What happened?

When saving a dndarray with dtype=ht.int32 using save_hdf5, the resulting h5 dataset has the type float32. This can be avoided by specifying dtype=int or dtype=np.int32 (not ht.int32 or torch.int32 as those throw errors) but it should use the dtype of the dndarray by default.

The same issue exists for load_hdf5, where it does not use the type of the h5 dataset but sets it to float32.

Code snippet triggering the error

import heat as ht
a = ht.arange(100, dtype=ht.int32)
ht.save_hdf5(a, 'test.h5', dataset='data')
b = ht.load_hdf5('test.h5', dataset='data')
print(b.dtype)

Error message or erroneous outcome

No error just the wrong type

Version

main (development branch)

Python version

3.12

PyTorch version

2.6

MPI version