-
Notifications
You must be signed in to change notification settings - Fork 3
Open
Description
Currently, we ignore the differences in array dimension ordering between frameworks.
We could add dimension ordering to each array in an array file, where we assume pytorch array dimensions are ordered sequentially like [1,2,3] as a baseline. If array dimensions from Julia are relatively ordered [3,1,2,], then dimension_order := [3,1,2] would appear next to the that array-from-Julia in the array file.
Or we can add some global flag like pytorch_dimension_order=True/False to the array file. If True, then we can easily port arrays to different frameworks. If False, then we can't. I would implement this simple solution. No plans for implementing dimension_order.
Reactions are currently unavailable
Metadata
Metadata
Assignees
Labels
No labels