It’s been marked as broken for over a year and requires CUDA
10. Even the non‐CUDA variant of the package refused to evaluate
without enabling broken packages due to `cudnn`, so I’m not sure
anyone is using this package at all…
The package has not been maintained aside from small bugfixes and
formatting fixes, and is not depended on by anything else, therefore it
should be good to drop now.
The repository moved out of the openai org, so it doesn't make sense to
prefix the package with it.
(cherry picked from commit af13bb4513647eec3c3790c5272dbd4aa190d208)
The proper name for a python package is the one in the setuptools
setup() call, which can also be seen on pypi.
Correct: https://pypi.org/project/torch/
Wrong: https://pypi.org/project/pytorch/
Includes a treewide rename of the attribute and creates aliases for the
old name.
There are many different versions of the `cudatoolkit` and related
cuda packages, and it can be tricky to ensure they remain compatible.
- `cudaPackages` is now a package set with `cudatoolkit`, `cudnn`, `cutensor`, `nccl`, as well as `cudatoolkit` split into smaller packages ("redist");
- expressions should now use `cudaPackages` as parameter instead of the individual cuda packages;
- `makeScope` is now used, so it is possible to use `.overrideScope'` to set e.g. a different `cudnn` version;
- `release-cuda.nix` is introduced to easily evaluate cuda packages using hydra.
