fix

* utils and test modifications to enable device agnostic testing

* device for manual seed in unet1d

* fix generator condition in vae test

* consistency changes to testing

* make style

* add device agnostic testing changes to source and one model test

* make dtype check fns private, log cuda fp16 case

* remove dtype checks from import utils, move to testing_utils

* adding tests for most model classes and one pipeline

* fix vae import

docs/source/en/using-diffusers/push_to_hub.md

@@ -174,10 +174,4 @@ Set `private=True` in the [`~diffusers.utils.PushToHubMixin.push_to_hub`] functi
 controlnet.push_to_hub("my-controlnet-model-private", private=True)
 ```
 
-Private repositories are only visible to you, and other users won't be able to clone the repository and your repository won't appear in search results. Even if a user has the URL to your private repository, they'll receive a `404 - Sorry, we can't find the page you are looking for.`
-
-To load a model, scheduler, or pipeline from private or gated repositories, set `use_auth_token=True`:
-
-```py
-model = ControlNetModel.from_pretrained("your-namespace/my-controlnet-model-private", use_auth_token=True)
-```
+Private repositories are only visible to you, and other users won't be able to clone the repository and your repository won't appear in search results. Even if a user has the URL to your private repository, they'll receive a `404 - Sorry, we can't find the page you are looking for`. You must be [logged in](https://huggingface.co/docs/huggingface_hub/quick-start#login) to load a model from a private repository.

examples/community/README.md

@@ -512,7 +512,6 @@ device = torch.device('cpu' if not has_cuda else 'cuda')
 pipe = DiffusionPipeline.from_pretrained(
     "CompVis/stable-diffusion-v1-4",
         safety_checker=None,
-    use_auth_token=True,
     custom_pipeline="imagic_stable_diffusion",
     scheduler = DDIMScheduler(beta_start=0.00085, beta_end=0.012, beta_schedule="scaled_linear", clip_sample=False, set_alpha_to_one=False)
 ).to(device)
@@ -552,7 +551,6 @@ device = th.device('cpu' if not has_cuda else 'cuda')
 
 pipe = DiffusionPipeline.from_pretrained(
     "CompVis/stable-diffusion-v1-4",
-    use_auth_token=True,
     custom_pipeline="seed_resize_stable_diffusion"
 ).to(device)
 
@@ -588,7 +586,6 @@ generator = th.Generator("cuda").manual_seed(0)
 
 pipe = DiffusionPipeline.from_pretrained(
     "CompVis/stable-diffusion-v1-4",
-    use_auth_token=True,
     custom_pipeline="/home/mark/open_source/diffusers/examples/community/"
 ).to(device)
 
@@ -607,7 +604,6 @@ image.save('./seed_resize/seed_resize_{w}_{h}_image.png'.format(w=width, h=heigh
 
 pipe_compare = DiffusionPipeline.from_pretrained(
     "CompVis/stable-diffusion-v1-4",
-    use_auth_token=True,
     custom_pipeline="/home/mark/open_source/diffusers/examples/community/"
 ).to(device)
 
@@ -2870,10 +2866,14 @@ The original repo can be found at [repo](https://github.com/PRIS-CV/DemoFusion).
 - `show_image` (`bool`, defaults to False):
   Determine whether to show intermediate results during generation.
 ```
-from pipeline_demofusion_sdxl import DemoFusionSDXLPipeline
+from diffusers import DiffusionPipeline
 
-model_ckpt = "stabilityai/stable-diffusion-xl-base-1.0"
-pipe = DemoFusionSDXLPipeline.from_pretrained(model_ckpt, torch_dtype=torch.float16)
+pipe = DiffusionPipeline.from_pretrained(
+    "stabilityai/stable-diffusion-xl-base-1.0",
+    custom_pipeline="pipeline_demofusion_sdxl",
+    custom_revision="main",
+    torch_dtype=torch.float16,
+)
 pipe = pipe.to("cuda")
 
 prompt = "Envision a portrait of an elderly woman, her face a canvas of time, framed by a headscarf with muted tones of rust and cream. Her eyes, blue like faded denim. Her attire, simple yet dignified."

examples/community/checkpoint_merger.py

@@ -5,10 +5,11 @@ from typing import Dict, List, Union
 import safetensors.torch
 import torch
 from huggingface_hub import snapshot_download
+from huggingface_hub.utils import validate_hf_hub_args
 
 from diffusers import DiffusionPipeline, __version__
 from diffusers.schedulers.scheduling_utils import SCHEDULER_CONFIG_NAME
-from diffusers.utils import CONFIG_NAME, DIFFUSERS_CACHE, ONNX_WEIGHTS_NAME, WEIGHTS_NAME
+from diffusers.utils import CONFIG_NAME, ONNX_WEIGHTS_NAME, WEIGHTS_NAME
 
 
 class CheckpointMergerPipeline(DiffusionPipeline):
@@ -57,6 +58,7 @@ class CheckpointMergerPipeline(DiffusionPipeline):
         return (temp_dict, meta_keys)
 
     @torch.no_grad()
+    @validate_hf_hub_args
     def merge(self, pretrained_model_name_or_path_list: List[Union[str, os.PathLike]], **kwargs):
         """
         Returns a new pipeline object of the class 'DiffusionPipeline' with the merged checkpoints(weights) of the models passed
@@ -69,7 +71,7 @@ class CheckpointMergerPipeline(DiffusionPipeline):
             **kwargs:
                 Supports all the default DiffusionPipeline.get_config_dict kwargs viz..
 
-                cache_dir, resume_download, force_download, proxies, local_files_only, use_auth_token, revision, torch_dtype, device_map.
+                cache_dir, resume_download, force_download, proxies, local_files_only, token, revision, torch_dtype, device_map.
 
                 alpha - The interpolation parameter. Ranges from 0 to 1.  It affects the ratio in which the checkpoints are merged. A 0.8 alpha
                     would mean that the first model checkpoints would affect the final result far less than an alpha of 0.2
@@ -81,12 +83,12 @@ class CheckpointMergerPipeline(DiffusionPipeline):
 
         """
         # Default kwargs from DiffusionPipeline
-        cache_dir = kwargs.pop("cache_dir", DIFFUSERS_CACHE)
+        cache_dir = kwargs.pop("cache_dir", None)
         resume_download = kwargs.pop("resume_download", False)
         force_download = kwargs.pop("force_download", False)
         proxies = kwargs.pop("proxies", None)
         local_files_only = kwargs.pop("local_files_only", False)
-        use_auth_token = kwargs.pop("use_auth_token", None)
+        token = kwargs.pop("token", None)
         revision = kwargs.pop("revision", None)
         torch_dtype = kwargs.pop("torch_dtype", None)
         device_map = kwargs.pop("device_map", None)
@@ -123,7 +125,7 @@ class CheckpointMergerPipeline(DiffusionPipeline):
                 force_download=force_download,
                 proxies=proxies,
                 local_files_only=local_files_only,
-                use_auth_token=use_auth_token,
+                token=token,
                 revision=revision,
             )
             config_dicts.append(config_dict)
@@ -159,7 +161,7 @@ class CheckpointMergerPipeline(DiffusionPipeline):
                     resume_download=resume_download,
                     proxies=proxies,
                     local_files_only=local_files_only,
-                    use_auth_token=use_auth_token,
+                    token=token,
                     revision=revision,
                     allow_patterns=allow_patterns,
                     user_agent=user_agent,

examples/community/pipeline_demofusion_sdxl.py

@@ -36,7 +36,9 @@ from diffusers.utils.torch_utils import randn_tensor
 
 
 if is_invisible_watermark_available():
-    from .watermark import StableDiffusionXLWatermarker
+    from diffusers.pipelines.stable_diffusion_xl.watermark import (
+        StableDiffusionXLWatermarker,
+    )
 
 
 logger = logging.get_logger(__name__)  # pylint: disable=invalid-name

examples/community/stable_diffusion_tensorrt_img2img.py

@@ -28,6 +28,7 @@ import PIL.Image
 import tensorrt as trt
 import torch
 from huggingface_hub import snapshot_download
+from huggingface_hub.utils import validate_hf_hub_args
 from onnx import shape_inference
 from polygraphy import cuda
 from polygraphy.backend.common import bytes_from_path
@@ -50,7 +51,7 @@ from diffusers.pipelines.stable_diffusion import (
     StableDiffusionSafetyChecker,
 )
 from diffusers.schedulers import DDIMScheduler
-from diffusers.utils import DIFFUSERS_CACHE, logging
+from diffusers.utils import logging
 
 
 """
@@ -778,12 +779,13 @@ class TensorRTStableDiffusionImg2ImgPipeline(StableDiffusionImg2ImgPipeline):
             self.models["vae_encoder"] = make_VAEEncoder(self.vae, **models_args)
 
     @classmethod
+    @validate_hf_hub_args
     def set_cached_folder(cls, pretrained_model_name_or_path: Optional[Union[str, os.PathLike]], **kwargs):
-        cache_dir = kwargs.pop("cache_dir", DIFFUSERS_CACHE)
+        cache_dir = kwargs.pop("cache_dir", None)
         resume_download = kwargs.pop("resume_download", False)
         proxies = kwargs.pop("proxies", None)
         local_files_only = kwargs.pop("local_files_only", False)
-        use_auth_token = kwargs.pop("use_auth_token", None)
+        token = kwargs.pop("token", None)
         revision = kwargs.pop("revision", None)
 
         cls.cached_folder = (
@@ -795,7 +797,7 @@ class TensorRTStableDiffusionImg2ImgPipeline(StableDiffusionImg2ImgPipeline):
                 resume_download=resume_download,
                 proxies=proxies,
                 local_files_only=local_files_only,
-                use_auth_token=use_auth_token,
+                token=token,
                 revision=revision,
             )
         )

examples/community/stable_diffusion_tensorrt_inpaint.py

@@ -28,6 +28,7 @@ import PIL.Image
 import tensorrt as trt
 import torch
 from huggingface_hub import snapshot_download
+from huggingface_hub.utils import validate_hf_hub_args
 from onnx import shape_inference
 from polygraphy import cuda
 from polygraphy.backend.common import bytes_from_path
@@ -51,7 +52,7 @@ from diffusers.pipelines.stable_diffusion import (
 )
 from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_inpaint import prepare_mask_and_masked_image
 from diffusers.schedulers import DDIMScheduler
-from diffusers.utils import DIFFUSERS_CACHE, logging
+from diffusers.utils import logging
 
 
 """
@@ -779,12 +780,13 @@ class TensorRTStableDiffusionInpaintPipeline(StableDiffusionInpaintPipeline):
             self.models["vae_encoder"] = make_VAEEncoder(self.vae, **models_args)
 
     @classmethod
+    @validate_hf_hub_args
     def set_cached_folder(cls, pretrained_model_name_or_path: Optional[Union[str, os.PathLike]], **kwargs):
-        cache_dir = kwargs.pop("cache_dir", DIFFUSERS_CACHE)
+        cache_dir = kwargs.pop("cache_dir", None)
         resume_download = kwargs.pop("resume_download", False)
         proxies = kwargs.pop("proxies", None)
         local_files_only = kwargs.pop("local_files_only", False)
-        use_auth_token = kwargs.pop("use_auth_token", None)
+        token = kwargs.pop("token", None)
         revision = kwargs.pop("revision", None)
 
         cls.cached_folder = (
@@ -796,7 +798,7 @@ class TensorRTStableDiffusionInpaintPipeline(StableDiffusionInpaintPipeline):
                 resume_download=resume_download,
                 proxies=proxies,
                 local_files_only=local_files_only,
-                use_auth_token=use_auth_token,
+                token=token,
                 revision=revision,
             )
         )

examples/community/stable_diffusion_tensorrt_txt2img.py

@@ -27,6 +27,7 @@ import onnx_graphsurgeon as gs
 import tensorrt as trt
 import torch
 from huggingface_hub import snapshot_download
+from huggingface_hub.utils import validate_hf_hub_args
 from onnx import shape_inference
 from polygraphy import cuda
 from polygraphy.backend.common import bytes_from_path
@@ -49,7 +50,7 @@ from diffusers.pipelines.stable_diffusion import (
     StableDiffusionSafetyChecker,
 )
 from diffusers.schedulers import DDIMScheduler
-from diffusers.utils import DIFFUSERS_CACHE, logging
+from diffusers.utils import logging
 
 
 """
@@ -691,12 +692,13 @@ class TensorRTStableDiffusionPipeline(StableDiffusionPipeline):
             self.models["vae"] = make_VAE(self.vae, **models_args)
 
     @classmethod
+    @validate_hf_hub_args
     def set_cached_folder(cls, pretrained_model_name_or_path: Optional[Union[str, os.PathLike]], **kwargs):
-        cache_dir = kwargs.pop("cache_dir", DIFFUSERS_CACHE)
+        cache_dir = kwargs.pop("cache_dir", None)
         resume_download = kwargs.pop("resume_download", False)
         proxies = kwargs.pop("proxies", None)
         local_files_only = kwargs.pop("local_files_only", False)
-        use_auth_token = kwargs.pop("use_auth_token", None)
+        token = kwargs.pop("token", None)
         revision = kwargs.pop("revision", None)
 
         cls.cached_folder = (
@@ -708,7 +710,7 @@ class TensorRTStableDiffusionPipeline(StableDiffusionPipeline):
                 resume_download=resume_download,
                 proxies=proxies,
                 local_files_only=local_files_only,
-                use_auth_token=use_auth_token,
+                token=token,
                 revision=revision,
             )
         )

examples/consistency_distillation/train_lcm_distill_lora_sd_wds.py

@@ -423,7 +423,7 @@ def import_model_class_from_model_name_or_path(
     pretrained_model_name_or_path: str, revision: str, subfolder: str = "text_encoder"
 ):
     text_encoder_config = PretrainedConfig.from_pretrained(
-        pretrained_model_name_or_path, subfolder=subfolder, revision=revision, use_auth_token=True
+        pretrained_model_name_or_path, subfolder=subfolder, revision=revision
     )
     model_class = text_encoder_config.architectures[0]
 

examples/consistency_distillation/train_lcm_distill_lora_sdxl_wds.py

@@ -392,7 +392,7 @@ def import_model_class_from_model_name_or_path(
     pretrained_model_name_or_path: str, revision: str, subfolder: str = "text_encoder"
 ):
     text_encoder_config = PretrainedConfig.from_pretrained(
-        pretrained_model_name_or_path, subfolder=subfolder, revision=revision, use_auth_token=True
+        pretrained_model_name_or_path, subfolder=subfolder, revision=revision
     )
     model_class = text_encoder_config.architectures[0]
 

examples/consistency_distillation/train_lcm_distill_sd_wds.py

@@ -400,7 +400,7 @@ def import_model_class_from_model_name_or_path(
     pretrained_model_name_or_path: str, revision: str, subfolder: str = "text_encoder"
 ):
     text_encoder_config = PretrainedConfig.from_pretrained(
-        pretrained_model_name_or_path, subfolder=subfolder, revision=revision, use_auth_token=True
+        pretrained_model_name_or_path, subfolder=subfolder, revision=revision
     )
     model_class = text_encoder_config.architectures[0]
 

examples/consistency_distillation/train_lcm_distill_sdxl_wds.py

@@ -414,7 +414,7 @@ def import_model_class_from_model_name_or_path(
     pretrained_model_name_or_path: str, revision: str, subfolder: str = "text_encoder"
 ):
     text_encoder_config = PretrainedConfig.from_pretrained(
-        pretrained_model_name_or_path, subfolder=subfolder, revision=revision, use_auth_token=True
+        pretrained_model_name_or_path, subfolder=subfolder, revision=revision
     )
     model_class = text_encoder_config.architectures[0]
 

examples/research_projects/controlnet/train_controlnet_webdataset.py

@@ -420,7 +420,7 @@ def import_model_class_from_model_name_or_path(
     pretrained_model_name_or_path: str, revision: str, subfolder: str = "text_encoder"
 ):
     text_encoder_config = PretrainedConfig.from_pretrained(
-        pretrained_model_name_or_path, subfolder=subfolder, revision=revision, use_auth_token=True
+        pretrained_model_name_or_path, subfolder=subfolder, revision=revision
     )
     model_class = text_encoder_config.architectures[0]
 
@@ -975,7 +975,7 @@ def main(args):
         revision=args.revision,
     )
     unet = UNet2DConditionModel.from_pretrained(
-        args.pretrained_model_name_or_path, subfolder="unet", revision=args.revision, use_auth_token=True
+        args.pretrained_model_name_or_path, subfolder="unet", revision=args.revision
     )
 
     if args.controlnet_model_name_or_path:

src/diffusers/commands/fp16_safetensors.py

@@ -19,6 +19,7 @@ Usage example:
 
 import glob
 import json
+import warnings
 from argparse import ArgumentParser, Namespace
 from importlib import import_module
 
@@ -32,12 +33,12 @@ from . import BaseDiffusersCLICommand
 
 
 def conversion_command_factory(args: Namespace):
-    return FP16SafetensorsCommand(
-        args.ckpt_id,
-        args.fp16,
-        args.use_safetensors,
-        args.use_auth_token,
-    )
+    if args.use_auth_token:
+        warnings.warn(
+            "The `--use_auth_token` flag is deprecated and will be removed in a future version. Authentication is now"
+            " handled automatically if user is logged in."
+        )
+    return FP16SafetensorsCommand(args.ckpt_id, args.fp16, args.use_safetensors)
 
 
 class FP16SafetensorsCommand(BaseDiffusersCLICommand):
@@ -62,7 +63,7 @@ class FP16SafetensorsCommand(BaseDiffusersCLICommand):
         )
         conversion_parser.set_defaults(func=conversion_command_factory)
 
-    def __init__(self, ckpt_id: str, fp16: bool, use_safetensors: bool, use_auth_token: bool):
+    def __init__(self, ckpt_id: str, fp16: bool, use_safetensors: bool):
         self.logger = logging.get_logger("diffusers-cli/fp16_safetensors")
         self.ckpt_id = ckpt_id
         self.local_ckpt_dir = f"/tmp/{ckpt_id}"
@@ -75,8 +76,6 @@ class FP16SafetensorsCommand(BaseDiffusersCLICommand):
                 "When `use_safetensors` and `fp16` both are False, then this command is of no use."
             )
 
-        self.use_auth_token = use_auth_token
-
     def run(self):
         if version.parse(huggingface_hub.__version__) < version.parse("0.9.0"):
             raise ImportError(
@@ -87,7 +86,7 @@ class FP16SafetensorsCommand(BaseDiffusersCLICommand):
             from huggingface_hub import create_commit
             from huggingface_hub._commit_api import CommitOperationAdd
 
-        model_index = hf_hub_download(repo_id=self.ckpt_id, filename="model_index.json", token=self.use_auth_token)
+        model_index = hf_hub_download(repo_id=self.ckpt_id, filename="model_index.json")
         with open(model_index, "r") as f:
             pipeline_class_name = json.load(f)["_class_name"]
         pipeline_class = getattr(import_module("diffusers"), pipeline_class_name)
@@ -96,7 +95,7 @@ class FP16SafetensorsCommand(BaseDiffusersCLICommand):
         # Load the appropriate pipeline. We could have use `DiffusionPipeline`
         # here, but just to avoid any rough edge cases.
         pipeline = pipeline_class.from_pretrained(
-            self.ckpt_id, torch_dtype=torch.float16 if self.fp16 else torch.float32, use_auth_token=self.use_auth_token
+            self.ckpt_id, torch_dtype=torch.float16 if self.fp16 else torch.float32
         )
         pipeline.save_pretrained(
             self.local_ckpt_dir,

src/diffusers/configuration_utils.py

@@ -27,12 +27,16 @@ from typing import Any, Dict, Tuple, Union
 
 import numpy as np
 from huggingface_hub import create_repo, hf_hub_download
-from huggingface_hub.utils import EntryNotFoundError, RepositoryNotFoundError, RevisionNotFoundError
+from huggingface_hub.utils import (
+    EntryNotFoundError,
+    RepositoryNotFoundError,
+    RevisionNotFoundError,
+    validate_hf_hub_args,
+)
 from requests import HTTPError
 
 from . import __version__
 from .utils import (
-    DIFFUSERS_CACHE,
     HUGGINGFACE_CO_RESOLVE_ENDPOINT,
     DummyObject,
     deprecate,
@@ -275,6 +279,7 @@ class ConfigMixin:
         return cls.load_config(*args, **kwargs)
 
     @classmethod
+    @validate_hf_hub_args
     def load_config(
         cls,
         pretrained_model_name_or_path: Union[str, os.PathLike],
@@ -311,7 +316,7 @@ class ConfigMixin:
             local_files_only (`bool`, *optional*, defaults to `False`):
                 Whether to only load local model weights and configuration files or not. If set to `True`, the model
                 won't be downloaded from the Hub.
-            use_auth_token (`str` or *bool*, *optional*):
+            token (`str` or *bool*, *optional*):
                 The token to use as HTTP bearer authorization for remote files. If `True`, the token generated from
                 `diffusers-cli login` (stored in `~/.huggingface`) is used.
             revision (`str`, *optional*, defaults to `"main"`):
@@ -329,11 +334,11 @@ class ConfigMixin:
                 A dictionary of all the parameters stored in a JSON configuration file.
 
         """
-        cache_dir = kwargs.pop("cache_dir", DIFFUSERS_CACHE)
+        cache_dir = kwargs.pop("cache_dir", None)
         force_download = kwargs.pop("force_download", False)
         resume_download = kwargs.pop("resume_download", False)
         proxies = kwargs.pop("proxies", None)
-        use_auth_token = kwargs.pop("use_auth_token", None)
+        token = kwargs.pop("token", None)
         local_files_only = kwargs.pop("local_files_only", False)
         revision = kwargs.pop("revision", None)
         _ = kwargs.pop("mirror", None)
@@ -376,7 +381,7 @@ class ConfigMixin:
                     proxies=proxies,
                     resume_download=resume_download,
                     local_files_only=local_files_only,
-                    use_auth_token=use_auth_token,
+                    token=token,
                     user_agent=user_agent,
                     subfolder=subfolder,
                     revision=revision,
@@ -385,8 +390,7 @@ class ConfigMixin:
                 raise EnvironmentError(
                     f"{pretrained_model_name_or_path} is not a local folder and is not a valid model identifier"
                     " listed on 'https://huggingface.co/models'\nIf this is a private repository, make sure to pass a"
-                    " token having permission to this repo with `use_auth_token` or log in with `huggingface-cli"
-                    " login`."
+                    " token having permission to this repo with `token` or log in with `huggingface-cli login`."
                 )
             except RevisionNotFoundError:
                 raise EnvironmentError(

src/diffusers/loaders/ip_adapter.py

@@ -15,11 +15,10 @@ import os
 from typing import Dict, Union
 
 import torch
+from huggingface_hub.utils import validate_hf_hub_args
 from safetensors import safe_open
 
 from ..utils import (
-    DIFFUSERS_CACHE,
-    HF_HUB_OFFLINE,
     _get_model_file,
     is_transformers_available,
     logging,
@@ -43,6 +42,7 @@ logger = logging.get_logger(__name__)
 class IPAdapterMixin:
     """Mixin for handling IP Adapters."""
 
+    @validate_hf_hub_args
     def load_ip_adapter(
         self,
         pretrained_model_name_or_path_or_dict: Union[str, Dict[str, torch.Tensor]],
@@ -77,7 +77,7 @@ class IPAdapterMixin:
             local_files_only (`bool`, *optional*, defaults to `False`):
                 Whether to only load local model weights and configuration files or not. If set to `True`, the model
                 won't be downloaded from the Hub.
-            use_auth_token (`str` or *bool*, *optional*):
+            token (`str` or *bool*, *optional*):
                 The token to use as HTTP bearer authorization for remote files. If `True`, the token generated from
                 `diffusers-cli login` (stored in `~/.huggingface`) is used.
             revision (`str`, *optional*, defaults to `"main"`):
@@ -88,12 +88,12 @@ class IPAdapterMixin:
         """
 
         # Load the main state dict first.
-        cache_dir = kwargs.pop("cache_dir", DIFFUSERS_CACHE)
+        cache_dir = kwargs.pop("cache_dir", None)
         force_download = kwargs.pop("force_download", False)
         resume_download = kwargs.pop("resume_download", False)
         proxies = kwargs.pop("proxies", None)
-        local_files_only = kwargs.pop("local_files_only", HF_HUB_OFFLINE)
-        use_auth_token = kwargs.pop("use_auth_token", None)
+        local_files_only = kwargs.pop("local_files_only", None)
+        token = kwargs.pop("token", None)
         revision = kwargs.pop("revision", None)
 
         user_agent = {
@@ -110,7 +110,7 @@ class IPAdapterMixin:
                 resume_download=resume_download,
                 proxies=proxies,
                 local_files_only=local_files_only,
-                use_auth_token=use_auth_token,
+                token=token,
                 revision=revision,
                 subfolder=subfolder,
                 user_agent=user_agent,

src/diffusers/loaders/lora.py

@@ -18,14 +18,13 @@ from typing import Callable, Dict, List, Optional, Union
 import safetensors
 import torch
 from huggingface_hub import model_info
+from huggingface_hub.utils import validate_hf_hub_args
 from packaging import version
 from torch import nn
 
 from .. import __version__
 from ..models.modeling_utils import _LOW_CPU_MEM_USAGE_DEFAULT, load_model_dict_into_meta
 from ..utils import (
-    DIFFUSERS_CACHE,
-    HF_HUB_OFFLINE,
     USE_PEFT_BACKEND,
     _get_model_file,
     convert_state_dict_to_diffusers,
@@ -132,6 +131,7 @@ class LoraLoaderMixin:
         )
 
     @classmethod
+    @validate_hf_hub_args
     def lora_state_dict(
         cls,
         pretrained_model_name_or_path_or_dict: Union[str, Dict[str, torch.Tensor]],
@@ -174,7 +174,7 @@ class LoraLoaderMixin:
             local_files_only (`bool`, *optional*, defaults to `False`):
                 Whether to only load local model weights and configuration files or not. If set to `True`, the model
                 won't be downloaded from the Hub.
-            use_auth_token (`str` or *bool*, *optional*):
+            token (`str` or *bool*, *optional*):
                 The token to use as HTTP bearer authorization for remote files. If `True`, the token generated from
                 `diffusers-cli login` (stored in `~/.huggingface`) is used.
             revision (`str`, *optional*, defaults to `"main"`):
@@ -195,12 +195,12 @@ class LoraLoaderMixin:
         """
         # Load the main state dict first which has the LoRA layers for either of
         # UNet and text encoder or both.
-        cache_dir = kwargs.pop("cache_dir", DIFFUSERS_CACHE)
+        cache_dir = kwargs.pop("cache_dir", None)
         force_download = kwargs.pop("force_download", False)
         resume_download = kwargs.pop("resume_download", False)
         proxies = kwargs.pop("proxies", None)
-        local_files_only = kwargs.pop("local_files_only", HF_HUB_OFFLINE)
-        use_auth_token = kwargs.pop("use_auth_token", None)
+        local_files_only = kwargs.pop("local_files_only", None)
+        token = kwargs.pop("token", None)
         revision = kwargs.pop("revision", None)
         subfolder = kwargs.pop("subfolder", None)
         weight_name = kwargs.pop("weight_name", None)
@@ -239,7 +239,7 @@ class LoraLoaderMixin:
                         resume_download=resume_download,
                         proxies=proxies,
                         local_files_only=local_files_only,
-                        use_auth_token=use_auth_token,
+                        token=token,
                         revision=revision,
                         subfolder=subfolder,
                         user_agent=user_agent,
@@ -265,7 +265,7 @@ class LoraLoaderMixin:
                     resume_download=resume_download,
                     proxies=proxies,
                     local_files_only=local_files_only,
-                    use_auth_token=use_auth_token,
+                    token=token,
                     revision=revision,
                     subfolder=subfolder,
                     user_agent=user_agent,

src/diffusers/loaders/single_file.py

@@ -18,10 +18,9 @@ from pathlib import Path
 import requests
 import torch
 from huggingface_hub import hf_hub_download
+from huggingface_hub.utils import validate_hf_hub_args
 
 from ..utils import (
-    DIFFUSERS_CACHE,
-    HF_HUB_OFFLINE,
     deprecate,
     is_accelerate_available,
     is_omegaconf_available,
@@ -52,6 +51,7 @@ class FromSingleFileMixin:
         return cls.from_single_file(*args, **kwargs)
 
     @classmethod
+    @validate_hf_hub_args
     def from_single_file(cls, pretrained_model_link_or_path, **kwargs):
         r"""
         Instantiate a [`DiffusionPipeline`] from pretrained pipeline weights saved in the `.ckpt` or `.safetensors`
@@ -81,7 +81,7 @@ class FromSingleFileMixin:
             local_files_only (`bool`, *optional*, defaults to `False`):
                 Whether to only load local model weights and configuration files or not. If set to `True`, the model
                 won't be downloaded from the Hub.
-            use_auth_token (`str` or *bool*, *optional*):
+            token (`str` or *bool*, *optional*):
                 The token to use as HTTP bearer authorization for remote files. If `True`, the token generated from
                 `diffusers-cli login` (stored in `~/.huggingface`) is used.
             revision (`str`, *optional*, defaults to `"main"`):
@@ -154,12 +154,12 @@ class FromSingleFileMixin:
 
         original_config_file = kwargs.pop("original_config_file", None)
         config_files = kwargs.pop("config_files", None)
-        cache_dir = kwargs.pop("cache_dir", DIFFUSERS_CACHE)
+        cache_dir = kwargs.pop("cache_dir", None)
         resume_download = kwargs.pop("resume_download", False)
         force_download = kwargs.pop("force_download", False)
         proxies = kwargs.pop("proxies", None)
-        local_files_only = kwargs.pop("local_files_only", HF_HUB_OFFLINE)
-        use_auth_token = kwargs.pop("use_auth_token", None)
+        local_files_only = kwargs.pop("local_files_only", None)
+        token = kwargs.pop("token", None)
         revision = kwargs.pop("revision", None)
         extract_ema = kwargs.pop("extract_ema", False)
         image_size = kwargs.pop("image_size", None)
@@ -253,7 +253,7 @@ class FromSingleFileMixin:
                 resume_download=resume_download,
                 proxies=proxies,
                 local_files_only=local_files_only,
-                use_auth_token=use_auth_token,
+                token=token,
                 revision=revision,
                 force_download=force_download,
             )
@@ -282,7 +282,7 @@ class FromSingleFileMixin:
         )
 
         if torch_dtype is not None:
-            pipe.to(torch_dtype=torch_dtype)
+            pipe.to(dtype=torch_dtype)
 
         return pipe
 
@@ -293,6 +293,7 @@ class FromOriginalVAEMixin:
     """
 
     @classmethod
+    @validate_hf_hub_args
     def from_single_file(cls, pretrained_model_link_or_path, **kwargs):
         r"""
         Instantiate a [`AutoencoderKL`] from pretrained ControlNet weights saved in the original `.ckpt` or
@@ -322,7 +323,7 @@ class FromOriginalVAEMixin:
             local_files_only (`bool`, *optional*, defaults to `False`):
                 Whether to only load local model weights and configuration files or not. If set to True, the model
                 won't be downloaded from the Hub.
-            use_auth_token (`str` or *bool*, *optional*):
+            token (`str` or *bool*, *optional*):
                 The token to use as HTTP bearer authorization for remote files. If `True`, the token generated from
                 `diffusers-cli login` (stored in `~/.huggingface`) is used.
             revision (`str`, *optional*, defaults to `"main"`):
@@ -379,12 +380,12 @@ class FromOriginalVAEMixin:
         )
 
         config_file = kwargs.pop("config_file", None)
-        cache_dir = kwargs.pop("cache_dir", DIFFUSERS_CACHE)
+        cache_dir = kwargs.pop("cache_dir", None)
         resume_download = kwargs.pop("resume_download", False)
         force_download = kwargs.pop("force_download", False)
         proxies = kwargs.pop("proxies", None)
-        local_files_only = kwargs.pop("local_files_only", HF_HUB_OFFLINE)
-        use_auth_token = kwargs.pop("use_auth_token", None)
+        local_files_only = kwargs.pop("local_files_only", None)
+        token = kwargs.pop("token", None)
         revision = kwargs.pop("revision", None)
         image_size = kwargs.pop("image_size", None)
         scaling_factor = kwargs.pop("scaling_factor", None)
@@ -425,7 +426,7 @@ class FromOriginalVAEMixin:
                 resume_download=resume_download,
                 proxies=proxies,
                 local_files_only=local_files_only,
-                use_auth_token=use_auth_token,
+                token=token,
                 revision=revision,
                 force_download=force_download,
             )
@@ -490,6 +491,7 @@ class FromOriginalControlnetMixin:
     """
 
     @classmethod
+    @validate_hf_hub_args
     def from_single_file(cls, pretrained_model_link_or_path, **kwargs):
         r"""
         Instantiate a [`ControlNetModel`] from pretrained ControlNet weights saved in the original `.ckpt` or
@@ -519,7 +521,7 @@ class FromOriginalControlnetMixin:
             local_files_only (`bool`, *optional*, defaults to `False`):
                 Whether to only load local model weights and configuration files or not. If set to True, the model
                 won't be downloaded from the Hub.
-            use_auth_token (`str` or *bool*, *optional*):
+            token (`str` or *bool*, *optional*):
                 The token to use as HTTP bearer authorization for remote files. If `True`, the token generated from
                 `diffusers-cli login` (stored in `~/.huggingface`) is used.
             revision (`str`, *optional*, defaults to `"main"`):
@@ -555,12 +557,12 @@ class FromOriginalControlnetMixin:
         from ..pipelines.stable_diffusion.convert_from_ckpt import download_controlnet_from_original_ckpt
 
         config_file = kwargs.pop("config_file", None)
-        cache_dir = kwargs.pop("cache_dir", DIFFUSERS_CACHE)
+        cache_dir = kwargs.pop("cache_dir", None)
         resume_download = kwargs.pop("resume_download", False)
         force_download = kwargs.pop("force_download", False)
         proxies = kwargs.pop("proxies", None)
-        local_files_only = kwargs.pop("local_files_only", HF_HUB_OFFLINE)
-        use_auth_token = kwargs.pop("use_auth_token", None)
+        local_files_only = kwargs.pop("local_files_only", None)
+        token = kwargs.pop("token", None)
         num_in_channels = kwargs.pop("num_in_channels", None)
         use_linear_projection = kwargs.pop("use_linear_projection", None)
         revision = kwargs.pop("revision", None)
@@ -603,7 +605,7 @@ class FromOriginalControlnetMixin:
                 resume_download=resume_download,
                 proxies=proxies,
                 local_files_only=local_files_only,
-                use_auth_token=use_auth_token,
+                token=token,
                 revision=revision,
                 force_download=force_download,
             )

src/diffusers/loaders/textual_inversion.py

@@ -15,16 +15,10 @@ from typing import Dict, List, Optional, Union
 
 import safetensors
 import torch
+from huggingface_hub.utils import validate_hf_hub_args
 from torch import nn
 
-from ..utils import (
-    DIFFUSERS_CACHE,
-    HF_HUB_OFFLINE,
-    _get_model_file,
-    is_accelerate_available,
-    is_transformers_available,
-    logging,
-)
+from ..utils import _get_model_file, is_accelerate_available, is_transformers_available, logging
 
 
 if is_transformers_available():
@@ -39,13 +33,14 @@ TEXT_INVERSION_NAME = "learned_embeds.bin"
 TEXT_INVERSION_NAME_SAFE = "learned_embeds.safetensors"
 
 
+@validate_hf_hub_args
 def load_textual_inversion_state_dicts(pretrained_model_name_or_paths, **kwargs):
-    cache_dir = kwargs.pop("cache_dir", DIFFUSERS_CACHE)
+    cache_dir = kwargs.pop("cache_dir", None)
     force_download = kwargs.pop("force_download", False)
     resume_download = kwargs.pop("resume_download", False)
     proxies = kwargs.pop("proxies", None)
-    local_files_only = kwargs.pop("local_files_only", HF_HUB_OFFLINE)
-    use_auth_token = kwargs.pop("use_auth_token", None)
+    local_files_only = kwargs.pop("local_files_only", None)
+    token = kwargs.pop("token", None)
     revision = kwargs.pop("revision", None)
     subfolder = kwargs.pop("subfolder", None)
     weight_name = kwargs.pop("weight_name", None)
@@ -79,7 +74,7 @@ def load_textual_inversion_state_dicts(pretrained_model_name_or_paths, **kwargs)
                         resume_download=resume_download,
                         proxies=proxies,
                         local_files_only=local_files_only,
-                        use_auth_token=use_auth_token,
+                        token=token,
                         revision=revision,
                         subfolder=subfolder,
                         user_agent=user_agent,
@@ -100,7 +95,7 @@ def load_textual_inversion_state_dicts(pretrained_model_name_or_paths, **kwargs)
                     resume_download=resume_download,
                     proxies=proxies,
                     local_files_only=local_files_only,
-                    use_auth_token=use_auth_token,
+                    token=token,
                     revision=revision,
                     subfolder=subfolder,
                     user_agent=user_agent,
@@ -267,6 +262,7 @@ class TextualInversionLoaderMixin:
 
         return all_tokens, all_embeddings
 
+    @validate_hf_hub_args
     def load_textual_inversion(
         self,
         pretrained_model_name_or_path: Union[str, List[str], Dict[str, torch.Tensor], List[Dict[str, torch.Tensor]]],
@@ -320,7 +316,7 @@ class TextualInversionLoaderMixin:
             local_files_only (`bool`, *optional*, defaults to `False`):
                 Whether to only load local model weights and configuration files or not. If set to `True`, the model
                 won't be downloaded from the Hub.
-            use_auth_token (`str` or *bool*, *optional*):
+            token (`str` or *bool*, *optional*):
                 The token to use as HTTP bearer authorization for remote files. If `True`, the token generated from
                 `diffusers-cli login` (stored in `~/.huggingface`) is used.
             revision (`str`, *optional*, defaults to `"main"`):

src/diffusers/loaders/unet.py

@@ -19,13 +19,12 @@ from typing import Callable, Dict, List, Optional, Union
 import safetensors
 import torch
 import torch.nn.functional as F
+from huggingface_hub.utils import validate_hf_hub_args
 from torch import nn
 
 from ..models.embeddings import ImageProjection, Resampler
 from ..models.modeling_utils import _LOW_CPU_MEM_USAGE_DEFAULT, load_model_dict_into_meta
 from ..utils import (
-    DIFFUSERS_CACHE,
-    HF_HUB_OFFLINE,
     USE_PEFT_BACKEND,
     _get_model_file,
     delete_adapter_layers,
@@ -62,6 +61,7 @@ class UNet2DConditionLoadersMixin:
     text_encoder_name = TEXT_ENCODER_NAME
     unet_name = UNET_NAME
 
+    @validate_hf_hub_args
     def load_attn_procs(self, pretrained_model_name_or_path_or_dict: Union[str, Dict[str, torch.Tensor]], **kwargs):
         r"""
         Load pretrained attention processor layers into [`UNet2DConditionModel`]. Attention processor layers have to be
@@ -95,7 +95,7 @@ class UNet2DConditionLoadersMixin:
             local_files_only (`bool`, *optional*, defaults to `False`):
                 Whether to only load local model weights and configuration files or not. If set to `True`, the model
                 won't be downloaded from the Hub.
-            use_auth_token (`str` or *bool*, *optional*):
+            token (`str` or *bool*, *optional*):
                 The token to use as HTTP bearer authorization for remote files. If `True`, the token generated from
                 `diffusers-cli login` (stored in `~/.huggingface`) is used.
             low_cpu_mem_usage (`bool`, *optional*, defaults to `True` if torch version >= 1.9.0 else `False`):
@@ -130,12 +130,12 @@ class UNet2DConditionLoadersMixin:
         from ..models.attention_processor import CustomDiffusionAttnProcessor
         from ..models.lora import LoRACompatibleConv, LoRACompatibleLinear, LoRAConv2dLayer, LoRALinearLayer
 
-        cache_dir = kwargs.pop("cache_dir", DIFFUSERS_CACHE)
+        cache_dir = kwargs.pop("cache_dir", None)
         force_download = kwargs.pop("force_download", False)
         resume_download = kwargs.pop("resume_download", False)
         proxies = kwargs.pop("proxies", None)
-        local_files_only = kwargs.pop("local_files_only", HF_HUB_OFFLINE)
-        use_auth_token = kwargs.pop("use_auth_token", None)
+        local_files_only = kwargs.pop("local_files_only", None)
+        token = kwargs.pop("token", None)
         revision = kwargs.pop("revision", None)
         subfolder = kwargs.pop("subfolder", None)
         weight_name = kwargs.pop("weight_name", None)
@@ -184,7 +184,7 @@ class UNet2DConditionLoadersMixin:
                         resume_download=resume_download,
                         proxies=proxies,
                         local_files_only=local_files_only,
-                        use_auth_token=use_auth_token,
+                        token=token,
                         revision=revision,
                         subfolder=subfolder,
                         user_agent=user_agent,
@@ -204,7 +204,7 @@ class UNet2DConditionLoadersMixin:
                     resume_download=resume_download,
                     proxies=proxies,
                     local_files_only=local_files_only,
-                    use_auth_token=use_auth_token,
+                    token=token,
                     revision=revision,
                     subfolder=subfolder,
                     user_agent=user_agent,

src/diffusers/models/__init__.py

@@ -33,8 +33,8 @@ if is_torch_available():
     _import_structure["consistency_decoder_vae"] = ["ConsistencyDecoderVAE"]
     _import_structure["controlnet"] = ["ControlNetModel"]
     _import_structure["dual_transformer_2d"] = ["DualTransformer2DModel"]
-    _import_structure["modeling_utils"] = ["ModelMixin"]
     _import_structure["embeddings"] = ["ImageProjection"]
+    _import_structure["modeling_utils"] = ["ModelMixin"]
     _import_structure["prior_transformer"] = ["PriorTransformer"]
     _import_structure["t5_film_transformer"] = ["T5FilmDecoder"]
     _import_structure["transformer_2d"] = ["Transformer2DModel"]

src/diffusers/models/modeling_flax_utils.py

@@ -24,13 +24,17 @@ from flax.core.frozen_dict import FrozenDict, unfreeze
 from flax.serialization import from_bytes, to_bytes
 from flax.traverse_util import flatten_dict, unflatten_dict
 from huggingface_hub import create_repo, hf_hub_download
-from huggingface_hub.utils import EntryNotFoundError, RepositoryNotFoundError, RevisionNotFoundError
+from huggingface_hub.utils import (
+    EntryNotFoundError,
+    RepositoryNotFoundError,
+    RevisionNotFoundError,
+    validate_hf_hub_args,
+)
 from requests import HTTPError
 
 from .. import __version__, is_torch_available
 from ..utils import (
     CONFIG_NAME,
-    DIFFUSERS_CACHE,
     FLAX_WEIGHTS_NAME,
     HUGGINGFACE_CO_RESOLVE_ENDPOINT,
     WEIGHTS_NAME,
@@ -197,6 +201,7 @@ class FlaxModelMixin(PushToHubMixin):
         raise NotImplementedError(f"init_weights method has to be implemented for {self}")
 
     @classmethod
+    @validate_hf_hub_args
     def from_pretrained(
         cls,
         pretrained_model_name_or_path: Union[str, os.PathLike],
@@ -288,13 +293,13 @@ class FlaxModelMixin(PushToHubMixin):
         ```
         """
         config = kwargs.pop("config", None)
-        cache_dir = kwargs.pop("cache_dir", DIFFUSERS_CACHE)
+        cache_dir = kwargs.pop("cache_dir", None)
         force_download = kwargs.pop("force_download", False)
         from_pt = kwargs.pop("from_pt", False)
         resume_download = kwargs.pop("resume_download", False)
         proxies = kwargs.pop("proxies", None)
         local_files_only = kwargs.pop("local_files_only", False)
-        use_auth_token = kwargs.pop("use_auth_token", None)
+        token = kwargs.pop("token", None)
         revision = kwargs.pop("revision", None)
         subfolder = kwargs.pop("subfolder", None)
 
@@ -314,7 +319,7 @@ class FlaxModelMixin(PushToHubMixin):
                 resume_download=resume_download,
                 proxies=proxies,
                 local_files_only=local_files_only,
-                use_auth_token=use_auth_token,
+                token=token,
                 revision=revision,
                 subfolder=subfolder,
                 **kwargs,
@@ -359,7 +364,7 @@ class FlaxModelMixin(PushToHubMixin):
                     proxies=proxies,
                     resume_download=resume_download,
                     local_files_only=local_files_only,
-                    use_auth_token=use_auth_token,
+                    token=token,
                     user_agent=user_agent,
                     subfolder=subfolder,
                     revision=revision,
@@ -369,7 +374,7 @@ class FlaxModelMixin(PushToHubMixin):
                 raise EnvironmentError(
                     f"{pretrained_model_name_or_path} is not a local folder and is not a valid model identifier "
                     "listed on 'https://huggingface.co/models'\nIf this is a private repository, make sure to pass a "
-                    "token having permission to this repo with `use_auth_token` or log in with `huggingface-cli "
+                    "token having permission to this repo with `token` or log in with `huggingface-cli "
                     "login`."
                 )
             except RevisionNotFoundError:

src/diffusers/models/modeling_utils.py

@@ -25,14 +25,13 @@ from typing import Any, Callable, List, Optional, Tuple, Union
 import safetensors
 import torch
 from huggingface_hub import create_repo
+from huggingface_hub.utils import validate_hf_hub_args
 from torch import Tensor, nn
 
 from .. import __version__
 from ..utils import (
     CONFIG_NAME,
-    DIFFUSERS_CACHE,
     FLAX_WEIGHTS_NAME,
-    HF_HUB_OFFLINE,
     MIN_PEFT_VERSION,
     SAFETENSORS_WEIGHTS_NAME,
     WEIGHTS_NAME,
@@ -535,6 +534,7 @@ class ModelMixin(torch.nn.Module, PushToHubMixin):
             )
 
     @classmethod
+    @validate_hf_hub_args
     def from_pretrained(cls, pretrained_model_name_or_path: Optional[Union[str, os.PathLike]], **kwargs):
         r"""
         Instantiate a pretrained PyTorch model from a pretrained model configuration.
@@ -571,7 +571,7 @@ class ModelMixin(torch.nn.Module, PushToHubMixin):
             local_files_only(`bool`, *optional*, defaults to `False`):
                 Whether to only load local model weights and configuration files or not. If set to `True`, the model
                 won't be downloaded from the Hub.
-            use_auth_token (`str` or *bool*, *optional*):
+            token (`str` or *bool*, *optional*):
                 The token to use as HTTP bearer authorization for remote files. If `True`, the token generated from
                 `diffusers-cli login` (stored in `~/.huggingface`) is used.
             revision (`str`, *optional*, defaults to `"main"`):
@@ -640,15 +640,15 @@ class ModelMixin(torch.nn.Module, PushToHubMixin):
         You should probably TRAIN this model on a down-stream task to be able to use it for predictions and inference.
         ```
         """
-        cache_dir = kwargs.pop("cache_dir", DIFFUSERS_CACHE)
+        cache_dir = kwargs.pop("cache_dir", None)
         ignore_mismatched_sizes = kwargs.pop("ignore_mismatched_sizes", False)
         force_download = kwargs.pop("force_download", False)
         from_flax = kwargs.pop("from_flax", False)
         resume_download = kwargs.pop("resume_download", False)
         proxies = kwargs.pop("proxies", None)
         output_loading_info = kwargs.pop("output_loading_info", False)
-        local_files_only = kwargs.pop("local_files_only", HF_HUB_OFFLINE)
-        use_auth_token = kwargs.pop("use_auth_token", None)
+        local_files_only = kwargs.pop("local_files_only", None)
+        token = kwargs.pop("token", None)
         revision = kwargs.pop("revision", None)
         torch_dtype = kwargs.pop("torch_dtype", None)
         subfolder = kwargs.pop("subfolder", None)
@@ -718,7 +718,7 @@ class ModelMixin(torch.nn.Module, PushToHubMixin):
             resume_download=resume_download,
             proxies=proxies,
             local_files_only=local_files_only,
-            use_auth_token=use_auth_token,
+            token=token,
             revision=revision,
             subfolder=subfolder,
             device_map=device_map,
@@ -740,7 +740,7 @@ class ModelMixin(torch.nn.Module, PushToHubMixin):
                 resume_download=resume_download,
                 proxies=proxies,
                 local_files_only=local_files_only,
-                use_auth_token=use_auth_token,
+                token=token,
                 revision=revision,
                 subfolder=subfolder,
                 user_agent=user_agent,
@@ -763,7 +763,7 @@ class ModelMixin(torch.nn.Module, PushToHubMixin):
                         resume_download=resume_download,
                         proxies=proxies,
                         local_files_only=local_files_only,
-                        use_auth_token=use_auth_token,
+                        token=token,
                         revision=revision,
                         subfolder=subfolder,
                         user_agent=user_agent,
@@ -782,7 +782,7 @@ class ModelMixin(torch.nn.Module, PushToHubMixin):
                     resume_download=resume_download,
                     proxies=proxies,
                     local_files_only=local_files_only,
-                    use_auth_token=use_auth_token,
+                    token=token,
                     revision=revision,
                     subfolder=subfolder,
                     user_agent=user_agent,

src/diffusers/pipelines/auto_pipeline.py

@@ -16,8 +16,9 @@
 import inspect
 from collections import OrderedDict
 
+from huggingface_hub.utils import validate_hf_hub_args
+
 from ..configuration_utils import ConfigMixin
-from ..utils import DIFFUSERS_CACHE
 from .controlnet import (
     StableDiffusionControlNetImg2ImgPipeline,
     StableDiffusionControlNetInpaintPipeline,
@@ -195,6 +196,7 @@ class AutoPipelineForText2Image(ConfigMixin):
         )
 
     @classmethod
+    @validate_hf_hub_args
     def from_pretrained(cls, pretrained_model_or_path, **kwargs):
         r"""
         Instantiates a text-to-image Pytorch diffusion pipeline from pretrained pipeline weight.
@@ -246,7 +248,7 @@ class AutoPipelineForText2Image(ConfigMixin):
             local_files_only (`bool`, *optional*, defaults to `False`):
                 Whether to only load local model weights and configuration files or not. If set to `True`, the model
                 won't be downloaded from the Hub.
-            use_auth_token (`str` or *bool*, *optional*):
+            token (`str` or *bool*, *optional*):
                 The token to use as HTTP bearer authorization for remote files. If `True`, the token generated from
                 `diffusers-cli login` (stored in `~/.huggingface`) is used.
             revision (`str`, *optional*, defaults to `"main"`):
@@ -310,11 +312,11 @@ class AutoPipelineForText2Image(ConfigMixin):
         >>> image = pipeline(prompt).images[0]
         ```
         """
-        cache_dir = kwargs.pop("cache_dir", DIFFUSERS_CACHE)
+        cache_dir = kwargs.pop("cache_dir", None)
         force_download = kwargs.pop("force_download", False)
         resume_download = kwargs.pop("resume_download", False)
         proxies = kwargs.pop("proxies", None)
-        use_auth_token = kwargs.pop("use_auth_token", None)
+        token = kwargs.pop("token", None)
         local_files_only = kwargs.pop("local_files_only", False)
         revision = kwargs.pop("revision", None)
 
@@ -323,7 +325,7 @@ class AutoPipelineForText2Image(ConfigMixin):
             "force_download": force_download,
             "resume_download": resume_download,
             "proxies": proxies,
-            "use_auth_token": use_auth_token,
+            "token": token,
             "local_files_only": local_files_only,
             "revision": revision,
         }
@@ -466,6 +468,7 @@ class AutoPipelineForImage2Image(ConfigMixin):
         )
 
     @classmethod
+    @validate_hf_hub_args
     def from_pretrained(cls, pretrained_model_or_path, **kwargs):
         r"""
         Instantiates a image-to-image Pytorch diffusion pipeline from pretrained pipeline weight.
@@ -518,7 +521,7 @@ class AutoPipelineForImage2Image(ConfigMixin):
             local_files_only (`bool`, *optional*, defaults to `False`):
                 Whether to only load local model weights and configuration files or not. If set to `True`, the model
                 won't be downloaded from the Hub.
-            use_auth_token (`str` or *bool*, *optional*):
+            token (`str` or *bool*, *optional*):
                 The token to use as HTTP bearer authorization for remote files. If `True`, the token generated from
                 `diffusers-cli login` (stored in `~/.huggingface`) is used.
             revision (`str`, *optional*, defaults to `"main"`):
@@ -582,11 +585,11 @@ class AutoPipelineForImage2Image(ConfigMixin):
         >>> image = pipeline(prompt, image).images[0]
         ```
         """
-        cache_dir = kwargs.pop("cache_dir", DIFFUSERS_CACHE)
+        cache_dir = kwargs.pop("cache_dir", None)
         force_download = kwargs.pop("force_download", False)
         resume_download = kwargs.pop("resume_download", False)
         proxies = kwargs.pop("proxies", None)
-        use_auth_token = kwargs.pop("use_auth_token", None)
+        token = kwargs.pop("token", None)
         local_files_only = kwargs.pop("local_files_only", False)
         revision = kwargs.pop("revision", None)
 
@@ -595,7 +598,7 @@ class AutoPipelineForImage2Image(ConfigMixin):
             "force_download": force_download,
             "resume_download": resume_download,
             "proxies": proxies,
-            "use_auth_token": use_auth_token,
+            "token": token,
             "local_files_only": local_files_only,
             "revision": revision,
         }
@@ -742,6 +745,7 @@ class AutoPipelineForInpainting(ConfigMixin):
         )
 
     @classmethod
+    @validate_hf_hub_args
     def from_pretrained(cls, pretrained_model_or_path, **kwargs):
         r"""
         Instantiates a inpainting Pytorch diffusion pipeline from pretrained pipeline weight.
@@ -793,7 +797,7 @@ class AutoPipelineForInpainting(ConfigMixin):
             local_files_only (`bool`, *optional*, defaults to `False`):
                 Whether to only load local model weights and configuration files or not. If set to `True`, the model
                 won't be downloaded from the Hub.
-            use_auth_token (`str` or *bool*, *optional*):
+            token (`str` or *bool*, *optional*):
                 The token to use as HTTP bearer authorization for remote files. If `True`, the token generated from
                 `diffusers-cli login` (stored in `~/.huggingface`) is used.
             revision (`str`, *optional*, defaults to `"main"`):
@@ -857,11 +861,11 @@ class AutoPipelineForInpainting(ConfigMixin):
         >>> image = pipeline(prompt, image=init_image, mask_image=mask_image).images[0]
         ```
         """
-        cache_dir = kwargs.pop("cache_dir", DIFFUSERS_CACHE)
+        cache_dir = kwargs.pop("cache_dir", None)
         force_download = kwargs.pop("force_download", False)
         resume_download = kwargs.pop("resume_download", False)
         proxies = kwargs.pop("proxies", None)
-        use_auth_token = kwargs.pop("use_auth_token", None)
+        token = kwargs.pop("token", None)
         local_files_only = kwargs.pop("local_files_only", False)
         revision = kwargs.pop("revision", None)
 
@@ -870,7 +874,7 @@ class AutoPipelineForInpainting(ConfigMixin):
             "force_download": force_download,
             "resume_download": resume_download,
             "proxies": proxies,
-            "use_auth_token": use_auth_token,
+            "token": token,
             "local_files_only": local_files_only,
             "revision": revision,
         }

src/diffusers/pipelines/onnx_utils.py

@@ -22,6 +22,7 @@ from typing import Optional, Union
 
 import numpy as np
 from huggingface_hub import hf_hub_download
+from huggingface_hub.utils import validate_hf_hub_args
 
 from ..utils import ONNX_EXTERNAL_WEIGHTS_NAME, ONNX_WEIGHTS_NAME, is_onnx_available, logging
 
@@ -130,10 +131,11 @@ class OnnxRuntimeModel:
         self._save_pretrained(save_directory, **kwargs)
 
     @classmethod
+    @validate_hf_hub_args
     def _from_pretrained(
         cls,
         model_id: Union[str, Path],
-        use_auth_token: Optional[Union[bool, str, None]] = None,
+        token: Optional[Union[bool, str, None]] = None,
         revision: Optional[Union[str, None]] = None,
         force_download: bool = False,
         cache_dir: Optional[str] = None,
@@ -148,7 +150,7 @@ class OnnxRuntimeModel:
         Arguments:
             model_id (`str` or `Path`):
                 Directory from which to load
-            use_auth_token (`str` or `bool`):
+            token (`str` or `bool`):
                 Is needed to load models from a private or gated repository
             revision (`str`):
                 Revision is the specific model version to use. It can be a branch name, a tag name, or a commit id
@@ -179,7 +181,7 @@ class OnnxRuntimeModel:
             model_cache_path = hf_hub_download(
                 repo_id=model_id,
                 filename=model_file_name,
-                use_auth_token=use_auth_token,
+                token=token,
                 revision=revision,
                 cache_dir=cache_dir,
                 force_download=force_download,
@@ -190,11 +192,12 @@ class OnnxRuntimeModel:
         return cls(model=model, **kwargs)
 
     @classmethod
+    @validate_hf_hub_args
     def from_pretrained(
         cls,
         model_id: Union[str, Path],
         force_download: bool = True,
-        use_auth_token: Optional[str] = None,
+        token: Optional[str] = None,
         cache_dir: Optional[str] = None,
         **model_kwargs,
     ):
@@ -207,6 +210,6 @@ class OnnxRuntimeModel:
             revision=revision,
             cache_dir=cache_dir,
             force_download=force_download,
-            use_auth_token=use_auth_token,
+            token=token,
             **model_kwargs,
         )

src/diffusers/pipelines/pipeline_flax_utils.py

@@ -24,6 +24,7 @@ import numpy as np
 import PIL.Image
 from flax.core.frozen_dict import FrozenDict
 from huggingface_hub import create_repo, snapshot_download
+from huggingface_hub.utils import validate_hf_hub_args
 from PIL import Image
 from tqdm.auto import tqdm
 
@@ -32,7 +33,6 @@ from ..models.modeling_flax_utils import FLAX_WEIGHTS_NAME, FlaxModelMixin
 from ..schedulers.scheduling_utils_flax import SCHEDULER_CONFIG_NAME, FlaxSchedulerMixin
 from ..utils import (
     CONFIG_NAME,
-    DIFFUSERS_CACHE,
     BaseOutput,
     PushToHubMixin,
     http_user_agent,
@@ -227,6 +227,7 @@ class FlaxDiffusionPipeline(ConfigMixin, PushToHubMixin):
                 )
 
     @classmethod
+    @validate_hf_hub_args
     def from_pretrained(cls, pretrained_model_name_or_path: Optional[Union[str, os.PathLike]], **kwargs):
         r"""
         Instantiate a Flax-based diffusion pipeline from pretrained pipeline weights.
@@ -264,7 +265,7 @@ class FlaxDiffusionPipeline(ConfigMixin, PushToHubMixin):
             local_files_only (`bool`, *optional*, defaults to `False`):
                 Whether to only load local model weights and configuration files or not. If set to `True`, the model
                 won't be downloaded from the Hub.
-            use_auth_token (`str` or *bool*, *optional*):
+            token (`str` or *bool*, *optional*):
                 The token to use as HTTP bearer authorization for remote files. If `True`, the token generated from
                 `diffusers-cli login` (stored in `~/.huggingface`) is used.
             revision (`str`, *optional*, defaults to `"main"`):
@@ -314,11 +315,11 @@ class FlaxDiffusionPipeline(ConfigMixin, PushToHubMixin):
         >>> dpm_params["scheduler"] = dpmpp_state
         ```
         """
-        cache_dir = kwargs.pop("cache_dir", DIFFUSERS_CACHE)
+        cache_dir = kwargs.pop("cache_dir", None)
         resume_download = kwargs.pop("resume_download", False)
         proxies = kwargs.pop("proxies", None)
         local_files_only = kwargs.pop("local_files_only", False)
-        use_auth_token = kwargs.pop("use_auth_token", None)
+        token = kwargs.pop("token", None)
         revision = kwargs.pop("revision", None)
         from_pt = kwargs.pop("from_pt", False)
         use_memory_efficient_attention = kwargs.pop("use_memory_efficient_attention", False)
@@ -334,7 +335,7 @@ class FlaxDiffusionPipeline(ConfigMixin, PushToHubMixin):
                 resume_download=resume_download,
                 proxies=proxies,
                 local_files_only=local_files_only,
-                use_auth_token=use_auth_token,
+                token=token,
                 revision=revision,
             )
             # make sure we only download sub-folders and `diffusers` filenames
@@ -365,7 +366,7 @@ class FlaxDiffusionPipeline(ConfigMixin, PushToHubMixin):
                 resume_download=resume_download,
                 proxies=proxies,
                 local_files_only=local_files_only,
-                use_auth_token=use_auth_token,
+                token=token,
                 revision=revision,
                 allow_patterns=allow_patterns,
                 ignore_patterns=ignore_patterns,

src/diffusers/pipelines/pipeline_utils.py

@@ -28,7 +28,14 @@ from typing import Any, Callable, Dict, List, Optional, Union
 import numpy as np
 import PIL.Image
 import torch
-from huggingface_hub import ModelCard, create_repo, hf_hub_download, model_info, snapshot_download
+from huggingface_hub import (
+    ModelCard,
+    create_repo,
+    hf_hub_download,
+    model_info,
+    snapshot_download,
+)
+from huggingface_hub.utils import validate_hf_hub_args
 from packaging import version
 from requests.exceptions import HTTPError
 from tqdm.auto import tqdm
@@ -40,8 +47,6 @@ from ..schedulers.scheduling_utils import SCHEDULER_CONFIG_NAME
 from ..utils import (
     CONFIG_NAME,
     DEPRECATED_REVISION_ARGS,
-    DIFFUSERS_CACHE,
-    HF_HUB_OFFLINE,
     SAFETENSORS_WEIGHTS_NAME,
     WEIGHTS_NAME,
     BaseOutput,
@@ -249,10 +254,11 @@ def variant_compatible_siblings(filenames, variant=None) -> Union[List[os.PathLi
     return usable_filenames, variant_filenames
 
 
-def warn_deprecated_model_variant(pretrained_model_name_or_path, use_auth_token, variant, revision, model_filenames):
+@validate_hf_hub_args
+def warn_deprecated_model_variant(pretrained_model_name_or_path, token, variant, revision, model_filenames):
     info = model_info(
         pretrained_model_name_or_path,
-        use_auth_token=use_auth_token,
+        token=token,
         revision=None,
     )
     filenames = {sibling.rfilename for sibling in info.siblings}
@@ -375,7 +381,6 @@ def _get_pipeline_class(
             custom_pipeline,
             module_file=file_name,
             class_name=class_name,
-            repo_id=repo_id,
             cache_dir=cache_dir,
             revision=revision,
         )
@@ -909,6 +914,7 @@ class DiffusionPipeline(ConfigMixin, PushToHubMixin):
         return torch.float32
 
     @classmethod
+    @validate_hf_hub_args
     def from_pretrained(cls, pretrained_model_name_or_path: Optional[Union[str, os.PathLike]], **kwargs):
         r"""
         Instantiate a PyTorch diffusion pipeline from pretrained pipeline weights.
@@ -976,7 +982,7 @@ class DiffusionPipeline(ConfigMixin, PushToHubMixin):
             local_files_only (`bool`, *optional*, defaults to `False`):
                 Whether to only load local model weights and configuration files or not. If set to `True`, the model
                 won't be downloaded from the Hub.
-            use_auth_token (`str` or *bool*, *optional*):
+            token (`str` or *bool*, *optional*):
                 The token to use as HTTP bearer authorization for remote files. If `True`, the token generated from
                 `diffusers-cli login` (stored in `~/.huggingface`) is used.
             revision (`str`, *optional*, defaults to `"main"`):
@@ -1056,12 +1062,12 @@ class DiffusionPipeline(ConfigMixin, PushToHubMixin):
         >>> pipeline.scheduler = scheduler
         ```
         """
-        cache_dir = kwargs.pop("cache_dir", DIFFUSERS_CACHE)
+        cache_dir = kwargs.pop("cache_dir", None)
         resume_download = kwargs.pop("resume_download", False)
         force_download = kwargs.pop("force_download", False)
         proxies = kwargs.pop("proxies", None)
-        local_files_only = kwargs.pop("local_files_only", HF_HUB_OFFLINE)
-        use_auth_token = kwargs.pop("use_auth_token", None)
+        local_files_only = kwargs.pop("local_files_only", None)
+        token = kwargs.pop("token", None)
         revision = kwargs.pop("revision", None)
         from_flax = kwargs.pop("from_flax", False)
         torch_dtype = kwargs.pop("torch_dtype", None)
@@ -1094,7 +1100,7 @@ class DiffusionPipeline(ConfigMixin, PushToHubMixin):
                 force_download=force_download,
                 proxies=proxies,
                 local_files_only=local_files_only,
-                use_auth_token=use_auth_token,
+                token=token,
                 revision=revision,
                 from_flax=from_flax,
                 use_safetensors=use_safetensors,
@@ -1299,7 +1305,7 @@ class DiffusionPipeline(ConfigMixin, PushToHubMixin):
                 "force_download": force_download,
                 "proxies": proxies,
                 "local_files_only": local_files_only,
-                "use_auth_token": use_auth_token,
+                "token": token,
                 "revision": revision,
                 "torch_dtype": torch_dtype,
                 "custom_pipeline": custom_pipeline,
@@ -1529,6 +1535,7 @@ class DiffusionPipeline(ConfigMixin, PushToHubMixin):
                 cpu_offload(model, device, offload_buffers=offload_buffers)
 
     @classmethod
+    @validate_hf_hub_args
     def download(cls, pretrained_model_name, **kwargs) -> Union[str, os.PathLike]:
         r"""
         Download and cache a PyTorch diffusion pipeline from pretrained pipeline weights.
@@ -1576,7 +1583,7 @@ class DiffusionPipeline(ConfigMixin, PushToHubMixin):
             local_files_only (`bool`, *optional*, defaults to `False`):
                 Whether to only load local model weights and configuration files or not. If set to `True`, the model
                 won't be downloaded from the Hub.
-            use_auth_token (`str` or *bool*, *optional*):
+            token (`str` or *bool*, *optional*):
                 The token to use as HTTP bearer authorization for remote files. If `True`, the token generated from
                 `diffusers-cli login` (stored in `~/.huggingface`) is used.
             revision (`str`, *optional*, defaults to `"main"`):
@@ -1619,12 +1626,12 @@ class DiffusionPipeline(ConfigMixin, PushToHubMixin):
         </Tip>
 
         """
-        cache_dir = kwargs.pop("cache_dir", DIFFUSERS_CACHE)
+        cache_dir = kwargs.pop("cache_dir", None)
         resume_download = kwargs.pop("resume_download", False)
         force_download = kwargs.pop("force_download", False)
         proxies = kwargs.pop("proxies", None)
-        local_files_only = kwargs.pop("local_files_only", HF_HUB_OFFLINE)
-        use_auth_token = kwargs.pop("use_auth_token", None)
+        local_files_only = kwargs.pop("local_files_only", None)
+        token = kwargs.pop("token", None)
         revision = kwargs.pop("revision", None)
         from_flax = kwargs.pop("from_flax", False)
         custom_pipeline = kwargs.pop("custom_pipeline", None)
@@ -1646,11 +1653,7 @@ class DiffusionPipeline(ConfigMixin, PushToHubMixin):
         model_info_call_error: Optional[Exception] = None
         if not local_files_only:
             try:
-                info = model_info(
-                    pretrained_model_name,
-                    use_auth_token=use_auth_token,
-                    revision=revision,
-                )
+                info = model_info(pretrained_model_name, token=token, revision=revision)
             except HTTPError as e:
                 logger.warn(f"Couldn't connect to the Hub: {e}.\nWill try to load from local cache.")
                 local_files_only = True
@@ -1665,7 +1668,7 @@ class DiffusionPipeline(ConfigMixin, PushToHubMixin):
                 proxies=proxies,
                 force_download=force_download,
                 resume_download=resume_download,
-                use_auth_token=use_auth_token,
+                token=token,
             )
 
             config_dict = cls._dict_from_json_file(config_file)
@@ -1715,9 +1718,7 @@ class DiffusionPipeline(ConfigMixin, PushToHubMixin):
             if revision in DEPRECATED_REVISION_ARGS and version.parse(
                 version.parse(__version__).base_version
             ) >= version.parse("0.22.0"):
-                warn_deprecated_model_variant(
-                    pretrained_model_name, use_auth_token, variant, revision, model_filenames
-                )
+                warn_deprecated_model_variant(pretrained_model_name, token, variant, revision, model_filenames)
 
             model_folder_names = {os.path.split(f)[0] for f in model_filenames if os.path.split(f)[0] in folder_names}
 
@@ -1859,7 +1860,7 @@ class DiffusionPipeline(ConfigMixin, PushToHubMixin):
                 resume_download=resume_download,
                 proxies=proxies,
                 local_files_only=local_files_only,
-                use_auth_token=use_auth_token,
+                token=token,
                 revision=revision,
                 allow_patterns=allow_patterns,
                 ignore_patterns=ignore_patterns,
@@ -1883,7 +1884,7 @@ class DiffusionPipeline(ConfigMixin, PushToHubMixin):
                         "force_download": force_download,
                         "proxies": proxies,
                         "local_files_only": local_files_only,
-                        "use_auth_token": use_auth_token,
+                        "token": token,
                         "variant": variant,
                         "use_safetensors": use_safetensors,
                     }

src/diffusers/schedulers/scheduling_utils.py

@@ -18,6 +18,7 @@ from enum import Enum
 from typing import Optional, Union
 
 import torch
+from huggingface_hub.utils import validate_hf_hub_args
 
 from ..utils import BaseOutput, PushToHubMixin
 
@@ -81,6 +82,7 @@ class SchedulerMixin(PushToHubMixin):
     has_compatibles = True
 
     @classmethod
+    @validate_hf_hub_args
     def from_pretrained(
         cls,
         pretrained_model_name_or_path: Optional[Union[str, os.PathLike]] = None,
@@ -120,7 +122,7 @@ class SchedulerMixin(PushToHubMixin):
             local_files_only(`bool`, *optional*, defaults to `False`):
                 Whether to only load local model weights and configuration files or not. If set to `True`, the model
                 won't be downloaded from the Hub.
-            use_auth_token (`str` or *bool*, *optional*):
+            token (`str` or *bool*, *optional*):
                 The token to use as HTTP bearer authorization for remote files. If `True`, the token generated from
                 `diffusers-cli login` (stored in `~/.huggingface`) is used.
             revision (`str`, *optional*, defaults to `"main"`):

src/diffusers/schedulers/scheduling_utils_flax.py

@@ -20,6 +20,7 @@ from typing import Optional, Tuple, Union
 
 import flax
 import jax.numpy as jnp
+from huggingface_hub.utils import validate_hf_hub_args
 
 from ..utils import BaseOutput, PushToHubMixin
 
@@ -70,6 +71,7 @@ class FlaxSchedulerMixin(PushToHubMixin):
     has_compatibles = True
 
     @classmethod
+    @validate_hf_hub_args
     def from_pretrained(
         cls,
         pretrained_model_name_or_path: Optional[Union[str, os.PathLike]] = None,
@@ -110,7 +112,7 @@ class FlaxSchedulerMixin(PushToHubMixin):
                 Whether or not to also return a dictionary containing missing keys, unexpected keys and error messages.
             local_files_only(`bool`, *optional*, defaults to `False`):
                 Whether or not to only look at local files (i.e., do not try to download the model).
-            use_auth_token (`str` or *bool*, *optional*):
+            token (`str` or *bool*, *optional*):
                 The token to use as HTTP bearer authorization for remote files. If `True`, will use the token generated
                 when running `transformers-cli login` (stored in `~/.huggingface`).
             revision (`str`, *optional*, defaults to `"main"`):

src/diffusers/utils/__init__.py

@@ -21,7 +21,6 @@ from .. import __version__
 from .constants import (
     CONFIG_NAME,
     DEPRECATED_REVISION_ARGS,
-    DIFFUSERS_CACHE,
     DIFFUSERS_DYNAMIC_MODULE_NAME,
     FLAX_WEIGHTS_NAME,
     HF_MODULES_CACHE,
@@ -38,7 +37,6 @@ from .doc_utils import replace_example_docstring
 from .dynamic_modules_utils import get_class_from_dynamic_module
 from .export_utils import export_to_gif, export_to_obj, export_to_ply, export_to_video
 from .hub_utils import (
-    HF_HUB_OFFLINE,
     PushToHubMixin,
     _add_variant,
     _get_model_file,

src/diffusers/utils/constants.py

@@ -14,15 +14,13 @@
 import importlib
 import os
 
-from huggingface_hub.constants import HUGGINGFACE_HUB_CACHE, hf_cache_home
+from huggingface_hub.constants import HF_HOME
 from packaging import version
 
 from ..dependency_versions_check import dep_version_check
 from .import_utils import ENV_VARS_TRUE_VALUES, is_peft_available, is_transformers_available
 
 
-default_cache_path = HUGGINGFACE_HUB_CACHE
-
 MIN_PEFT_VERSION = "0.6.0"
 MIN_TRANSFORMERS_VERSION = "4.34.0"
 _CHECK_PEFT = os.environ.get("_CHECK_PEFT", "1") in ENV_VARS_TRUE_VALUES
@@ -35,9 +33,8 @@ ONNX_WEIGHTS_NAME = "model.onnx"
 SAFETENSORS_WEIGHTS_NAME = "diffusion_pytorch_model.safetensors"
 ONNX_EXTERNAL_WEIGHTS_NAME = "weights.pb"
 HUGGINGFACE_CO_RESOLVE_ENDPOINT = os.environ.get("HF_ENDPOINT", "https://huggingface.co")
-DIFFUSERS_CACHE = default_cache_path
 DIFFUSERS_DYNAMIC_MODULE_NAME = "diffusers_modules"
-HF_MODULES_CACHE = os.getenv("HF_MODULES_CACHE", os.path.join(hf_cache_home, "modules"))
+HF_MODULES_CACHE = os.getenv("HF_MODULES_CACHE", os.path.join(HF_HOME, "modules"))
 DEPRECATED_REVISION_ARGS = ["fp16", "non-ema"]
 
 # Below should be `True` if the current version of `peft` and `transformers` are compatible with

src/diffusers/utils/dynamic_modules_utils.py

@@ -25,7 +25,8 @@ from pathlib import Path
 from typing import Dict, Optional, Union
 from urllib import request
 
-from huggingface_hub import HfFolder, cached_download, hf_hub_download, model_info
+from huggingface_hub import cached_download, hf_hub_download, model_info
+from huggingface_hub.utils import validate_hf_hub_args
 from packaging import version
 
 from .. import __version__
@@ -194,6 +195,7 @@ def find_pipeline_class(loaded_module):
     return pipeline_class
 
 
+@validate_hf_hub_args
 def get_cached_module_file(
     pretrained_model_name_or_path: Union[str, os.PathLike],
     module_file: str,
@@ -201,7 +203,7 @@ def get_cached_module_file(
     force_download: bool = False,
     resume_download: bool = False,
     proxies: Optional[Dict[str, str]] = None,
-    use_auth_token: Optional[Union[bool, str]] = None,
+    token: Optional[Union[bool, str]] = None,
     revision: Optional[str] = None,
     local_files_only: bool = False,
 ):
@@ -232,7 +234,7 @@ def get_cached_module_file(
         proxies (`Dict[str, str]`, *optional*):
             A dictionary of proxy servers to use by protocol or endpoint, e.g., `{'http': 'foo.bar:3128',
             'http://hostname': 'foo.bar:4012'}.` The proxies are used on each request.
-        use_auth_token (`str` or *bool*, *optional*):
+        token (`str` or *bool*, *optional*):
             The token to use as HTTP bearer authorization for remote files. If `True`, will use the token generated
             when running `transformers-cli login` (stored in `~/.huggingface`).
         revision (`str`, *optional*, defaults to `"main"`):
@@ -244,7 +246,7 @@ def get_cached_module_file(
 
     <Tip>
 
-    You may pass a token in `use_auth_token` if you are not logged in (`huggingface-cli long`) and want to use private
+    You may pass a token in `token` if you are not logged in (`huggingface-cli login`) and want to use private
     or [gated models](https://huggingface.co/docs/hub/models-gated#gated-models).
 
     </Tip>
@@ -289,7 +291,7 @@ def get_cached_module_file(
                 proxies=proxies,
                 resume_download=resume_download,
                 local_files_only=local_files_only,
-                use_auth_token=False,
+                token=False,
             )
             submodule = "git"
             module_file = pretrained_model_name_or_path + ".py"
@@ -307,7 +309,7 @@ def get_cached_module_file(
                 proxies=proxies,
                 resume_download=resume_download,
                 local_files_only=local_files_only,
-                use_auth_token=use_auth_token,
+                token=token,
             )
             submodule = os.path.join("local", "--".join(pretrained_model_name_or_path.split("/")))
         except EnvironmentError:
@@ -332,13 +334,6 @@ def get_cached_module_file(
     else:
         # Get the commit hash
         # TODO: we will get this info in the etag soon, so retrieve it from there and not here.
-        if isinstance(use_auth_token, str):
-            token = use_auth_token
-        elif use_auth_token is True:
-            token = HfFolder.get_token()
-        else:
-            token = None
-
         commit_hash = model_info(pretrained_model_name_or_path, revision=revision, token=token).sha
 
         # The module file will end up being placed in a subfolder with the git hash of the repo. This way we get the
@@ -359,13 +354,14 @@ def get_cached_module_file(
                     force_download=force_download,
                     resume_download=resume_download,
                     proxies=proxies,
-                    use_auth_token=use_auth_token,
+                    token=token,
                     revision=revision,
                     local_files_only=local_files_only,
                 )
     return os.path.join(full_submodule, module_file)
 
 
+@validate_hf_hub_args
 def get_class_from_dynamic_module(
     pretrained_model_name_or_path: Union[str, os.PathLike],
     module_file: str,
@@ -374,7 +370,7 @@ def get_class_from_dynamic_module(
     force_download: bool = False,
     resume_download: bool = False,
     proxies: Optional[Dict[str, str]] = None,
-    use_auth_token: Optional[Union[bool, str]] = None,
+    token: Optional[Union[bool, str]] = None,
     revision: Optional[str] = None,
     local_files_only: bool = False,
     **kwargs,
@@ -414,7 +410,7 @@ def get_class_from_dynamic_module(
         proxies (`Dict[str, str]`, *optional*):
             A dictionary of proxy servers to use by protocol or endpoint, e.g., `{'http': 'foo.bar:3128',
             'http://hostname': 'foo.bar:4012'}.` The proxies are used on each request.
-        use_auth_token (`str` or `bool`, *optional*):
+        token (`str` or `bool`, *optional*):
             The token to use as HTTP bearer authorization for remote files. If `True`, will use the token generated
             when running `transformers-cli login` (stored in `~/.huggingface`).
         revision (`str`, *optional*, defaults to `"main"`):
@@ -426,7 +422,7 @@ def get_class_from_dynamic_module(
 
     <Tip>
 
-    You may pass a token in `use_auth_token` if you are not logged in (`huggingface-cli long`) and want to use private
+    You may pass a token in `token` if you are not logged in (`huggingface-cli login`) and want to use private
     or [gated models](https://huggingface.co/docs/hub/models-gated#gated-models).
 
     </Tip>
@@ -449,7 +445,7 @@ def get_class_from_dynamic_module(
         force_download=force_download,
         resume_download=resume_download,
         proxies=proxies,
-        use_auth_token=use_auth_token,
+        token=token,
         revision=revision,
         local_files_only=local_files_only,
     )

src/diffusers/utils/hub_utils.py

@@ -25,20 +25,21 @@ from typing import Dict, Optional, Union
 from uuid import uuid4
 
 from huggingface_hub import (
-    HfFolder,
     ModelCard,
     ModelCardData,
     create_repo,
+    get_full_repo_name,
     hf_hub_download,
     upload_folder,
-    whoami,
 )
+from huggingface_hub.constants import HF_HUB_CACHE, HF_HUB_DISABLE_TELEMETRY, HF_HUB_OFFLINE
 from huggingface_hub.file_download import REGEX_COMMIT_HASH
 from huggingface_hub.utils import (
     EntryNotFoundError,
     RepositoryNotFoundError,
     RevisionNotFoundError,
     is_jinja_available,
+    validate_hf_hub_args,
 )
 from packaging import version
 from requests import HTTPError
@@ -46,7 +47,6 @@ from requests import HTTPError
 from .. import __version__
 from .constants import (
     DEPRECATED_REVISION_ARGS,
-    DIFFUSERS_CACHE,
     HUGGINGFACE_CO_RESOLVE_ENDPOINT,
     SAFETENSORS_WEIGHTS_NAME,
     WEIGHTS_NAME,
@@ -69,9 +69,6 @@ logger = get_logger(__name__)
 
 MODEL_CARD_TEMPLATE_PATH = Path(__file__).parent / "model_card_template.md"
 SESSION_ID = uuid4().hex
-HF_HUB_OFFLINE = os.getenv("HF_HUB_OFFLINE", "").upper() in ENV_VARS_TRUE_VALUES
-DISABLE_TELEMETRY = os.getenv("DISABLE_TELEMETRY", "").upper() in ENV_VARS_TRUE_VALUES
-HUGGINGFACE_CO_TELEMETRY = HUGGINGFACE_CO_RESOLVE_ENDPOINT + "/api/telemetry/"
 
 
 def http_user_agent(user_agent: Union[Dict, str, None] = None) -> str:
@@ -79,7 +76,7 @@ def http_user_agent(user_agent: Union[Dict, str, None] = None) -> str:
     Formats a user-agent string with basic info about a request.
     """
     ua = f"diffusers/{__version__}; python/{sys.version.split()[0]}; session_id/{SESSION_ID}"
-    if DISABLE_TELEMETRY or HF_HUB_OFFLINE:
+    if HF_HUB_DISABLE_TELEMETRY or HF_HUB_OFFLINE:
         return ua + "; telemetry/off"
     if is_torch_available():
         ua += f"; torch/{_torch_version}"
@@ -98,16 +95,6 @@ def http_user_agent(user_agent: Union[Dict, str, None] = None) -> str:
     return ua
 
 
-def get_full_repo_name(model_id: str, organization: Optional[str] = None, token: Optional[str] = None):
-    if token is None:
-        token = HfFolder.get_token()
-    if organization is None:
-        username = whoami(token)["name"]
-        return f"{username}/{model_id}"
-    else:
-        return f"{organization}/{model_id}"
-
-
 def create_model_card(args, model_name):
     if not is_jinja_available():
         raise ValueError(
@@ -183,7 +170,7 @@ old_diffusers_cache = os.path.join(hf_cache_home, "diffusers")
 
 def move_cache(old_cache_dir: Optional[str] = None, new_cache_dir: Optional[str] = None) -> None:
     if new_cache_dir is None:
-        new_cache_dir = DIFFUSERS_CACHE
+        new_cache_dir = HF_HUB_CACHE
     if old_cache_dir is None:
         old_cache_dir = old_diffusers_cache
 
@@ -203,7 +190,7 @@ def move_cache(old_cache_dir: Optional[str] = None, new_cache_dir: Optional[str]
     # At this point, old_cache_dir contains symlinks to the new cache (it can still be used).
 
 
-cache_version_file = os.path.join(DIFFUSERS_CACHE, "version_diffusers_cache.txt")
+cache_version_file = os.path.join(HF_HUB_CACHE, "version_diffusers_cache.txt")
 if not os.path.isfile(cache_version_file):
     cache_version = 0
 else:
@@ -233,12 +220,12 @@ if cache_version < 1:
 
 if cache_version < 1:
     try:
-        os.makedirs(DIFFUSERS_CACHE, exist_ok=True)
+        os.makedirs(HF_HUB_CACHE, exist_ok=True)
         with open(cache_version_file, "w") as f:
             f.write("1")
     except Exception:
         logger.warning(
-            f"There was a problem when trying to write in your cache folder ({DIFFUSERS_CACHE}). Please, ensure "
+            f"There was a problem when trying to write in your cache folder ({HF_HUB_CACHE}). Please, ensure "
             "the directory exists and can be written to."
         )
 
@@ -252,20 +239,21 @@ def _add_variant(weights_name: str, variant: Optional[str] = None) -> str:
     return weights_name
 
 
+@validate_hf_hub_args
 def _get_model_file(
-    pretrained_model_name_or_path,
+    pretrained_model_name_or_path: Union[str, Path],
     *,
-    weights_name,
-    subfolder,
-    cache_dir,
-    force_download,
-    proxies,
-    resume_download,
-    local_files_only,
-    use_auth_token,
-    user_agent,
-    revision,
-    commit_hash=None,
+    weights_name: str,
+    subfolder: Optional[str],
+    cache_dir: Optional[str],
+    force_download: bool,
+    proxies: Optional[Dict],
+    resume_download: bool,
+    local_files_only: bool,
+    token: Optional[str],
+    user_agent: Union[Dict, str, None],
+    revision: Optional[str],
+    commit_hash: Optional[str] = None,
 ):
     pretrained_model_name_or_path = str(pretrained_model_name_or_path)
     if os.path.isfile(pretrained_model_name_or_path):
@@ -300,7 +288,7 @@ def _get_model_file(
                     proxies=proxies,
                     resume_download=resume_download,
                     local_files_only=local_files_only,
-                    use_auth_token=use_auth_token,
+                    token=token,
                     user_agent=user_agent,
                     subfolder=subfolder,
                     revision=revision or commit_hash,
@@ -325,7 +313,7 @@ def _get_model_file(
                 proxies=proxies,
                 resume_download=resume_download,
                 local_files_only=local_files_only,
-                use_auth_token=use_auth_token,
+                token=token,
                 user_agent=user_agent,
                 subfolder=subfolder,
                 revision=revision or commit_hash,
@@ -336,7 +324,7 @@ def _get_model_file(
             raise EnvironmentError(
                 f"{pretrained_model_name_or_path} is not a local folder and is not a valid model identifier "
                 "listed on 'https://huggingface.co/models'\nIf this is a private repository, make sure to pass a "
-                "token having permission to this repo with `use_auth_token` or log in with `huggingface-cli "
+                "token having permission to this repo with `token` or log in with `huggingface-cli "
                 "login`."
             )
         except RevisionNotFoundError:

src/diffusers/utils/testing_utils.py

@@ -17,7 +17,7 @@ from contextlib import contextmanager
 from distutils.util import strtobool
 from io import BytesIO, StringIO
 from pathlib import Path
-from typing import List, Optional, Union
+from typing import Callable, Dict, List, Optional, Union
 
 import numpy as np
 import PIL.Image
@@ -58,6 +58,17 @@ USE_PEFT_BACKEND = _required_peft_version and _required_transformers_version
 if is_torch_available():
     import torch
 
+    # Set a backend environment variable for any extra module import required for a custom accelerator
+    if "DIFFUSERS_TEST_BACKEND" in os.environ:
+        backend = os.environ["DIFFUSERS_TEST_BACKEND"]
+        try:
+            _ = importlib.import_module(backend)
+        except ModuleNotFoundError as e:
+            raise ModuleNotFoundError(
+                f"Failed to import `DIFFUSERS_TEST_BACKEND` '{backend}'! This should be the name of an installed module \
+                    to enable a specified backend.):\n{e}"
+            ) from e
+
     if "DIFFUSERS_TEST_DEVICE" in os.environ:
         torch_device = os.environ["DIFFUSERS_TEST_DEVICE"]
         try:
@@ -210,6 +221,36 @@ def require_torch_gpu(test_case):
     )
 
 
+# These decorators are for accelerator-specific behaviours that are not GPU-specific
+def require_torch_accelerator(test_case):
+    """Decorator marking a test that requires an accelerator backend and PyTorch."""
+    return unittest.skipUnless(is_torch_available() and torch_device != "cpu", "test requires accelerator+PyTorch")(
+        test_case
+    )
+
+
+def require_torch_accelerator_with_fp16(test_case):
+    """Decorator marking a test that requires an accelerator with support for the FP16 data type."""
+    return unittest.skipUnless(_is_torch_fp16_available(torch_device), "test requires accelerator with fp16 support")(
+        test_case
+    )
+
+
+def require_torch_accelerator_with_fp64(test_case):
+    """Decorator marking a test that requires an accelerator with support for the FP64 data type."""
+    return unittest.skipUnless(_is_torch_fp64_available(torch_device), "test requires accelerator with fp64 support")(
+        test_case
+    )
+
+
+def require_torch_accelerator_with_training(test_case):
+    """Decorator marking a test that requires an accelerator with support for training."""
+    return unittest.skipUnless(
+        is_torch_available() and backend_supports_training(torch_device),
+        "test requires accelerator with training support",
+    )(test_case)
+
+
 def skip_mps(test_case):
     """Decorator marking a test to skip if torch_device is 'mps'"""
     return unittest.skipUnless(torch_device != "mps", "test requires non 'mps' device")(test_case)
@@ -766,3 +807,139 @@ def disable_full_determinism():
     os.environ["CUDA_LAUNCH_BLOCKING"] = "0"
     os.environ["CUBLAS_WORKSPACE_CONFIG"] = ""
     torch.use_deterministic_algorithms(False)
+
+
+# Utils for custom and alternative accelerator devices
+def _is_torch_fp16_available(device):
+    if not is_torch_available():
+        return False
+
+    import torch
+
+    device = torch.device(device)
+
+    try:
+        x = torch.zeros((2, 2), dtype=torch.float16).to(device)
+        _ = x @ x
+    except Exception as e:
+        if device.type == "cuda":
+            raise ValueError(
+                f"You have passed a device of type 'cuda' which should work with 'fp16', but 'cuda' does not seem to be correctly installed on your machine: {e}"
+            )
+
+        return False
+
+
+def _is_torch_fp64_available(device):
+    if not is_torch_available():
+        return False
+
+    import torch
+
+    try:
+        x = torch.zeros((2, 2), dtype=torch.float64).to(device)
+        _ = x @ x
+    except Exception as e:
+        if device.type == "cuda":
+            raise ValueError(
+                f"You have passed a device of type 'cuda' which should work with 'fp64', but 'cuda' does not seem to be correctly installed on your machine: {e}"
+            )
+
+        return False
+
+
+# Guard these lookups for when Torch is not used - alternative accelerator support is for PyTorch
+if is_torch_available():
+    # Behaviour flags
+    BACKEND_SUPPORTS_TRAINING = {"cuda": True, "cpu": True, "mps": False, "default": True}
+
+    # Function definitions
+    BACKEND_EMPTY_CACHE = {"cuda": torch.cuda.empty_cache, "cpu": None, "mps": None, "default": None}
+    BACKEND_DEVICE_COUNT = {"cuda": torch.cuda.device_count, "cpu": lambda: 0, "mps": lambda: 0, "default": 0}
+    BACKEND_MANUAL_SEED = {"cuda": torch.cuda.manual_seed, "cpu": torch.manual_seed, "default": torch.manual_seed}
+
+
+# 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 fn is None:
+        return None
+
+    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_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)
+
+
+# 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]
+
+
+# Guard for when Torch is not available
+if is_torch_available():
+    # Update device function dict mapping
+    def update_mapping_from_spec(device_fn_dict: Dict[str, Callable], attribute_name: str):
+        try:
+            # Try to import the function directly
+            spec_fn = getattr(device_spec_module, attribute_name)
+            device_fn_dict[torch_device] = spec_fn
+        except AttributeError as e:
+            # If the function doesn't exist, and there is no default, throw an error
+            if "default" not in device_fn_dict:
+                raise AttributeError(
+                    f"`{attribute_name}` not found in '{device_spec_path}' and no default fallback function found."
+                ) from e
+
+    if "DIFFUSERS_TEST_DEVICE_SPEC" in os.environ:
+        device_spec_path = os.environ["DIFFUSERS_TEST_DEVICE_SPEC"]
+        if not Path(device_spec_path).is_file():
+            raise ValueError(f"Specified path to device specification file is not found. Received {device_spec_path}")
+
+        try:
+            import_name = device_spec_path[: device_spec_path.index(".py")]
+        except ValueError as e:
+            raise ValueError(f"Provided device spec file is not a Python file! Received {device_spec_path}") from e
+
+        device_spec_module = importlib.import_module(import_name)
+
+        try:
+            device_name = device_spec_module.DEVICE_NAME
+        except AttributeError:
+            raise AttributeError("Device spec file did not contain `DEVICE_NAME`")
+
+        if "DIFFUSERS_TEST_DEVICE" in os.environ and torch_device != device_name:
+            msg = f"Mismatch between environment variable `DIFFUSERS_TEST_DEVICE` '{torch_device}' and device found in spec '{device_name}'\n"
+            msg += "Either unset `DIFFUSERS_TEST_DEVICE` or ensure it matches device spec name."
+            raise ValueError(msg)
+
+        torch_device = device_name
+
+        # Add one entry here for each `BACKEND_*` dictionary.
+        update_mapping_from_spec(BACKEND_MANUAL_SEED, "MANUAL_SEED_FN")
+        update_mapping_from_spec(BACKEND_EMPTY_CACHE, "EMPTY_CACHE_FN")
+        update_mapping_from_spec(BACKEND_DEVICE_COUNT, "DEVICE_COUNT_FN")
+        update_mapping_from_spec(BACKEND_SUPPORTS_TRAINING, "SUPPORTS_TRAINING")

tests/models/test_layers_utils.py

@@ -25,7 +25,11 @@ from diffusers.models.embeddings import get_timestep_embedding
 from diffusers.models.lora import LoRACompatibleLinear
 from diffusers.models.resnet import Downsample2D, ResnetBlock2D, Upsample2D
 from diffusers.models.transformer_2d import Transformer2DModel
-from diffusers.utils.testing_utils import torch_device
+from diffusers.utils.testing_utils import (
+    backend_manual_seed,
+    require_torch_accelerator_with_fp64,
+    torch_device,
+)
 
 
 class EmbeddingsTests(unittest.TestCase):
@@ -315,8 +319,7 @@ class ResnetBlock2DTests(unittest.TestCase):
 class Transformer2DModelTests(unittest.TestCase):
     def test_spatial_transformer_default(self):
         torch.manual_seed(0)
-        if torch.cuda.is_available():
-            torch.cuda.manual_seed_all(0)
+        backend_manual_seed(torch_device, 0)
 
         sample = torch.randn(1, 32, 64, 64).to(torch_device)
         spatial_transformer_block = Transformer2DModel(
@@ -339,8 +342,7 @@ class Transformer2DModelTests(unittest.TestCase):
 
     def test_spatial_transformer_cross_attention_dim(self):
         torch.manual_seed(0)
-        if torch.cuda.is_available():
-            torch.cuda.manual_seed_all(0)
+        backend_manual_seed(torch_device, 0)
 
         sample = torch.randn(1, 64, 64, 64).to(torch_device)
         spatial_transformer_block = Transformer2DModel(
@@ -363,8 +365,7 @@ class Transformer2DModelTests(unittest.TestCase):
 
     def test_spatial_transformer_timestep(self):
         torch.manual_seed(0)
-        if torch.cuda.is_available():
-            torch.cuda.manual_seed_all(0)
+        backend_manual_seed(torch_device, 0)
 
         num_embeds_ada_norm = 5
 
@@ -401,8 +402,7 @@ class Transformer2DModelTests(unittest.TestCase):
 
     def test_spatial_transformer_dropout(self):
         torch.manual_seed(0)
-        if torch.cuda.is_available():
-            torch.cuda.manual_seed_all(0)
+        backend_manual_seed(torch_device, 0)
 
         sample = torch.randn(1, 32, 64, 64).to(torch_device)
         spatial_transformer_block = (
@@ -427,11 +427,10 @@ class Transformer2DModelTests(unittest.TestCase):
         )
         assert torch.allclose(output_slice.flatten(), expected_slice, atol=1e-3)
 
-    @unittest.skipIf(torch_device == "mps", "MPS does not support float64")
+    @require_torch_accelerator_with_fp64
     def test_spatial_transformer_discrete(self):
         torch.manual_seed(0)
-        if torch.cuda.is_available():
-            torch.cuda.manual_seed_all(0)
+        backend_manual_seed(torch_device, 0)
 
         num_embed = 5
 

tests/models/test_modeling_common.py

@@ -35,6 +35,7 @@ from diffusers.utils.testing_utils import (
     CaptureLogger,
     require_python39_or_higher,
     require_torch_2,
+    require_torch_accelerator_with_training,
     require_torch_gpu,
     run_test_in_subprocess,
     torch_device,
@@ -536,7 +537,7 @@ class ModelTesterMixin:
 
         self.assertEqual(output_1.shape, output_2.shape)
 
-    @unittest.skipIf(torch_device == "mps", "Training is not supported in mps")
+    @require_torch_accelerator_with_training
     def test_training(self):
         init_dict, inputs_dict = self.prepare_init_args_and_inputs_for_common()
 
@@ -553,7 +554,7 @@ class ModelTesterMixin:
         loss = torch.nn.functional.mse_loss(output, noise)
         loss.backward()
 
-    @unittest.skipIf(torch_device == "mps", "Training is not supported in mps")
+    @require_torch_accelerator_with_training
     def test_ema_training(self):
         init_dict, inputs_dict = self.prepare_init_args_and_inputs_for_common()
 
@@ -624,7 +625,7 @@ class ModelTesterMixin:
 
         recursive_check(outputs_tuple, outputs_dict)
 
-    @unittest.skipIf(torch_device == "mps", "Gradient checkpointing skipped on MPS")
+    @require_torch_accelerator_with_training
     def test_enable_disable_gradient_checkpointing(self):
         if not self.model_class._supports_gradient_checkpointing:
             return  # Skip test if model does not support gradient checkpointing

tests/models/test_models_prior.py

@@ -21,7 +21,14 @@ import torch
 from parameterized import parameterized
 
 from diffusers import PriorTransformer
-from diffusers.utils.testing_utils import enable_full_determinism, floats_tensor, slow, torch_all_close, torch_device
+from diffusers.utils.testing_utils import (
+    backend_empty_cache,
+    enable_full_determinism,
+    floats_tensor,
+    slow,
+    torch_all_close,
+    torch_device,
+)
 
 from .test_modeling_common import ModelTesterMixin
 
@@ -157,7 +164,7 @@ class PriorTransformerIntegrationTests(unittest.TestCase):
         # clean up the VRAM after each test
         super().tearDown()
         gc.collect()
-        torch.cuda.empty_cache()
+        backend_empty_cache()
 
     @parameterized.expand(
         [

tests/models/test_models_unet_1d.py

@@ -18,7 +18,12 @@ import unittest
 import torch
 
 from diffusers import UNet1DModel
-from diffusers.utils.testing_utils import floats_tensor, slow, torch_device
+from diffusers.utils.testing_utils import (
+    backend_manual_seed,
+    floats_tensor,
+    slow,
+    torch_device,
+)
 
 from .test_modeling_common import ModelTesterMixin, UNetTesterMixin
 
@@ -103,8 +108,7 @@ class UNet1DModelTests(ModelTesterMixin, UNetTesterMixin, unittest.TestCase):
     def test_output_pretrained(self):
         model = UNet1DModel.from_pretrained("bglick13/hopper-medium-v2-value-function-hor32", subfolder="unet")
         torch.manual_seed(0)
-        if torch.cuda.is_available():
-            torch.cuda.manual_seed_all(0)
+        backend_manual_seed(torch_device, 0)
 
         num_features = model.config.in_channels
         seq_len = 16
@@ -244,8 +248,7 @@ class UNetRLModelTests(ModelTesterMixin, UNetTesterMixin, unittest.TestCase):
             "bglick13/hopper-medium-v2-value-function-hor32", output_loading_info=True, subfolder="value_function"
         )
         torch.manual_seed(0)
-        if torch.cuda.is_available():
-            torch.cuda.manual_seed_all(0)
+        backend_manual_seed(torch_device, 0)
 
         num_features = value_function.config.in_channels
         seq_len = 14

tests/models/test_models_unet_2d.py

@@ -24,6 +24,7 @@ from diffusers.utils import logging
 from diffusers.utils.testing_utils import (
     enable_full_determinism,
     floats_tensor,
+    require_torch_accelerator,
     slow,
     torch_all_close,
     torch_device,
@@ -153,7 +154,7 @@ class UNetLDMModelTests(ModelTesterMixin, UNetTesterMixin, unittest.TestCase):
 
         assert image is not None, "Make sure output is not None"
 
-    @unittest.skipIf(torch_device != "cuda", "This test is supposed to run on GPU")
+    @require_torch_accelerator
     def test_from_pretrained_accelerate(self):
         model, _ = UNet2DModel.from_pretrained("fusing/unet-ldm-dummy-update", output_loading_info=True)
         model.to(torch_device)
@@ -161,7 +162,7 @@ class UNetLDMModelTests(ModelTesterMixin, UNetTesterMixin, unittest.TestCase):
 
         assert image is not None, "Make sure output is not None"
 
-    @unittest.skipIf(torch_device != "cuda", "This test is supposed to run on GPU")
+    @require_torch_accelerator
     def test_from_pretrained_accelerate_wont_change_results(self):
         # by defautl model loading will use accelerate as `low_cpu_mem_usage=True`
         model_accelerate, _ = UNet2DModel.from_pretrained("fusing/unet-ldm-dummy-update", output_loading_info=True)

tests/models/test_models_unet_2d_condition.py

@@ -30,10 +30,15 @@ from diffusers.models.embeddings import ImageProjection, Resampler
 from diffusers.utils import logging
 from diffusers.utils.import_utils import is_xformers_available
 from diffusers.utils.testing_utils import (
+    backend_empty_cache,
     enable_full_determinism,
     floats_tensor,
     load_hf_numpy,
+    require_torch_accelerator,
+    require_torch_accelerator_with_fp16,
+    require_torch_accelerator_with_training,
     require_torch_gpu,
+    skip_mps,
     slow,
     torch_all_close,
     torch_device,
@@ -280,7 +285,7 @@ class UNet2DConditionModelTests(ModelTesterMixin, UNetTesterMixin, unittest.Test
             == "XFormersAttnProcessor"
         ), "xformers is not enabled"
 
-    @unittest.skipIf(torch_device == "mps", "Gradient checkpointing skipped on MPS")
+    @require_torch_accelerator_with_training
     def test_gradient_checkpointing(self):
         # enable deterministic behavior for gradient checkpointing
         init_dict, inputs_dict = self.prepare_init_args_and_inputs_for_common()
@@ -864,7 +869,7 @@ class UNet2DConditionModelIntegrationTests(unittest.TestCase):
         # clean up the VRAM after each test
         super().tearDown()
         gc.collect()
-        torch.cuda.empty_cache()
+        backend_empty_cache()
 
     def get_latents(self, seed=0, shape=(4, 4, 64, 64), fp16=False):
         dtype = torch.float16 if fp16 else torch.float32
@@ -882,6 +887,7 @@ class UNet2DConditionModelIntegrationTests(unittest.TestCase):
 
         return model
 
+    @require_torch_gpu
     def test_set_attention_slice_auto(self):
         torch.cuda.empty_cache()
         torch.cuda.reset_max_memory_allocated()
@@ -901,6 +907,7 @@ class UNet2DConditionModelIntegrationTests(unittest.TestCase):
 
         assert mem_bytes < 5 * 10**9
 
+    @require_torch_gpu
     def test_set_attention_slice_max(self):
         torch.cuda.empty_cache()
         torch.cuda.reset_max_memory_allocated()
@@ -920,6 +927,7 @@ class UNet2DConditionModelIntegrationTests(unittest.TestCase):
 
         assert mem_bytes < 5 * 10**9
 
+    @require_torch_gpu
     def test_set_attention_slice_int(self):
         torch.cuda.empty_cache()
         torch.cuda.reset_max_memory_allocated()
@@ -939,6 +947,7 @@ class UNet2DConditionModelIntegrationTests(unittest.TestCase):
 
         assert mem_bytes < 5 * 10**9
 
+    @require_torch_gpu
     def test_set_attention_slice_list(self):
         torch.cuda.empty_cache()
         torch.cuda.reset_max_memory_allocated()
@@ -975,7 +984,7 @@ class UNet2DConditionModelIntegrationTests(unittest.TestCase):
             # fmt: on
         ]
     )
-    @require_torch_gpu
+    @require_torch_accelerator_with_fp16
     def test_compvis_sd_v1_4(self, seed, timestep, expected_slice):
         model = self.get_unet_model(model_id="CompVis/stable-diffusion-v1-4")
         latents = self.get_latents(seed)
@@ -1003,7 +1012,7 @@ class UNet2DConditionModelIntegrationTests(unittest.TestCase):
             # fmt: on
         ]
     )
-    @require_torch_gpu
+    @require_torch_accelerator_with_fp16
     def test_compvis_sd_v1_4_fp16(self, seed, timestep, expected_slice):
         model = self.get_unet_model(model_id="CompVis/stable-diffusion-v1-4", fp16=True)
         latents = self.get_latents(seed, fp16=True)
@@ -1031,7 +1040,8 @@ class UNet2DConditionModelIntegrationTests(unittest.TestCase):
             # fmt: on
         ]
     )
-    @require_torch_gpu
+    @require_torch_accelerator
+    @skip_mps
     def test_compvis_sd_v1_5(self, seed, timestep, expected_slice):
         model = self.get_unet_model(model_id="runwayml/stable-diffusion-v1-5")
         latents = self.get_latents(seed)
@@ -1059,7 +1069,7 @@ class UNet2DConditionModelIntegrationTests(unittest.TestCase):
             # fmt: on
         ]
     )
-    @require_torch_gpu
+    @require_torch_accelerator_with_fp16
     def test_compvis_sd_v1_5_fp16(self, seed, timestep, expected_slice):
         model = self.get_unet_model(model_id="runwayml/stable-diffusion-v1-5", fp16=True)
         latents = self.get_latents(seed, fp16=True)
@@ -1087,7 +1097,8 @@ class UNet2DConditionModelIntegrationTests(unittest.TestCase):
             # fmt: on
         ]
     )
-    @require_torch_gpu
+    @require_torch_accelerator
+    @skip_mps
     def test_compvis_sd_inpaint(self, seed, timestep, expected_slice):
         model = self.get_unet_model(model_id="runwayml/stable-diffusion-inpainting")
         latents = self.get_latents(seed, shape=(4, 9, 64, 64))
@@ -1115,7 +1126,7 @@ class UNet2DConditionModelIntegrationTests(unittest.TestCase):
             # fmt: on
         ]
     )
-    @require_torch_gpu
+    @require_torch_accelerator_with_fp16
     def test_compvis_sd_inpaint_fp16(self, seed, timestep, expected_slice):
         model = self.get_unet_model(model_id="runwayml/stable-diffusion-inpainting", fp16=True)
         latents = self.get_latents(seed, shape=(4, 9, 64, 64), fp16=True)
@@ -1143,7 +1154,7 @@ class UNet2DConditionModelIntegrationTests(unittest.TestCase):
             # fmt: on
         ]
     )
-    @require_torch_gpu
+    @require_torch_accelerator_with_fp16
     def test_stabilityai_sd_v2_fp16(self, seed, timestep, expected_slice):
         model = self.get_unet_model(model_id="stabilityai/stable-diffusion-2", fp16=True)
         latents = self.get_latents(seed, shape=(4, 4, 96, 96), fp16=True)

tests/models/test_models_vae.py

@@ -31,10 +31,15 @@ from diffusers import (
 from diffusers.utils.import_utils import is_xformers_available
 from diffusers.utils.loading_utils import load_image
 from diffusers.utils.testing_utils import (
+    backend_empty_cache,
     enable_full_determinism,
     floats_tensor,
     load_hf_numpy,
+    require_torch_accelerator,
+    require_torch_accelerator_with_fp16,
+    require_torch_accelerator_with_training,
     require_torch_gpu,
+    skip_mps,
     slow,
     torch_all_close,
     torch_device,
@@ -157,7 +162,7 @@ class AutoencoderKLTests(ModelTesterMixin, UNetTesterMixin, unittest.TestCase):
     def test_training(self):
         pass
 
-    @unittest.skipIf(torch_device == "mps", "Gradient checkpointing skipped on MPS")
+    @require_torch_accelerator_with_training
     def test_gradient_checkpointing(self):
         # enable deterministic behavior for gradient checkpointing
         init_dict, inputs_dict = self.prepare_init_args_and_inputs_for_common()
@@ -213,10 +218,12 @@ class AutoencoderKLTests(ModelTesterMixin, UNetTesterMixin, unittest.TestCase):
         model = model.to(torch_device)
         model.eval()
 
-        if torch_device == "mps":
-            generator = torch.manual_seed(0)
+        # Keep generator on CPU for non-CUDA devices to compare outputs with CPU result tensors
+        generator_device = "cpu" if not torch_device.startswith("cuda") else "cuda"
+        if torch_device != "mps":
+            generator = torch.Generator(device=generator_device).manual_seed(0)
         else:
-            generator = torch.Generator(device=torch_device).manual_seed(0)
+            generator = torch.manual_seed(0)
 
         image = torch.randn(
             1,
@@ -247,7 +254,7 @@ class AutoencoderKLTests(ModelTesterMixin, UNetTesterMixin, unittest.TestCase):
                     -9.8644e-03,
                 ]
             )
-        elif torch_device == "cpu":
+        elif generator_device == "cpu":
             expected_output_slice = torch.tensor(
                 [
                     -0.1352,
@@ -478,7 +485,7 @@ class AutoencoderTinyIntegrationTests(unittest.TestCase):
         # clean up the VRAM after each test
         super().tearDown()
         gc.collect()
-        torch.cuda.empty_cache()
+        backend_empty_cache()
 
     def get_file_format(self, seed, shape):
         return f"gaussian_noise_s={seed}_shape={'_'.join([str(s) for s in shape])}.npy"
@@ -558,7 +565,7 @@ class AutoencoderKLIntegrationTests(unittest.TestCase):
         # clean up the VRAM after each test
         super().tearDown()
         gc.collect()
-        torch.cuda.empty_cache()
+        backend_empty_cache()
 
     def get_sd_image(self, seed=0, shape=(4, 3, 512, 512), fp16=False):
         dtype = torch.float16 if fp16 else torch.float32
@@ -580,9 +587,10 @@ class AutoencoderKLIntegrationTests(unittest.TestCase):
         return model
 
     def get_generator(self, seed=0):
-        if torch_device == "mps":
-            return torch.manual_seed(seed)
-        return torch.Generator(device=torch_device).manual_seed(seed)
+        generator_device = "cpu" if not torch_device.startswith("cuda") else "cuda"
+        if torch_device != "mps":
+            return torch.Generator(device=generator_device).manual_seed(seed)
+        return torch.manual_seed(seed)
 
     @parameterized.expand(
         [
@@ -623,7 +631,7 @@ class AutoencoderKLIntegrationTests(unittest.TestCase):
             # fmt: on
         ]
     )
-    @require_torch_gpu
+    @require_torch_accelerator_with_fp16
     def test_stable_diffusion_fp16(self, seed, expected_slice):
         model = self.get_sd_vae_model(fp16=True)
         image = self.get_sd_image(seed, fp16=True)
@@ -677,7 +685,8 @@ class AutoencoderKLIntegrationTests(unittest.TestCase):
             # fmt: on
         ]
     )
-    @require_torch_gpu
+    @require_torch_accelerator
+    @skip_mps
     def test_stable_diffusion_decode(self, seed, expected_slice):
         model = self.get_sd_vae_model()
         encoding = self.get_sd_image(seed, shape=(3, 4, 64, 64))
@@ -700,7 +709,7 @@ class AutoencoderKLIntegrationTests(unittest.TestCase):
             # fmt: on
         ]
     )
-    @require_torch_gpu
+    @require_torch_accelerator_with_fp16
     def test_stable_diffusion_decode_fp16(self, seed, expected_slice):
         model = self.get_sd_vae_model(fp16=True)
         encoding = self.get_sd_image(seed, shape=(3, 4, 64, 64), fp16=True)
@@ -811,7 +820,7 @@ class AsymmetricAutoencoderKLIntegrationTests(unittest.TestCase):
         # clean up the VRAM after each test
         super().tearDown()
         gc.collect()
-        torch.cuda.empty_cache()
+        backend_empty_cache()
 
     def get_sd_image(self, seed=0, shape=(4, 3, 512, 512), fp16=False):
         dtype = torch.float16 if fp16 else torch.float32
@@ -832,9 +841,10 @@ class AsymmetricAutoencoderKLIntegrationTests(unittest.TestCase):
         return model
 
     def get_generator(self, seed=0):
-        if torch_device == "mps":
-            return torch.manual_seed(seed)
-        return torch.Generator(device=torch_device).manual_seed(seed)
+        generator_device = "cpu" if not torch_device.startswith("cuda") else "cuda"
+        if torch_device != "mps":
+            return torch.Generator(device=generator_device).manual_seed(seed)
+        return torch.manual_seed(seed)
 
     @parameterized.expand(
         [
@@ -905,7 +915,8 @@ class AsymmetricAutoencoderKLIntegrationTests(unittest.TestCase):
             # fmt: on
         ]
     )
-    @require_torch_gpu
+    @require_torch_accelerator
+    @skip_mps
     def test_stable_diffusion_decode(self, seed, expected_slice):
         model = self.get_sd_vae_model()
         encoding = self.get_sd_image(seed, shape=(3, 4, 64, 64))

tests/models/test_models_vq.py

@@ -18,7 +18,12 @@ import unittest
 import torch
 
 from diffusers import VQModel
-from diffusers.utils.testing_utils import enable_full_determinism, floats_tensor, torch_device
+from diffusers.utils.testing_utils import (
+    backend_manual_seed,
+    enable_full_determinism,
+    floats_tensor,
+    torch_device,
+)
 
 from .test_modeling_common import ModelTesterMixin, UNetTesterMixin
 
@@ -80,8 +85,7 @@ class VQModelTests(ModelTesterMixin, UNetTesterMixin, unittest.TestCase):
         model.to(torch_device).eval()
 
         torch.manual_seed(0)
-        if torch.cuda.is_available():
-            torch.cuda.manual_seed_all(0)
+        backend_manual_seed(torch_device, 0)
 
         image = torch.randn(1, model.config.in_channels, model.config.sample_size, model.config.sample_size)
         image = image.to(torch_device)

tests/models/test_unet_blocks_common.py

@@ -12,12 +12,17 @@
 # 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.
-import unittest
 from typing import Tuple
 
 import torch
 
-from diffusers.utils.testing_utils import floats_tensor, require_torch, torch_all_close, torch_device
+from diffusers.utils.testing_utils import (
+    floats_tensor,
+    require_torch,
+    require_torch_accelerator_with_training,
+    torch_all_close,
+    torch_device,
+)
 from diffusers.utils.torch_utils import randn_tensor
 
 
@@ -104,7 +109,7 @@ class UNetBlockTesterMixin:
         expected_slice = torch.tensor(expected_slice).to(torch_device)
         assert torch_all_close(output_slice.flatten(), expected_slice, atol=5e-3)
 
-    @unittest.skipIf(torch_device == "mps", "Training is not supported in mps")
+    @require_torch_accelerator_with_training
     def test_training(self):
         init_dict, inputs_dict = self.prepare_init_args_and_inputs_for_common()
         model = self.block_class(**init_dict)

tests/pipelines/stable_diffusion_2/test_stable_diffusion.py

@@ -34,11 +34,14 @@ from diffusers import (
 )
 from diffusers.utils.testing_utils import (
     CaptureLogger,
+    backend_empty_cache,
     enable_full_determinism,
     load_numpy,
     nightly,
     numpy_cosine_similarity_distance,
+    require_torch_accelerator,
     require_torch_gpu,
+    skip_mps,
     slow,
     torch_device,
 )
@@ -128,10 +131,12 @@ class StableDiffusion2PipelineFastTests(
         return components
 
     def get_dummy_inputs(self, device, seed=0):
-        if str(device).startswith("mps"):
-            generator = torch.manual_seed(seed)
+        generator_device = "cpu" if not device.startswith("cuda") else "cuda"
+        if not str(device).startswith("mps"):
+            generator = torch.Generator(device=generator_device).manual_seed(seed)
         else:
-            generator = torch.Generator(device=device).manual_seed(seed)
+            generator = torch.manual_seed(seed)
+
         inputs = {
             "prompt": "A painting of a squirrel eating a burger",
             "generator": generator,
@@ -299,15 +304,21 @@ class StableDiffusion2PipelineFastTests(
 
 
 @slow
-@require_torch_gpu
+@require_torch_accelerator
+@skip_mps
 class StableDiffusion2PipelineSlowTests(unittest.TestCase):
     def tearDown(self):
         super().tearDown()
         gc.collect()
-        torch.cuda.empty_cache()
+        backend_empty_cache()
 
     def get_inputs(self, device, generator_device="cpu", dtype=torch.float32, seed=0):
-        generator = torch.Generator(device=generator_device).manual_seed(seed)
+        _generator_device = "cpu" if not generator_device.startswith("cuda") else "cuda"
+        if not str(device).startswith("mps"):
+            generator = torch.Generator(device=_generator_device).manual_seed(seed)
+        else:
+            generator = torch.manual_seed(seed)
+
         latents = np.random.RandomState(seed).standard_normal((1, 4, 64, 64))
         latents = torch.from_numpy(latents).to(device=device, dtype=dtype)
         inputs = {
@@ -361,6 +372,7 @@ class StableDiffusion2PipelineSlowTests(unittest.TestCase):
         expected_slice = np.array([0.10440, 0.13115, 0.11100, 0.10141, 0.11440, 0.07215, 0.11332, 0.09693, 0.10006])
         assert np.abs(image_slice - expected_slice).max() < 3e-3
 
+    @require_torch_gpu
     def test_stable_diffusion_attention_slicing(self):
         torch.cuda.reset_peak_memory_stats()
         pipe = StableDiffusionPipeline.from_pretrained(
@@ -432,6 +444,7 @@ class StableDiffusion2PipelineSlowTests(unittest.TestCase):
         assert callback_fn.has_been_called
         assert number_of_steps == inputs["num_inference_steps"]
 
+    @require_torch_gpu
     def test_stable_diffusion_pipeline_with_sequential_cpu_offloading(self):
         torch.cuda.empty_cache()
         torch.cuda.reset_max_memory_allocated()
@@ -452,6 +465,7 @@ class StableDiffusion2PipelineSlowTests(unittest.TestCase):
         # make sure that less than 2.8 GB is allocated
         assert mem_bytes < 2.8 * 10**9
 
+    @require_torch_gpu
     def test_stable_diffusion_pipeline_with_model_offloading(self):
         torch.cuda.empty_cache()
         torch.cuda.reset_max_memory_allocated()
@@ -511,15 +525,21 @@ class StableDiffusion2PipelineSlowTests(unittest.TestCase):
 
 
 @nightly
-@require_torch_gpu
+@require_torch_accelerator
+@skip_mps
 class StableDiffusion2PipelineNightlyTests(unittest.TestCase):
     def tearDown(self):
         super().tearDown()
         gc.collect()
-        torch.cuda.empty_cache()
+        backend_empty_cache()
 
     def get_inputs(self, device, generator_device="cpu", dtype=torch.float32, seed=0):
-        generator = torch.Generator(device=generator_device).manual_seed(seed)
+        _generator_device = "cpu" if not generator_device.startswith("cuda") else "cuda"
+        if not str(device).startswith("mps"):
+            generator = torch.Generator(device=_generator_device).manual_seed(seed)
+        else:
+            generator = torch.manual_seed(seed)
+
         latents = np.random.RandomState(seed).standard_normal((1, 4, 64, 64))
         latents = torch.from_numpy(latents).to(device=device, dtype=dtype)
         inputs = {