Add your issue here

We need to figure out a way to test the new Array API implementations. Ideally, this should be something that can run on any system and only checks the mechanics of the code without passing in data (as we do not have any GPU instances yet, which will change in the future - ARCHER2). In any case, we cannot keep running simulations on ARCHER2 indefinitely, as we will run out of ARCHER2 time once the grant reaches its end (if I am not wrong).

We should look at how SciPy does this. At the moment I am testing the GPU implementations locally (like how I did in my previous job - merge the PR if the tests pass on 2 independent local CUDA machines 🥲).