It’s been quite a while since I last dabbled myself in deep learning and therefore decided to revisit from the basics. And what better way than to start doing that than learning from fastai? :D In this post, we will see how to quickly setup your dev environment for running notebooks locally to put your hard earned GPUs to use :p
Of course, you can run your notebooks on cloud with free GPU support on platforms such as Google Colab, Paperspace Gradient or even kaggle notebooks but sometimes, it feels good to run things locally without worrying too much about quotas or network issues etc. If you’re starting new in this field, it’s highly recommended to try the above platforms first.
Firstly, you need mamba. Use it wherever you use conda because it’s much faster. Once you install it, run the following script:
# create a conda environment
mamba create -n fastai python=3.10
# install suitable version of `pytorch-cuda` at your time of installation
mamba install pytorch torchvision torchaudio pytorch-cuda=11.7 -c pytorch -c nvidia
# install fastai
mamba install -c fastchan fastai
mamba install -c conda-forge jupyterlab ipywidgetsMake sure you can use GPU with pytorch by running this in your python session:
import torch
assert torch.cuda.is_available()That’s it. Now you can run notebooks with GPU support locally simply by doing mamba activate fastai and launching jupyter ! 💚