The NumPy fallback mode in CuPy was first introduced in CuPy v7 (2019), aiming to ease the porting of NumPy-based code to CuPy by providing a mechanism to automatically delegate "missing" APIs in CuPy to NumPy. However, now that we have covered almost all the NumPy API, I think this layer is no longer playing its role.

As far as I can tell from the GitHub search resutls, no public code is relying on cupyx.fallback_mode, so I'm thinking of removing the feature in CuPy v14. Please let us know here if you have use-cases relying on the fallback mode.