Torch sparse tensor unsupported at mps

mps
1

I have sparse tensor for handling graph representation, however it can’t be processed by ‘mps’, I wonder if there is a workaround ?
Error I get:
NotImplementedError: Unsupported device type for sparse layout: mps

2

Hi,

Sparse support is not planned yet I’m afraid.
You should be able to run that part on the CPU and then transfer to do dense ops on the MPS side.

3

Hello,

I am facing the same problem. Still no support for spare tensors on MPS planned?

Right now I am doing the spare matrix multiplication on the CPU, but this is really slow because my tensors are quite big.

Any way to fix this?