Hi all! I am developing a new package integrating CUDA.jl as a dependency. However, I encountered errors indicating CUDA driver not found
in CI when testing. I am using Standard GitHub-hosted runners and do not want to switch to Larger runners. What is the typical way to test the whole package on GitHub?
I have seen some similar packages being developed without a CI file. Does this mean that it is acceptable to test locally and push to GitHub without official CI testing?
1 Like
For GPU testing you need GPU infrastructure, which is not available in freely accessible GH runners/CI. You can either self-host GPU runners, or use the JuliaGPU Buildkite runners (somewhat separate CI from GH actions). You can see in packages like KernelAbstractions.jl how both CPU and GPU CI is handled (via .github
and .buildkite
files).
One alternative could be, if your package supports both CPU and GPU backend, to for now only have CPU testing in CI, while running GPU tests locally…
2 Likes