@behxyz, do you think you could try this PR? I'm fairly confident it will have solved the problem.

@wyli do you understand the error? https://github.com/Project-MONAI/MONAI/pull/1801/checks?check_run_id=2141079751#step:11:5166

It seems that the following code is being executed on the windows and osx test:

# skip this test if multiprocessing uses 'spawn', as the check is only basic anyway
if torch.multiprocessing.get_start_method(allow_none=True) == "spawn":
    # Check that error is thrown when inverse are used out of order.
    t = SpatialPadd("image", [10, 5])
    with self.assertRaises(RuntimeError):
        t.inverse(forwards[-1])  # <-- windows and osx shouldn't reach here, but they do, causing error

Seems to be unrelated CUDA error: https://github.com/Project-MONAI/MONAI/pull/1801/checks?check_run_id=2141451226#step:6:7682

@wyli do you understand the error? https://github.com/Project-MONAI/MONAI/pull/1801/checks?check_run_id=2141079751#step:11:5166

It seems that the following code is being executed on the windows and osx test:

# skip this test if multiprocessing uses 'spawn', as the check is only basic anyway
if torch.multiprocessing.get_start_method(allow_none=True) == "spawn":
    # Check that error is thrown when inverse are used out of order.
    t = SpatialPadd("image", [10, 5])
    with self.assertRaises(RuntimeError):
        t.inverse(forwards[-1])  # <-- windows and osx shouldn't reach here, but they do, causing error

I've changed allow_none to False, it'll set a specific start method otherwise I think it's None...

@behxyz, do you think you could try this PR? I'm fairly confident it will have solved the problem.

Sure, let me check this PR before merging.

It works perfectly! Thank you, @rijobro, for the quick fix.