Please do not install Python of 3.10 and above, as they do not support PyTorch versions 1.x. If you have already installed, you can create a suitable Python virtual environment through Anaconda. It is recommended to use Python of versions from 3.7 to 3.9.
First, install Chocolatey (Windows package manager):
Right-click on the Windows icon, select "Command Prompt (Admin)" or "Terminal (Admin)" (if your terminal default is cmd), then input:
@"%SystemRoot%\System32\WindowsPowerShell\v1.0\powershell.exe" -NoProfile -InputFormat None -ExecutionPolicy Bypass -Command "iex ((New-Object System.Net.WebClient).DownloadString('https://chocolatey.org/install.ps1'))" && SET "PATH=%PATH%;%ALLUSERSPROFILE%\chocolatey\bin"Or select "PowerShell (Admin)" or "Terminal (Admin)" (if your terminal default is powershell), then input:
Set-ExecutionPolicy Bypass -Scope Process -Force; iex ((New-Object System.Net.WebClient).DownloadString('https://chocolatey.org/install.ps1'))Then follow the prompted steps to install. After installation, close the current command line and open a new one.
Then, input the following command in the command line:
choco install gitOnce GIT installation is complete, go to https://mirrors.tuna.tsinghua.edu.cn/anaconda/archive/Anaconda3-2020.11-Windows-x86_64.exe to download and install Anaconda (base environment is Python 3.8). Note: Be sure to add the conda environment to PATH (when "add ... to PATH" appears, please check it).
Testing
Then, input the following commands in the command line:
git --help
python --help
pip --helpCheck if they are installed successfully. If there is no error, it means the installation was successful.
First, install Homebrew (MacOS package manager):
Open the "Terminal" application, and then input:
# Without GFW (if you don't know GFW, use this)
/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"
# With GFW (if errors occur, use this)
/bin/zsh -c "$(curl -fsSL https://gitee.com/cunkai/HomebrewCN/raw/master/Homebrew.sh)"Follow the prompted steps to install. After installation, close the current command line and open a new one.
Then, input the following command in the command line:
brew install gitOnce GIT installation is complete, go to https://mirrors.tuna.tsinghua.edu.cn/anaconda/archive/Anaconda3-2020.11-MacOSX-x86_64.pkg to download and install Anaconda (base environment is Python 3.8).
Testing
Then, input the following commands in the command line:
git --help
python --help
pip --helpCheck if they are installed successfully. If there is no error, it means the installation was successful.
First, update APT sources. Input the following command in the command line:
sudo apt updateThen, input the following command in the command line:
sudo apt install git python3-pip python-is-python3Install GIT and PIP. Note: Ubuntu 20.04 comes with Python version 3.8, so there is no need to reinstall Anaconda; however, the built-in Python does not include PIP and needs to be installed manually.
Alternatively, you can go to https://mirrors.tuna.tsinghua.edu.cn/anaconda/archive/Anaconda3-2020.11-Linux-x86_64.sh to download and install Anaconda (base environment is Python 3.8) through:
sudo path/to/Anaconda3-2020.11-Linux-x86_64.shTesting
Then, input the following commands in the command line:
git --help
python3 --help
pip3 --helpCheck if they are installed successfully. If there is no error, it means the installation was successful.
If you do not have a CUDA-supported NVIDIA™ graphics card, you can skip this step. CUDA versions 11.6 and above are all installable, and it is recommended to install CUDA 11.7.1 version.
Click on the link https://developer.nvidia.com/cuda-11-7-1-download-archive, and choose the following options for installation:
If you are a dual-boot or virtual machine user, please make the following selection:
If you are a WSL user, please make the following selection:
Then, relevant command lines will pop up at the bottom of the page:
Please copy and execute the commands in sequence for CUDA installation.
After CUDA installation, please proceed to https://developer.nvidia.com/zh-cn/cudnn and download and install CUDNN suitable for CUDA 11.x according to the page instructions.
Note: Although CUDA and CUDNN are installed after completing the above steps, they will not be activated automatically when executed. At this point, we need to add self-startup code in ~/.bashrc:
(1) Use vim (or text editor, etc.) to open ~/.bashrc;
sudo apt install vim
sudo vim ~/.bashrc(2) Then, the command line will display the content of ~/.bashrc. Move the cursor to the end of the file, press the I key to start editing, and insert the following code at the end;
function switch_cuda {
v=$1
export PATH=/usr/local/cuda-$v/bin:$PATH
export CUDADIR=/usr/local/cuda-$v
export CUDA_HOME=/usr/local/cuda-$v
export LD_LIBRARY_PATH=/usr/local/cuda-$v/lib64:$LD_LIBRARY_PATH
}
switch_cuda 11.7(3) After insertion, press the esc key to enter command input mode, and input:
:wqto save and exit;
(4) Close the command line or input:
source ~/.bashrcto make CUDA effective in the command line.
Testing
After installation, input the following command in the command line:
nvcc --versionIf there is no error, and the version number is observed to be 11.7.1, it means the installation was successful.
Please refer to the tutorial https://zhuanlan.zhihu.com/p/99880204 for installation. It is recommended to install CUDA 11.7.1.
Testing
After installation, input the following command in the command line:
nvcc --versionIf there is no error, and the version number is observed to be 11.7.1, it means the installation was successful.
Copy from PyTorch Historical Versions page, do not install PyTorch versions above 2.0, as we cannot determine if there are compatibility issues. It is recommended to install version 1.13.1.
pip install torch==1.13.1 torchvision==0.14.1 torchaudio==0.13.1# For ROCM 5.2
pip install torch==1.13.1+rocm5.2 torchvision==0.14.1+rocm5.2 torchaudio==0.13.1 --extra-index-url https://download.pytorch.org/whl/rocm5.2
# For CUDA 11.6
pip install torch==1.13.1+cu116 torchvision==0.14.1+cu116 torchaudio==0.13.1 --extra-index-url https://download.pytorch.org/whl/cu116
# For CUDA 11.7
pip install torch==1.13.1+cu117 torchvision==0.14.1+cu117 torchaudio==0.13.1 --extra-index-url https://download.pytorch.org/whl/cu117
# Otherwise
pip install torch==1.13.1+cpu torchvision==0.14.1+cpu torchaudio==0.13.1 --extra-index-url https://download.pytorch.org/whl/cpuTesting
After installation, open the command line, input:
# Windows or MacOS
python
# Ubuntu 20.04
python3When Python is started, input:
import torchIf there are no errors, it means PyTorch has been successfully installed.
If you have installed the CUDA version of PyTorch, continue by inputting:
torch.cuda.is_available()If the command line returns:
Truethen it indicates that the CUDA extension of PyTorch is available and machine learning can be performed using the GPU.
Matterhorn relies on C++ and CUDA for accelerating neurons. If you want Matterhorn to run at full speed and get the best experience, please install GCC and G++.
If you input
gcc -v
g++ -vin the command line without errors, you can skip this step.
Use the provided mingw-get-setup.exe in this repository, and refer to this tutorial for installing GCC and G++.
Refer to this tutorial for installing Command Line Tools, which includes the GCC and G++ provided by Apple.
Install GCC and G++ via APT:
sudo apt install gcc g++Testing
Open the command line, input:
gcc -v
g++ -vIf there are no errors, it means the installation was successful.
Clone the repository:
cd your/path
git clone https://github.com/xjtuiair-cag/Matterhorn.gitThen install:
cd Matterhorn
python setup.py developIf an error occurs, please check if you are running in administrator mode (reopen terminal in administrator mode for Windows, add sudo before the command for Linux).
Testing
Open the command line, input:
pythonWhen Python is started, input:
import matterhorn_pytorch as mthIf there are no errors, it means Matterhorn has been successfully installed.