Functions ht.empty_like(a), ht.full_like(a), ht.ones_like(a) and ht.zeros_like(a) return new DNDarrays with the same type, shape and data distribution of a but not with the same device. The device is mistakenly set to the globally-set default, this explains why the CI didn't catch it, as @mrfh92 correctly pointed out.
The equivalent torch functions return new tensors with the same attributes including the same device as the input tensor.
Code snippet triggering the error
import heat as ht
a = ht.arange(10, device="gpu")
b = ht.ones_like(a)
a + b
Error message or erroneous outcome
RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cpu!
What happened?
Functions
ht.empty_like(a),ht.full_like(a),ht.ones_like(a)andht.zeros_like(a)return new DNDarrays with the same type, shape and data distribution ofabut not with the same device. The device is mistakenly set to the globally-set default, this explains why the CI didn't catch it, as @mrfh92 correctly pointed out.The equivalent
torchfunctions return new tensors with the same attributes including the same device as the input tensor.Code snippet triggering the error
Error message or erroneous outcome
Version
main (development branch)
Python version
3.11
PyTorch version
2.1
MPI version
No response