update

docs: Fix VAE scale factor calculation

docs/source/en/api/pipelines/qwenimage.md

@@ -120,6 +120,12 @@ The `guidance_scale` parameter in the pipeline is there to support future guidan
   - all
   - __call__
 
+## QwenImageEditInpaintPipeline
+
+[[autodoc]] QwenImageEditInpaintPipeline
+  - all
+  - __call__
+
 ## QwenImaggeControlNetPipeline
   - all
   - __call__

docs/source/en/modular_diffusers/components_manager.md

@@ -51,10 +51,10 @@ t2i_pipeline = t2i_blocks.init_pipeline(modular_repo_id, components_manager=comp
 </hfoption>
 </hfoptions>
 
-Components are only loaded and registered when using [`~ModularPipeline.load_components`] or [`~ModularPipeline.load_default_components`]. The example below uses [`~ModularPipeline.load_default_components`] to create a second pipeline that reuses all the components from the first one, and assigns it to a different collection
+Components are only loaded and registered when using [`~ModularPipeline.load_components`] or [`~ModularPipeline.load_components`]. The example below uses [`~ModularPipeline.load_components`] to create a second pipeline that reuses all the components from the first one, and assigns it to a different collection
 
 ```py
-pipe.load_default_components()
+pipe.load_components()
 pipe2 = ModularPipeline.from_pretrained("YiYiXu/modular-demo-auto", components_manager=comp, collection="test2")
 ```
 
@@ -187,4 +187,4 @@ comp.enable_auto_cpu_offload(device="cuda")
 
 All models begin on the CPU and [`ComponentsManager`] moves them to the appropriate device right before they're needed, and moves other models back to the CPU when GPU memory is low.
 
-You can set your own rules for which models to offload first.
+You can set your own rules for which models to offload first.

docs/source/en/modular_diffusers/guiders.md

@@ -75,13 +75,13 @@ Guiders that are already saved on the Hub with a `modular_model_index.json` file
 }
 ```
 
-The guider is only created after calling [`~ModularPipeline.load_default_components`] based on the loading specification in `modular_model_index.json`.
+The guider is only created after calling [`~ModularPipeline.load_components`] based on the loading specification in `modular_model_index.json`.
 
 ```py
 t2i_pipeline = t2i_blocks.init_pipeline("YiYiXu/modular-doc-guider")
 # not created during init
 assert t2i_pipeline.guider is None
-t2i_pipeline.load_default_components()
+t2i_pipeline.load_components()
 # loaded as PAG guider
 t2i_pipeline.guider
 ```
@@ -172,4 +172,4 @@ t2i_pipeline.push_to_hub("YiYiXu/modular-doc-guider")
 ```
 
 </hfoption>
-</hfoptions>
+</hfoptions>

docs/source/en/modular_diffusers/modular_pipeline.md

@@ -29,7 +29,7 @@ blocks = SequentialPipelineBlocks.from_blocks_dict(TEXT2IMAGE_BLOCKS)
 modular_repo_id = "YiYiXu/modular-loader-t2i-0704"
 pipeline = blocks.init_pipeline(modular_repo_id)
 
-pipeline.load_default_components(torch_dtype=torch.float16)
+pipeline.load_components(torch_dtype=torch.float16)
 pipeline.to("cuda")
 
 image = pipeline(prompt="Astronaut in a jungle, cold color palette, muted colors, detailed, 8k", output="images")[0]
@@ -49,7 +49,7 @@ blocks = SequentialPipelineBlocks.from_blocks_dict(IMAGE2IMAGE_BLOCKS)
 modular_repo_id = "YiYiXu/modular-loader-t2i-0704"
 pipeline = blocks.init_pipeline(modular_repo_id)
 
-pipeline.load_default_components(torch_dtype=torch.float16)
+pipeline.load_components(torch_dtype=torch.float16)
 pipeline.to("cuda")
 
 url = "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/sdxl-text2img.png"
@@ -73,7 +73,7 @@ blocks = SequentialPipelineBlocks.from_blocks_dict(INPAINT_BLOCKS)
 modular_repo_id = "YiYiXu/modular-loader-t2i-0704"
 pipeline = blocks.init_pipeline(modular_repo_id)
 
-pipeline.load_default_components(torch_dtype=torch.float16)
+pipeline.load_components(torch_dtype=torch.float16)
 pipeline.to("cuda")
 
 img_url = "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/sdxl-text2img.png"
@@ -176,15 +176,15 @@ diffdiff_pipeline = ModularPipeline.from_pretrained(modular_repo_id, trust_remot
 
 ## Loading components
 
-A [`ModularPipeline`] doesn't automatically instantiate with components. It only loads the configuration and component specifications. You can load all components with [`~ModularPipeline.load_default_components`] or only load specific components with [`~ModularPipeline.load_components`].
+A [`ModularPipeline`] doesn't automatically instantiate with components. It only loads the configuration and component specifications. You can load all components with [`~ModularPipeline.load_components`] or only load specific components with [`~ModularPipeline.load_components`].
 
 <hfoptions id="load">
-<hfoption id="load_default_components">
+<hfoption id="load_components">
 
 ```py
 import torch
 
-t2i_pipeline.load_default_components(torch_dtype=torch.float16)
+t2i_pipeline.load_components(torch_dtype=torch.float16)
 t2i_pipeline.to("cuda")
 ```
 
@@ -355,4 +355,4 @@ The [config.json](https://huggingface.co/YiYiXu/modular-diffdiff-0704/blob/main/
     "ModularPipelineBlocks": "block.DiffDiffBlocks"
   }
 }
-```
+```

docs/source/en/modular_diffusers/quickstart.md

@@ -173,9 +173,9 @@ print(dd_blocks)
 
 ## ModularPipeline
 
-Convert the [`SequentialPipelineBlocks`] into a [`ModularPipeline`] with the [`ModularPipeline.init_pipeline`] method. This initializes the expected components to load from a `modular_model_index.json` file. Explicitly load the components by calling [`ModularPipeline.load_default_components`].
+Convert the [`SequentialPipelineBlocks`] into a [`ModularPipeline`] with the [`ModularPipeline.init_pipeline`] method. This initializes the expected components to load from a `modular_model_index.json` file. Explicitly load the components by calling [`ModularPipeline.load_components`].
 
-It is a good idea to initialize the [`ComponentManager`] with the pipeline to help manage the different components. Once you call [`~ModularPipeline.load_default_components`], the components are registered to the [`ComponentManager`] and can be shared between workflows. The example below uses the `collection` argument to assign the components a `"diffdiff"` label for better organization.
+It is a good idea to initialize the [`ComponentManager`] with the pipeline to help manage the different components. Once you call [`~ModularPipeline.load_components`], the components are registered to the [`ComponentManager`] and can be shared between workflows. The example below uses the `collection` argument to assign the components a `"diffdiff"` label for better organization.
 
 ```py
 from diffusers.modular_pipelines import ComponentsManager
@@ -209,11 +209,11 @@ Use the [`sub_blocks.insert`] method to insert it into the [`ModularPipeline`].
 dd_blocks.sub_blocks.insert("ip_adapter", ip_adapter_block, 0)
 ```
 
-Call [`~ModularPipeline.init_pipeline`] to initialize a [`ModularPipeline`] and use [`~ModularPipeline.load_default_components`] to load the model components. Load and set the IP-Adapter to run the pipeline.
+Call [`~ModularPipeline.init_pipeline`] to initialize a [`ModularPipeline`] and use [`~ModularPipeline.load_components`] to load the model components. Load and set the IP-Adapter to run the pipeline.
 
 ```py
 dd_pipeline = dd_blocks.init_pipeline("YiYiXu/modular-demo-auto", collection="diffdiff")
-dd_pipeline.load_default_components(torch_dtype=torch.float16)
+dd_pipeline.load_components(torch_dtype=torch.float16)
 dd_pipeline.loader.load_ip_adapter("h94/IP-Adapter", subfolder="sdxl_models", weight_name="ip-adapter_sdxl.bin")
 dd_pipeline.loader.set_ip_adapter_scale(0.6)
 dd_pipeline = dd_pipeline.to(device)
@@ -260,14 +260,14 @@ class SDXLDiffDiffControlNetDenoiseStep(StableDiffusionXLDenoiseLoopWrapper):
 controlnet_denoise_block = SDXLDiffDiffControlNetDenoiseStep()
 ```
 
-Insert the `controlnet_input` block and replace the `denoise` block with the new `controlnet_denoise_block`. Initialize a [`ModularPipeline`] and [`~ModularPipeline.load_default_components`] into it.
+Insert the `controlnet_input` block and replace the `denoise` block with the new `controlnet_denoise_block`. Initialize a [`ModularPipeline`] and [`~ModularPipeline.load_components`] into it.
 
 ```py
 dd_blocks.sub_blocks.insert("controlnet_input", control_input_block, 7)
 dd_blocks.sub_blocks["denoise"] = controlnet_denoise_block
 
 dd_pipeline = dd_blocks.init_pipeline("YiYiXu/modular-demo-auto", collection="diffdiff")
-dd_pipeline.load_default_components(torch_dtype=torch.float16)
+dd_pipeline.load_components(torch_dtype=torch.float16)
 dd_pipeline = dd_pipeline.to(device)
 
 control_image = load_image("https://huggingface.co/datasets/YiYiXu/testing-images/resolve/main/diffdiff_tomato_canny.jpeg")
@@ -320,7 +320,7 @@ Call [`SequentialPipelineBlocks.from_blocks_dict`] to create a [`SequentialPipel
 ```py
 dd_auto_blocks = SequentialPipelineBlocks.from_blocks_dict(DIFFDIFF_AUTO_BLOCKS)
 dd_pipeline = dd_auto_blocks.init_pipeline("YiYiXu/modular-demo-auto", collection="diffdiff")
-dd_pipeline.load_default_components(torch_dtype=torch.float16)
+dd_pipeline.load_components(torch_dtype=torch.float16)
 ```
 
 ## Share
@@ -340,5 +340,5 @@ from diffusers.modular_pipelines import ModularPipeline, ComponentsManager
 components = ComponentsManager()
 
 diffdiff_pipeline = ModularPipeline.from_pretrained("YiYiXu/modular-diffdiff-0704", trust_remote_code=True, components_manager=components, collection="diffdiff")
-diffdiff_pipeline.load_default_components(torch_dtype=torch.float16)
-```
+diffdiff_pipeline.load_components(torch_dtype=torch.float16)
+```

docs/source/en/training/distributed_inference.md

@@ -223,7 +223,7 @@ from diffusers.image_processor import VaeImageProcessor
 import torch 
 
 vae = AutoencoderKL.from_pretrained(ckpt_id, subfolder="vae", torch_dtype=torch.bfloat16).to("cuda")
-vae_scale_factor = 2 ** (len(vae.config.block_out_channels))
+vae_scale_factor = 2 ** (len(vae.config.block_out_channels) - 1)
 image_processor = VaeImageProcessor(vae_scale_factor=vae_scale_factor)
 
 with torch.no_grad():

docs/source/zh/modular_diffusers/components_manager.md

@@ -48,10 +48,10 @@ t2i_pipeline = t2i_blocks.init_pipeline(modular_repo_id, components_manager=comp
 </hfoption>
 </hfoptions>
 
-组件仅在调用 [`~ModularPipeline.load_components`] 或 [`~ModularPipeline.load_default_components`] 时加载和注册。以下示例使用 [`~ModularPipeline.load_default_components`] 创建第二个管道，重用第一个管道的所有组件，并将其分配到不同的集合。
+组件仅在调用 [`~ModularPipeline.load_components`] 或 [`~ModularPipeline.load_components`] 时加载和注册。以下示例使用 [`~ModularPipeline.load_components`] 创建第二个管道，重用第一个管道的所有组件，并将其分配到不同的集合。
 
 ```py
-pipe.load_default_components()
+pipe.load_components()
 pipe2 = ModularPipeline.from_pretrained("YiYiXu/modular-demo-auto", components_manager=comp, collection="test2")
 ```
 
@@ -185,4 +185,4 @@ comp.enable_auto_cpu_offload(device="cuda")
 
 所有模型开始时都在 CPU 上，[`ComponentsManager`] 在需要它们之前将它们移动到适当的设备，并在 GPU 内存不足时将其他模型移回 CPU。
 
-您可以设置自己的规则来决定哪些模型要卸载。
+您可以设置自己的规则来决定哪些模型要卸载。

docs/source/zh/modular_diffusers/guiders.md

@@ -73,13 +73,13 @@ ComponentSpec(name='guider', type_hint=<class 'diffusers.guiders.perturbed_atten
 }
 ```
 
-引导器只有在调用 [`~ModularPipeline.load_default_components`] 之后才会创建，基于 `modular_model_index.json` 中的加载规范。
+引导器只有在调用 [`~ModularPipeline.load_components`] 之后才会创建，基于 `modular_model_index.json` 中的加载规范。
 
 ```py
 t2i_pipeline = t2i_blocks.init_pipeline("YiYiXu/modular-doc-guider")
 # 在初始化时未创建
 assert t2i_pipeline.guider is None
-t2i_pipeline.load_default_components()
+t2i_pipeline.load_components()
 # 加载为 PAG 引导器
 t2i_pipeline.guider
 ```
@@ -170,4 +170,4 @@ t2i_pipeline.push_to_hub("YiYiXu/modular-doc-guider")
 ```
 
 </hfoption>
-</hfoptions>
+</hfoptions>

docs/source/zh/modular_diffusers/modular_pipeline.md

@@ -28,7 +28,7 @@ blocks = SequentialPipelineBlocks.from_blocks_dict(TEXT2IMAGE_BLOCKS)
 modular_repo_id = "YiYiXu/modular-loader-t2i-0704"
 pipeline = blocks.init_pipeline(modular_repo_id)
 
-pipeline.load_default_components(torch_dtype=torch.float16)
+pipeline.load_components(torch_dtype=torch.float16)
 pipeline.to("cuda")
 
 image = pipeline(prompt="Astronaut in a jungle, cold color palette, muted colors, detailed, 8k", output="images")[0]
@@ -48,7 +48,7 @@ blocks = SequentialPipelineBlocks.from_blocks_dict(IMAGE2IMAGE_BLOCKS)
 modular_repo_id = "YiYiXu/modular-loader-t2i-0704"
 pipeline = blocks.init_pipeline(modular_repo_id)
 
-pipeline.load_default_components(torch_dtype=torch.float16)
+pipeline.load_components(torch_dtype=torch.float16)
 pipeline.to("cuda")
 
 url = "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/sdxl-text2img.png"
@@ -72,7 +72,7 @@ blocks = SequentialPipelineBlocks.from_blocks_dict(INPAINT_BLOCKS)
 modular_repo_id = "YiYiXu/modular-loader-t2i-0704"
 pipeline = blocks.init_pipeline(modular_repo_id)
 
-pipeline.load_default_components(torch_dtype=torch.float16)
+pipeline.load_components(torch_dtype=torch.float16)
 pipeline.to("cuda")
 
 img_url = "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/sdxl-text2img.png"
@@ -176,15 +176,15 @@ diffdiff_pipeline = ModularPipeline.from_pretrained(modular_repo_id, trust_remot
 
 ## 加载组件
 
-一个[`ModularPipeline`]不会自动实例化组件。它只加载配置和组件规范。您可以使用[`~ModularPipeline.load_default_components`]加载所有组件，或仅使用[`~ModularPipeline.load_components`]加载特定组件。
+一个[`ModularPipeline`]不会自动实例化组件。它只加载配置和组件规范。您可以使用[`~ModularPipeline.load_components`]加载所有组件，或仅使用[`~ModularPipeline.load_components`]加载特定组件。
 
 <hfoptions id="load">
-<hfoption id="load_default_components">
+<hfoption id="load_components">
 
 ```py
 import torch
 
-t2i_pipeline.load_default_components(torch_dtype=torch.float16)
+t2i_pipeline.load_components(torch_dtype=torch.float16)
 t2i_pipeline.to("cuda")
 ```
 

docs/source/zh/modular_diffusers/quickstart.md

@@ -175,7 +175,7 @@ print(dd_blocks)
 将 [`SequentialPipelineBlocks`] 转换为 [`ModularPipeline`]，使用 [`ModularPipeline.init_pipeline`] 方法。这会初始化从 `modular_model_index.json` 文件加载的预期组件。通过调用 [`ModularPipeline.load_defau
 lt_components`]。
 
-初始化[`ComponentManager`]时传入pipeline是一个好主意，以帮助管理不同的组件。一旦调用[`~ModularPipeline.load_default_components`]，组件就会被注册到[`ComponentManager`]中，并且可以在工作流之间共享。下面的例子使用`collection`参数为组件分配了一个`"diffdiff"`标签，以便更好地组织。
+初始化[`ComponentManager`]时传入pipeline是一个好主意，以帮助管理不同的组件。一旦调用[`~ModularPipeline.load_components`]，组件就会被注册到[`ComponentManager`]中，并且可以在工作流之间共享。下面的例子使用`collection`参数为组件分配了一个`"diffdiff"`标签，以便更好地组织。
 
 ```py
 from diffusers.modular_pipelines import ComponentsManager
@@ -209,11 +209,11 @@ ip_adapter_block = StableDiffusionXLAutoIPAdapterStep()
 dd_blocks.sub_blocks.insert("ip_adapter", ip_adapter_block, 0)
 ```
 
-调用[`~ModularPipeline.init_pipeline`]来初始化一个[`ModularPipeline`]，并使用[`~ModularPipeline.load_default_components`]加载模型组件。加载并设置IP-Adapter以运行pipeline。
+调用[`~ModularPipeline.init_pipeline`]来初始化一个[`ModularPipeline`]，并使用[`~ModularPipeline.load_components`]加载模型组件。加载并设置IP-Adapter以运行pipeline。
 
 ```py
 dd_pipeline = dd_blocks.init_pipeline("YiYiXu/modular-demo-auto", collection="diffdiff")
-dd_pipeline.load_default_components(torch_dtype=torch.float16)
+dd_pipeline.load_components(torch_dtype=torch.float16)
 dd_pipeline.loader.load_ip_adapter("h94/IP-Adapter", subfolder="sdxl_models", weight_name="ip-adapter_sdxl.bin")
 dd_pipeline.loader.set_ip_adapter_scale(0.6)
 dd_pipeline = dd_pipeline.to(device)
@@ -261,14 +261,14 @@ class SDXLDiffDiffControlNetDenoiseStep(StableDiffusionXLDenoiseLoopWrapper):
 controlnet_denoise_block = SDXLDiffDiffControlNetDenoiseStep()
 ```
 
-插入 `controlnet_input` 块并用新的 `controlnet_denoise_block` 替换 `denoise` 块。初始化一个 [`ModularPipeline`] 并将 [`~ModularPipeline.load_default_components`] 加载到其中。
+插入 `controlnet_input` 块并用新的 `controlnet_denoise_block` 替换 `denoise` 块。初始化一个 [`ModularPipeline`] 并将 [`~ModularPipeline.load_components`] 加载到其中。
 
 ```py
 dd_blocks.sub_blocks.insert("controlnet_input", control_input_block, 7)
 dd_blocks.sub_blocks["denoise"] = controlnet_denoise_block
 
 dd_pipeline = dd_blocks.init_pipeline("YiYiXu/modular-demo-auto", collection="diffdiff")
-dd_pipeline.load_default_components(torch_dtype=torch.float16)
+dd_pipeline.load_components(torch_dtype=torch.float16)
 dd_pipeline = dd_pipeline.to(device)
 
 control_image = load_image("https://huggingface.co/datasets/YiYiXu/testing-images/resolve/main/diffdiff_tomato_canny.jpeg")
@@ -322,7 +322,7 @@ DIFFDIFF_AUTO_BLOCKS.insert("controlnet_input",StableDiffusionXLControlNetAutoIn
 ```py
 dd_auto_blocks = SequentialPipelineBlocks.from_blocks_dict(DIFFDIFF_AUTO_BLOCKS)
 dd_pipeline = dd_auto_blocks.init_pipeline("YiYiXu/modular-demo-auto", collection="diffdiff")
-dd_pipeline.load_default_components(torch_dtype=torch.float16)
+dd_pipeline.load_components(torch_dtype=torch.float16)
 ```
 
 ## 分享
@@ -342,5 +342,5 @@ from diffusers.modular_pipelines import ModularPipeline, ComponentsManager
 components = ComponentsManager()
 
 diffdiff_pipeline = ModularPipeline.from_pretrained("YiYiXu/modular-diffdiff-0704", trust_remote_code=True, components_manager=components, collection="diffdiff")
-diffdiff_pipeline.load_default_components(torch_dtype=torch.float16)
+diffdiff_pipeline.load_components(torch_dtype=torch.float16)
 ```

docs/source/zh/training/distributed_inference.md

@@ -223,7 +223,7 @@ from diffusers.image_processor import VaeImageProcessor
 import torch 
 
 vae = AutoencoderKL.from_pretrained(ckpt_id, subfolder="vae", torch_dtype=torch.bfloat16).to("cuda")
-vae_scale_factor = 2 ** (len(vae.config.block_out_channels))
+vae_scale_factor = 2 ** (len(vae.config.block_out_channels) - 1)
 image_processor = VaeImageProcessor(vae_scale_factor=vae_scale_factor)
 
 with torch.no_grad():

examples/conftest.py

@@ -25,6 +25,11 @@ from os.path import abspath, dirname, join
 git_repo_path = abspath(join(dirname(dirname(dirname(__file__))), "src"))
 sys.path.insert(1, git_repo_path)
 
+# Add parent directory to path so we can import from tests
+repo_root = abspath(dirname(dirname(__file__)))
+if repo_root not in sys.path:
+    sys.path.insert(0, repo_root)
+
 
 # silence FutureWarning warnings in tests since often we can't act on them until
 # they become normal warnings - i.e. the tests still need to test the current functionality
@@ -32,13 +37,13 @@ warnings.simplefilter(action="ignore", category=FutureWarning)
 
 
 def pytest_addoption(parser):
-    from diffusers.utils.testing_utils import pytest_addoption_shared
+    from tests.testing_utils import pytest_addoption_shared
 
     pytest_addoption_shared(parser)
 
 
 def pytest_terminal_summary(terminalreporter):
-    from diffusers.utils.testing_utils import pytest_terminal_summary_main
+    from tests.testing_utils import pytest_terminal_summary_main
 
     make_reports = terminalreporter.config.getoption("--make-reports")
     if make_reports:

examples/controlnet/train_controlnet_sd3.py

@@ -24,6 +24,8 @@ import math
 import os
 import random
 import shutil
+
+# Add repo root to path to import from tests
 from pathlib import Path
 
 import accelerate
@@ -54,8 +56,7 @@ from diffusers.optimization import get_scheduler
 from diffusers.training_utils import compute_density_for_timestep_sampling, compute_loss_weighting_for_sd3, free_memory
 from diffusers.utils import check_min_version, is_wandb_available, make_image_grid
 from diffusers.utils.hub_utils import load_or_create_model_card, populate_model_card
-from diffusers.utils.testing_utils import backend_empty_cache
-from diffusers.utils.torch_utils import is_compiled_module
+from diffusers.utils.torch_utils import backend_empty_cache, is_compiled_module
 
 
 if is_wandb_available():

examples/dreambooth/train_dreambooth_lora_flux_kontext.py

@@ -1270,6 +1270,7 @@ def main(args):
                 subfolder="transformer",
                 revision=args.revision,
                 variant=args.variant,
+                torch_dtype=torch_dtype,
             )
             pipeline = FluxKontextPipeline.from_pretrained(
                 args.pretrained_model_name_or_path,
@@ -1292,7 +1293,8 @@ def main(args):
             for example in tqdm(
                 sample_dataloader, desc="Generating class images", disable=not accelerator.is_local_main_process
             ):
-                images = pipeline(example["prompt"]).images
+                with torch.autocast(device_type=accelerator.device.type, dtype=torch_dtype):
+                    images = pipeline(prompt=example["prompt"]).images
 
                 for i, image in enumerate(images):
                     hash_image = insecure_hashlib.sha1(image.tobytes()).hexdigest()
@@ -1899,6 +1901,10 @@ def main(args):
                             device=accelerator.device,
                             prompt=args.instance_prompt,
                         )
+                    else:
+                        prompt_embeds, pooled_prompt_embeds, text_ids = compute_text_embeddings(
+                            prompts, text_encoders, tokenizers
+                        )
 
                 # Convert images to latent space
                 if args.cache_latents:

examples/vqgan/test_vqgan.py

@@ -24,12 +24,18 @@ import tempfile
 import torch
 
 from diffusers import VQModel
-from diffusers.utils.testing_utils import require_timm
 
 
+# Add parent directories to path to import from tests
 sys.path.append("..")
+repo_root = os.path.abspath(os.path.join(os.path.dirname(__file__), "../.."))
+if repo_root not in sys.path:
+    sys.path.insert(0, repo_root)
+
 from test_examples_utils import ExamplesTestsAccelerate, run_command  # noqa: E402
 
+from tests.testing_utils import require_timm  # noqa
+
 
 logging.basicConfig(level=logging.DEBUG)
 

src/diffusers/__init__.py

@@ -494,6 +494,7 @@ else:
             "PixArtSigmaPAGPipeline",
             "PixArtSigmaPipeline",
             "QwenImageControlNetPipeline",
+            "QwenImageEditInpaintPipeline",
             "QwenImageEditPipeline",
             "QwenImageImg2ImgPipeline",
             "QwenImageInpaintPipeline",
@@ -1134,6 +1135,7 @@ if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
             PixArtSigmaPAGPipeline,
             PixArtSigmaPipeline,
             QwenImageControlNetPipeline,
+            QwenImageEditInpaintPipeline,
             QwenImageEditPipeline,
             QwenImageImg2ImgPipeline,
             QwenImageInpaintPipeline,

src/diffusers/guiders/auto_guidance.py

@@ -82,15 +82,15 @@ class AutoGuidance(BaseGuidance):
         self.guidance_rescale = guidance_rescale
         self.use_original_formulation = use_original_formulation
 
-        if auto_guidance_layers is None and auto_guidance_config is None:
+        is_layer_or_config_provided = auto_guidance_layers is not None or auto_guidance_config is not None
+        is_layer_and_config_provided = auto_guidance_layers is not None and auto_guidance_config is not None
+        if not is_layer_or_config_provided:
             raise ValueError(
-                "Either `auto_guidance_layers` or `auto_guidance_config` must be provided to enable Skip Layer Guidance."
+                "Either `auto_guidance_layers` or `auto_guidance_config` must be provided to enable AutoGuidance."
             )
-        if auto_guidance_layers is not None and auto_guidance_config is not None:
+        if is_layer_and_config_provided:
             raise ValueError("Only one of `auto_guidance_layers` or `auto_guidance_config` can be provided.")
-        if (dropout is None and auto_guidance_layers is not None) or (
-            dropout is not None and auto_guidance_layers is None
-        ):
+        if auto_guidance_config is None and dropout is None:
             raise ValueError("`dropout` must be provided if `auto_guidance_layers` is provided.")
 
         if auto_guidance_layers is not None:

src/diffusers/loaders/lora_conversion_utils.py

@@ -2129,6 +2129,10 @@ def _convert_non_diffusers_ltxv_lora_to_diffusers(state_dict, non_diffusers_pref
 
 
 def _convert_non_diffusers_qwen_lora_to_diffusers(state_dict):
+    has_diffusion_model = any(k.startswith("diffusion_model.") for k in state_dict)
+    if has_diffusion_model:
+        state_dict = {k.removeprefix("diffusion_model."): v for k, v in state_dict.items()}
+
     has_lora_unet = any(k.startswith("lora_unet_") for k in state_dict)
     if has_lora_unet:
         state_dict = {k.removeprefix("lora_unet_"): v for k, v in state_dict.items()}
@@ -2201,29 +2205,44 @@ def _convert_non_diffusers_qwen_lora_to_diffusers(state_dict):
     all_keys = list(state_dict.keys())
     down_key = ".lora_down.weight"
     up_key = ".lora_up.weight"
+    a_key = ".lora_A.weight"
+    b_key = ".lora_B.weight"
 
-    def get_alpha_scales(down_weight, alpha_key):
-        rank = down_weight.shape[0]
-        alpha = state_dict.pop(alpha_key).item()
-        scale = alpha / rank  # LoRA is scaled by 'alpha / rank' in forward pass, so we need to scale it back here
-        scale_down = scale
-        scale_up = 1.0
-        while scale_down * 2 < scale_up:
-            scale_down *= 2
-            scale_up /= 2
-        return scale_down, scale_up
+    has_non_diffusers_lora_id = any(down_key in k or up_key in k for k in all_keys)
+    has_diffusers_lora_id = any(a_key in k or b_key in k for k in all_keys)
 
-    for k in all_keys:
-        if k.endswith(down_key):
-            diffusers_down_key = k.replace(down_key, ".lora_A.weight")
-            diffusers_up_key = k.replace(down_key, up_key).replace(up_key, ".lora_B.weight")
-            alpha_key = k.replace(down_key, ".alpha")
+    if has_non_diffusers_lora_id:
 
-            down_weight = state_dict.pop(k)
-            up_weight = state_dict.pop(k.replace(down_key, up_key))
-            scale_down, scale_up = get_alpha_scales(down_weight, alpha_key)
-            converted_state_dict[diffusers_down_key] = down_weight * scale_down
-            converted_state_dict[diffusers_up_key] = up_weight * scale_up
+        def get_alpha_scales(down_weight, alpha_key):
+            rank = down_weight.shape[0]
+            alpha = state_dict.pop(alpha_key).item()
+            scale = alpha / rank  # LoRA is scaled by 'alpha / rank' in forward pass, so we need to scale it back here
+            scale_down = scale
+            scale_up = 1.0
+            while scale_down * 2 < scale_up:
+                scale_down *= 2
+                scale_up /= 2
+            return scale_down, scale_up
+
+        for k in all_keys:
+            if k.endswith(down_key):
+                diffusers_down_key = k.replace(down_key, ".lora_A.weight")
+                diffusers_up_key = k.replace(down_key, up_key).replace(up_key, ".lora_B.weight")
+                alpha_key = k.replace(down_key, ".alpha")
+
+                down_weight = state_dict.pop(k)
+                up_weight = state_dict.pop(k.replace(down_key, up_key))
+                scale_down, scale_up = get_alpha_scales(down_weight, alpha_key)
+                converted_state_dict[diffusers_down_key] = down_weight * scale_down
+                converted_state_dict[diffusers_up_key] = up_weight * scale_up
+
+    # Already in diffusers format (lora_A/lora_B), just pop
+    elif has_diffusers_lora_id:
+        for k in all_keys:
+            if a_key in k or b_key in k:
+                converted_state_dict[k] = state_dict.pop(k)
+            elif ".alpha" in k:
+                state_dict.pop(k)
 
     if len(state_dict) > 0:
         raise ValueError(f"`state_dict` should be empty at this point but has {state_dict.keys()=}")

src/diffusers/loaders/lora_pipeline.py

@@ -6684,7 +6684,8 @@ class QwenImageLoraLoaderMixin(LoraBaseMixin):
 
         has_alphas_in_sd = any(k.endswith(".alpha") for k in state_dict)
         has_lora_unet = any(k.startswith("lora_unet_") for k in state_dict)
-        if has_alphas_in_sd or has_lora_unet:
+        has_diffusion_model = any(k.startswith("diffusion_model.") for k in state_dict)
+        if has_alphas_in_sd or has_lora_unet or has_diffusion_model:
             state_dict = _convert_non_diffusers_qwen_lora_to_diffusers(state_dict)
 
         out = (state_dict, metadata) if return_lora_metadata else state_dict

src/diffusers/models/attention_dispatch.py

@@ -955,12 +955,13 @@ def _native_npu_attention(
     dropout_p: float = 0.0,
     scale: Optional[float] = None,
 ) -> torch.Tensor:
-    return npu_fusion_attention(
+    query, key, value = (x.transpose(1, 2).contiguous() for x in (query, key, value))
+    out = npu_fusion_attention(
         query,
         key,
         value,
-        query.size(2),  # num_heads
-        input_layout="BSND",
+        query.size(1),  # num_heads
+        input_layout="BNSD",
         pse=None,
         scale=1.0 / math.sqrt(query.shape[-1]) if scale is None else scale,
         pre_tockens=65536,
@@ -969,6 +970,8 @@ def _native_npu_attention(
         sync=False,
         inner_precise=0,
     )[0]
+    out = out.transpose(1, 2).contiguous()
+    return out
 
 
 # Reference: https://github.com/pytorch/xla/blob/06c5533de6588f6b90aa1655d9850bcf733b90b4/torch_xla/experimental/custom_kernel.py#L853

src/diffusers/modular_pipelines/modular_pipeline.py

@@ -299,7 +299,7 @@ class ModularPipelineBlocks(ConfigMixin, PushToHubMixin):
     def from_pretrained(
         cls,
         pretrained_model_name_or_path: str,
-        trust_remote_code: Optional[bool] = None,
+        trust_remote_code: bool = False,
         **kwargs,
     ):
         hub_kwargs_names = [
@@ -1418,7 +1418,7 @@ class LoopSequentialPipelineBlocks(ModularPipelineBlocks):
 # YiYi TODO:
 # 1. look into the serialization of modular_model_index.json, make sure the items are properly ordered like model_index.json (currently a mess)
 # 2. do we need ConfigSpec? the are basically just key/val kwargs
-# 3. imnprove docstring and potentially add validator for methods where we accpet kwargs to be passed to from_pretrained/save_pretrained/load_default_components(), load_components()
+# 3. imnprove docstring and potentially add validator for methods where we accpet kwargs to be passed to from_pretrained/save_pretrained/load_components()
 class ModularPipeline(ConfigMixin, PushToHubMixin):
     """
     Base class for all Modular pipelines.
@@ -1488,7 +1488,7 @@ class ModularPipeline(ConfigMixin, PushToHubMixin):
             - Components with default_creation_method="from_config" are created immediately, its specs are not included
               in config dict and will not be saved in `modular_model_index.json`
             - Components with default_creation_method="from_pretrained" are set to None and can be loaded later with
-              `load_default_components()`/`load_components()`
+              `load_components()` (with or without specific component names)
             - The pipeline's config dict is populated with component specs (only for from_pretrained components) and
               config values, which will be saved as `modular_model_index.json` during `save_pretrained`
             - The pipeline's config dict is also used to store the pipeline blocks's class name, which will be saved as
@@ -1603,20 +1603,6 @@ class ModularPipeline(ConfigMixin, PushToHubMixin):
             params[input_param.name] = input_param.default
         return params
 
-    def load_default_components(self, **kwargs):
-        """
-        Load from_pretrained components using the loading specs in the config dict.
-
-        Args:
-            **kwargs: Additional arguments passed to `from_pretrained` method, e.g. torch_dtype, cache_dir, etc.
-        """
-        names = [
-            name
-            for name in self._component_specs.keys()
-            if self._component_specs[name].default_creation_method == "from_pretrained"
-        ]
-        self.load_components(names=names, **kwargs)
-
     @classmethod
     @validate_hf_hub_args
     def from_pretrained(
@@ -1770,8 +1756,8 @@ class ModularPipeline(ConfigMixin, PushToHubMixin):
            - non from_pretrained components are created during __init__ and registered as the object itself
         - Components are updated with the `update_components()` method: e.g. loader.update_components(unet=unet) or
           loader.update_components(guider=guider_spec)
-        - (from_pretrained) Components are loaded with the `load_default_components()` method: e.g.
-          loader.load_default_components(names=["unet"])
+        - (from_pretrained) Components are loaded with the `load_components()` method: e.g.
+          loader.load_components(names=["unet"]) or loader.load_components() to load all default components
 
         Args:
             **kwargs: Keyword arguments where keys are component names and values are component objects.
@@ -2097,13 +2083,14 @@ class ModularPipeline(ConfigMixin, PushToHubMixin):
         self.register_to_config(**config_to_register)
 
     # YiYi TODO: support map for additional from_pretrained kwargs
-    # YiYi/Dhruv TODO: consolidate load_components and load_default_components?
-    def load_components(self, names: Union[List[str], str], **kwargs):
+    def load_components(self, names: Optional[Union[List[str], str]] = None, **kwargs):
         """
         Load selected components from specs.
 
         Args:
-            names: List of component names to load; by default will not load any components
+            names: List of component names to load. If None, will load all components with
+                   default_creation_method == "from_pretrained". If provided as a list or string, will load only the
+                   specified components.
             **kwargs: additional kwargs to be passed to `from_pretrained()`.Can be:
              - a single value to be applied to all components to be loaded, e.g. torch_dtype=torch.bfloat16
              - a dict, e.g. torch_dtype={"unet": torch.bfloat16, "default": torch.float32}
@@ -2111,7 +2098,13 @@ class ModularPipeline(ConfigMixin, PushToHubMixin):
                `variant`, `revision`, etc.
         """
 
-        if isinstance(names, str):
+        if names is None:
+            names = [
+                name
+                for name in self._component_specs.keys()
+                if self._component_specs[name].default_creation_method == "from_pretrained"
+            ]
+        elif isinstance(names, str):
             names = [names]
         elif not isinstance(names, list):
             raise ValueError(f"Invalid type for names: {type(names)}")

src/diffusers/pipelines/__init__.py

@@ -393,6 +393,7 @@ else:
         "QwenImageImg2ImgPipeline",
         "QwenImageInpaintPipeline",
         "QwenImageEditPipeline",
+        "QwenImageEditInpaintPipeline",
         "QwenImageControlNetPipeline",
     ]
 try:
@@ -714,6 +715,7 @@ if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
         from .pixart_alpha import PixArtAlphaPipeline, PixArtSigmaPipeline
         from .qwenimage import (
             QwenImageControlNetPipeline,
+            QwenImageEditInpaintPipeline,
             QwenImageEditPipeline,
             QwenImageImg2ImgPipeline,
             QwenImageInpaintPipeline,

src/diffusers/pipelines/qwenimage/__init__.py

@@ -26,6 +26,7 @@ else:
     _import_structure["pipeline_qwenimage"] = ["QwenImagePipeline"]
     _import_structure["pipeline_qwenimage_controlnet"] = ["QwenImageControlNetPipeline"]
     _import_structure["pipeline_qwenimage_edit"] = ["QwenImageEditPipeline"]
+    _import_structure["pipeline_qwenimage_edit_inpaint"] = ["QwenImageEditInpaintPipeline"]
     _import_structure["pipeline_qwenimage_img2img"] = ["QwenImageImg2ImgPipeline"]
     _import_structure["pipeline_qwenimage_inpaint"] = ["QwenImageInpaintPipeline"]
 
@@ -39,6 +40,7 @@ if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
         from .pipeline_qwenimage import QwenImagePipeline
         from .pipeline_qwenimage_controlnet import QwenImageControlNetPipeline
         from .pipeline_qwenimage_edit import QwenImageEditPipeline
+        from .pipeline_qwenimage_edit_inpaint import QwenImageEditInpaintPipeline
         from .pipeline_qwenimage_img2img import QwenImageImg2ImgPipeline
         from .pipeline_qwenimage_inpaint import QwenImageInpaintPipeline
 else:

src/diffusers/pipelines/qwenimage/pipeline_qwenimage_edit.py

@@ -551,6 +551,12 @@ class QwenImageEditPipeline(DiffusionPipeline, QwenImageLoraLoaderMixin):
         Function invoked when calling the pipeline for generation.
 
         Args:
+            image (`torch.Tensor`, `PIL.Image.Image`, `np.ndarray`, `List[torch.Tensor]`, `List[PIL.Image.Image]`, or `List[np.ndarray]`):
+                `Image`, numpy array or tensor representing an image batch to be used as the starting point. For both
+                numpy array and pytorch tensor, the expected value range is between `[0, 1]` If it's a tensor or a list
+                or tensors, the expected shape should be `(B, C, H, W)` or `(C, H, W)`. If it is a numpy array or a
+                list of arrays, the expected shape should be `(B, H, W, C)` or `(H, W, C)` It can also accept image
+                latents as `image`, but if passing latents directly it is not encoded again.
             prompt (`str` or `List[str]`, *optional*):
                 The prompt or prompts to guide the image generation. If not defined, one has to pass `prompt_embeds`.
                 instead.

src/diffusers/utils/constants.py

@@ -45,6 +45,7 @@ DIFFUSERS_ATTN_BACKEND = os.getenv("DIFFUSERS_ATTN_BACKEND", "native")
 DIFFUSERS_ATTN_CHECKS = os.getenv("DIFFUSERS_ATTN_CHECKS", "0") in ENV_VARS_TRUE_VALUES
 DEFAULT_HF_PARALLEL_LOADING_WORKERS = 8
 HF_ENABLE_PARALLEL_LOADING = os.environ.get("HF_ENABLE_PARALLEL_LOADING", "").upper() in ENV_VARS_TRUE_VALUES
+DIFFUSERS_DISABLE_REMOTE_CODE = os.getenv("DIFFUSERS_DISABLE_REMOTE_CODE", "false").lower() in ENV_VARS_TRUE_VALUES
 
 # Below should be `True` if the current version of `peft` and `transformers` are compatible with
 # PEFT backend. Will automatically fall back to PEFT backend if the correct versions of the libraries are

src/diffusers/utils/dummy_torch_and_transformers_objects.py

@@ -1772,6 +1772,21 @@ class QwenImageControlNetPipeline(metaclass=DummyObject):
         requires_backends(cls, ["torch", "transformers"])
 
 
+class QwenImageEditInpaintPipeline(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 QwenImageEditPipeline(metaclass=DummyObject):
     _backends = ["torch", "transformers"]
 

src/diffusers/utils/dynamic_modules_utils.py

@@ -20,7 +20,6 @@ import json
 import os
 import re
 import shutil
-import signal
 import sys
 import threading
 from pathlib import Path
@@ -34,6 +33,7 @@ from packaging import version
 
 from .. import __version__
 from . import DIFFUSERS_DYNAMIC_MODULE_NAME, HF_MODULES_CACHE, logging
+from .constants import DIFFUSERS_DISABLE_REMOTE_CODE
 
 
 logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
@@ -159,52 +159,25 @@ def check_imports(filename):
     return get_relative_imports(filename)
 
 
-def _raise_timeout_error(signum, frame):
-    raise ValueError(
-        "Loading this model requires you to execute custom code contained in the model repository on your local "
-        "machine. Please set the option `trust_remote_code=True` to permit loading of this model."
-    )
-
-
 def resolve_trust_remote_code(trust_remote_code, model_name, has_remote_code):
-    if trust_remote_code is None:
-        if has_remote_code and TIME_OUT_REMOTE_CODE > 0:
-            prev_sig_handler = None
-            try:
-                prev_sig_handler = signal.signal(signal.SIGALRM, _raise_timeout_error)
-                signal.alarm(TIME_OUT_REMOTE_CODE)
-                while trust_remote_code is None:
-                    answer = input(
-                        f"The repository for {model_name} contains custom code which must be executed to correctly "
-                        f"load the model. You can inspect the repository content at https://hf.co/{model_name}.\n"
-                        f"You can avoid this prompt in future by passing the argument `trust_remote_code=True`.\n\n"
-                        f"Do you wish to run the custom code? [y/N] "
-                    )
-                    if answer.lower() in ["yes", "y", "1"]:
-                        trust_remote_code = True
-                    elif answer.lower() in ["no", "n", "0", ""]:
-                        trust_remote_code = False
-                signal.alarm(0)
-            except Exception:
-                # OS which does not support signal.SIGALRM
-                raise ValueError(
-                    f"The repository for {model_name} contains custom code which must be executed to correctly "
-                    f"load the model. You can inspect the repository content at https://hf.co/{model_name}.\n"
-                    f"Please pass the argument `trust_remote_code=True` to allow custom code to be run."
-                )
-            finally:
-                if prev_sig_handler is not None:
-                    signal.signal(signal.SIGALRM, prev_sig_handler)
-                    signal.alarm(0)
-        elif has_remote_code:
-            # For the CI which puts the timeout at 0
-            _raise_timeout_error(None, None)
+    trust_remote_code = trust_remote_code and not DIFFUSERS_DISABLE_REMOTE_CODE
+    if DIFFUSERS_DISABLE_REMOTE_CODE:
+        logger.warning(
+            "Downloading remote code is disabled globally via the DIFFUSERS_DISABLE_REMOTE_CODE environment variable. Ignoring `trust_remote_code`."
+        )
 
     if has_remote_code and not trust_remote_code:
-        raise ValueError(
-            f"Loading {model_name} requires you to execute the configuration file in that"
-            " repo on your local machine. Make sure you have read the code there to avoid malicious use, then"
-            " set the option `trust_remote_code=True` to remove this error."
+        error_msg = f"The repository for {model_name} contains custom code. "
+        error_msg += (
+            "Downloading remote code is disabled globally via the DIFFUSERS_DISABLE_REMOTE_CODE environment variable."
+            if DIFFUSERS_DISABLE_REMOTE_CODE
+            else "Pass `trust_remote_code=True` to allow loading remote code modules."
+        )
+        raise ValueError(error_msg)
+
+    elif has_remote_code and trust_remote_code:
+        logger.warning(
+            f"`trust_remote_code` is enabled. Downloading code from {model_name}. Please ensure you trust the contents of this repository"
         )
 
     return trust_remote_code

src/diffusers/utils/testing_utils.py

@@ -66,7 +66,10 @@ else:
 global_rng = random.Random()
 
 logger = get_logger(__name__)
-
+logger.warning(
+    "diffusers.utils.testing_utils' is deprecated and will be removed in a future version. "
+    "Determinism and device backend utilities have been moved to `diffusers.utils.torch_utils`. "
+)
 _required_peft_version = is_peft_available() and version.parse(
     version.parse(importlib.metadata.version("peft")).base_version
 ) > version.parse("0.5")
@@ -801,10 +804,9 @@ def export_to_ply(mesh, output_ply_path: str = None):
                 f.write(format.pack(*vertex))
 
         if faces is not None:
-            format = struct.Struct("<B3I")
             for tri in faces.tolist():
                 f.write(format.pack(len(tri), *tri))
-
+            format = struct.Struct("<B3I")
     return output_ply_path
 
 
@@ -1144,23 +1146,23 @@ def enable_full_determinism():
     Helper function for reproducible behavior during distributed training. See
     - https://pytorch.org/docs/stable/notes/randomness.html for pytorch
     """
-    #  Enable PyTorch deterministic mode. This potentially requires either the environment
-    #  variable 'CUDA_LAUNCH_BLOCKING' or 'CUBLAS_WORKSPACE_CONFIG' to be set,
-    # depending on the CUDA version, so we set them both here
-    os.environ["CUDA_LAUNCH_BLOCKING"] = "1"
-    os.environ["CUBLAS_WORKSPACE_CONFIG"] = ":16:8"
-    torch.use_deterministic_algorithms(True)
+    from .torch_utils import enable_full_determinism as _enable_full_determinism
 
-    # Enable CUDNN deterministic mode
-    torch.backends.cudnn.deterministic = True
-    torch.backends.cudnn.benchmark = False
-    torch.backends.cuda.matmul.allow_tf32 = False
+    logger.warning(
+        "enable_full_determinism has been moved to diffusers.utils.torch_utils. "
+        "Importing from diffusers.utils.testing_utils is deprecated and will be removed in a future version."
+    )
+    return _enable_full_determinism()
 
 
 def disable_full_determinism():
-    os.environ["CUDA_LAUNCH_BLOCKING"] = "0"
-    os.environ["CUBLAS_WORKSPACE_CONFIG"] = ""
-    torch.use_deterministic_algorithms(False)
+    from .torch_utils import disable_full_determinism as _disable_full_determinism
+
+    logger.warning(
+        "disable_full_determinism has been moved to diffusers.utils.torch_utils. "
+        "Importing from diffusers.utils.testing_utils is deprecated and will be removed in a future version."
+    )
+    return _disable_full_determinism()
 
 
 # Utils for custom and alternative accelerator devices
@@ -1282,43 +1284,85 @@ def _device_agnostic_dispatch(device: str, dispatch_table: Dict[str, Callable],
 
 # These are callables which automatically dispatch the function specific to the accelerator
 def backend_manual_seed(device: str, seed: int):
-    return _device_agnostic_dispatch(device, BACKEND_MANUAL_SEED, seed)
+    from .torch_utils import backend_manual_seed as _backend_manual_seed
+
+    logger.warning(
+        "backend_manual_seed has been moved to diffusers.utils.torch_utils. "
+        "diffusers.utils.testing_utils is deprecated and will be removed in a future version."
+    )
+    return _backend_manual_seed(device, seed)
 
 
 def backend_synchronize(device: str):
-    return _device_agnostic_dispatch(device, BACKEND_SYNCHRONIZE)
+    from .torch_utils import backend_synchronize as _backend_synchronize
+
+    logger.warning(
+        "backend_synchronize has been moved to diffusers.utils.torch_utils. "
+        "diffusers.utils.testing_utils is deprecated and will be removed in a future version."
+    )
+    return _backend_synchronize(device)
 
 
 def backend_empty_cache(device: str):
-    return _device_agnostic_dispatch(device, BACKEND_EMPTY_CACHE)
+    from .torch_utils import backend_empty_cache as _backend_empty_cache
+
+    logger.warning(
+        "backend_empty_cache has been moved to diffusers.utils.torch_utils. "
+        "diffusers.utils.testing_utils is deprecated and will be removed in a future version."
+    )
+    return _backend_empty_cache(device)
 
 
 def backend_device_count(device: str):
-    return _device_agnostic_dispatch(device, BACKEND_DEVICE_COUNT)
+    from .torch_utils import backend_device_count as _backend_device_count
+
+    logger.warning(
+        "backend_device_count has been moved to diffusers.utils.torch_utils. "
+        "diffusers.utils.testing_utils is deprecated and will be removed in a future version."
+    )
+    return _backend_device_count(device)
 
 
 def backend_reset_peak_memory_stats(device: str):
-    return _device_agnostic_dispatch(device, BACKEND_RESET_PEAK_MEMORY_STATS)
+    from .torch_utils import backend_reset_peak_memory_stats as _backend_reset_peak_memory_stats
+
+    logger.warning(
+        "backend_reset_peak_memory_stats has been moved to diffusers.utils.torch_utils. "
+        "diffusers.utils.testing_utils is deprecated and will be removed in a future version."
+    )
+    return _backend_reset_peak_memory_stats(device)
 
 
 def backend_reset_max_memory_allocated(device: str):
-    return _device_agnostic_dispatch(device, BACKEND_RESET_MAX_MEMORY_ALLOCATED)
+    from .torch_utils import backend_reset_max_memory_allocated as _backend_reset_max_memory_allocated
+
+    logger.warning(
+        "backend_reset_max_memory_allocated has been moved to diffusers.utils.torch_utils. "
+        "diffusers.utils.testing_utils is deprecated and will be removed in a future version."
+    )
+    return _backend_reset_max_memory_allocated(device)
 
 
 def backend_max_memory_allocated(device: str):
-    return _device_agnostic_dispatch(device, BACKEND_MAX_MEMORY_ALLOCATED)
+    from .torch_utils import backend_max_memory_allocated as _backend_max_memory_allocated
+
+    logger.warning(
+        "backend_max_memory_allocated has been moved to diffusers.utils.torch_utils. "
+        "diffusers.utils.testing_utils is deprecated and will be removed in a future version."
+    )
+    return _backend_max_memory_allocated(device)
 
 
 # These are callables which return boolean behaviour flags and can be used to specify some
 # device agnostic alternative where the feature is unsupported.
 def backend_supports_training(device: str):
-    if not is_torch_available():
-        return False
+    from .torch_utils import backend_supports_training as _backend_supports_training
 
-    if device not in BACKEND_SUPPORTS_TRAINING:
-        device = "default"
-
-    return BACKEND_SUPPORTS_TRAINING[device]
+    logger.warning(
+        "backend_supports_training has been moved to diffusers.utils.torch_utils. "
+        "diffusers.utils.testing_utils is deprecated and will be removed in a future version."
+    )
+    return _backend_supports_training(device)
 
 
 # Guard for when Torch is not available

src/diffusers/utils/torch_utils.py

@@ -16,7 +16,8 @@ PyTorch utilities: Utilities related to PyTorch
 """
 
 import functools
-from typing import List, Optional, Tuple, Union
+import os
+from typing import Callable, Dict, List, Optional, Tuple, Union
 
 from . import logging
 from .import_utils import is_torch_available, is_torch_npu_available, is_torch_version
@@ -26,6 +27,56 @@ if is_torch_available():
     import torch
     from torch.fft import fftn, fftshift, ifftn, ifftshift
 
+    BACKEND_SUPPORTS_TRAINING = {"cuda": True, "xpu": True, "cpu": True, "mps": False, "default": True}
+    BACKEND_EMPTY_CACHE = {
+        "cuda": torch.cuda.empty_cache,
+        "xpu": torch.xpu.empty_cache,
+        "cpu": None,
+        "mps": torch.mps.empty_cache,
+        "default": None,
+    }
+    BACKEND_DEVICE_COUNT = {
+        "cuda": torch.cuda.device_count,
+        "xpu": torch.xpu.device_count,
+        "cpu": lambda: 0,
+        "mps": lambda: 0,
+        "default": 0,
+    }
+    BACKEND_MANUAL_SEED = {
+        "cuda": torch.cuda.manual_seed,
+        "xpu": torch.xpu.manual_seed,
+        "cpu": torch.manual_seed,
+        "mps": torch.mps.manual_seed,
+        "default": torch.manual_seed,
+    }
+    BACKEND_RESET_PEAK_MEMORY_STATS = {
+        "cuda": torch.cuda.reset_peak_memory_stats,
+        "xpu": getattr(torch.xpu, "reset_peak_memory_stats", None),
+        "cpu": None,
+        "mps": None,
+        "default": None,
+    }
+    BACKEND_RESET_MAX_MEMORY_ALLOCATED = {
+        "cuda": torch.cuda.reset_max_memory_allocated,
+        "xpu": getattr(torch.xpu, "reset_peak_memory_stats", None),
+        "cpu": None,
+        "mps": None,
+        "default": None,
+    }
+    BACKEND_MAX_MEMORY_ALLOCATED = {
+        "cuda": torch.cuda.max_memory_allocated,
+        "xpu": getattr(torch.xpu, "max_memory_allocated", None),
+        "cpu": 0,
+        "mps": 0,
+        "default": 0,
+    }
+    BACKEND_SYNCHRONIZE = {
+        "cuda": torch.cuda.synchronize,
+        "xpu": getattr(torch.xpu, "synchronize", None),
+        "cpu": None,
+        "mps": None,
+        "default": None,
+    }
 logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
 
 try:
@@ -36,6 +87,62 @@ except (ImportError, ModuleNotFoundError):
         return cls
 
 
+# This dispatches a defined function according to the accelerator from the function definitions.
+def _device_agnostic_dispatch(device: str, dispatch_table: Dict[str, Callable], *args, **kwargs):
+    if device not in dispatch_table:
+        return dispatch_table["default"](*args, **kwargs)
+
+    fn = dispatch_table[device]
+
+    # Some device agnostic functions return values. Need to guard against 'None' instead at
+    # user level
+    if not callable(fn):
+        return fn
+
+    return fn(*args, **kwargs)
+
+
+# These are callables which automatically dispatch the function specific to the accelerator
+def backend_manual_seed(device: str, seed: int):
+    return _device_agnostic_dispatch(device, BACKEND_MANUAL_SEED, seed)
+
+
+def backend_synchronize(device: str):
+    return _device_agnostic_dispatch(device, BACKEND_SYNCHRONIZE)
+
+
+def backend_empty_cache(device: str):
+    return _device_agnostic_dispatch(device, BACKEND_EMPTY_CACHE)
+
+
+def backend_device_count(device: str):
+    return _device_agnostic_dispatch(device, BACKEND_DEVICE_COUNT)
+
+
+def backend_reset_peak_memory_stats(device: str):
+    return _device_agnostic_dispatch(device, BACKEND_RESET_PEAK_MEMORY_STATS)
+
+
+def backend_reset_max_memory_allocated(device: str):
+    return _device_agnostic_dispatch(device, BACKEND_RESET_MAX_MEMORY_ALLOCATED)
+
+
+def backend_max_memory_allocated(device: str):
+    return _device_agnostic_dispatch(device, BACKEND_MAX_MEMORY_ALLOCATED)
+
+
+# These are callables which return boolean behaviour flags and can be used to specify some
+# device agnostic alternative where the feature is unsupported.
+def backend_supports_training(device: str):
+    if not is_torch_available():
+        return False
+
+    if device not in BACKEND_SUPPORTS_TRAINING:
+        device = "default"
+
+    return BACKEND_SUPPORTS_TRAINING[device]
+
+
 def randn_tensor(
     shape: Union[Tuple, List],
     generator: Optional[Union[List["torch.Generator"], "torch.Generator"]] = None,
@@ -197,3 +304,31 @@ def device_synchronize(device_type: Optional[str] = None):
         device_type = get_device()
     device_mod = getattr(torch, device_type, torch.cuda)
     device_mod.synchronize()
+
+
+def enable_full_determinism():
+    """
+    Helper function for reproducible behavior during distributed training. See
+    - https://pytorch.org/docs/stable/notes/randomness.html for pytorch
+    """
+    #  Enable PyTorch deterministic mode. This potentially requires either the environment
+    #  variable 'CUDA_LAUNCH_BLOCKING' or 'CUBLAS_WORKSPACE_CONFIG' to be set,
+    # depending on the CUDA version, so we set them both here
+    os.environ["CUDA_LAUNCH_BLOCKING"] = "1"
+    os.environ["CUBLAS_WORKSPACE_CONFIG"] = ":16:8"
+    torch.use_deterministic_algorithms(True)
+
+    # Enable CUDNN deterministic mode
+    torch.backends.cudnn.deterministic = True
+    torch.backends.cudnn.benchmark = False
+    torch.backends.cuda.matmul.allow_tf32 = False
+
+
+def disable_full_determinism():
+    os.environ["CUDA_LAUNCH_BLOCKING"] = "0"
+    os.environ["CUBLAS_WORKSPACE_CONFIG"] = ""
+    torch.use_deterministic_algorithms(False)
+
+
+if is_torch_available():
+    torch_device = get_device()

tests/conftest.py

@@ -35,13 +35,13 @@ def pytest_configure(config):
 
 
 def pytest_addoption(parser):
-    from diffusers.utils.testing_utils import pytest_addoption_shared
+    from .testing_utils import pytest_addoption_shared
 
     pytest_addoption_shared(parser)
 
 
 def pytest_terminal_summary(terminalreporter):
-    from diffusers.utils.testing_utils import pytest_terminal_summary_main
+    from .testing_utils import pytest_terminal_summary_main
 
     make_reports = terminalreporter.config.getoption("--make-reports")
     if make_reports:

tests/hooks/test_group_offloading.py

@@ -24,7 +24,8 @@ from diffusers.models import ModelMixin
 from diffusers.pipelines.pipeline_utils import DiffusionPipeline
 from diffusers.utils import get_logger
 from diffusers.utils.import_utils import compare_versions
-from diffusers.utils.testing_utils import (
+
+from ..testing_utils import (
     backend_empty_cache,
     backend_max_memory_allocated,
     backend_reset_peak_memory_stats,

tests/hooks/test_hooks.py

@@ -20,7 +20,8 @@ import torch
 from diffusers.hooks import HookRegistry, ModelHook
 from diffusers.training_utils import free_memory
 from diffusers.utils.logging import get_logger
-from diffusers.utils.testing_utils import CaptureLogger, torch_device
+
+from ..testing_utils import CaptureLogger, torch_device
 
 
 logger = get_logger(__name__)  # pylint: disable=invalid-name

tests/lora/test_lora_layers_auraflow.py

@@ -23,7 +23,8 @@ from diffusers import (
     AuraFlowTransformer2DModel,
     FlowMatchEulerDiscreteScheduler,
 )
-from diffusers.utils.testing_utils import (
+
+from ..testing_utils import (
     floats_tensor,
     is_peft_available,
     require_peft_backend,
@@ -35,7 +36,7 @@ if is_peft_available():
 
 sys.path.append(".")
 
-from utils import PeftLoraLoaderMixinTests  # noqa: E402
+from .utils import PeftLoraLoaderMixinTests  # noqa: E402
 
 
 @require_peft_backend

tests/lora/test_lora_layers_cogvideox.py

@@ -26,7 +26,8 @@ from diffusers import (
     CogVideoXPipeline,
     CogVideoXTransformer3DModel,
 )
-from diffusers.utils.testing_utils import (
+
+from ..testing_utils import (
     floats_tensor,
     require_peft_backend,
     require_torch_accelerator,
@@ -35,7 +36,7 @@ from diffusers.utils.testing_utils import (
 
 sys.path.append(".")
 
-from utils import PeftLoraLoaderMixinTests  # noqa: E402
+from .utils import PeftLoraLoaderMixinTests  # noqa: E402
 
 
 @require_peft_backend

tests/lora/test_lora_layers_cogview4.py

@@ -22,7 +22,8 @@ from parameterized import parameterized
 from transformers import AutoTokenizer, GlmModel
 
 from diffusers import AutoencoderKL, CogView4Pipeline, CogView4Transformer2DModel, FlowMatchEulerDiscreteScheduler
-from diffusers.utils.testing_utils import (
+
+from ..testing_utils import (
     floats_tensor,
     require_peft_backend,
     require_torch_accelerator,
@@ -33,7 +34,7 @@ from diffusers.utils.testing_utils import (
 
 sys.path.append(".")
 
-from utils import PeftLoraLoaderMixinTests  # noqa: E402
+from .utils import PeftLoraLoaderMixinTests  # noqa: E402
 
 
 class TokenizerWrapper:

tests/lora/test_lora_layers_flux.py

@@ -28,7 +28,8 @@ from transformers import AutoTokenizer, CLIPTextModel, CLIPTokenizer, T5EncoderM
 
 from diffusers import FlowMatchEulerDiscreteScheduler, FluxControlPipeline, FluxPipeline, FluxTransformer2DModel
 from diffusers.utils import load_image, logging
-from diffusers.utils.testing_utils import (
+
+from ..testing_utils import (
     CaptureLogger,
     backend_empty_cache,
     floats_tensor,
@@ -48,7 +49,7 @@ if is_peft_available():
 
 sys.path.append(".")
 
-from utils import PeftLoraLoaderMixinTests, check_if_lora_correctly_set  # noqa: E402
+from .utils import PeftLoraLoaderMixinTests, check_if_lora_correctly_set  # noqa: E402
 
 
 @require_peft_backend

tests/lora/test_lora_layers_hunyuanvideo.py

@@ -26,7 +26,8 @@ from diffusers import (
     HunyuanVideoPipeline,
     HunyuanVideoTransformer3DModel,
 )
-from diffusers.utils.testing_utils import (
+
+from ..testing_utils import (
     Expectations,
     backend_empty_cache,
     floats_tensor,
@@ -42,7 +43,7 @@ from diffusers.utils.testing_utils import (
 
 sys.path.append(".")
 
-from utils import PeftLoraLoaderMixinTests  # noqa: E402
+from .utils import PeftLoraLoaderMixinTests  # noqa: E402
 
 
 @require_peft_backend

tests/lora/test_lora_layers_ltx_video.py

@@ -24,12 +24,13 @@ from diffusers import (
     LTXPipeline,
     LTXVideoTransformer3DModel,
 )
-from diffusers.utils.testing_utils import floats_tensor, require_peft_backend
+
+from ..testing_utils import floats_tensor, require_peft_backend
 
 
 sys.path.append(".")
 
-from utils import PeftLoraLoaderMixinTests  # noqa: E402
+from .utils import PeftLoraLoaderMixinTests  # noqa: E402
 
 
 @require_peft_backend

tests/lora/test_lora_layers_lumina2.py

@@ -26,12 +26,13 @@ from diffusers import (
     Lumina2Pipeline,
     Lumina2Transformer2DModel,
 )
-from diffusers.utils.testing_utils import floats_tensor, is_torch_version, require_peft_backend, skip_mps, torch_device
+
+from ..testing_utils import floats_tensor, is_torch_version, require_peft_backend, skip_mps, torch_device
 
 
 sys.path.append(".")
 
-from utils import PeftLoraLoaderMixinTests, check_if_lora_correctly_set  # noqa: E402
+from .utils import PeftLoraLoaderMixinTests, check_if_lora_correctly_set  # noqa: E402
 
 
 @require_peft_backend

tests/lora/test_lora_layers_mochi.py

@@ -19,7 +19,8 @@ import torch
 from transformers import AutoTokenizer, T5EncoderModel
 
 from diffusers import AutoencoderKLMochi, FlowMatchEulerDiscreteScheduler, MochiPipeline, MochiTransformer3DModel
-from diffusers.utils.testing_utils import (
+
+from ..testing_utils import (
     floats_tensor,
     require_peft_backend,
     skip_mps,
@@ -28,7 +29,7 @@ from diffusers.utils.testing_utils import (
 
 sys.path.append(".")
 
-from utils import PeftLoraLoaderMixinTests  # noqa: E402
+from .utils import PeftLoraLoaderMixinTests  # noqa: E402
 
 
 @require_peft_backend

tests/lora/test_lora_layers_qwenimage.py

@@ -24,12 +24,13 @@ from diffusers import (
     QwenImagePipeline,
     QwenImageTransformer2DModel,
 )
-from diffusers.utils.testing_utils import floats_tensor, require_peft_backend
+
+from ..testing_utils import floats_tensor, require_peft_backend
 
 
 sys.path.append(".")
 
-from utils import PeftLoraLoaderMixinTests  # noqa: E402
+from .utils import PeftLoraLoaderMixinTests  # noqa: E402
 
 
 @require_peft_backend

tests/lora/test_lora_layers_sana.py

@@ -19,12 +19,13 @@ import torch
 from transformers import Gemma2Model, GemmaTokenizer
 
 from diffusers import AutoencoderDC, FlowMatchEulerDiscreteScheduler, SanaPipeline, SanaTransformer2DModel
-from diffusers.utils.testing_utils import floats_tensor, require_peft_backend
+
+from ..testing_utils import floats_tensor, require_peft_backend
 
 
 sys.path.append(".")
 
-from utils import PeftLoraLoaderMixinTests  # noqa: E402
+from .utils import PeftLoraLoaderMixinTests  # noqa: E402
 
 
 @require_peft_backend

tests/lora/test_lora_layers_sd.py

@@ -32,7 +32,8 @@ from diffusers import (
     StableDiffusionPipeline,
 )
 from diffusers.utils.import_utils import is_accelerate_available
-from diffusers.utils.testing_utils import (
+
+from ..testing_utils import (
     Expectations,
     backend_empty_cache,
     load_image,
@@ -47,7 +48,7 @@ from diffusers.utils.testing_utils import (
 
 sys.path.append(".")
 
-from utils import PeftLoraLoaderMixinTests, check_if_lora_correctly_set  # noqa: E402
+from .utils import PeftLoraLoaderMixinTests, check_if_lora_correctly_set  # noqa: E402
 
 
 if is_accelerate_available():

tests/lora/test_lora_layers_sd3.py

@@ -28,7 +28,8 @@ from diffusers import (
 )
 from diffusers.utils import load_image
 from diffusers.utils.import_utils import is_accelerate_available
-from diffusers.utils.testing_utils import (
+
+from ..testing_utils import (
     backend_empty_cache,
     is_flaky,
     nightly,
@@ -42,7 +43,7 @@ from diffusers.utils.testing_utils import (
 
 sys.path.append(".")
 
-from utils import PeftLoraLoaderMixinTests  # noqa: E402
+from .utils import PeftLoraLoaderMixinTests  # noqa: E402
 
 
 if is_accelerate_available():

tests/lora/test_lora_layers_sdxl.py

@@ -35,7 +35,8 @@ from diffusers import (
 )
 from diffusers.utils import logging
 from diffusers.utils.import_utils import is_accelerate_available
-from diffusers.utils.testing_utils import (
+
+from ..testing_utils import (
     CaptureLogger,
     backend_empty_cache,
     is_flaky,
@@ -51,7 +52,7 @@ from diffusers.utils.testing_utils import (
 
 sys.path.append(".")
 
-from utils import PeftLoraLoaderMixinTests, check_if_lora_correctly_set, state_dicts_almost_equal  # noqa: E402
+from .utils import PeftLoraLoaderMixinTests, check_if_lora_correctly_set, state_dicts_almost_equal  # noqa: E402
 
 
 if is_accelerate_available():

tests/lora/test_lora_layers_wan.py

@@ -24,7 +24,8 @@ from diffusers import (
     WanPipeline,
     WanTransformer3DModel,
 )
-from diffusers.utils.testing_utils import (
+
+from ..testing_utils import (
     floats_tensor,
     require_peft_backend,
     skip_mps,
@@ -33,7 +34,7 @@ from diffusers.utils.testing_utils import (
 
 sys.path.append(".")
 
-from utils import PeftLoraLoaderMixinTests  # noqa: E402
+from .utils import PeftLoraLoaderMixinTests  # noqa: E402
 
 
 @require_peft_backend

tests/lora/test_lora_layers_wanvace.py

@@ -25,7 +25,8 @@ from transformers import AutoTokenizer, T5EncoderModel
 
 from diffusers import AutoencoderKLWan, FlowMatchEulerDiscreteScheduler, WanVACEPipeline, WanVACETransformer3DModel
 from diffusers.utils.import_utils import is_peft_available
-from diffusers.utils.testing_utils import (
+
+from ..testing_utils import (
     floats_tensor,
     is_flaky,
     require_peft_backend,
@@ -40,7 +41,7 @@ if is_peft_available():
 
 sys.path.append(".")
 
-from utils import PeftLoraLoaderMixinTests  # noqa: E402
+from .utils import PeftLoraLoaderMixinTests  # noqa: E402
 
 
 @require_peft_backend

tests/lora/utils.py

@@ -32,7 +32,8 @@ from diffusers import (
 )
 from diffusers.utils import logging
 from diffusers.utils.import_utils import is_peft_available
-from diffusers.utils.testing_utils import (
+
+from ..testing_utils import (
     CaptureLogger,
     check_if_dicts_are_equal,
     floats_tensor,

tests/models/autoencoders/test_models_asymmetric_autoencoder_kl.py

@@ -21,7 +21,8 @@ from parameterized import parameterized
 
 from diffusers import AsymmetricAutoencoderKL
 from diffusers.utils.import_utils import is_xformers_available
-from diffusers.utils.testing_utils import (
+
+from ...testing_utils import (
     Expectations,
     backend_empty_cache,
     enable_full_determinism,
@@ -34,7 +35,6 @@ from diffusers.utils.testing_utils import (
     torch_all_close,
     torch_device,
 )
-
 from ..test_modeling_common import ModelTesterMixin, UNetTesterMixin
 
 

tests/models/autoencoders/test_models_autoencoder_cosmos.py

@@ -15,8 +15,8 @@
 import unittest
 
 from diffusers import AutoencoderKLCosmos
-from diffusers.utils.testing_utils import enable_full_determinism, floats_tensor, torch_device
 
+from ...testing_utils import enable_full_determinism, floats_tensor, torch_device
 from ..test_modeling_common import ModelTesterMixin, UNetTesterMixin
 
 

tests/models/autoencoders/test_models_autoencoder_dc.py

@@ -16,12 +16,12 @@
 import unittest
 
 from diffusers import AutoencoderDC
-from diffusers.utils.testing_utils import (
+
+from ...testing_utils import (
     enable_full_determinism,
     floats_tensor,
     torch_device,
 )
-
 from ..test_modeling_common import ModelTesterMixin, UNetTesterMixin
 
 

tests/models/autoencoders/test_models_autoencoder_hunyuan_video.py

@@ -19,12 +19,12 @@ import torch
 
 from diffusers import AutoencoderKLHunyuanVideo
 from diffusers.models.autoencoders.autoencoder_kl_hunyuan_video import prepare_causal_attention_mask
-from diffusers.utils.testing_utils import (
+
+from ...testing_utils import (
     enable_full_determinism,
     floats_tensor,
     torch_device,
 )
-
 from ..test_modeling_common import ModelTesterMixin, UNetTesterMixin
 
 

tests/models/autoencoders/test_models_autoencoder_kl.py

@@ -21,7 +21,8 @@ from parameterized import parameterized
 
 from diffusers import AutoencoderKL
 from diffusers.utils.import_utils import is_xformers_available
-from diffusers.utils.testing_utils import (
+
+from ...testing_utils import (
     backend_empty_cache,
     enable_full_determinism,
     floats_tensor,
@@ -34,7 +35,6 @@ from diffusers.utils.testing_utils import (
     torch_all_close,
     torch_device,
 )
-
 from ..test_modeling_common import ModelTesterMixin, UNetTesterMixin
 
 

tests/models/autoencoders/test_models_autoencoder_kl_cogvideox.py

@@ -18,12 +18,12 @@ import unittest
 import torch
 
 from diffusers import AutoencoderKLCogVideoX
-from diffusers.utils.testing_utils import (
+
+from ...testing_utils import (
     enable_full_determinism,
     floats_tensor,
     torch_device,
 )
-
 from ..test_modeling_common import ModelTesterMixin, UNetTesterMixin
 
 

tests/models/autoencoders/test_models_autoencoder_kl_temporal_decoder.py

@@ -16,12 +16,12 @@
 import unittest
 
 from diffusers import AutoencoderKLTemporalDecoder
-from diffusers.utils.testing_utils import (
+
+from ...testing_utils import (
     enable_full_determinism,
     floats_tensor,
     torch_device,
 )
-
 from ..test_modeling_common import ModelTesterMixin, UNetTesterMixin
 
 

tests/models/autoencoders/test_models_autoencoder_ltx_video.py

@@ -18,12 +18,12 @@ import unittest
 import torch
 
 from diffusers import AutoencoderKLLTXVideo
-from diffusers.utils.testing_utils import (
+
+from ...testing_utils import (
     enable_full_determinism,
     floats_tensor,
     torch_device,
 )
-
 from ..test_modeling_common import ModelTesterMixin, UNetTesterMixin
 
 

tests/models/autoencoders/test_models_autoencoder_magvit.py

@@ -16,8 +16,8 @@
 import unittest
 
 from diffusers import AutoencoderKLMagvit
-from diffusers.utils.testing_utils import enable_full_determinism, floats_tensor, torch_device
 
+from ...testing_utils import enable_full_determinism, floats_tensor, torch_device
 from ..test_modeling_common import ModelTesterMixin, UNetTesterMixin
 
 

tests/models/autoencoders/test_models_autoencoder_mochi.py

@@ -16,12 +16,12 @@
 import unittest
 
 from diffusers import AutoencoderKLMochi
-from diffusers.utils.testing_utils import (
+
+from ...testing_utils import (
     enable_full_determinism,
     floats_tensor,
     torch_device,
 )
-
 from ..test_modeling_common import ModelTesterMixin, UNetTesterMixin
 
 

tests/models/autoencoders/test_models_autoencoder_oobleck.py

@@ -21,7 +21,8 @@ from datasets import load_dataset
 from parameterized import parameterized
 
 from diffusers import AutoencoderOobleck
-from diffusers.utils.testing_utils import (
+
+from ...testing_utils import (
     backend_empty_cache,
     enable_full_determinism,
     floats_tensor,
@@ -29,7 +30,6 @@ from diffusers.utils.testing_utils import (
     torch_all_close,
     torch_device,
 )
-
 from ..test_modeling_common import ModelTesterMixin, UNetTesterMixin
 
 

tests/models/autoencoders/test_models_autoencoder_tiny.py

@@ -21,7 +21,8 @@ import torch
 from parameterized import parameterized
 
 from diffusers import AutoencoderTiny
-from diffusers.utils.testing_utils import (
+
+from ...testing_utils import (
     backend_empty_cache,
     enable_full_determinism,
     floats_tensor,
@@ -30,7 +31,6 @@ from diffusers.utils.testing_utils import (
     torch_all_close,
     torch_device,
 )
-
 from ..test_modeling_common import ModelTesterMixin, UNetTesterMixin
 
 

tests/models/autoencoders/test_models_autoencoder_wan.py

@@ -18,8 +18,8 @@ import unittest
 import torch
 
 from diffusers import AutoencoderKLWan
-from diffusers.utils.testing_utils import enable_full_determinism, floats_tensor, torch_device
 
+from ...testing_utils import enable_full_determinism, floats_tensor, torch_device
 from ..test_modeling_common import ModelTesterMixin, UNetTesterMixin
 
 

tests/models/autoencoders/test_models_consistency_decoder_vae.py

@@ -20,7 +20,9 @@ import numpy as np
 import torch
 
 from diffusers import ConsistencyDecoderVAE, StableDiffusionPipeline
-from diffusers.utils.testing_utils import (
+from diffusers.utils.torch_utils import randn_tensor
+
+from ...testing_utils import (
     backend_empty_cache,
     enable_full_determinism,
     load_image,
@@ -28,8 +30,6 @@ from diffusers.utils.testing_utils import (
     torch_all_close,
     torch_device,
 )
-from diffusers.utils.torch_utils import randn_tensor
-
 from ..test_modeling_common import ModelTesterMixin
 
 

tests/models/autoencoders/test_models_vq.py

@@ -18,13 +18,13 @@ import unittest
 import torch
 
 from diffusers import VQModel
-from diffusers.utils.testing_utils import (
+
+from ...testing_utils import (
     backend_manual_seed,
     enable_full_determinism,
     floats_tensor,
     torch_device,
 )
-
 from ..test_modeling_common import ModelTesterMixin, UNetTesterMixin
 
 

tests/models/test_attention_processor.py

@@ -7,7 +7,8 @@ import torch
 
 from diffusers import DiffusionPipeline
 from diffusers.models.attention_processor import Attention, AttnAddedKVProcessor
-from diffusers.utils.testing_utils import torch_device
+
+from ..testing_utils import torch_device
 
 
 class AttnAddedKVProcessorTests(unittest.TestCase):

tests/models/test_layers_utils.py

@@ -24,7 +24,8 @@ from diffusers.models.attention import GEGLU, AdaLayerNorm, ApproximateGELU
 from diffusers.models.embeddings import get_timestep_embedding
 from diffusers.models.resnet import Downsample2D, ResnetBlock2D, Upsample2D
 from diffusers.models.transformers.transformer_2d import Transformer2DModel
-from diffusers.utils.testing_utils import (
+
+from ..testing_utils import (
     backend_manual_seed,
     require_torch_accelerator_with_fp64,
     require_torch_version_greater_equal,

tests/models/test_modeling_common.py

@@ -59,7 +59,10 @@ from diffusers.utils import (
     logging,
 )
 from diffusers.utils.hub_utils import _add_variant
-from diffusers.utils.testing_utils import (
+from diffusers.utils.torch_utils import get_torch_cuda_device_capability
+
+from ..others.test_utils import TOKEN, USER, is_staging_test
+from ..testing_utils import (
     CaptureLogger,
     _check_safetensors_serialization,
     backend_empty_cache,
@@ -82,9 +85,6 @@ from diffusers.utils.testing_utils import (
     torch_all_close,
     torch_device,
 )
-from diffusers.utils.torch_utils import get_torch_cuda_device_capability
-
-from ..others.test_utils import TOKEN, USER, is_staging_test
 
 
 if is_peft_available():

tests/models/transformers/test_models_dit_transformer2d.py

@@ -18,13 +18,13 @@ import unittest
 import torch
 
 from diffusers import DiTTransformer2DModel, Transformer2DModel
-from diffusers.utils.testing_utils import (
+
+from ...testing_utils import (
     enable_full_determinism,
     floats_tensor,
     slow,
     torch_device,
 )
-
 from ..test_modeling_common import ModelTesterMixin
 
 

tests/models/transformers/test_models_pixart_transformer2d.py

@@ -18,13 +18,13 @@ import unittest
 import torch
 
 from diffusers import PixArtTransformer2DModel, Transformer2DModel
-from diffusers.utils.testing_utils import (
+
+from ...testing_utils import (
     enable_full_determinism,
     floats_tensor,
     slow,
     torch_device,
 )
-
 from ..test_modeling_common import ModelTesterMixin
 
 

tests/models/transformers/test_models_prior.py

@@ -21,7 +21,8 @@ import torch
 from parameterized import parameterized
 
 from diffusers import PriorTransformer
-from diffusers.utils.testing_utils import (
+
+from ...testing_utils import (
     backend_empty_cache,
     enable_full_determinism,
     floats_tensor,
@@ -29,7 +30,6 @@ from diffusers.utils.testing_utils import (
     torch_all_close,
     torch_device,
 )
-
 from ..test_modeling_common import ModelTesterMixin
 
 

tests/models/transformers/test_models_transformer_allegro.py

@@ -17,11 +17,11 @@ import unittest
 import torch
 
 from diffusers import AllegroTransformer3DModel
-from diffusers.utils.testing_utils import (
+
+from ...testing_utils import (
     enable_full_determinism,
     torch_device,
 )
-
 from ..test_modeling_common import ModelTesterMixin
 
 

tests/models/transformers/test_models_transformer_aura_flow.py

@@ -18,8 +18,8 @@ import unittest
 import torch
 
 from diffusers import AuraFlowTransformer2DModel
-from diffusers.utils.testing_utils import enable_full_determinism, torch_device
 
+from ...testing_utils import enable_full_determinism, torch_device
 from ..test_modeling_common import ModelTesterMixin
 
 

tests/models/transformers/test_models_transformer_bria.py

@@ -20,8 +20,8 @@ import torch
 from diffusers import BriaTransformer2DModel
 from diffusers.models.attention_processor import FluxIPAdapterJointAttnProcessor2_0
 from diffusers.models.embeddings import ImageProjection
-from diffusers.utils.testing_utils import enable_full_determinism, torch_device
 
+from ...testing_utils import enable_full_determinism, torch_device
 from ..test_modeling_common import LoraHotSwappingForModelTesterMixin, ModelTesterMixin, TorchCompileTesterMixin
 
 

tests/models/transformers/test_models_transformer_chroma.py

@@ -20,8 +20,8 @@ import torch
 from diffusers import ChromaTransformer2DModel
 from diffusers.models.attention_processor import FluxIPAdapterJointAttnProcessor2_0
 from diffusers.models.embeddings import ImageProjection
-from diffusers.utils.testing_utils import enable_full_determinism, torch_device
 
+from ...testing_utils import enable_full_determinism, torch_device
 from ..test_modeling_common import LoraHotSwappingForModelTesterMixin, ModelTesterMixin, TorchCompileTesterMixin
 
 

tests/models/transformers/test_models_transformer_cogvideox.py

@@ -18,11 +18,11 @@ import unittest
 import torch
 
 from diffusers import CogVideoXTransformer3DModel
-from diffusers.utils.testing_utils import (
+
+from ...testing_utils import (
     enable_full_determinism,
     torch_device,
 )
-
 from ..test_modeling_common import ModelTesterMixin
 
 

tests/models/transformers/test_models_transformer_cogview3plus.py

@@ -18,11 +18,11 @@ import unittest
 import torch
 
 from diffusers import CogView3PlusTransformer2DModel
-from diffusers.utils.testing_utils import (
+
+from ...testing_utils import (
     enable_full_determinism,
     torch_device,
 )
-
 from ..test_modeling_common import ModelTesterMixin
 
 

tests/models/transformers/test_models_transformer_cogview4.py

@@ -17,8 +17,8 @@ import unittest
 import torch
 
 from diffusers import CogView4Transformer2DModel
-from diffusers.utils.testing_utils import enable_full_determinism, torch_device
 
+from ...testing_utils import enable_full_determinism, torch_device
 from ..test_modeling_common import ModelTesterMixin
 
 

tests/models/transformers/test_models_transformer_consisid.py

@@ -18,11 +18,11 @@ import unittest
 import torch
 
 from diffusers import ConsisIDTransformer3DModel
-from diffusers.utils.testing_utils import (
+
+from ...testing_utils import (
     enable_full_determinism,
     torch_device,
 )
-
 from ..test_modeling_common import ModelTesterMixin
 
 

tests/models/transformers/test_models_transformer_cosmos.py

@@ -17,8 +17,8 @@ import unittest
 import torch
 
 from diffusers import CosmosTransformer3DModel
-from diffusers.utils.testing_utils import enable_full_determinism, torch_device
 
+from ...testing_utils import enable_full_determinism, torch_device
 from ..test_modeling_common import ModelTesterMixin
 
 

tests/models/transformers/test_models_transformer_easyanimate.py

@@ -18,8 +18,8 @@ import unittest
 import torch
 
 from diffusers import EasyAnimateTransformer3DModel
-from diffusers.utils.testing_utils import enable_full_determinism, torch_device
 
+from ...testing_utils import enable_full_determinism, torch_device
 from ..test_modeling_common import ModelTesterMixin
 
 

tests/models/transformers/test_models_transformer_flux.py

@@ -20,8 +20,8 @@ import torch
 from diffusers import FluxTransformer2DModel
 from diffusers.models.attention_processor import FluxIPAdapterJointAttnProcessor2_0
 from diffusers.models.embeddings import ImageProjection
-from diffusers.utils.testing_utils import enable_full_determinism, is_peft_available, torch_device
 
+from ...testing_utils import enable_full_determinism, is_peft_available, torch_device
 from ..test_modeling_common import LoraHotSwappingForModelTesterMixin, ModelTesterMixin, TorchCompileTesterMixin
 
 

tests/models/transformers/test_models_transformer_hidream.py

@@ -18,11 +18,11 @@ import unittest
 import torch
 
 from diffusers import HiDreamImageTransformer2DModel
-from diffusers.utils.testing_utils import (
+
+from ...testing_utils import (
     enable_full_determinism,
     torch_device,
 )
-
 from ..test_modeling_common import ModelTesterMixin
 
 

tests/models/transformers/test_models_transformer_hunyuan_dit.py

@@ -18,11 +18,11 @@ import unittest
 import torch
 
 from diffusers import HunyuanDiT2DModel
-from diffusers.utils.testing_utils import (
+
+from ...testing_utils import (
     enable_full_determinism,
     torch_device,
 )
-
 from ..test_modeling_common import ModelTesterMixin
 
 

tests/models/transformers/test_models_transformer_hunyuan_video.py

@@ -17,11 +17,11 @@ import unittest
 import torch
 
 from diffusers import HunyuanVideoTransformer3DModel
-from diffusers.utils.testing_utils import (
+
+from ...testing_utils import (
     enable_full_determinism,
     torch_device,
 )
-
 from ..test_modeling_common import ModelTesterMixin, TorchCompileTesterMixin
 
 

tests/models/transformers/test_models_transformer_hunyuan_video_framepack.py

@@ -17,11 +17,11 @@ import unittest
 import torch
 
 from diffusers import HunyuanVideoFramepackTransformer3DModel
-from diffusers.utils.testing_utils import (
+
+from ...testing_utils import (
     enable_full_determinism,
     torch_device,
 )
-
 from ..test_modeling_common import ModelTesterMixin
 
 

tests/models/transformers/test_models_transformer_latte.py

@@ -18,11 +18,11 @@ import unittest
 import torch
 
 from diffusers import LatteTransformer3DModel
-from diffusers.utils.testing_utils import (
+
+from ...testing_utils import (
     enable_full_determinism,
     torch_device,
 )
-
 from ..test_modeling_common import ModelTesterMixin
 
 

tests/models/transformers/test_models_transformer_ltx.py

@@ -18,8 +18,8 @@ import unittest
 import torch
 
 from diffusers import LTXVideoTransformer3DModel
-from diffusers.utils.testing_utils import enable_full_determinism, torch_device
 
+from ...testing_utils import enable_full_determinism, torch_device
 from ..test_modeling_common import ModelTesterMixin, TorchCompileTesterMixin
 
 

tests/models/transformers/test_models_transformer_lumina.py

@@ -18,11 +18,11 @@ import unittest
 import torch
 
 from diffusers import LuminaNextDiT2DModel
-from diffusers.utils.testing_utils import (
+
+from ...testing_utils import (
     enable_full_determinism,
     torch_device,
 )
-
 from ..test_modeling_common import ModelTesterMixin
 
 

tests/models/transformers/test_models_transformer_lumina2.py

@@ -18,11 +18,11 @@ import unittest
 import torch
 
 from diffusers import Lumina2Transformer2DModel
-from diffusers.utils.testing_utils import (
+
+from ...testing_utils import (
     enable_full_determinism,
     torch_device,
 )
-
 from ..test_modeling_common import ModelTesterMixin
 
 

tests/models/transformers/test_models_transformer_mochi.py

@@ -18,8 +18,8 @@ import unittest
 import torch
 
 from diffusers import MochiTransformer3DModel
-from diffusers.utils.testing_utils import enable_full_determinism, torch_device
 
+from ...testing_utils import enable_full_determinism, torch_device
 from ..test_modeling_common import ModelTesterMixin
 
 

tests/models/transformers/test_models_transformer_omnigen.py

@@ -18,8 +18,8 @@ import unittest
 import torch
 
 from diffusers import OmniGenTransformer2DModel
-from diffusers.utils.testing_utils import enable_full_determinism, torch_device
 
+from ...testing_utils import enable_full_determinism, torch_device
 from ..test_modeling_common import ModelTesterMixin
 
 

tests/models/transformers/test_models_transformer_qwenimage.py

@@ -19,8 +19,8 @@ import pytest
 import torch
 
 from diffusers import QwenImageTransformer2DModel
-from diffusers.utils.testing_utils import enable_full_determinism, torch_device
 
+from ...testing_utils import enable_full_determinism, torch_device
 from ..test_modeling_common import ModelTesterMixin, TorchCompileTesterMixin
 
 

tests/models/transformers/test_models_transformer_sana.py

@@ -17,11 +17,11 @@ import unittest
 import torch
 
 from diffusers import SanaTransformer2DModel
-from diffusers.utils.testing_utils import (
+
+from ...testing_utils import (
     enable_full_determinism,
     torch_device,
 )
-
 from ..test_modeling_common import ModelTesterMixin
 
 

tests/models/transformers/test_models_transformer_sd3.py

@@ -19,11 +19,11 @@ import torch
 
 from diffusers import SD3Transformer2DModel
 from diffusers.utils.import_utils import is_xformers_available
-from diffusers.utils.testing_utils import (
+
+from ...testing_utils import (
     enable_full_determinism,
     torch_device,
 )
-
 from ..test_modeling_common import ModelTesterMixin
 
 

tests/models/transformers/test_models_transformer_skyreels_v2.py

@@ -17,11 +17,11 @@ import unittest
 import torch
 
 from diffusers import SkyReelsV2Transformer3DModel
-from diffusers.utils.testing_utils import (
+
+from ...testing_utils import (
     enable_full_determinism,
     torch_device,
 )
-
 from ..test_modeling_common import ModelTesterMixin, TorchCompileTesterMixin
 
 

tests/models/transformers/test_models_transformer_temporal.py

@@ -18,11 +18,11 @@ import unittest
 import torch
 
 from diffusers.models.transformers import TransformerTemporalModel
-from diffusers.utils.testing_utils import (
+
+from ...testing_utils import (
     enable_full_determinism,
     torch_device,
 )
-
 from ..test_modeling_common import ModelTesterMixin
 
 

tests/models/transformers/test_models_transformer_wan.py

@@ -17,11 +17,11 @@ import unittest
 import torch
 
 from diffusers import WanTransformer3DModel
-from diffusers.utils.testing_utils import (
+
+from ...testing_utils import (
     enable_full_determinism,
     torch_device,
 )
-
 from ..test_modeling_common import ModelTesterMixin, TorchCompileTesterMixin
 
 

tests/models/unets/test_models_unet_1d.py

@@ -19,13 +19,13 @@ import pytest
 import torch
 
 from diffusers import UNet1DModel
-from diffusers.utils.testing_utils import (
+
+from ...testing_utils import (
     backend_manual_seed,
     floats_tensor,
     slow,
     torch_device,
 )
-
 from ..test_modeling_common import ModelTesterMixin, UNetTesterMixin