move test_hooks.py to pytest

src/diffusers/quantizers/quantization_config.py

@@ -36,7 +36,7 @@ from typing import Any, Callable
 
 from packaging import version
 
-from ..utils import deprecate, is_torch_available, is_torchao_available, is_torchao_version, logging
+from ..utils import is_torch_available, is_torchao_available, is_torchao_version, logging
 
 
 if is_torch_available():
@@ -844,8 +844,6 @@ class QuantoConfig(QuantizationConfigMixin):
         modules_to_not_convert: list[str] | None = None,
         **kwargs,
     ):
-        deprecation_message = "`QuantoConfig` is deprecated and will be removed in version 1.0.0."
-        deprecate("QuantoConfig", "1.0.0", deprecation_message)
         self.quant_method = QuantizationMethod.QUANTO
         self.weights_dtype = weights_dtype
         self.modules_to_not_convert = modules_to_not_convert

src/diffusers/quantizers/quanto/quanto_quantizer.py

@@ -3,7 +3,6 @@ from typing import TYPE_CHECKING, Any
 from diffusers.utils.import_utils import is_optimum_quanto_version
 
 from ...utils import (
-    deprecate,
     get_module_from_name,
     is_accelerate_available,
     is_accelerate_version,
@@ -43,9 +42,6 @@ class QuantoQuantizer(DiffusersQuantizer):
         super().__init__(quantization_config, **kwargs)
 
     def validate_environment(self, *args, **kwargs):
-        deprecation_message = "The Quanto quantizer is deprecated and will be removed in version 1.0.0."
-        deprecate("QuantoQuantizer", "1.0.0", deprecation_message)
-
         if not is_optimum_quanto_available():
             raise ImportError(
                 "Loading an optimum-quanto quantized model requires optimum-quanto library (`pip install optimum-quanto`)"

tests/hooks/test_hooks.py

@@ -13,8 +13,8 @@
 # limitations under the License.
 
 import gc
-import unittest
 
+import pytest
 import torch
 
 from diffusers.hooks import HookRegistry, ModelHook
@@ -134,20 +134,18 @@ class SkipLayerHook(ModelHook):
         return output
 
 
-class HookTests(unittest.TestCase):
+class TestHooks:
     in_features = 4
     hidden_features = 8
     out_features = 4
     num_layers = 2
 
-    def setUp(self):
+    def setup_method(self):
         params = self.get_module_parameters()
         self.model = DummyModel(**params)
         self.model.to(torch_device)
 
-    def tearDown(self):
-        super().tearDown()
-
+    def teardown_method(self):
         del self.model
         gc.collect()
         free_memory()
@@ -171,20 +169,20 @@ class HookTests(unittest.TestCase):
         registry_repr = repr(registry)
         expected_repr = "HookRegistry(\n  (0) add_hook - AddHook\n  (1) multiply_hook - MultiplyHook(value=2)\n)"
 
-        self.assertEqual(len(registry.hooks), 2)
-        self.assertEqual(registry._hook_order, ["add_hook", "multiply_hook"])
-        self.assertEqual(registry_repr, expected_repr)
+        assert len(registry.hooks) == 2
+        assert registry._hook_order == ["add_hook", "multiply_hook"]
+        assert registry_repr == expected_repr
 
         registry.remove_hook("add_hook")
 
-        self.assertEqual(len(registry.hooks), 1)
-        self.assertEqual(registry._hook_order, ["multiply_hook"])
+        assert len(registry.hooks) == 1
+        assert registry._hook_order == ["multiply_hook"]
 
     def test_stateful_hook(self):
         registry = HookRegistry.check_if_exists_or_initialize(self.model)
         registry.register_hook(StatefulAddHook(1), "stateful_add_hook")
 
-        self.assertEqual(registry.hooks["stateful_add_hook"].increment, 0)
+        assert registry.hooks["stateful_add_hook"].increment == 0
 
         input = torch.randn(1, 4, device=torch_device, generator=self.get_generator())
         num_repeats = 3
@@ -194,13 +192,13 @@ class HookTests(unittest.TestCase):
             if i == 0:
                 output1 = result
 
-        self.assertEqual(registry.get_hook("stateful_add_hook").increment, num_repeats)
+        assert registry.get_hook("stateful_add_hook").increment == num_repeats
 
         registry.reset_stateful_hooks()
         output2 = self.model(input)
 
-        self.assertEqual(registry.get_hook("stateful_add_hook").increment, 1)
-        self.assertTrue(torch.allclose(output1, output2))
+        assert registry.get_hook("stateful_add_hook").increment == 1
+        assert torch.allclose(output1, output2)
 
     def test_inference(self):
         registry = HookRegistry.check_if_exists_or_initialize(self.model)
@@ -218,9 +216,9 @@ class HookTests(unittest.TestCase):
         new_input = input * 2 + 1
         output3 = self.model(new_input).mean().detach().cpu().item()
 
-        self.assertAlmostEqual(output1, output2, places=5)
-        self.assertAlmostEqual(output1, output3, places=5)
-        self.assertAlmostEqual(output2, output3, places=5)
+        assert output1 == pytest.approx(output2, abs=5e-6)
+        assert output1 == pytest.approx(output3, abs=5e-6)
+        assert output2 == pytest.approx(output3, abs=5e-6)
 
     def test_skip_layer_hook(self):
         registry = HookRegistry.check_if_exists_or_initialize(self.model)
@@ -228,30 +226,29 @@ class HookTests(unittest.TestCase):
 
         input = torch.zeros(1, 4, device=torch_device)
         output = self.model(input).mean().detach().cpu().item()
-        self.assertEqual(output, 0.0)
+        assert output == 0.0
 
         registry.remove_hook("skip_layer_hook")
         registry.register_hook(SkipLayerHook(skip_layer=False), "skip_layer_hook")
         output = self.model(input).mean().detach().cpu().item()
-        self.assertNotEqual(output, 0.0)
+        assert output != 0.0
 
     def test_skip_layer_internal_block(self):
         registry = HookRegistry.check_if_exists_or_initialize(self.model.linear_1)
         input = torch.zeros(1, 4, device=torch_device)
 
         registry.register_hook(SkipLayerHook(skip_layer=True), "skip_layer_hook")
-        with self.assertRaises(RuntimeError) as cm:
+        with pytest.raises(RuntimeError, match="mat1 and mat2 shapes cannot be multiplied"):
             self.model(input).mean().detach().cpu().item()
-        self.assertIn("mat1 and mat2 shapes cannot be multiplied", str(cm.exception))
 
         registry.remove_hook("skip_layer_hook")
         output = self.model(input).mean().detach().cpu().item()
-        self.assertNotEqual(output, 0.0)
+        assert output != 0.0
 
         registry = HookRegistry.check_if_exists_or_initialize(self.model.blocks[1])
         registry.register_hook(SkipLayerHook(skip_layer=True), "skip_layer_hook")
         output = self.model(input).mean().detach().cpu().item()
-        self.assertNotEqual(output, 0.0)
+        assert output != 0.0
 
     def test_invocation_order_stateful_first(self):
         registry = HookRegistry.check_if_exists_or_initialize(self.model)
@@ -278,7 +275,7 @@ class HookTests(unittest.TestCase):
             .replace(" ", "")
             .replace("\n", "")
         )
-        self.assertEqual(output, expected_invocation_order_log)
+        assert output == expected_invocation_order_log
 
         registry.remove_hook("add_hook")
         with CaptureLogger(logger) as cap_logger:
@@ -289,7 +286,7 @@ class HookTests(unittest.TestCase):
             .replace(" ", "")
             .replace("\n", "")
         )
-        self.assertEqual(output, expected_invocation_order_log)
+        assert output == expected_invocation_order_log
 
     def test_invocation_order_stateful_middle(self):
         registry = HookRegistry.check_if_exists_or_initialize(self.model)
@@ -316,7 +313,7 @@ class HookTests(unittest.TestCase):
             .replace(" ", "")
             .replace("\n", "")
         )
-        self.assertEqual(output, expected_invocation_order_log)
+        assert output == expected_invocation_order_log
 
         registry.remove_hook("add_hook")
         with CaptureLogger(logger) as cap_logger:
@@ -327,7 +324,7 @@ class HookTests(unittest.TestCase):
             .replace(" ", "")
             .replace("\n", "")
         )
-        self.assertEqual(output, expected_invocation_order_log)
+        assert output == expected_invocation_order_log
 
         registry.remove_hook("add_hook_2")
         with CaptureLogger(logger) as cap_logger:
@@ -336,7 +333,7 @@ class HookTests(unittest.TestCase):
         expected_invocation_order_log = (
             ("MultiplyHook pre_forward\nMultiplyHook post_forward\n").replace(" ", "").replace("\n", "")
         )
-        self.assertEqual(output, expected_invocation_order_log)
+        assert output == expected_invocation_order_log
 
     def test_invocation_order_stateful_last(self):
         registry = HookRegistry.check_if_exists_or_initialize(self.model)
@@ -363,7 +360,7 @@ class HookTests(unittest.TestCase):
             .replace(" ", "")
             .replace("\n", "")
         )
-        self.assertEqual(output, expected_invocation_order_log)
+        assert output == expected_invocation_order_log
 
         registry.remove_hook("add_hook")
         with CaptureLogger(logger) as cap_logger:
@@ -374,4 +371,4 @@ class HookTests(unittest.TestCase):
             .replace(" ", "")
             .replace("\n", "")
         )
-        self.assertEqual(output, expected_invocation_order_log)
+        assert output == expected_invocation_order_log

tests/models/transformers/test_models_transformer_ltx.py

@@ -1,3 +1,4 @@
+# coding=utf-8
 # Copyright 2025 HuggingFace Inc.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
@@ -12,47 +13,59 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+import unittest
+
 import torch
 
 from diffusers import LTXVideoTransformer3DModel
-from diffusers.utils.torch_utils import randn_tensor
 
 from ...testing_utils import enable_full_determinism, torch_device
-from ..testing_utils import (
-    BaseModelTesterConfig,
-    MemoryTesterMixin,
-    ModelTesterMixin,
-    TorchCompileTesterMixin,
-    TrainingTesterMixin,
-)
+from ..test_modeling_common import ModelTesterMixin, TorchCompileTesterMixin
 
 
 enable_full_determinism()
 
 
-class LTXTransformerTesterConfig(BaseModelTesterConfig):
-    @property
-    def model_class(self):
-        return LTXVideoTransformer3DModel
+class LTXTransformerTests(ModelTesterMixin, unittest.TestCase):
+    model_class = LTXVideoTransformer3DModel
+    main_input_name = "hidden_states"
+    uses_custom_attn_processor = True
 
     @property
-    def output_shape(self) -> tuple[int, int]:
-        return (512, 4)
+    def dummy_input(self):
+        batch_size = 2
+        num_channels = 4
+        num_frames = 2
+        height = 16
+        width = 16
+        embedding_dim = 16
+        sequence_length = 16
 
-    @property
-    def input_shape(self) -> tuple[int, int]:
-        return (512, 4)
+        hidden_states = torch.randn((batch_size, num_frames * height * width, num_channels)).to(torch_device)
+        encoder_hidden_states = torch.randn((batch_size, sequence_length, embedding_dim)).to(torch_device)
+        encoder_attention_mask = torch.ones((batch_size, sequence_length)).bool().to(torch_device)
+        timestep = torch.randint(0, 1000, size=(batch_size,)).to(torch_device)
 
-    @property
-    def main_input_name(self) -> str:
-        return "hidden_states"
-
-    @property
-    def generator(self):
-        return torch.Generator("cpu").manual_seed(0)
-
-    def get_init_dict(self):
         return {
+            "hidden_states": hidden_states,
+            "encoder_hidden_states": encoder_hidden_states,
+            "timestep": timestep,
+            "encoder_attention_mask": encoder_attention_mask,
+            "num_frames": num_frames,
+            "height": height,
+            "width": width,
+        }
+
+    @property
+    def input_shape(self):
+        return (512, 4)
+
+    @property
+    def output_shape(self):
+        return (512, 4)
+
+    def prepare_init_args_and_inputs_for_common(self):
+        init_dict = {
             "in_channels": 4,
             "out_channels": 4,
             "num_attention_heads": 2,
@@ -62,57 +75,16 @@ class LTXTransformerTesterConfig(BaseModelTesterConfig):
             "qk_norm": "rms_norm_across_heads",
             "caption_channels": 16,
         }
-
-    def get_dummy_inputs(self) -> dict[str, torch.Tensor]:
-        batch_size = 2
-        num_channels = 4
-        num_frames = 2
-        height = 16
-        width = 16
-        embedding_dim = 16
-        sequence_length = 16
-
-        return {
-            "hidden_states": randn_tensor(
-                (batch_size, num_frames * height * width, num_channels),
-                generator=self.generator,
-                device=torch_device,
-            ),
-            "encoder_hidden_states": randn_tensor(
-                (batch_size, sequence_length, embedding_dim), generator=self.generator, device=torch_device
-            ),
-            "timestep": torch.randint(0, 1000, size=(batch_size,), generator=self.generator).to(torch_device),
-            "encoder_attention_mask": torch.ones((batch_size, sequence_length)).bool().to(torch_device),
-            "num_frames": num_frames,
-            "height": height,
-            "width": width,
-        }
-
-
-class TestLTXTransformer(LTXTransformerTesterConfig, ModelTesterMixin):
-    """Core model tests for LTX Video Transformer."""
-
-
-class TestLTXTransformerMemory(LTXTransformerTesterConfig, MemoryTesterMixin):
-    """Memory optimization tests for LTX Video Transformer."""
-
-
-class TestLTXTransformerTraining(LTXTransformerTesterConfig, TrainingTesterMixin):
-    """Training tests for LTX Video Transformer."""
+        inputs_dict = self.dummy_input
+        return init_dict, inputs_dict
 
     def test_gradient_checkpointing_is_applied(self):
-        super().test_gradient_checkpointing_is_applied(expected_set={"LTXVideoTransformer3DModel"})
+        expected_set = {"LTXVideoTransformer3DModel"}
+        super().test_gradient_checkpointing_is_applied(expected_set=expected_set)
 
 
-class TestLTXTransformerCompile(LTXTransformerTesterConfig, TorchCompileTesterMixin):
-    """Torch compile tests for LTX Video Transformer."""
+class LTXTransformerCompileTests(TorchCompileTesterMixin, unittest.TestCase):
+    model_class = LTXVideoTransformer3DModel
 
-
-# TODO: Add pretrained_model_name_or_path once a tiny LTX model is available on the Hub
-# class TestLTXTransformerBitsAndBytes(LTXTransformerTesterConfig, BitsAndBytesTesterMixin):
-#     """BitsAndBytes quantization tests for LTX Video Transformer."""
-
-
-# TODO: Add pretrained_model_name_or_path once a tiny LTX model is available on the Hub
-# class TestLTXTransformerTorchAo(LTXTransformerTesterConfig, TorchAoTesterMixin):
-#     """TorchAo quantization tests for LTX Video Transformer."""
+    def prepare_init_args_and_inputs_for_common(self):
+        return LTXTransformerTests().prepare_init_args_and_inputs_for_common()

tests/models/transformers/test_models_transformer_ltx2.py

@@ -1,3 +1,4 @@
+# coding=utf-8
 # Copyright 2025 HuggingFace Inc.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
@@ -12,49 +13,77 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-import pytest
+import unittest
+
 import torch
 
 from diffusers import LTX2VideoTransformer3DModel
-from diffusers.utils.torch_utils import randn_tensor
 
 from ...testing_utils import enable_full_determinism, torch_device
-from ..testing_utils import (
-    AttentionTesterMixin,
-    BaseModelTesterConfig,
-    MemoryTesterMixin,
-    ModelTesterMixin,
-    TorchCompileTesterMixin,
-    TrainingTesterMixin,
-)
+from ..test_modeling_common import ModelTesterMixin, TorchCompileTesterMixin
 
 
 enable_full_determinism()
 
 
-class LTX2TransformerTesterConfig(BaseModelTesterConfig):
-    @property
-    def model_class(self):
-        return LTX2VideoTransformer3DModel
+class LTX2TransformerTests(ModelTesterMixin, unittest.TestCase):
+    model_class = LTX2VideoTransformer3DModel
+    main_input_name = "hidden_states"
+    uses_custom_attn_processor = True
 
     @property
-    def output_shape(self) -> tuple[int, int]:
-        return (512, 4)
+    def dummy_input(self):
+        # Common
+        batch_size = 2
 
-    @property
-    def input_shape(self) -> tuple[int, int]:
-        return (512, 4)
+        # Video
+        num_frames = 2
+        num_channels = 4
+        height = 16
+        width = 16
 
-    @property
-    def main_input_name(self) -> str:
-        return "hidden_states"
+        # Audio
+        audio_num_frames = 9
+        audio_num_channels = 2
+        num_mel_bins = 2
 
-    @property
-    def generator(self):
-        return torch.Generator("cpu").manual_seed(0)
+        # Text
+        embedding_dim = 16
+        sequence_length = 16
+
+        hidden_states = torch.randn((batch_size, num_frames * height * width, num_channels)).to(torch_device)
+        audio_hidden_states = torch.randn((batch_size, audio_num_frames, audio_num_channels * num_mel_bins)).to(
+            torch_device
+        )
+        encoder_hidden_states = torch.randn((batch_size, sequence_length, embedding_dim)).to(torch_device)
+        audio_encoder_hidden_states = torch.randn((batch_size, sequence_length, embedding_dim)).to(torch_device)
+        encoder_attention_mask = torch.ones((batch_size, sequence_length)).bool().to(torch_device)
+        timestep = torch.rand((batch_size,)).to(torch_device) * 1000
 
-    def get_init_dict(self):
         return {
+            "hidden_states": hidden_states,
+            "audio_hidden_states": audio_hidden_states,
+            "encoder_hidden_states": encoder_hidden_states,
+            "audio_encoder_hidden_states": audio_encoder_hidden_states,
+            "timestep": timestep,
+            "encoder_attention_mask": encoder_attention_mask,
+            "num_frames": num_frames,
+            "height": height,
+            "width": width,
+            "audio_num_frames": audio_num_frames,
+            "fps": 25.0,
+        }
+
+    @property
+    def input_shape(self):
+        return (512, 4)
+
+    @property
+    def output_shape(self):
+        return (512, 4)
+
+    def prepare_init_args_and_inputs_for_common(self):
+        init_dict = {
             "in_channels": 4,
             "out_channels": 4,
             "patch_size": 1,
@@ -72,80 +101,122 @@ class LTX2TransformerTesterConfig(BaseModelTesterConfig):
             "caption_channels": 16,
             "rope_double_precision": False,
         }
-
-    def get_dummy_inputs(self) -> dict[str, torch.Tensor]:
-        batch_size = 2
-        num_frames = 2
-        num_channels = 4
-        height = 16
-        width = 16
-        audio_num_frames = 9
-        audio_num_channels = 2
-        num_mel_bins = 2
-        embedding_dim = 16
-        sequence_length = 16
-
-        return {
-            "hidden_states": randn_tensor(
-                (batch_size, num_frames * height * width, num_channels),
-                generator=self.generator,
-                device=torch_device,
-            ),
-            "audio_hidden_states": randn_tensor(
-                (batch_size, audio_num_frames, audio_num_channels * num_mel_bins),
-                generator=self.generator,
-                device=torch_device,
-            ),
-            "encoder_hidden_states": randn_tensor(
-                (batch_size, sequence_length, embedding_dim), generator=self.generator, device=torch_device
-            ),
-            "audio_encoder_hidden_states": randn_tensor(
-                (batch_size, sequence_length, embedding_dim), generator=self.generator, device=torch_device
-            ),
-            "timestep": (randn_tensor((batch_size,), generator=self.generator, device=torch_device).abs() * 1000),
-            "encoder_attention_mask": torch.ones((batch_size, sequence_length)).bool().to(torch_device),
-            "num_frames": num_frames,
-            "height": height,
-            "width": width,
-            "audio_num_frames": audio_num_frames,
-            "fps": 25.0,
-        }
-
-
-class TestLTX2Transformer(LTX2TransformerTesterConfig, ModelTesterMixin):
-    """Core model tests for LTX2 Video Transformer."""
-
-
-class TestLTX2TransformerMemory(LTX2TransformerTesterConfig, MemoryTesterMixin):
-    """Memory optimization tests for LTX2 Video Transformer."""
-
-
-class TestLTX2TransformerTraining(LTX2TransformerTesterConfig, TrainingTesterMixin):
-    """Training tests for LTX2 Video Transformer."""
+        inputs_dict = self.dummy_input
+        return init_dict, inputs_dict
 
     def test_gradient_checkpointing_is_applied(self):
-        super().test_gradient_checkpointing_is_applied(expected_set={"LTX2VideoTransformer3DModel"})
+        expected_set = {"LTX2VideoTransformer3DModel"}
+        super().test_gradient_checkpointing_is_applied(expected_set=expected_set)
+
+    # def test_ltx2_consistency(self, seed=0, dtype=torch.float32):
+    #     torch.manual_seed(seed)
+    #     init_dict, _ = self.prepare_init_args_and_inputs_for_common()
+
+    #     # Calculate dummy inputs in a custom manner to ensure compatibility with original code
+    #     batch_size = 2
+    #     num_frames = 9
+    #     latent_frames = 2
+    #     text_embedding_dim = 16
+    #     text_seq_len = 16
+    #     fps = 25.0
+    #     sampling_rate = 16000.0
+    #     hop_length = 160.0
+
+    #     sigma = torch.rand((1,), generator=torch.manual_seed(seed), dtype=dtype, device="cpu") * 1000
+    #     timestep = (sigma * torch.ones((batch_size,), dtype=dtype, device="cpu")).to(device=torch_device)
+
+    #     num_channels = 4
+    #     latent_height = 4
+    #     latent_width = 4
+    #     hidden_states = torch.randn(
+    #         (batch_size, num_channels, latent_frames, latent_height, latent_width),
+    #         generator=torch.manual_seed(seed),
+    #         dtype=dtype,
+    #         device="cpu",
+    #     )
+    #     # Patchify video latents (with patch_size (1, 1, 1))
+    #     hidden_states = hidden_states.reshape(batch_size, -1, latent_frames, 1, latent_height, 1, latent_width, 1)
+    #     hidden_states = hidden_states.permute(0, 2, 4, 6, 1, 3, 5, 7).flatten(4, 7).flatten(1, 3)
+    #     encoder_hidden_states = torch.randn(
+    #         (batch_size, text_seq_len, text_embedding_dim),
+    #         generator=torch.manual_seed(seed),
+    #         dtype=dtype,
+    #         device="cpu",
+    #     )
+
+    #     audio_num_channels = 2
+    #     num_mel_bins = 2
+    #     latent_length = int((sampling_rate / hop_length / 4) * (num_frames / fps))
+    #     audio_hidden_states = torch.randn(
+    #         (batch_size, audio_num_channels, latent_length, num_mel_bins),
+    #         generator=torch.manual_seed(seed),
+    #         dtype=dtype,
+    #         device="cpu",
+    #     )
+    #     # Patchify audio latents
+    #     audio_hidden_states = audio_hidden_states.transpose(1, 2).flatten(2, 3)
+    #     audio_encoder_hidden_states = torch.randn(
+    #         (batch_size, text_seq_len, text_embedding_dim),
+    #         generator=torch.manual_seed(seed),
+    #         dtype=dtype,
+    #         device="cpu",
+    #     )
+
+    #     inputs_dict = {
+    #         "hidden_states": hidden_states.to(device=torch_device),
+    #         "audio_hidden_states": audio_hidden_states.to(device=torch_device),
+    #         "encoder_hidden_states": encoder_hidden_states.to(device=torch_device),
+    #         "audio_encoder_hidden_states": audio_encoder_hidden_states.to(device=torch_device),
+    #         "timestep": timestep,
+    #         "num_frames": latent_frames,
+    #         "height": latent_height,
+    #         "width": latent_width,
+    #         "audio_num_frames": num_frames,
+    #         "fps": 25.0,
+    #     }
+
+    #     model = self.model_class.from_pretrained(
+    #         "diffusers-internal-dev/dummy-ltx2",
+    #         subfolder="transformer",
+    #         device_map="cpu",
+    #     )
+    #     # torch.manual_seed(seed)
+    #     # model = self.model_class(**init_dict)
+    #     model.to(torch_device)
+    #     model.eval()
+
+    #     with attention_backend("native"):
+    #         with torch.no_grad():
+    #             output = model(**inputs_dict)
+
+    #             video_output, audio_output = output.to_tuple()
+
+    #     self.assertIsNotNone(video_output)
+    #     self.assertIsNotNone(audio_output)
+
+    #     # input & output have to have the same shape
+    #     video_expected_shape = (batch_size, latent_frames * latent_height * latent_width, num_channels)
+    #     self.assertEqual(video_output.shape, video_expected_shape, "Video input and output shapes do not match")
+    #     audio_expected_shape = (batch_size, latent_length, audio_num_channels * num_mel_bins)
+    #     self.assertEqual(audio_output.shape, audio_expected_shape, "Audio input and output shapes do not match")
+
+    #     # Check against expected slice
+    #     # fmt: off
+    #     video_expected_slice = torch.tensor([0.4783, 1.6954, -1.2092, 0.1762, 0.7801, 1.2025, -1.4525, -0.2721, 0.3354, 1.9144, -1.5546, 0.0831, 0.4391, 1.7012, -1.7373, -0.2676])
+    #     audio_expected_slice = torch.tensor([-0.4236, 0.4750, 0.3901, -0.4339, -0.2782, 0.4357, 0.4526, -0.3927, -0.0980, 0.4870, 0.3964, -0.3169, -0.3974, 0.4408, 0.3809, -0.4692])
+    #     # fmt: on
+
+    #     video_output_flat = video_output.cpu().flatten().float()
+    #     video_generated_slice = torch.cat([video_output_flat[:8], video_output_flat[-8:]])
+    #     self.assertTrue(torch.allclose(video_generated_slice, video_expected_slice, atol=1e-4))
+
+    #     audio_output_flat = audio_output.cpu().flatten().float()
+    #     audio_generated_slice = torch.cat([audio_output_flat[:8], audio_output_flat[-8:]])
+    #     self.assertTrue(torch.allclose(audio_generated_slice, audio_expected_slice, atol=1e-4))
 
 
-class TestLTX2TransformerAttention(LTX2TransformerTesterConfig, AttentionTesterMixin):
-    """Attention processor tests for LTX2 Video Transformer."""
+class LTX2TransformerCompileTests(TorchCompileTesterMixin, unittest.TestCase):
+    model_class = LTX2VideoTransformer3DModel
 
-    @pytest.mark.skip(
-        "LTX2Attention does not set is_cross_attention, so fuse_projections tries to fuse Q+K+V together even for cross-attention modules with different input dimensions."
-    )
-    def test_fuse_unfuse_qkv_projections(self, atol=1e-3, rtol=0):
-        pass
-
-
-class TestLTX2TransformerCompile(LTX2TransformerTesterConfig, TorchCompileTesterMixin):
-    """Torch compile tests for LTX2 Video Transformer."""
-
-
-# TODO: Add pretrained_model_name_or_path once a tiny LTX2 model is available on the Hub
-# class TestLTX2TransformerBitsAndBytes(LTX2TransformerTesterConfig, BitsAndBytesTesterMixin):
-#     """BitsAndBytes quantization tests for LTX2 Video Transformer."""
-
-
-# TODO: Add pretrained_model_name_or_path once a tiny LTX2 model is available on the Hub
-# class TestLTX2TransformerTorchAo(LTX2TransformerTesterConfig, TorchAoTesterMixin):
-#     """TorchAo quantization tests for LTX2 Video Transformer."""
+    def prepare_init_args_and_inputs_for_common(self):
+        return LTX2TransformerTests().prepare_init_args_and_inputs_for_common()