search

rapidsai / dask-cuda

Utilities for Dask and CUDA interactions
https://docs.rapids.ai/api/dask-cuda/stable/
Apache License 2.0
292 stars 93 forks source link

Publish nightly wheels to NVIDIA index instead of PyPI #1294

Closed pentschev closed 11 months ago

pentschev commented 11 months ago

Nightly wheels also require rapids-dask-dependency which is only available in NVIDIA's PyPI index and cannot be published to PyPI as it installs Dask/Distributed from GitHub, which is forbidden by PyPI. Therefore, we're switching to publishing nightlies only to NVIDIA index as it doesn't seem external projects currently rely on nightlies. Release packages will continue to be published to PyPI.

pentschev commented 11 months ago

Good catch @charlesbluca , updated.

pentschev commented 11 months ago

cc @vyasr @raydouglass

pentschev commented 11 months ago

Looks like the wheel build is still failing - my guess is that this is because the provided nightly version of 24.02.00a22 isn't PEP440 compliant? Perhaps we need to do something like rapidsai/dask-cuda#1279 here?

You're right, but honestly I don't know what's the "right way" to fix this. The version being generated 24.02.00a22 seems in accordance to the version defined by rapids-dask-dependency where I took these changes from. Maybe @vyasr can point us to how it should be handled here, I believe we may also need to be compliant with release versions (which won't include a0 suffix).

charlesbluca commented 11 months ago

On second thought, looks like maybe what we want to do here is bump the version in dask_cuda/VERSION rather than in the pyproject file, similar to what's happening here:

https://github.com/rapidsai/cudf/blob/8f7cbe69d4c2f670b97decc63e73b08e0eef7329/ci/build_wheel.sh#L25-L26

charlesbluca commented 11 months ago

/merge