Patch release: v0.10.2

* do not automatically enable xformers

* uP

examples/dreambooth/train_dreambooth.py

@@ -17,6 +17,7 @@ from accelerate.utils import set_seed
 from diffusers import AutoencoderKL, DDPMScheduler, DiffusionPipeline, UNet2DConditionModel
 from diffusers.optimization import get_scheduler
 from diffusers.utils import check_min_version
+from diffusers.utils.import_utils import is_xformers_available
 from huggingface_hub import HfFolder, Repository, whoami
 from PIL import Image
 from torchvision import transforms
@@ -488,6 +489,15 @@ def main(args):
         revision=args.revision,
     )
 
+    if is_xformers_available():
+        try:
+            unet.enable_xformers_memory_efficient_attention(True)
+        except Exception as e:
+            logger.warning(
+                "Could not enable memory efficient attention. Make sure xformers is installed"
+                f" correctly and a GPU is available: {e}"
+            )
+
     vae.requires_grad_(False)
     if not args.train_text_encoder:
         text_encoder.requires_grad_(False)

examples/text_to_image/train_text_to_image.py

@@ -18,6 +18,7 @@ from datasets import load_dataset
 from diffusers import AutoencoderKL, DDPMScheduler, StableDiffusionPipeline, UNet2DConditionModel
 from diffusers.optimization import get_scheduler
 from diffusers.utils import check_min_version
+from diffusers.utils.import_utils import is_xformers_available
 from huggingface_hub import HfFolder, Repository, whoami
 from torchvision import transforms
 from tqdm.auto import tqdm
@@ -364,6 +365,15 @@ def main():
         revision=args.revision,
     )
 
+    if is_xformers_available():
+        try:
+            unet.enable_xformers_memory_efficient_attention(True)
+        except Exception as e:
+            logger.warning(
+                "Could not enable memory efficient attention. Make sure xformers is installed"
+                f" correctly and a GPU is available: {e}"
+            )
+
     # Freeze vae and text_encoder
     vae.requires_grad_(False)
     text_encoder.requires_grad_(False)

examples/textual_inversion/textual_inversion.py

@@ -20,6 +20,7 @@ from diffusers import AutoencoderKL, DDPMScheduler, PNDMScheduler, StableDiffusi
 from diffusers.optimization import get_scheduler
 from diffusers.pipelines.stable_diffusion import StableDiffusionSafetyChecker
 from diffusers.utils import check_min_version
+from diffusers.utils.import_utils import is_xformers_available
 from huggingface_hub import HfFolder, Repository, whoami
 
 # TODO: remove and import from diffusers.utils when the new version of diffusers is released
@@ -439,6 +440,15 @@ def main():
         revision=args.revision,
     )
 
+    if is_xformers_available():
+        try:
+            unet.enable_xformers_memory_efficient_attention(True)
+        except Exception as e:
+            logger.warning(
+                "Could not enable memory efficient attention. Make sure xformers is installed"
+                f" correctly and a GPU is available: {e}"
+            )
+
     # Resize the token embeddings as we are adding new special tokens to the tokenizer
     text_encoder.resize_token_embeddings(len(tokenizer))
 

setup.py

@@ -218,7 +218,7 @@ install_requires = [
 
 setup(
     name="diffusers",
-    version="0.10.0",  # expected format is one of x.y.z.dev0, or x.y.z.rc1 or x.y.z (no to dashes, yes to dots)
+    version="0.10.2",  # expected format is one of x.y.z.dev0, or x.y.z.rc1 or x.y.z (no to dashes, yes to dots)
     description="Diffusers",
     long_description=open("README.md", "r", encoding="utf-8").read(),
     long_description_content_type="text/markdown",

src/diffusers/__init__.py

@@ -1,4 +1,4 @@
-__version__ = "0.10.0"
+__version__ = "0.10.2"
 
 from .configuration_utils import ConfigMixin
 from .onnx_utils import OnnxRuntimeModel
@@ -18,18 +18,6 @@ from .utils import (
 )
 
 
-# Make sure `transformers` is up to date
-if is_transformers_available():
-    import transformers
-
-    if is_transformers_version("<", "4.25.1"):
-        raise ImportError(
-            f"`diffusers` requires transformers >= 4.25.1 to function correctly, but {transformers.__version__} was"
-            " found in your environment. You can upgrade it with pip: `pip install transformers --upgrade`"
-        )
-else:
-    pass
-
 try:
     if not is_torch_available():
         raise OptionalDependencyNotAvailable()

src/diffusers/modeling_utils.py

@@ -188,6 +188,39 @@ class ModelMixin(torch.nn.Module):
         if self._supports_gradient_checkpointing:
             self.apply(partial(self._set_gradient_checkpointing, value=False))
 
+    def set_use_memory_efficient_attention_xformers(self, valid: bool) -> None:
+        # Recursively walk through all the children.
+        # Any children which exposes the set_use_memory_efficient_attention_xformers method
+        # gets the message
+        def fn_recursive_set_mem_eff(module: torch.nn.Module):
+            if hasattr(module, "set_use_memory_efficient_attention_xformers"):
+                module.set_use_memory_efficient_attention_xformers(valid)
+
+            for child in module.children():
+                fn_recursive_set_mem_eff(child)
+
+        for module in self.children():
+            if isinstance(module, torch.nn.Module):
+                fn_recursive_set_mem_eff(module)
+
+    def enable_xformers_memory_efficient_attention(self):
+        r"""
+        Enable memory efficient attention as implemented in xformers.
+
+        When this option is enabled, you should observe lower GPU memory usage and a potential speed up at inference
+        time. Speed up at training time is not guaranteed.
+
+        Warning: When Memory Efficient Attention and Sliced attention are both enabled, the Memory Efficient Attention
+        is used.
+        """
+        self.set_use_memory_efficient_attention_xformers(True)
+
+    def disable_xformers_memory_efficient_attention(self):
+        r"""
+        Disable memory efficient attention as implemented in xformers.
+        """
+        self.set_use_memory_efficient_attention_xformers(False)
+
     def save_pretrained(
         self,
         save_directory: Union[str, os.PathLike],

src/diffusers/models/attention.py

@@ -12,7 +12,6 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 import math
-import warnings
 from dataclasses import dataclass
 from typing import Optional
 
@@ -447,16 +446,6 @@ class BasicTransformerBlock(nn.Module):
         # 3. Feed-forward
         self.norm3 = nn.LayerNorm(dim)
 
-        # if xformers is installed try to use memory_efficient_attention by default
-        if is_xformers_available():
-            try:
-                self.set_use_memory_efficient_attention_xformers(True)
-            except Exception as e:
-                warnings.warn(
-                    "Could not enable memory efficient attention. Make sure xformers is installed"
-                    f" correctly and a GPU is available: {e}"
-                )
-
     def set_use_memory_efficient_attention_xformers(self, use_memory_efficient_attention_xformers: bool):
         if not is_xformers_available():
             print("Here is how to install it")

src/diffusers/pipelines/stable_diffusion/__init__.py

@@ -46,7 +46,7 @@ if is_transformers_available() and is_torch_available():
     from .safety_checker import StableDiffusionSafetyChecker
 
 try:
-    if not (is_transformers_available() and is_torch_available()):
+    if not (is_transformers_available() and is_torch_available() and is_transformers_version(">=", "4.25.0")):
         raise OptionalDependencyNotAvailable()
 except OptionalDependencyNotAvailable:
     from ...utils.dummy_torch_and_transformers_objects import StableDiffusionImageVariationPipeline

src/diffusers/pipelines/versatile_diffusion/__init__.py

@@ -7,7 +7,7 @@ from ...utils import (
 
 
 try:
-    if not (is_transformers_available() and is_torch_available()):
+    if not (is_transformers_available() and is_torch_available() and is_transformers_version(">=", "4.25.0")):
         raise OptionalDependencyNotAvailable()
 except OptionalDependencyNotAvailable:
     from ...utils.dummy_torch_and_transformers_objects import (

src/diffusers/utils/import_utils.py

@@ -354,7 +354,20 @@ def requires_backends(obj, backends):
     if failed:
         raise ImportError("".join(failed))
 
-    if name in ["StableDiffusionDepth2ImgPipeline"] and is_transformers_version("<", "4.26.0.dev0"):
+    if name in [
+        "VersatileDiffusionTextToImagePipeline",
+        "VersatileDiffusionPipeline",
+        "VersatileDiffusionDualGuidedPipeline",
+        "StableDiffusionImageVariationPipeline",
+    ] and is_transformers_version("<", "4.25.0"):
+        raise ImportError(
+            f"You need to install `transformers>=4.25` in order to use {name}: \n```\n pip install"
+            " --upgrade transformers \n```"
+        )
+
+    if name in [
+        "StableDiffusionDepth2ImgPipeline",
+    ] and is_transformers_version("<", "4.26.0.dev0"):
         raise ImportError(
             f"You need to install `transformers` from 'main' in order to use {name}: \n```\n pip install"
             " git+https://github.com/huggingface/transformers \n```"

tests/models/test_models_unet_2d.py

@@ -30,6 +30,7 @@ from diffusers.utils import (
     torch_all_close,
     torch_device,
 )
+from diffusers.utils.import_utils import is_xformers_available
 from parameterized import parameterized
 
 from ..test_modeling_common import ModelTesterMixin
@@ -255,6 +256,20 @@ class UNet2DConditionModelTests(ModelTesterMixin, unittest.TestCase):
         inputs_dict = self.dummy_input
         return init_dict, inputs_dict
 
+    @unittest.skipIf(
+        torch_device != "cuda" or not is_xformers_available(),
+        reason="XFormers attention is only available with CUDA and `xformers` installed",
+    )
+    def test_xformers_enable_works(self):
+        init_dict, inputs_dict = self.prepare_init_args_and_inputs_for_common()
+        model = self.model_class(**init_dict)
+
+        model.enable_xformers_memory_efficient_attention()
+
+        assert (
+            model.mid_block.attentions[0].transformer_blocks[0].attn1._use_memory_efficient_attention_xformers
+        ), "xformers is not enabled"
+
     @unittest.skipIf(torch_device == "mps", "Gradient checkpointing skipped on MPS")
     def test_gradient_checkpointing(self):
         # enable deterministic behavior for gradient checkpointing