NV驱动下载安装/Download/index.aspx多卡的话，安装 NCCL /nccl/nccl-download安装conda，并配置通道vim ~/.condarc
show_channel_urls: truessl_verify: truechannels:- https://mirrors.tuna./anaconda/cloud/pytorch/- https://mirrors.tuna./anaconda/pkgs/main/- https://mirrors.tuna./anaconda/pkgs/free/- https://mirrors./anaconda/cloud/peterjc123/- https://mirrors./anaconda/cloud/pytorch/- https://mirrors./anaconda/cloud/menpo/- https://mirrors./anaconda/cloud/bioconda/- https://mirrors./anaconda/cloud/msys2/- https://mirrors./anaconda/cloud/conda-forge/- https://mirrors./anaconda/pkgs/main/- https://mirrors./anaconda/pkgs/free/- https://mirrors./anaconda/cloud/menpo/- https://mirrors./anaconda/cloud/bioconda/- https://mirrors./anaconda/cloud/msys2/- https://mirrors./anaconda/cloud/conda-forge/- https://mirrors./anaconda/pkgs/free/- https://mirrors./anaconda/pkgs/main/- https://mirrors.tuna./anaconda/cloud/peterjc123/- https://mirrors.tuna./anaconda/cloud/menpo/- https://mirrors.tuna./anaconda/cloud/bioconda/- https://mirrors.tuna./anaconda/cloud/msys2/- https://mirrors.tuna./anaconda/cloud/conda-forge/- https://mirrors.tuna./anaconda/cloud/conda-forge- defaults
配置 pip 源vim ~/.pip/pip.conf
[global]index-url = https://pypi.tuna./simple
创建虚拟环境
conda create -n torch_paddle python=3.8conda activate torch_paddle
安装 torch，注意都是用的 conda 安装方式，pip 的安装方式见文末参考链接
速度慢
conda install pytorch torchvision torchaudio cudatoolkit=10.2 -c pytorch
速度快
conda install pytorch torchvision torchaudio cudatoolkit=10.2 -c https://mirrors.tuna./anaconda/cloud/pytorch/
安装 paddle
conda install paddlepaddle-gpu==2.2.1 cudatoolkit=10.2 --channel https://mirrors.tuna./anaconda/cloud/Paddle/
conda 包列表
# packages in environment at /anaconda3/envs/torch_paddle:## NameVersion Build Channel_libgcc_mutex 0.1 main https://mirrors./anaconda/pkgs/main_openmp_mutex 4.5 1_gnu https://mirrors./anaconda/pkgs/mainastor 0.8.1 py38h06a4308_0 https://mirrors./anaconda/pkgs/mainblas 1.0mkl https://mirrors./anaconda/pkgs/mainbrotlipy 0.7.0 py38h27cfd23_1003 https://mirrors./anaconda/pkgs/mainbzip2 1.0.8h7b6447c_0 https://mirrors./anaconda/pkgs/mainca-certificates .10.26 h06a4308_2 https://mirrors./anaconda/pkgs/maincertifi .10.8 py38h06a4308_2 https://mirrors./anaconda/pkgs/maincffi 1.14.6 py38h400218f_0 https://mirrors./anaconda/pkgs/maincharset-normalizer 2.0.4 pyhd3eb1b0_0 https://mirrors./anaconda/pkgs/maincryptography 36.0.0 py38h9ce1e76_0 https://mirrors./anaconda/pkgs/maincudatoolkit10.2.89 hfd86e86_1 https://mirrors./anaconda/pkgs/maincudnn 7.6.5cuda10.2_0 https://mirrors./anaconda/pkgs/maindecorator 5.1.0 pyhd3eb1b0_0 https://mirrors./anaconda/pkgs/mainffmpeg4.3 hf484d3e_0 https://mirrors./anaconda/cloud/pytorchfreetype 2.11.0h70c0345_0 https://mirrors./anaconda/pkgs/maingast 0.3.3 py_0 https://mirrors./anaconda/pkgs/maingiflib5.2.1h7b6447c_0 https://mirrors./anaconda/pkgs/maingmp 6.2.1h2531618_2 https://mirrors./anaconda/pkgs/maingnutls3.6.15he1e5248_0 https://mirrors./anaconda/pkgs/mainidna 3.3pyhd3eb1b0_0 https://mirrors./anaconda/pkgs/mainintel-openmp .4.0h06a4308_3561 https://mirrors./anaconda/pkgs/mainjpeg 9d h7f8727e_0 https://mirrors./anaconda/pkgs/mainlame 3.100h7b6447c_0 https://mirrors./anaconda/pkgs/mainlcms2 2.12 h3be6417_0 https://mirrors./anaconda/pkgs/mainld_impl_linux-642.35.1h7274673_9 https://mirrors./anaconda/pkgs/mainlibffi3.3 he6710b0_2 https://mirrors./anaconda/pkgs/mainlibgcc-ng 9.3.0h5101ec6_17 https://mirrors./anaconda/pkgs/mainlibgomp 9.3.0h5101ec6_17 https://mirrors./anaconda/pkgs/mainlibiconv 1.15 h63c8f33_5 https://mirrors./anaconda/pkgs/mainlibidn2 2.3.2h7f8727e_0 https://mirrors./anaconda/pkgs/mainlibpng1.6.37hbc83047_0 https://mirrors./anaconda/pkgs/mainlibprotobuf3.19.1h4ff587b_0 https://mirrors./anaconda/pkgs/mainlibstdcxx-ng 9.3.0hd4cf53a_17 https://mirrors./anaconda/pkgs/mainlibtasn1 4.16.0h27cfd23_0 https://mirrors./anaconda/pkgs/mainlibtiff 4.2.0h85742a9_0 https://mirrors./anaconda/pkgs/mainlibunistring 0.9.10h27cfd23_0 https://mirrors./anaconda/pkgs/mainlibuv 1.40.0h7b6447c_0 https://mirrors./anaconda/pkgs/mainlibwebp 1.2.0h89dd481_0 https://mirrors./anaconda/pkgs/mainlibwebp-base 1.2.0h27cfd23_0 https://mirrors./anaconda/pkgs/mainlz4-c 1.9.3h295c915_1 https://mirrors./anaconda/pkgs/mainmkl .4.0 h06a4308_640 https://mirrors./anaconda/pkgs/mainmkl-service2.4.0 py38h7f8727e_0 https://mirrors./anaconda/pkgs/mainmkl_fft 1.3.1 py38hd3c417c_0 https://mirrors./anaconda/pkgs/mainmkl_random1.2.2 py38h51133e4_0 https://mirrors./anaconda/pkgs/mainncurses 6.3 h7f8727e_2 https://mirrors./anaconda/pkgs/mainnettle3.7.3hbbd107a_1 https://mirrors./anaconda/pkgs/mainnumpy 1.21.2 py38h20f2e39_0 https://mirrors./anaconda/pkgs/mainnumpy-base1.21.2 py38h79a1101_0 https://mirrors./anaconda/pkgs/mainolefile 0.46pyhd3eb1b0_0 https://mirrors./anaconda/pkgs/mainopenh264 2.1.1h4ff587b_0 https://mirrors./anaconda/pkgs/mainopenssl 1.1.1mh7f8727e_0 https://mirrors./anaconda/pkgs/mainpaddlepaddle-gpu2.2.1 py38_gpu_cuda10.2_many_linux https://mirrors.tuna./anaconda/cloud/Paddlepillow8.4.0 py38h5aabda8_0 https://mirrors./anaconda/pkgs/mainpip 21.2.4 py38h06a4308_0 https://mirrors./anaconda/pkgs/mainprotobuf 3.19.1 py38h295c915_0 https://mirrors./anaconda/pkgs/mainpycparser 2.21pyhd3eb1b0_0 https://mirrors./anaconda/pkgs/mainpyopenssl 21.0.0 pyhd3eb1b0_1 https://mirrors./anaconda/pkgs/mainpysocks 1.7.1 py38h06a4308_0 https://mirrors./anaconda/pkgs/mainpython3.8.12h12debd9_0 https://mirrors./anaconda/pkgs/mainpytorch 1.10.1py3.8_cuda10.2_cudnn7.6.5_0 https://mirrors./anaconda/cloud/pytorchpytorch-mutex 1.0 cuda https://mirrors./anaconda/cloud/pytorchreadline 8.1.2h7f8727e_1 https://mirrors./anaconda/pkgs/mainrequests 2.27.1 pyhd3eb1b0_0 https://mirrors./anaconda/pkgs/mainsetuptools58.0.4 py38h06a4308_0 https://mirrors./anaconda/pkgs/mainsix 1.16.0 pyhd3eb1b0_0 https://mirrors./anaconda/pkgs/mainsqlite3.37.0hc218d9a_0 https://mirrors./anaconda/pkgs/maintk 8.6.11h1ccaba5_0 https://mirrors./anaconda/pkgs/maintorchaudio0.10.1py38_cu102 https://mirrors./anaconda/cloud/pytorchtorchvision0.11.2py38_cu102 https://mirrors./anaconda/cloud/pytorchtyping_extensions 3.10.0.2 pyh06a4308_0 https://mirrors./anaconda/pkgs/mainurllib3 1.26.7 pyhd3eb1b0_0 https://mirrors./anaconda/pkgs/mainwheel 0.37.1 pyhd3eb1b0_0 https://mirrors./anaconda/pkgs/mainxz 5.2.5h7b6447c_0 https://mirrors./anaconda/pkgs/mainzlib 1.2.11h7f8727e_4 https://mirrors./anaconda/pkgs/mainzstd 1.4.9haebb681_0 https://mirrors./anaconda/pkgs/main
这样两个框架就都被安装在一个虚拟环境下了。
参考：
paddle安装指导
torch安装指导
TensorFlow 2.x GPU版在conda虚拟环境下安装步骤
pip/conda导出 requirements.txt 注意事项
pip安装pytorch1.10.1+paddlepaddle-gpu2.2.1+cuda10.2+cudnn7.6.5
