Apply style fixes

* fix consisid imports

* fix opencv import

* fix

.github/workflows/pr_style_bot.yml

@@ -14,4 +14,4 @@ jobs:
     with:
       python_quality_dependencies: "[quality]"
     secrets:
-      bot_token: ${{ secrets.GITHUB_TOKEN }}
+      bot_token: ${{ secrets.DIFFUSERS_STYLE_BOT_ACTION_TOKEN }}

docs/source/en/_toctree.yml

@@ -265,6 +265,8 @@
     sections:
     - local: api/models/overview
       title: Overview
+    - local: api/models/auto_model
+      title: AutoModel
     - sections:
       - local: api/models/controlnet
         title: ControlNetModel

docs/source/en/api/models/auto_model.md

@@ -0,0 +1,29 @@
+<!--Copyright 2024 The HuggingFace Team. All rights reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
+the License. You may obtain a copy of the License at
+
+http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
+an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
+specific language governing permissions and limitations under the License.
+-->
+
+# AutoModel
+
+The `AutoModel` is designed to make it easy to load a checkpoint without needing to know the specific model class. `AutoModel` automatically retrieves the correct model class from the checkpoint `config.json` file.
+
+```python
+from diffusers import AutoModel, AutoPipelineForText2Image
+
+unet = AutoModel.from_pretrained("stable-diffusion-v1-5/stable-diffusion-v1-5", subfolder="unet")
+pipe = AutoPipelineForText2Image.from_pretrained("stable-diffusion-v1-5/stable-diffusion-v1-5", unet=unet)
+```
+
+
+## AutoModel
+
+[[autodoc]] AutoModel
+	- all
+	- from_pretrained

setup.py

@@ -142,6 +142,7 @@ _deps = [
     "urllib3<=2.0.0",
     "black",
     "phonemizer",
+    "opencv-python",
 ]
 
 # this is a lookup table with items like:

src/diffusers/__init__.py

@@ -14,6 +14,7 @@ from .utils import (
     is_librosa_available,
     is_note_seq_available,
     is_onnx_available,
+    is_opencv_available,
     is_optimum_quanto_available,
     is_scipy_available,
     is_sentencepiece_available,
@@ -352,7 +353,6 @@ else:
             "CogView3PlusPipeline",
             "CogView4ControlPipeline",
             "CogView4Pipeline",
-            "ConsisIDPipeline",
             "CycleDiffusionPipeline",
             "EasyAnimateControlPipeline",
             "EasyAnimateInpaintPipeline",
@@ -518,6 +518,19 @@ else:
         ]
     )
 
+try:
+    if not (is_torch_available() and is_transformers_available() and is_opencv_available()):
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    from .utils import dummy_torch_and_transformers_and_opencv_objects  # noqa F403
+
+    _import_structure["utils.dummy_torch_and_transformers_and_opencv_objects"] = [
+        name for name in dir(dummy_torch_and_transformers_and_opencv_objects) if not name.startswith("_")
+    ]
+
+else:
+    _import_structure["pipelines"].extend(["ConsisIDPipeline"])
+
 try:
     if not (is_torch_available() and is_transformers_available() and is_k_diffusion_available()):
         raise OptionalDependencyNotAvailable()
@@ -909,7 +922,6 @@ if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
             CogView3PlusPipeline,
             CogView4ControlPipeline,
             CogView4Pipeline,
-            ConsisIDPipeline,
             CycleDiffusionPipeline,
             EasyAnimateControlPipeline,
             EasyAnimateInpaintPipeline,
@@ -1088,6 +1100,15 @@ if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
         from .utils.dummy_torch_and_transformers_and_sentencepiece_objects import *  # noqa F403
     else:
         from .pipelines import KolorsImg2ImgPipeline, KolorsPAGPipeline, KolorsPipeline
+
+    try:
+        if not (is_torch_available() and is_transformers_available() and is_opencv_available()):
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        from .utils.dummy_torch_and_transformers_and_opencv_objects import *  # noqa F403
+    else:
+        from .pipelines import ConsisIDPipeline
+
     try:
         if not (is_torch_available() and is_transformers_available() and is_onnx_available()):
             raise OptionalDependencyNotAvailable()

src/diffusers/dependency_versions_table.py

@@ -49,4 +49,5 @@ deps = {
     "urllib3": "urllib3<=2.0.0",
     "black": "black",
     "phonemizer": "phonemizer",
+    "opencv-python": "opencv-python",
 }

src/diffusers/models/controlnets/controlnet_flux.py

@@ -298,15 +298,6 @@ class FluxControlNetModel(ModelMixin, ConfigMixin, PeftAdapterMixin):
         )
         encoder_hidden_states = self.context_embedder(encoder_hidden_states)
 
-        if self.union:
-            # union mode
-            if controlnet_mode is None:
-                raise ValueError("`controlnet_mode` cannot be `None` when applying ControlNet-Union")
-            # union mode emb
-            controlnet_mode_emb = self.controlnet_mode_embedder(controlnet_mode)
-            encoder_hidden_states = torch.cat([controlnet_mode_emb, encoder_hidden_states], dim=1)
-            txt_ids = torch.cat([txt_ids[:1], txt_ids], dim=0)
-
         if txt_ids.ndim == 3:
             logger.warning(
                 "Passing `txt_ids` 3d torch.Tensor is deprecated."
@@ -320,6 +311,15 @@ class FluxControlNetModel(ModelMixin, ConfigMixin, PeftAdapterMixin):
             )
             img_ids = img_ids[0]
 
+        if self.union:
+            # union mode
+            if controlnet_mode is None:
+                raise ValueError("`controlnet_mode` cannot be `None` when applying ControlNet-Union")
+            # union mode emb
+            controlnet_mode_emb = self.controlnet_mode_embedder(controlnet_mode)
+            encoder_hidden_states = torch.cat([controlnet_mode_emb, encoder_hidden_states], dim=1)
+            txt_ids = torch.cat([txt_ids[:1], txt_ids], dim=0)
+
         ids = torch.cat((txt_ids, img_ids), dim=0)
         image_rotary_emb = self.pos_embed(ids)
 

src/diffusers/pipelines/__init__.py

@@ -10,6 +10,7 @@ from ..utils import (
     is_librosa_available,
     is_note_seq_available,
     is_onnx_available,
+    is_opencv_available,
     is_sentencepiece_available,
     is_torch_available,
     is_torch_npu_available,
@@ -155,7 +156,6 @@ else:
     ]
     _import_structure["cogview3"] = ["CogView3PlusPipeline"]
     _import_structure["cogview4"] = ["CogView4Pipeline", "CogView4ControlPipeline"]
-    _import_structure["consisid"] = ["ConsisIDPipeline"]
     _import_structure["controlnet"].extend(
         [
             "BlipDiffusionControlNetPipeline",
@@ -414,6 +414,18 @@ else:
         "KolorsImg2ImgPipeline",
     ]
 
+try:
+    if not (is_torch_available() and is_transformers_available() and is_opencv_available()):
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    from ..utils import (
+        dummy_torch_and_transformers_and_opencv_objects,
+    )
+
+    _dummy_objects.update(get_objects_from_module(dummy_torch_and_transformers_and_opencv_objects))
+else:
+    _import_structure["consisid"] = ["ConsisIDPipeline"]
+
 try:
     if not is_flax_available():
         raise OptionalDependencyNotAvailable()
@@ -512,7 +524,6 @@ if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
         )
         from .cogview3 import CogView3PlusPipeline
         from .cogview4 import CogView4ControlPipeline, CogView4Pipeline
-        from .consisid import ConsisIDPipeline
         from .controlnet import (
             BlipDiffusionControlNetPipeline,
             StableDiffusionControlNetImg2ImgPipeline,
@@ -761,6 +772,14 @@ if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
                 KolorsPipeline,
             )
 
+        try:
+            if not (is_torch_available() and is_transformers_available() and is_opencv_available()):
+                raise OptionalDependencyNotAvailable()
+        except OptionalDependencyNotAvailable:
+            from ..utils.dummy_torch_and_transformers_and_opencv_objects import *
+        else:
+            from .consisid import ConsisIDPipeline
+
         try:
             if not is_flax_available():
                 raise OptionalDependencyNotAvailable()

src/diffusers/pipelines/consisid/__init__.py

@@ -5,6 +5,7 @@ from ...utils import (
     OptionalDependencyNotAvailable,
     _LazyModule,
     get_objects_from_module,
+    is_opencv_available,
     is_torch_available,
     is_transformers_available,
 )
@@ -15,12 +16,12 @@ _import_structure = {}
 
 
 try:
-    if not (is_transformers_available() and is_torch_available()):
+    if not (is_transformers_available() and is_torch_available() and is_opencv_available()):
         raise OptionalDependencyNotAvailable()
 except OptionalDependencyNotAvailable:
-    from ...utils import dummy_torch_and_transformers_objects  # noqa F403
+    from ...utils import dummy_torch_and_transformers_and_opencv_objects  # noqa F403
 
-    _dummy_objects.update(get_objects_from_module(dummy_torch_and_transformers_objects))
+    _dummy_objects.update(get_objects_from_module(dummy_torch_and_transformers_and_opencv_objects))
 else:
     _import_structure["pipeline_consisid"] = ["ConsisIDPipeline"]
 

src/diffusers/pipelines/consisid/pipeline_consisid.py

@@ -16,7 +16,6 @@ import inspect
 import math
 from typing import Any, Callable, Dict, List, Optional, Tuple, Union
 
-import cv2
 import numpy as np
 import PIL
 import torch
@@ -29,12 +28,16 @@ from ...models import AutoencoderKLCogVideoX, ConsisIDTransformer3DModel
 from ...models.embeddings import get_3d_rotary_pos_embed
 from ...pipelines.pipeline_utils import DiffusionPipeline
 from ...schedulers import CogVideoXDPMScheduler
-from ...utils import logging, replace_example_docstring
+from ...utils import is_opencv_available, logging, replace_example_docstring
 from ...utils.torch_utils import randn_tensor
 from ...video_processor import VideoProcessor
 from .pipeline_output import ConsisIDPipelineOutput
 
 
+if is_opencv_available():
+    import cv2
+
+
 logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
 
 

src/diffusers/utils/__init__.py

@@ -79,6 +79,7 @@ from .import_utils import (
     is_matplotlib_available,
     is_note_seq_available,
     is_onnx_available,
+    is_opencv_available,
     is_optimum_quanto_available,
     is_optimum_quanto_version,
     is_peft_available,

src/diffusers/utils/dummy_torch_and_transformers_and_opencv_objects.py

@@ -0,0 +1,17 @@
+# This file is autogenerated by the command `make fix-copies`, do not edit.
+from ..utils import DummyObject, requires_backends
+
+
+class ConsisIDPipeline(metaclass=DummyObject):
+    _backends = ["torch", "transformers", "opencv"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch", "transformers", "opencv"])
+
+    @classmethod
+    def from_config(cls, *args, **kwargs):
+        requires_backends(cls, ["torch", "transformers", "opencv"])
+
+    @classmethod
+    def from_pretrained(cls, *args, **kwargs):
+        requires_backends(cls, ["torch", "transformers", "opencv"])

src/diffusers/utils/dummy_torch_and_transformers_objects.py

@@ -392,21 +392,6 @@ class CogView4Pipeline(metaclass=DummyObject):
         requires_backends(cls, ["torch", "transformers"])
 
 
-class ConsisIDPipeline(metaclass=DummyObject):
-    _backends = ["torch", "transformers"]
-
-    def __init__(self, *args, **kwargs):
-        requires_backends(self, ["torch", "transformers"])
-
-    @classmethod
-    def from_config(cls, *args, **kwargs):
-        requires_backends(cls, ["torch", "transformers"])
-
-    @classmethod
-    def from_pretrained(cls, *args, **kwargs):
-        requires_backends(cls, ["torch", "transformers"])
-
-
 class CycleDiffusionPipeline(metaclass=DummyObject):
     _backends = ["torch", "transformers"]
 

tests/others/test_dependencies.py

@@ -37,6 +37,8 @@ class DependencyTester(unittest.TestCase):
                         backend = "k-diffusion"
                     elif backend == "invisible_watermark":
                         backend = "invisible-watermark"
+                    elif backend == "opencv":
+                        backend = "opencv-python"
                     assert backend in deps, f"{backend} is not in the deps table!"
 
     def test_pipeline_imports(self):

tests/pipelines/flux/test_pipeline_flux_redux.py

@@ -8,6 +8,7 @@ import torch
 from diffusers import FluxPipeline, FluxPriorReduxPipeline
 from diffusers.utils import load_image
 from diffusers.utils.testing_utils import (
+    Expectations,
     backend_empty_cache,
     numpy_cosine_similarity_distance,
     require_big_accelerator,
@@ -21,7 +22,7 @@ from diffusers.utils.testing_utils import (
 @pytest.mark.big_gpu_with_torch_cuda
 class FluxReduxSlowTests(unittest.TestCase):
     pipeline_class = FluxPriorReduxPipeline
-    repo_id = "YiYiXu/yiyi-redux"  # update to "black-forest-labs/FLUX.1-Redux-dev" once PR is merged
+    repo_id = "black-forest-labs/FLUX.1-Redux-dev"
     base_pipeline_class = FluxPipeline
     base_repo_id = "black-forest-labs/FLUX.1-schnell"
 
@@ -69,41 +70,82 @@ class FluxReduxSlowTests(unittest.TestCase):
         image = pipe_base(**base_pipeline_inputs, **redux_pipeline_output).images[0]
 
         image_slice = image[0, :10, :10]
-        expected_slice = np.array(
-            [
-                0.30078125,
-                0.37890625,
-                0.46875,
-                0.28125,
-                0.36914062,
-                0.47851562,
-                0.28515625,
-                0.375,
-                0.4765625,
-                0.28125,
-                0.375,
-                0.48046875,
-                0.27929688,
-                0.37695312,
-                0.47851562,
-                0.27734375,
-                0.38085938,
-                0.4765625,
-                0.2734375,
-                0.38085938,
-                0.47265625,
-                0.27539062,
-                0.37890625,
-                0.47265625,
-                0.27734375,
-                0.37695312,
-                0.47070312,
-                0.27929688,
-                0.37890625,
-                0.47460938,
-            ],
-            dtype=np.float32,
+        expected_slices = Expectations(
+            {
+                ("cuda", 7): np.array(
+                    [
+                        0.30078125,
+                        0.37890625,
+                        0.46875,
+                        0.28125,
+                        0.36914062,
+                        0.47851562,
+                        0.28515625,
+                        0.375,
+                        0.4765625,
+                        0.28125,
+                        0.375,
+                        0.48046875,
+                        0.27929688,
+                        0.37695312,
+                        0.47851562,
+                        0.27734375,
+                        0.38085938,
+                        0.4765625,
+                        0.2734375,
+                        0.38085938,
+                        0.47265625,
+                        0.27539062,
+                        0.37890625,
+                        0.47265625,
+                        0.27734375,
+                        0.37695312,
+                        0.47070312,
+                        0.27929688,
+                        0.37890625,
+                        0.47460938,
+                    ],
+                    dtype=np.float32,
+                ),
+                ("xpu", 3): np.array(
+                    [
+                        0.20507812,
+                        0.30859375,
+                        0.3984375,
+                        0.18554688,
+                        0.30078125,
+                        0.41015625,
+                        0.19921875,
+                        0.3125,
+                        0.40625,
+                        0.19726562,
+                        0.3125,
+                        0.41601562,
+                        0.19335938,
+                        0.31445312,
+                        0.4140625,
+                        0.1953125,
+                        0.3203125,
+                        0.41796875,
+                        0.19726562,
+                        0.32421875,
+                        0.41992188,
+                        0.19726562,
+                        0.32421875,
+                        0.41992188,
+                        0.20117188,
+                        0.32421875,
+                        0.41796875,
+                        0.203125,
+                        0.32617188,
+                        0.41796875,
+                    ],
+                    dtype=np.float32,
+                ),
+            }
         )
+        expected_slice = expected_slices.get_expectation()
+
         max_diff = numpy_cosine_similarity_distance(expected_slice.flatten(), image_slice.flatten())
 
         assert max_diff < 1e-4

tests/pipelines/test_pipelines_common.py

@@ -1347,7 +1347,7 @@ class PipelineTesterMixin:
 
     @unittest.skipIf(torch_device not in ["cuda", "xpu"], reason="float16 requires CUDA or XPU")
     @require_accelerator
-    def test_float16_inference(self, expected_max_diff=5e-2):
+    def test_float16_inference(self, expected_max_diff=6e-2):
         components = self.get_dummy_components()
         pipe = self.pipeline_class(**components)
         for component in pipe.components.values():

utils/fetch_latest_release_branch.py

@@ -17,8 +17,6 @@
 import requests
 from packaging.version import parse
 
-from ..src.diffusers.utils.constants import DIFFUSERS_REQUEST_TIMEOUT
-
 
 # GitHub repository details
 USER = "huggingface"
@@ -33,7 +31,7 @@ def fetch_all_branches(user, repo):
         response = requests.get(
             f"https://api.github.com/repos/{user}/{repo}/branches",
             params={"page": page},
-            timeout=DIFFUSERS_REQUEST_TIMEOUT,
+            timeout=60,
         )
 
         # Check if the request was successful