Open wolfv opened 1 week ago
Think this is because we test the different CUDA versions. That said, agree we can reduce this
Would suggest we reduce this to just CUDA 11.8 as it is the only pre-CUDA 12 version we still support ( https://github.com/conda-forge/conda-forge-pinning-feedstock/issues/5339 )
All other CUDA versions are dropped and CUDA 12.0+ doesn't need the same logic that we have here
There are >40 builds kicked off per PR but I am not sure that we actually need a CUDA variant for this repository? Maybe we could reduce this and make sure that the test suite covers cuda cases as well?