Fix EMAModel test_from_pretrained

[hlky]

What does this PR do?

Torch CUDA Tests (others) has been failing from EMAModel.

FAILED tests/others/test_ema.py::EMAModelTests::test_from_pretrained - RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cpu!

ema_unet is on the correct device as it's created from unet which is moved to torch_device

diffusers/tests/others/test_ema.py

Lines 39 to 41 in a6288a5

	unet = UNet2DConditionModel.from_pretrained(self.model_id, subfolder="unet")
	unet = unet.to(torch_device)
	ema_unet = EMAModel(unet.parameters(), decay=decay, model_cls=UNet2DConditionModel, model_config=unet.config)

loaded_ema_unet is on CPU.

This matches torch.allclose call with original_param, loaded_param and the order of the error cuda:0 and cpu!.

Who can review?

Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.

[sayakpaul]

Thanks!