For the cluster container, it would be nice to install cuda with a matching cuda version.
Maybe consider splitting the reco.sh into a cpuonly version for binder etc,
and a cuda version for our cluster (and other cuda hosts)
So maybe, before installing mrpro, we could install tensorflow (?) and pytorch with cuda (and related cuda libraries) via conda?
For the cluster container, it would be nice to install cuda with a matching cuda version.
Maybe consider splitting the reco.sh into a cpuonly version for binder etc,
and a cuda version for our cluster (and other cuda hosts)
So maybe, before installing mrpro, we could install tensorflow (?) and pytorch with cuda (and related cuda libraries) via conda?