update

src/diffusers/pipelines/pipeline_utils.py

@@ -764,7 +764,7 @@ class DiffusionPipeline(ConfigMixin, PushToHubMixin):
         token = kwargs.pop("token", None)
         revision = kwargs.pop("revision", None)
         from_flax = kwargs.pop("from_flax", False)
-        torch_dtype = kwargs.pop("torch_dtype", None)
+        torch_dtype = kwargs.pop("torch_dtype", torch.float32)
         custom_pipeline = kwargs.pop("custom_pipeline", None)
         custom_revision = kwargs.pop("custom_revision", None)
         provider = kwargs.pop("provider", None)

src/diffusers/quantizers/quantization_config.py

@@ -470,8 +470,8 @@ class TorchAoConfig(QuantizationConfigMixin):
         self.post_init()
 
     def post_init(self):
-        if is_torchao_version("<=", "0.9.0"):
-            raise ValueError("TorchAoConfig requires torchao > 0.9.0. Please upgrade with `pip install -U torchao`.")
+        if is_torchao_version("<", "0.15.0"):
+            raise ValueError("TorchAoConfig requires torchao >= 0.15.0. Please upgrade with `pip install -U torchao`.")
 
         from torchao.quantization.quant_api import AOBaseConfig
 
@@ -495,8 +495,8 @@ class TorchAoConfig(QuantizationConfigMixin):
     @classmethod
     def from_dict(cls, config_dict, return_unused_kwargs=False, **kwargs):
         """Create configuration from a dictionary."""
-        if not is_torchao_version(">", "0.9.0"):
-            raise NotImplementedError("TorchAoConfig requires torchao > 0.9.0 for construction from dict")
+        if not is_torchao_version(">=", "0.15.0"):
+            raise NotImplementedError("TorchAoConfig requires torchao >= 0.15.0 for construction from dict")
         config_dict = config_dict.copy()
         quant_type = config_dict.pop("quant_type")
 

src/diffusers/quantizers/torchao/torchao_quantizer.py

@@ -113,7 +113,7 @@ if (
     is_torch_available()
     and is_torch_version(">=", "2.6.0")
     and is_torchao_available()
-    and is_torchao_version(">=", "0.7.0")
+    and is_torchao_version(">=", "0.15.0")
 ):
     _update_torch_safe_globals()
 
@@ -168,10 +168,10 @@ class TorchAoHfQuantizer(DiffusersQuantizer):
             raise ImportError(
                 "Loading a TorchAO quantized model requires the torchao library. Please install with `pip install torchao`"
             )
-        torchao_version = version.parse(importlib.metadata.version("torch"))
-        if torchao_version < version.parse("0.7.0"):
+        torchao_version = version.parse(importlib.metadata.version("torchao"))
+        if torchao_version < version.parse("0.15.0"):
             raise RuntimeError(
-                f"The minimum required version of `torchao` is 0.7.0, but the current version is {torchao_version}. Please upgrade with `pip install -U torchao`."
+                f"The minimum required version of `torchao` is 0.15.0, but the current version is {torchao_version}. Please upgrade with `pip install -U torchao`."
             )
 
         self.offload = False

tests/models/transformers/test_models_transformer_bria.py

@@ -13,31 +13,23 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-from typing import Any
+import unittest
 
 import torch
 
 from diffusers import BriaTransformer2DModel
 from diffusers.models.attention_processor import FluxIPAdapterJointAttnProcessor2_0
 from diffusers.models.embeddings import ImageProjection
-from diffusers.utils.torch_utils import randn_tensor
 
 from ...testing_utils import enable_full_determinism, torch_device
-from ..testing_utils import (
-    BaseModelTesterConfig,
-    IPAdapterTesterMixin,
-    LoraHotSwappingForModelTesterMixin,
-    LoraTesterMixin,
-    ModelTesterMixin,
-    TorchCompileTesterMixin,
-    TrainingTesterMixin,
-)
+from ..test_modeling_common import LoraHotSwappingForModelTesterMixin, ModelTesterMixin, TorchCompileTesterMixin
 
 
 enable_full_determinism()
 
 
-def create_bria_ip_adapter_state_dict(model) -> dict[str, dict[str, Any]]:
+def create_bria_ip_adapter_state_dict(model):
+    # "ip_adapter" (cross-attention weights)
     ip_cross_attn_state_dict = {}
     key_id = 0
 
@@ -58,8 +50,11 @@ def create_bria_ip_adapter_state_dict(model) -> dict[str, dict[str, Any]]:
                 f"{key_id}.to_v_ip.bias": sd["to_v_ip.0.bias"],
             }
         )
+
         key_id += 1
 
+    # "image_proj" (ImageProjection layer weights)
+
     image_projection = ImageProjection(
         cross_attention_dim=model.config["joint_attention_dim"],
         image_embed_dim=model.config["pooled_projection_dim"],
@@ -78,36 +73,53 @@ def create_bria_ip_adapter_state_dict(model) -> dict[str, dict[str, Any]]:
     )
 
     del sd
-    return {"image_proj": ip_image_projection_state_dict, "ip_adapter": ip_cross_attn_state_dict}
+    ip_state_dict = {}
+    ip_state_dict.update({"image_proj": ip_image_projection_state_dict, "ip_adapter": ip_cross_attn_state_dict})
+    return ip_state_dict
 
 
-class BriaTransformerTesterConfig(BaseModelTesterConfig):
-    @property
-    def model_class(self):
-        return BriaTransformer2DModel
+class BriaTransformerTests(ModelTesterMixin, unittest.TestCase):
+    model_class = BriaTransformer2DModel
+    main_input_name = "hidden_states"
+    # We override the items here because the transformer under consideration is small.
+    model_split_percents = [0.8, 0.7, 0.7]
+
+    # Skip setting testing with default: AttnProcessor
+    uses_custom_attn_processor = True
 
     @property
-    def main_input_name(self) -> str:
-        return "hidden_states"
+    def dummy_input(self):
+        batch_size = 1
+        num_latent_channels = 4
+        num_image_channels = 3
+        height = width = 4
+        sequence_length = 48
+        embedding_dim = 32
 
-    @property
-    def model_split_percents(self) -> list:
-        return [0.8, 0.7, 0.7]
+        hidden_states = torch.randn((batch_size, height * width, num_latent_channels)).to(torch_device)
+        encoder_hidden_states = torch.randn((batch_size, sequence_length, embedding_dim)).to(torch_device)
+        text_ids = torch.randn((sequence_length, num_image_channels)).to(torch_device)
+        image_ids = torch.randn((height * width, num_image_channels)).to(torch_device)
+        timestep = torch.tensor([1.0]).to(torch_device).expand(batch_size)
 
-    @property
-    def output_shape(self) -> tuple:
-        return (16, 4)
-
-    @property
-    def input_shape(self) -> tuple:
-        return (16, 4)
-
-    @property
-    def generator(self):
-        return torch.Generator("cpu").manual_seed(0)
-
-    def get_init_dict(self) -> dict:
         return {
+            "hidden_states": hidden_states,
+            "encoder_hidden_states": encoder_hidden_states,
+            "img_ids": image_ids,
+            "txt_ids": text_ids,
+            "timestep": timestep,
+        }
+
+    @property
+    def input_shape(self):
+        return (16, 4)
+
+    @property
+    def output_shape(self):
+        return (16, 4)
+
+    def prepare_init_args_and_inputs_for_common(self):
+        init_dict = {
             "patch_size": 1,
             "in_channels": 4,
             "num_layers": 1,
@@ -119,35 +131,11 @@ class BriaTransformerTesterConfig(BaseModelTesterConfig):
             "axes_dims_rope": [0, 4, 4],
         }
 
-    def get_dummy_inputs(self, batch_size: int = 1) -> dict[str, torch.Tensor]:
-        num_latent_channels = 4
-        num_image_channels = 3
-        height = width = 4
-        sequence_length = 48
-        embedding_dim = 32
+        inputs_dict = self.dummy_input
+        return init_dict, inputs_dict
 
-        return {
-            "hidden_states": randn_tensor(
-                (batch_size, height * width, num_latent_channels), generator=self.generator, device=torch_device
-            ),
-            "encoder_hidden_states": randn_tensor(
-                (batch_size, sequence_length, embedding_dim), generator=self.generator, device=torch_device
-            ),
-            "img_ids": randn_tensor(
-                (height * width, num_image_channels), generator=self.generator, device=torch_device
-            ),
-            "txt_ids": randn_tensor(
-                (sequence_length, num_image_channels), generator=self.generator, device=torch_device
-            ),
-            "timestep": torch.tensor([1.0]).to(torch_device).expand(batch_size),
-        }
-
-
-class TestBriaTransformer(BriaTransformerTesterConfig, ModelTesterMixin):
     def test_deprecated_inputs_img_txt_ids_3d(self):
-        init_dict = self.get_init_dict()
-        inputs_dict = self.get_dummy_inputs()
-
+        init_dict, inputs_dict = self.prepare_init_args_and_inputs_for_common()
         model = self.model_class(**init_dict)
         model.to(torch_device)
         model.eval()
@@ -155,6 +143,7 @@ class TestBriaTransformer(BriaTransformerTesterConfig, ModelTesterMixin):
         with torch.no_grad():
             output_1 = model(**inputs_dict).to_tuple()[0]
 
+        # update inputs_dict with txt_ids and img_ids as 3d tensors (deprecated)
         text_ids_3d = inputs_dict["txt_ids"].unsqueeze(0)
         image_ids_3d = inputs_dict["img_ids"].unsqueeze(0)
 
@@ -167,59 +156,26 @@ class TestBriaTransformer(BriaTransformerTesterConfig, ModelTesterMixin):
         with torch.no_grad():
             output_2 = model(**inputs_dict).to_tuple()[0]
 
-        assert output_1.shape == output_2.shape
-        assert torch.allclose(output_1, output_2, atol=1e-5), (
-            "output with deprecated inputs (img_ids and txt_ids as 3d torch tensors) "
-            "are not equal as them as 2d inputs"
+        self.assertEqual(output_1.shape, output_2.shape)
+        self.assertTrue(
+            torch.allclose(output_1, output_2, atol=1e-5),
+            msg="output with deprecated inputs (img_ids and txt_ids as 3d torch tensors) are not equal as them as 2d inputs",
         )
 
-
-class TestBriaTransformerTraining(BriaTransformerTesterConfig, TrainingTesterMixin):
     def test_gradient_checkpointing_is_applied(self):
         expected_set = {"BriaTransformer2DModel"}
         super().test_gradient_checkpointing_is_applied(expected_set=expected_set)
 
 
-class TestBriaTransformerCompile(BriaTransformerTesterConfig, TorchCompileTesterMixin):
-    pass
+class BriaTransformerCompileTests(TorchCompileTesterMixin, unittest.TestCase):
+    model_class = BriaTransformer2DModel
+
+    def prepare_init_args_and_inputs_for_common(self):
+        return BriaTransformerTests().prepare_init_args_and_inputs_for_common()
 
 
-class TestBriaTransformerIPAdapter(BriaTransformerTesterConfig, IPAdapterTesterMixin):
-    @property
-    def ip_adapter_processor_cls(self):
-        return FluxIPAdapterJointAttnProcessor2_0
+class BriaTransformerLoRAHotSwapTests(LoraHotSwappingForModelTesterMixin, unittest.TestCase):
+    model_class = BriaTransformer2DModel
 
-    def modify_inputs_for_ip_adapter(self, model, inputs_dict):
-        torch.manual_seed(0)
-        cross_attention_dim = getattr(model.config, "joint_attention_dim", 32)
-        image_embeds = torch.randn(1, 1, cross_attention_dim).to(torch_device)
-        inputs_dict.update({"joint_attention_kwargs": {"ip_adapter_image_embeds": image_embeds}})
-        return inputs_dict
-
-    def create_ip_adapter_state_dict(self, model: Any) -> dict[str, dict[str, Any]]:
-        return create_bria_ip_adapter_state_dict(model)
-
-
-class TestBriaTransformerLoRA(BriaTransformerTesterConfig, LoraTesterMixin):
-    pass
-
-
-class TestBriaTransformerLoRAHotSwap(BriaTransformerTesterConfig, LoraHotSwappingForModelTesterMixin):
-    @property
-    def different_shapes_for_compilation(self):
-        return [(4, 4), (4, 8), (8, 8)]
-
-    def get_dummy_inputs(self, height: int = 4, width: int = 4) -> dict[str, torch.Tensor]:
-        batch_size = 1
-        num_latent_channels = 4
-        num_image_channels = 3
-        sequence_length = 24
-        embedding_dim = 32
-
-        return {
-            "hidden_states": randn_tensor((batch_size, height * width, num_latent_channels), device=torch_device),
-            "encoder_hidden_states": randn_tensor((batch_size, sequence_length, embedding_dim), device=torch_device),
-            "img_ids": randn_tensor((height * width, num_image_channels), device=torch_device),
-            "txt_ids": randn_tensor((sequence_length, num_image_channels), device=torch_device),
-            "timestep": torch.tensor([1.0]).to(torch_device).expand(batch_size),
-        }
+    def prepare_init_args_and_inputs_for_common(self):
+        return BriaTransformerTests().prepare_init_args_and_inputs_for_common()

tests/models/transformers/test_models_transformer_bria_fibo.py

@@ -13,50 +13,62 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+import unittest
+
 import torch
 
 from diffusers import BriaFiboTransformer2DModel
-from diffusers.utils.torch_utils import randn_tensor
 
 from ...testing_utils import enable_full_determinism, torch_device
-from ..testing_utils import (
-    BaseModelTesterConfig,
-    ModelTesterMixin,
-    TorchCompileTesterMixin,
-    TrainingTesterMixin,
-)
+from ..test_modeling_common import ModelTesterMixin
 
 
 enable_full_determinism()
 
 
-class BriaFiboTransformerTesterConfig(BaseModelTesterConfig):
-    @property
-    def model_class(self):
-        return BriaFiboTransformer2DModel
+class BriaFiboTransformerTests(ModelTesterMixin, unittest.TestCase):
+    model_class = BriaFiboTransformer2DModel
+    main_input_name = "hidden_states"
+    # We override the items here because the transformer under consideration is small.
+    model_split_percents = [0.8, 0.7, 0.7]
+
+    # Skip setting testing with default: AttnProcessor
+    uses_custom_attn_processor = True
 
     @property
-    def main_input_name(self) -> str:
-        return "hidden_states"
+    def dummy_input(self):
+        batch_size = 1
+        num_latent_channels = 48
+        num_image_channels = 3
+        height = width = 16
+        sequence_length = 32
+        embedding_dim = 64
+
+        hidden_states = torch.randn((batch_size, height * width, num_latent_channels)).to(torch_device)
+        encoder_hidden_states = torch.randn((batch_size, sequence_length, embedding_dim)).to(torch_device)
+        text_ids = torch.randn((sequence_length, num_image_channels)).to(torch_device)
+        image_ids = torch.randn((height * width, num_image_channels)).to(torch_device)
+        timestep = torch.tensor([1.0]).to(torch_device).expand(batch_size)
+
+        return {
+            "hidden_states": hidden_states,
+            "encoder_hidden_states": encoder_hidden_states,
+            "img_ids": image_ids,
+            "txt_ids": text_ids,
+            "timestep": timestep,
+            "text_encoder_layers": [encoder_hidden_states[:, :, :32], encoder_hidden_states[:, :, :32]],
+        }
 
     @property
-    def model_split_percents(self) -> list:
-        return [0.8, 0.7, 0.7]
-
-    @property
-    def output_shape(self) -> tuple:
-        return (256, 48)
-
-    @property
-    def input_shape(self) -> tuple:
+    def input_shape(self):
         return (16, 16)
 
     @property
-    def generator(self):
-        return torch.Generator("cpu").manual_seed(0)
+    def output_shape(self):
+        return (256, 48)
 
-    def get_init_dict(self) -> dict:
-        return {
+    def prepare_init_args_and_inputs_for_common(self):
+        init_dict = {
             "patch_size": 1,
             "in_channels": 48,
             "num_layers": 1,
@@ -69,41 +81,9 @@ class BriaFiboTransformerTesterConfig(BaseModelTesterConfig):
             "axes_dims_rope": [0, 4, 4],
         }
 
-    def get_dummy_inputs(self, batch_size: int = 1) -> dict[str, torch.Tensor]:
-        num_latent_channels = 48
-        num_image_channels = 3
-        height = width = 16
-        sequence_length = 32
-        embedding_dim = 64
+        inputs_dict = self.dummy_input
+        return init_dict, inputs_dict
 
-        encoder_hidden_states = randn_tensor(
-            (batch_size, sequence_length, embedding_dim), generator=self.generator, device=torch_device
-        )
-        return {
-            "hidden_states": randn_tensor(
-                (batch_size, height * width, num_latent_channels), generator=self.generator, device=torch_device
-            ),
-            "encoder_hidden_states": encoder_hidden_states,
-            "img_ids": randn_tensor(
-                (height * width, num_image_channels), generator=self.generator, device=torch_device
-            ),
-            "txt_ids": randn_tensor(
-                (sequence_length, num_image_channels), generator=self.generator, device=torch_device
-            ),
-            "timestep": torch.tensor([1.0]).to(torch_device).expand(batch_size),
-            "text_encoder_layers": [encoder_hidden_states[:, :, :32], encoder_hidden_states[:, :, :32]],
-        }
-
-
-class TestBriaFiboTransformer(BriaFiboTransformerTesterConfig, ModelTesterMixin):
-    pass
-
-
-class TestBriaFiboTransformerTraining(BriaFiboTransformerTesterConfig, TrainingTesterMixin):
     def test_gradient_checkpointing_is_applied(self):
         expected_set = {"BriaFiboTransformer2DModel"}
         super().test_gradient_checkpointing_is_applied(expected_set=expected_set)
-
-
-class TestBriaFiboTransformerCompile(BriaFiboTransformerTesterConfig, TorchCompileTesterMixin):
-    pass

tests/quantization/torchao/test_torchao.py

@@ -14,13 +14,11 @@
 # limitations under the License.
 
 import gc
-import importlib.metadata
 import tempfile
 import unittest
 from typing import List
 
 import numpy as np
-from packaging import version
 from parameterized import parameterized
 from transformers import AutoTokenizer, CLIPTextModel, CLIPTokenizer, T5EncoderModel
 
@@ -82,18 +80,17 @@ if is_torchao_available():
         Float8WeightOnlyConfig,
         Int4WeightOnlyConfig,
         Int8DynamicActivationInt8WeightConfig,
+        Int8DynamicActivationIntxWeightConfig,
         Int8WeightOnlyConfig,
+        IntxWeightOnlyConfig,
     )
     from torchao.quantization.linear_activation_quantized_tensor import LinearActivationQuantizedTensor
     from torchao.utils import get_model_size_in_bytes
 
-    if version.parse(importlib.metadata.version("torchao")) >= version.Version("0.10.0"):
-        from torchao.quantization import Int8DynamicActivationIntxWeightConfig, IntxWeightOnlyConfig
-
 
 @require_torch
 @require_torch_accelerator
-@require_torchao_version_greater_or_equal("0.14.0")
+@require_torchao_version_greater_or_equal("0.15.0")
 class TorchAoConfigTest(unittest.TestCase):
     def test_to_dict(self):
         """
@@ -128,7 +125,7 @@ class TorchAoConfigTest(unittest.TestCase):
 # Slices for these tests have been obtained on our aws-g6e-xlarge-plus runners
 @require_torch
 @require_torch_accelerator
-@require_torchao_version_greater_or_equal("0.14.0")
+@require_torchao_version_greater_or_equal("0.15.0")
 class TorchAoTest(unittest.TestCase):
     def tearDown(self):
         gc.collect()
@@ -527,7 +524,7 @@ class TorchAoTest(unittest.TestCase):
         inputs = self.get_dummy_inputs(torch_device)
         _ = pipe(**inputs)
 
-    @require_torchao_version_greater_or_equal("0.9.0")
+    @require_torchao_version_greater_or_equal("0.15.0")
     def test_aobase_config(self):
         quantization_config = TorchAoConfig(Int8WeightOnlyConfig())
         components = self.get_dummy_components(quantization_config)
@@ -540,7 +537,7 @@ class TorchAoTest(unittest.TestCase):
 # Slices for these tests have been obtained on our aws-g6e-xlarge-plus runners
 @require_torch
 @require_torch_accelerator
-@require_torchao_version_greater_or_equal("0.14.0")
+@require_torchao_version_greater_or_equal("0.15.0")
 class TorchAoSerializationTest(unittest.TestCase):
     model_name = "hf-internal-testing/tiny-flux-pipe"
 
@@ -650,7 +647,7 @@ class TorchAoSerializationTest(unittest.TestCase):
         self._check_serialization_expected_slice(quant_type, expected_slice, device)
 
 
-@require_torchao_version_greater_or_equal("0.14.0")
+@require_torchao_version_greater_or_equal("0.15.0")
 class TorchAoCompileTest(QuantCompileTests, unittest.TestCase):
     @property
     def quantization_config(self):
@@ -696,7 +693,7 @@ class TorchAoCompileTest(QuantCompileTests, unittest.TestCase):
 # Slices for these tests have been obtained on our aws-g6e-xlarge-plus runners
 @require_torch
 @require_torch_accelerator
-@require_torchao_version_greater_or_equal("0.14.0")
+@require_torchao_version_greater_or_equal("0.15.0")
 @slow
 @nightly
 class SlowTorchAoTests(unittest.TestCase):
@@ -854,7 +851,7 @@ class SlowTorchAoTests(unittest.TestCase):
 
 @require_torch
 @require_torch_accelerator
-@require_torchao_version_greater_or_equal("0.14.0")
+@require_torchao_version_greater_or_equal("0.15.0")
 @slow
 @nightly
 class SlowTorchAoPreserializedModelTests(unittest.TestCase):