This writeup is my notes on setting up a deep-learning server (some aspects applicable for a desktop machine too). Pullrequests are welcome.
All writeup is for Ubuntu 18.x LTS (Bionic Beaver). But maybe applicable for newer versions.
Server installation is assumed, but everything is applicable for desktop installation too.
Desktop and server instalations
Needed for remote ssh access. During Server installation, it will ask if openssh-server should be installed. If it was not installed, or for the Desktop installation, run:
sudo apt-get install openssh-server
sudo service ssh start
Intructions direct ethernet connection between machines
Intructions for Nvidia drivers, CUDA, cuDNN installation
Intructions for IPMI tool and fan control
Whether you need to build some native code or not, this might be helpful to build python packages (e.g., dlib)
sudo apt install build-essential
sudo apt install cmake
Ubuntu 18.04 does not include /usr/bin/python (Python 2) by default, but it does include /usr/bin/python3 (Python 3).
Because of that, running command python will result in error Command 'python' not found.
To address that, since python2 is obsolete, I prefer to make a symlink using update-alternatives.
Then, to make command python refer to python3:
sudo update-alternatives --install /usr/bin/python python /usr/bin/python3 10
This will essentially make a symlink /usr/bin/python3 to /usr/bin/python. later on, you can change that with sudo update-alternatives --config python
To install pip:
sudo install python3-pip
Similarly, make a symlink pip
sudo update-alternatives --install /usr/bin/pip pip /usr/bin/pip3 10
If you have installed both python2 along with python3, then you might need to do (on older distributives, prior 18.x, it may break things):
sudo update-alternatives --install /usr/bin/python python /usr/bin/python2 1
sudo update-alternatives --install /usr/bin/python python /usr/bin/python3 10
To change the version of python:
sudo update-alternatives --config python
I prefer to install frequently used packages globally with pip using sudo. It is not recommended using sudo with pip, but if adhere to a rule to never install python packages with apt, that works fine. This way, all main packages will be immediately available for a new user, and disk space in home directory will not be wasted.
Some packages that make sence to install globally:
sudo pip install torch torchvision tensorflow-gpu Keras pandas Pillow scipy dlib imageio ninja yacs cython matplotlib tqdm opencv-python scipy scikit-learn scikit-image sklearn
To add a new user:
sudo adduser <username>
To add a new user to sooders:
sudo usermod -aG sudo <username>
This should be enabled by default unless it was disabled during installation:
sudo add-apt-repository universe
sudo add-apt-repository multiverse
sudo apt update