Optional: Make your own deep learning box

Here are some key points if you decide to do the course using your own hardware instead of a cloud-based solution. These are meant to indicate whether this approach is for you. You'll find plenty of detailed instructions online, of course. If the below seems uninteresting to you, to be too much work, or impossible with the hardware you already have, starting out with a cloud-based solution is recommended. See https://course.fast.ai/ Links to an external site. and the forums at https://forums.fast.ai/ Links to an external site. for recommended cloud solutions.

In the last parts of the course (those not based on fastai) it will be convenient to use your own hardware (running either GNU/Linux or Windows). This will be discussed later when we get to the relevant modules.

 

  • You should use a PC, not a Mac.
  • You'll need an NVIDIA GPU with a minimum compute capability of 5.0: https://developer.nvidia.com/cuda-gpus#compute Links to an external site.. The more GPU memory, the better.
  • You should run GNU/Linux, preferably Ubuntu 20.04. A dual boot setup with Ubuntu alongside Windows is fine. If you insist on using Windows you can find multiple guides describing how to install PyTorch and fastai with GPU support online. But note that this is less supported than GNU/Linux.
  • Install fastai and the other necessary Python libraries using Anaconda, as detailed here: https://docs.fast.ai.  Links to an external site.. It's convenient to install the libraries in a separate conda environment, not in the base environment. F.ex. conda create -n fastai ipykernel.
  • To follow along with the fastai course it's convenient to clone the fastbook repository: git clone https://github.com/fastai/fastbook.git
  • Remember to launch the notebooks using the correct conda environment (or install an ipykernel pointing to the environment you created above: i.e. python -m ipykernel install --user --name fastai ).