Here are custom builds of dependency packages that we needed to make ourselves and upload to anaconda.org, either because they weren't there, or they weren't packaged the way we dependencies need to get resolved.
To install these packages users will need to add: -c fastai/label/test to their "conda install" command.
This one is built to depend on pytorch-nightly from the pytorch channel
conda build torchvision -c pytorch --python=3.6
anaconda upload ~/anaconda3/conda-bld/noarch/torchvision-0.2.1-pyhe7f20fa_0.tar.bz2 -u fastai --label test
then need to ask users to install torchvision=0.2.1=pyhe7f20fa_0
conda build dataclasses
conda build nvidia-ml-py3
In case there is a new version out there, the recipe was auto-generated with: conda skeleton pypi nvidia-ml-py3, but remember to add noarch: python inside build:, since by default it'll set it to 'linux-64'.
See libjpeg-turbo/conda-build.txt
See libtiff/conda-build.txt
See pillow/conda-build.txt
See pillow-simd/conda-build.txt
conda uninstall fastai pytorch-nightly pytorch torchvision dataclasses nvidia-ml-py3
pip uninstall fastai pytorch-nightly pytorch torchvision dataclasses nvidia-ml-py3