update

src/diffusers/pipelines/cosmos/pipeline_cosmos2_5_predict.py

@@ -16,22 +16,29 @@ from typing import Callable
 
 import numpy as np
 import torch
-import torchvision
-import torchvision.transforms
-import torchvision.transforms.functional
 from transformers import AutoTokenizer, Qwen2_5_VLForConditionalGeneration
 
 from ...callbacks import MultiPipelineCallbacks, PipelineCallback
 from ...image_processor import PipelineImageInput
 from ...models import AutoencoderKLWan, CosmosTransformer3DModel
 from ...schedulers import UniPCMultistepScheduler
-from ...utils import is_cosmos_guardrail_available, is_torch_xla_available, logging, replace_example_docstring
+from ...utils import (
+    is_cosmos_guardrail_available,
+    is_torch_xla_available,
+    is_torchvision_available,
+    logging,
+    replace_example_docstring,
+)
 from ...utils.torch_utils import randn_tensor
 from ...video_processor import VideoProcessor
 from ..pipeline_utils import DiffusionPipeline
 from .pipeline_output import CosmosPipelineOutput
 
 
+if is_torchvision_available():
+    import torchvision.transforms.functional
+
+
 if is_cosmos_guardrail_available():
     from cosmos_guardrail import CosmosSafetyChecker
 else:

tests/models/transformers/test_models_transformer_sana.py

@@ -1,4 +1,3 @@
-# coding=utf-8
 # Copyright 2025 HuggingFace Inc.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
@@ -13,58 +12,57 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+import unittest
+
 import torch
 
 from diffusers import SanaTransformer2DModel
-from diffusers.utils.torch_utils import randn_tensor
 
-from ...testing_utils import enable_full_determinism, torch_device
-from ..testing_utils import (
-    AttentionTesterMixin,
-    BaseModelTesterConfig,
-    BitsAndBytesTesterMixin,
-    MemoryTesterMixin,
-    ModelTesterMixin,
-    TorchAoTesterMixin,
-    TorchCompileTesterMixin,
-    TrainingTesterMixin,
+from ...testing_utils import (
+    enable_full_determinism,
+    torch_device,
 )
+from ..test_modeling_common import ModelTesterMixin
 
 
 enable_full_determinism()
 
 
-class SanaTransformer2DTesterConfig(BaseModelTesterConfig):
-    @property
-    def model_class(self):
-        return SanaTransformer2DModel
+class SanaTransformerTests(ModelTesterMixin, unittest.TestCase):
+    model_class = SanaTransformer2DModel
+    main_input_name = "hidden_states"
+    uses_custom_attn_processor = True
+    model_split_percents = [0.7, 0.7, 0.9]
 
     @property
-    def output_shape(self) -> tuple[int, ...]:
-        return (4, 32, 32)
+    def dummy_input(self):
+        batch_size = 2
+        num_channels = 4
+        height = 32
+        width = 32
+        embedding_dim = 8
+        sequence_length = 8
 
-    @property
-    def input_shape(self) -> tuple[int, ...]:
-        return (4, 32, 32)
+        hidden_states = torch.randn((batch_size, num_channels, height, width)).to(torch_device)
+        encoder_hidden_states = torch.randn((batch_size, sequence_length, embedding_dim)).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 uses_custom_attn_processor(self) -> bool:
-        return True
-
-    @property
-    def model_split_percents(self) -> list:
-        return [0.7, 0.7, 0.9]
-
-    @property
-    def generator(self):
-        return torch.Generator("cpu").manual_seed(0)
-
-    def get_init_dict(self) -> dict[str, int | list[int] | tuple | str | bool]:
         return {
+            "hidden_states": hidden_states,
+            "encoder_hidden_states": encoder_hidden_states,
+            "timestep": timestep,
+        }
+
+    @property
+    def input_shape(self):
+        return (4, 32, 32)
+
+    @property
+    def output_shape(self):
+        return (4, 32, 32)
+
+    def prepare_init_args_and_inputs_for_common(self):
+        init_dict = {
             "patch_size": 1,
             "in_channels": 4,
             "out_channels": 4,
@@ -77,53 +75,9 @@ class SanaTransformer2DTesterConfig(BaseModelTesterConfig):
             "caption_channels": 8,
             "sample_size": 32,
         }
-
-    def get_dummy_inputs(self) -> dict[str, torch.Tensor]:
-        batch_size = 2
-        num_channels = 4
-        height = 32
-        width = 32
-        embedding_dim = 8
-        sequence_length = 8
-
-        return {
-            "hidden_states": randn_tensor(
-                (batch_size, num_channels, height, width), 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,)).to(torch_device),
-        }
-
-
-class TestSanaTransformer2D(SanaTransformer2DTesterConfig, ModelTesterMixin):
-    """Core model tests for Sana Transformer 2D."""
-
-
-class TestSanaTransformer2DMemory(SanaTransformer2DTesterConfig, MemoryTesterMixin):
-    """Memory optimization tests for Sana Transformer 2D."""
-
-
-class TestSanaTransformer2DTraining(SanaTransformer2DTesterConfig, TrainingTesterMixin):
-    """Training tests for Sana Transformer 2D."""
+        inputs_dict = self.dummy_input
+        return init_dict, inputs_dict
 
     def test_gradient_checkpointing_is_applied(self):
         expected_set = {"SanaTransformer2DModel"}
         super().test_gradient_checkpointing_is_applied(expected_set=expected_set)
-
-
-class TestSanaTransformer2DAttention(SanaTransformer2DTesterConfig, AttentionTesterMixin):
-    """Attention processor tests for Sana Transformer 2D."""
-
-
-class TestSanaTransformer2DCompile(SanaTransformer2DTesterConfig, TorchCompileTesterMixin):
-    """Torch compile tests for Sana Transformer 2D."""
-
-
-class TestSanaTransformer2DBitsAndBytes(SanaTransformer2DTesterConfig, BitsAndBytesTesterMixin):
-    """BitsAndBytes quantization tests for Sana Transformer 2D."""
-
-
-class TestSanaTransformer2DTorchAo(SanaTransformer2DTesterConfig, TorchAoTesterMixin):
-    """TorchAO quantization tests for Sana Transformer 2D."""

tests/models/transformers/test_models_transformer_sana_video.py

@@ -1,4 +1,3 @@
-# coding=utf-8
 # Copyright 2025 HuggingFace Inc.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
@@ -13,54 +12,57 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+import unittest
+
 import torch
 
 from diffusers import SanaVideoTransformer3DModel
-from diffusers.utils.torch_utils import randn_tensor
 
-from ...testing_utils import enable_full_determinism, torch_device
-from ..testing_utils import (
-    AttentionTesterMixin,
-    BaseModelTesterConfig,
-    BitsAndBytesTesterMixin,
-    MemoryTesterMixin,
-    ModelTesterMixin,
-    TorchAoTesterMixin,
-    TorchCompileTesterMixin,
-    TrainingTesterMixin,
+from ...testing_utils import (
+    enable_full_determinism,
+    torch_device,
 )
+from ..test_modeling_common import ModelTesterMixin, TorchCompileTesterMixin
 
 
 enable_full_determinism()
 
 
-class SanaVideoTransformer3DTesterConfig(BaseModelTesterConfig):
-    @property
-    def model_class(self):
-        return SanaVideoTransformer3DModel
+class SanaVideoTransformer3DTests(ModelTesterMixin, unittest.TestCase):
+    model_class = SanaVideoTransformer3DModel
+    main_input_name = "hidden_states"
+    uses_custom_attn_processor = True
 
     @property
-    def output_shape(self) -> tuple[int, ...]:
-        return (16, 2, 16, 16)
+    def dummy_input(self):
+        batch_size = 1
+        num_channels = 16
+        num_frames = 2
+        height = 16
+        width = 16
+        text_encoder_embedding_dim = 16
+        sequence_length = 12
 
-    @property
-    def input_shape(self) -> tuple[int, ...]:
-        return (16, 2, 16, 16)
+        hidden_states = torch.randn((batch_size, num_channels, num_frames, height, width)).to(torch_device)
+        timestep = torch.randint(0, 1000, size=(batch_size,)).to(torch_device)
+        encoder_hidden_states = torch.randn((batch_size, sequence_length, text_encoder_embedding_dim)).to(torch_device)
 
-    @property
-    def main_input_name(self) -> str:
-        return "hidden_states"
-
-    @property
-    def uses_custom_attn_processor(self) -> bool:
-        return True
-
-    @property
-    def generator(self):
-        return torch.Generator("cpu").manual_seed(0)
-
-    def get_init_dict(self) -> dict[str, int | float | list[int] | tuple | str | bool]:
         return {
+            "hidden_states": hidden_states,
+            "encoder_hidden_states": encoder_hidden_states,
+            "timestep": timestep,
+        }
+
+    @property
+    def input_shape(self):
+        return (16, 2, 16, 16)
+
+    @property
+    def output_shape(self):
+        return (16, 2, 16, 16)
+
+    def prepare_init_args_and_inputs_for_common(self):
+        init_dict = {
             "in_channels": 16,
             "out_channels": 16,
             "num_attention_heads": 2,
@@ -80,56 +82,16 @@ class SanaVideoTransformer3DTesterConfig(BaseModelTesterConfig):
             "qk_norm": "rms_norm_across_heads",
             "rope_max_seq_len": 32,
         }
-
-    def get_dummy_inputs(self) -> dict[str, torch.Tensor]:
-        batch_size = 1
-        num_channels = 16
-        num_frames = 2
-        height = 16
-        width = 16
-        text_encoder_embedding_dim = 16
-        sequence_length = 12
-
-        return {
-            "hidden_states": randn_tensor(
-                (batch_size, num_channels, num_frames, height, width), generator=self.generator, device=torch_device
-            ),
-            "encoder_hidden_states": randn_tensor(
-                (batch_size, sequence_length, text_encoder_embedding_dim),
-                generator=self.generator,
-                device=torch_device,
-            ),
-            "timestep": torch.randint(0, 1000, size=(batch_size,)).to(torch_device),
-        }
-
-
-class TestSanaVideoTransformer3D(SanaVideoTransformer3DTesterConfig, ModelTesterMixin):
-    """Core model tests for Sana Video Transformer 3D."""
-
-
-class TestSanaVideoTransformer3DMemory(SanaVideoTransformer3DTesterConfig, MemoryTesterMixin):
-    """Memory optimization tests for Sana Video Transformer 3D."""
-
-
-class TestSanaVideoTransformer3DTraining(SanaVideoTransformer3DTesterConfig, TrainingTesterMixin):
-    """Training tests for Sana Video Transformer 3D."""
+        inputs_dict = self.dummy_input
+        return init_dict, inputs_dict
 
     def test_gradient_checkpointing_is_applied(self):
         expected_set = {"SanaVideoTransformer3DModel"}
         super().test_gradient_checkpointing_is_applied(expected_set=expected_set)
 
 
-class TestSanaVideoTransformer3DAttention(SanaVideoTransformer3DTesterConfig, AttentionTesterMixin):
-    """Attention processor tests for Sana Video Transformer 3D."""
+class SanaVideoTransformerCompileTests(TorchCompileTesterMixin, unittest.TestCase):
+    model_class = SanaVideoTransformer3DModel
 
-
-class TestSanaVideoTransformer3DCompile(SanaVideoTransformer3DTesterConfig, TorchCompileTesterMixin):
-    """Torch compile tests for Sana Video Transformer 3D."""
-
-
-class TestSanaVideoTransformer3DBitsAndBytes(SanaVideoTransformer3DTesterConfig, BitsAndBytesTesterMixin):
-    """BitsAndBytes quantization tests for Sana Video Transformer 3D."""
-
-
-class TestSanaVideoTransformer3DTorchAo(SanaVideoTransformer3DTesterConfig, TorchAoTesterMixin):
-    """TorchAO quantization tests for Sana Video Transformer 3D."""
+    def prepare_init_args_and_inputs_for_common(self):
+        return SanaVideoTransformer3DTests().prepare_init_args_and_inputs_for_common()