[Contributor Experience] Test Collection fails on MPS

[UmerHA]

Describe the bug

On MPS, the test collection fails. Therefore not all tests are shown e.g. in VS Code, which makes the contributor experience more tedious.

Currently, 4515-4486 = 29 tests are missed.

---

The error is caused because in utls/testing_utils.py the variable torch_device is defined as a string ("cpu", "cuda" or "mps"), but used as a device (i.e., torch_device.type is called).

I'll submit a small fix PR.

Reproduction

pytest --collect-only tests

Logs

============================================================================ ERRORS ============================================================================
______________________________________________________ ERROR collecting tests/models/test_layers_utils.py ______________________________________________________
src/diffusers/utils/testing_utils.py:862: in _is_torch_fp64_available
    x = torch.zeros((2, 2), dtype=torch.float64).to(device)
E   TypeError: Cannot convert a MPS Tensor to float64 dtype as the MPS framework doesn't support float64. Please use float32 instead.

During handling of the above exception, another exception occurred:
tests/models/test_layers_utils.py:319: in <module>
    class Transformer2DModelTests(unittest.TestCase):
tests/models/test_layers_utils.py:430: in Transformer2DModelTests
    @require_torch_accelerator_with_fp64
src/diffusers/utils/testing_utils.py:241: in require_torch_accelerator_with_fp64
    return unittest.skipUnless(_is_torch_fp64_available(torch_device), "test requires accelerator with fp64 support")(
src/diffusers/utils/testing_utils.py:867: in _is_torch_fp64_available
    if device.type == "cuda":
E   AttributeError: 'str' object has no attribute 'type'

System Info

• diffusers version: 0.26.0.dev0
• Platform: macOS-14.2.1-arm64-arm-64bit
• Python version: 3.11.7
• PyTorch version (GPU?): 2.1.2 (False)
• Huggingface_hub version: 0.20.2
• Transformers version: 4.36.2
• Accelerate version: 0.26.1
• xFormers version: not installed
• Using GPU in script?: no
• Using distributed or parallel set-up in script?: no

Who can help?

Questions on Tests: @DN6 @sayakpaul

[sayakpaul]

@pcuenca