[Tests] Skip layerwise casting tests on devices without float8_e4m3fn support (#14073)

* Skip layerwise casting tests on devices without float8_e4m3fn support

* Xfail float8 layerwise casting tests on MPS

---------

Co-authored-by: Sayak Paul <spsayakpaul@gmail.com>

Author: GiGiKoneti

tests/lora/utils.py

@@ -2105,6 +2105,11 @@ def test_correct_lora_configs_with_different_ranks(self):
         self.assertTrue(not np.allclose(original_output, lora_output_diff_alpha, atol=1e-3, rtol=1e-3))
         self.assertTrue(not np.allclose(lora_output_diff_alpha, lora_output_same_rank, atol=1e-3, rtol=1e-3))
 
+    @pytest.mark.xfail(
+        condition=torch_device == "mps",
+        reason="MPS does not support float8 casting.",
+        strict=True,
+    )
     def test_layerwise_casting_inference_denoiser(self):
         from diffusers.hooks._common import _GO_LC_SUPPORTED_PYTORCH_LAYERS
         from diffusers.hooks.layerwise_casting import DEFAULT_SKIP_MODULES_PATTERN
@@ -2149,6 +2154,11 @@ def initialize_pipeline(storage_dtype=None, compute_dtype=torch.float32):
         pipe_float8_e4m3_bf16 = initialize_pipeline(storage_dtype=torch.float8_e4m3fn, compute_dtype=torch.bfloat16)
         pipe_float8_e4m3_bf16(**inputs, generator=torch.manual_seed(0))[0]
 
+    @pytest.mark.xfail(
+        condition=torch_device == "mps",
+        reason="MPS does not support float8 casting.",
+        strict=True,
+    )
     @require_peft_version_greater("0.14.0")
     def test_layerwise_casting_peft_input_autocast_denoiser(self):
         r"""

tests/models/testing_utils/memory.py

@@ -383,6 +383,11 @@ class LayerwiseCastingTesterMixin:
         - get_dummy_inputs(): Returns dict of inputs to pass to the model forward pass
     """
 
+    @pytest.mark.xfail(
+        condition=torch_device == "mps",
+        reason="MPS does not support float8 casting.",
+        strict=True,
+    )
     @torch.no_grad()
     def test_layerwise_casting_memory(self):
         MB_TOLERANCE = 0.2
@@ -436,6 +441,11 @@ def get_memory_usage(storage_dtype, compute_dtype):
             or abs(fp8_e4m3_fp32_max_memory - fp32_max_memory) < MB_TOLERANCE
         ), "Peak memory should be lower or within tolerance with fp8 storage"
 
+    @pytest.mark.xfail(
+        condition=torch_device == "mps",
+        reason="MPS does not support float8 casting.",
+        strict=True,
+    )
     def test_layerwise_casting_training(self):
         def test_fn(storage_dtype, compute_dtype):
             if torch.device(torch_device).type == "cpu" and compute_dtype == torch.bfloat16:

tests/pipelines/test_pipelines_common.py

@@ -2292,6 +2292,11 @@ def test_save_load_dduf(self, atol=1e-4, rtol=1e-4):
         elif isinstance(pipeline_out, torch.Tensor) and isinstance(loaded_pipeline_out, torch.Tensor):
             assert torch.allclose(pipeline_out, loaded_pipeline_out, atol=atol, rtol=rtol)
 
+    @pytest.mark.xfail(
+        condition=torch_device == "mps",
+        reason="MPS does not support float8 casting.",
+        strict=True,
+    )
     def test_layerwise_casting_inference(self):
         if not self.test_layerwise_casting:
             return