This is just my working notes for going through the course Practical Deep Learning For Coders.
This uses a conda environment called pdl_p (practical deep learning practice) This enviroment doesn't use the GPU (so should be portable) and can be created either by:
conda install jupyter
conda install -c fastai fastai
conda install -c fastai nbdev
conda install gradio
Or
conda env create -f environment.yml
(The environment.yml file installs also numba and graphviz, that are used sparingly)
For my testing (on WSL2), I also have an enviroment named pdl_gpu which was created like this:
conda install pytorch torchvision torchaudio pytorch-cuda=12.1 -c pytorch -c nvidia
conda install jupyter
conda install -c fastai fastai
This is not portable, and the particular version of cuda you need may vary. (Use nvidia-smi)
Note make sure to update / upgrade with conda before doing this .
For part two, I sometimes need more compute, so I will be using Paperspace Gradient. I used a 'start from scratch' notebook. To setup I created a simple notebook that does
!pip install -Uq diffusers transformers fastcore
And can do the huggingspace login if needed but not sure when it is needed, I did not seem to need it for lesson 9.
For working locally with minai I also create a pyproject.toml so that one can do pip install -e . to install the library.