search

pytorch / builder

Continuous builder and binary build scripts for pytorch
BSD 2-Clause "Simplified" License
328 stars 220 forks source link

Cannot install scikit-learn from conda-forge channel alongside pytorch from the pytorch channel #1951

Open anjos opened 1 month ago

anjos commented 1 month ago

(Cross-linking related issue from https://github.com/conda-forge/scikit-learn-feedstock/issues/271)

There seems to be no way to have these 2 packages installed together with the current conda builds for pytorch:

scikit-learn >=1.4 (from the conda-forge channel) pytorch >=2.3 (from pytorch channel)

Test command:

mamba create --strict-channel-priority --override-channels -c pytorch -c conda-forge -n test --dry-run 'scikit-learn>=1.4' 'pytorch>=2.3'

The problem seems related to this package's very restrictive pinning on llvm-openmp<16 and the way this dependency is treated on conda-forge (c.f.: https://github.com/conda-forge/scikit-learn-feedstock/pull/265, https://github.com/conda-forge/openmp-feedstock/issues/126).

The offending line seems to come from: https://github.com/pytorch/builder/blob/main/conda/pytorch-nightly/meta.yaml#L49

I wonder if the issue is still present with versions 17 or 18 of LLVM, and if one could update this pin to allow for a more inclusive software stack.