie.txt －-keep-session-cookies <https://jiutian.10086.cn/edu/ingress/cc4b3bc9-6862-4009-96c9-c59b50483903-dl-221116221939m68-pod/files/share/7.zip>
mv 1.zip cudatoolkit-11.0.3-h88f8997_10.tar.bz2
mv 2.zip cudnn-8.4.1.50-hed8a83a_0.tar.bz2
mv 3.zip nccl-2.14.3.1-h17a0586_0.tar.bz2
mv 4.zip tensorflow-base-2.8.1-cuda110py310h9e8cd52_0.tar.bz2
mv 5.zip torch-1.7.1+cu110-cp37-cp37m-linux_x86_64.whl
mv 6.zip torchvision-0.8.2+cu110-cp37-cp37m-linux_x86_64
mv 7.zip xformers-0.0.14.dev0-cp310-cp310-linux_x86_64.whl

网速大概是这样

Untitled

1.conda相关

1.1 conda添加镜像源

<aside> 💡 使用一个就行，检测不到需要的包时再添加其他。阿里的速度在10kb-10MB之间，经常抽风。清华稳定1M-5M。

</aside>

清华（推荐）

conda config --add channels <https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/free>
conda config --add channels <https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud/conda-forge>
conda config --add channels <https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud/bioconda>
conda config --set show_channel_urls yes

<aside> 💡 远离阿里源，不然你会变得不幸

</aside>

如果想删除并恢复默认的conda源

<aside> 💡 conda的超大段报错似乎没有影响，但不乱动应该也不会报错

</aside>