You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Notes #711 showed that our code couldn't run on an M1 MBP. #761 fixed this by explicitly using float32's instead of float64's. Perhaps we should run our tests on an M1 image so that future changes don't accidentally reintroduce float64 usage.
The text was updated successfully, but these errors were encountered:
Thanks for pointing that out. Our tests have been working on macos-latest even with using some float64's. Do you know why our tests worked while #711 had the issue? Is that something we can reproduce in the GitHub tests so it isn't missed in the future?
I looked back at the issue and realized that it's actually because the user was running using MPS (see PyTorch docs), basically the apple version of CUDA/ROCM acceleration. Our unit tests probably would have found this if anyone had run them on a machine using MPS-pytorch. The github actions runners don't have GPUs unless we upgrade to this so we have no way to catch it right now. But really it's a pretty small problem - we can probably safely ignore this and just fix these issues as they arise.
Notes
#711 showed that our code couldn't run on an M1 MBP. #761 fixed this by explicitly using float32's instead of float64's. Perhaps we should run our tests on an M1 image so that future changes don't accidentally reintroduce float64 usage.
The text was updated successfully, but these errors were encountered: