You can follow these steps to make a Linux system ready for AI from scratch.
You need CUDA to use NVIDIA GPUs for general purpose processing. As NVIDIA puts it: “CUDA is a parallel computing platform and programming model that makes using a GPU for general purpose computing simple and elegant.” More information can be found under this link.
Let’s check to see if CUDA is installed or not and if yes, which version. We can do that by:
If nvcc is not available, you can install it by the following command:
sudo apt install nvidia-cuda-toolkit
If CUDA is not installed or you want an updated version installed, you can follow the steps on the NVIDIA website to install it.
For example, here are the instructions for installing CUDA Toolkit version 11.0. We are assuming Ubuntu 18.04 with x86_64 architecture. The installation method is runfile (local). You can find further instructions and options under this link.
wget http://developer.download.nvidia.com/compute/cuda/11.0.2/local_installers/cuda_11.0.2_450.51.05_linux.runsudo sh cuda_11.0.2_450.51.05_linux.run
If you like to install CUDA 10.1, the instructions can be found here. I have found the runfile (local) method of installation to be prone to less errors.
The last step is to make sure that the path to CUDA is correct. Let’s quickly check the CUDA version again by running the following command:
If the version is what you expected, we are good to go. Otherwise, let’s fix the path. You can find the list of installed CUDA versions under /usr/local/ folder by running the following command:
1. How to automatically deskew (straighten) a text image using OpenCV
2. Explanation of YOLO V4 a one stage detector
3. 5 Best Artificial Intelligence Online Courses for Beginners in 2020
4. A Non Mathematical guide to the mathematics behind Machine Learning
Assuming that you want to use cuda-11.0. You can run the following commands to fix the path:
sudo rm -rf /usr/local/cudasudo ln -sf /usr/local/cuda-11.0 /usr/local/cudaecho 'export CUDA_HOME=/usr/local/cuda' >> ~/.bashrcecho 'export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/cuda/lib64:/usr/local/cuda/extras/CUPTI/lib64' >> ~/.bashrcecho 'export PATH=$CUDA_HOME/bin:$PATH' >> ~/.bashrcsource ~/.bashrc
What we are doing here is that we remove the current cuda folder and we symlink the newly installed cuda-11.0 in its place. We then add the path to the .bashrc file (assuming using bash).
To have independent environments for different projects, we can use Anaconda. It is especially useful when you are working on multiple projects with different dependencies and you want to switch between them quickly.
If you do not have pip installed, you can install it via:
sudo apt install python-pip
To install Anaconda, you can follow the steps under this link. For example, to install the 2020.07 distribution for x86_64 architecture, you can run the following commands:
sudo apt-get install libgl1-mesa-glx libegl1-mesa libxrandr2 libxrandr2 libxss1 libxcursor1 libxcomposite1 libasound2 libxi6 libxtst6wget https://repo.anaconda.com/archive/Anaconda3-2020.07-Linux-x86_64.shbash Anaconda3-2020.07-Linux-x86_64.sh
You can choose different distributions of Anaconda from this repository.
This is the last step to create a conda environment and install the dependencies.
Let’s open a new terminal and create and activate a new conda environment, named ai, with the following commands:
conda create --name ai python=3.7 --yes
conda activate ai
Please note that you can edit the choice of name and Python version for the environment.
PyTorch can be installed by following the instructions under the official website. For example, you can install the latest version of PyTorch with assumption of CUDA 11.0 using the following command:
conda install pytorch torchvision torchaudio cudatoolkit=11.0 -c pytorch
TensorFlow can be installed by following the instructions of the official website, by simply using the command:
pip install tensorflow
Other dependencies can be installed as follows:
conda install -c fastai fastai
pip install opencv-python
# Jupyter Notebook
conda install -c conda-forge notebook
pip install transformers
You can install further dependencies.
To deactivate the conda environment, you can run:
conda deactivate ai
Then you can switch to another environment, say ai2 by the following command:
conda activate ai2