fix

.github/workflows/pr_test_fetcher.yml

@@ -1,6 +1,12 @@
 name: Fast tests for PRs - Test Fetcher
 
-on: workflow_dispatch
+on:
+  pull_request:
+    branches:
+      - main
+  push:
+    branches:
+      - ci-*
 
 env:
   DIFFUSERS_IS_CI: yes

docs/source/en/api/attnprocessor.md

@@ -20,9 +20,6 @@ An attention processor is a class for applying different types of attention mech
 ## AttnProcessor2_0
 [[autodoc]] models.attention_processor.AttnProcessor2_0
 
-## FusedAttnProcessor2_0
-[[autodoc]] models.attention_processor.FusedAttnProcessor2_0
-
 ## LoRAAttnProcessor
 [[autodoc]] models.attention_processor.LoRAAttnProcessor
 

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/advanced_diffusion_training/train_dreambooth_lora_sdxl_advanced.py

@@ -133,7 +133,7 @@ def save_model_card(
         diffusers_imports_pivotal = """from huggingface_hub import hf_hub_download
 from safetensors.torch import load_file
         """
-        diffusers_example_pivotal = f"""embedding_path = hf_hub_download(repo_id='{repo_id}', filename="embeddings.safetensors", repo_type="model")
+        diffusers_example_pivotal = f"""embedding_path = hf_hub_download(repo_id="{repo_id}", filename="embeddings.safetensors", repo_type="model")
 state_dict = load_file(embedding_path)
 pipeline.load_textual_inversion(state_dict["clip_l"], token=["<s0>", "<s1>"], text_encoder=pipe.text_encoder, tokenizer=pipe.tokenizer)
 pipeline.load_textual_inversion(state_dict["clip_g"], token=["<s0>", "<s1>"], text_encoder=pipe.text_encoder_2, tokenizer=pipe.tokenizer_2)
@@ -145,7 +145,8 @@ pipeline.load_textual_inversion(state_dict["clip_g"], token=["<s0>", "<s1>"], te
 to trigger concept `{key}` → use `{tokens}` in your prompt \n
 """
 
-    yaml = f"""---
+    yaml = f"""
+---
 tags:
 - stable-diffusion-xl
 - stable-diffusion-xl-diffusers
@@ -158,7 +159,7 @@ base_model: {base_model}
 instance_prompt: {instance_prompt}
 license: openrail++
 ---
-"""
+    """
 
     model_card = f"""
 # SDXL LoRA DreamBooth - {repo_id}
@@ -169,6 +170,14 @@ license: openrail++
 
 ### These are {repo_id} LoRA adaption weights for {base_model}.
 
+The weights were trained  using [DreamBooth](https://dreambooth.github.io/).
+
+LoRA for the text encoder was enabled: {train_text_encoder}.
+
+Pivotal tuning was enabled: {train_text_encoder_ti}.
+
+Special VAE used for training: {vae_path}.
+
 ## Trigger words
 
 {trigger_str}
@@ -187,24 +196,11 @@ image = pipeline('{validation_prompt if validation_prompt else instance_prompt}'
 
 For more details, including weighting, merging and fusing LoRAs, check the [documentation on loading LoRAs in diffusers](https://huggingface.co/docs/diffusers/main/en/using-diffusers/loading_adapters)
 
-## Download model
+## Download model (use it with UIs such as AUTO1111, Comfy, SD.Next, Invoke)
 
-### Use it with UIs such as AUTOMATIC1111, Comfy UI, SD.Next, Invoke
+Weights for this model are available in Safetensors format.
 
-- Download the LoRA *.safetensors [here](/{repo_id}/blob/main/pytorch_lora_weights.safetensors). Rename it and place it on your Lora folder.
-- Download the text embeddings *.safetensors [here](/{repo_id}/blob/main/embeddings.safetensors). Rename it and place it on it on your embeddings folder.
-
-All [Files & versions](/{repo_id}/tree/main).
-
-## Details
-
-The weights were trained using [🧨 diffusers Advanced Dreambooth Training Script](https://github.com/huggingface/diffusers/blob/main/examples/advanced_diffusion_training/train_dreambooth_lora_sdxl_advanced.py).
-
-LoRA for the text encoder was enabled. {train_text_encoder}.
-
-Pivotal tuning was enabled: {train_text_encoder_ti}.
-
-Special VAE used for training: {vae_path}.
+[Download]({repo_id}/tree/main) them in the Files & versions tab.
 
 """
     with open(os.path.join(repo_folder, "README.md"), "w") as f:
@@ -671,12 +667,6 @@ def parse_args(input_args=None):
         default=4,
         help=("The dimension of the LoRA update matrices."),
     )
-    parser.add_argument(
-        "--cache_latents",
-        action="store_true",
-        default=False,
-        help="Cache the VAE latents",
-    )
 
     if input_args is not None:
         args = parser.parse_args(input_args)
@@ -1180,7 +1170,6 @@ def main(args):
         revision=args.revision,
         variant=args.variant,
     )
-    vae_scaling_factor = vae.config.scaling_factor
     unet = UNet2DConditionModel.from_pretrained(
         args.pretrained_model_name_or_path, subfolder="unet", revision=args.revision, variant=args.variant
     )
@@ -1611,20 +1600,6 @@ def main(args):
             args.validation_prompt = args.validation_prompt.replace(token_abs, "".join(token_replacement))
     print("validation prompt:", args.validation_prompt)
 
-    if args.cache_latents:
-        latents_cache = []
-        for batch in tqdm(train_dataloader, desc="Caching latents"):
-            with torch.no_grad():
-                batch["pixel_values"] = batch["pixel_values"].to(
-                    accelerator.device, non_blocking=True, dtype=torch.float32
-                )
-                latents_cache.append(vae.encode(batch["pixel_values"]).latent_dist)
-
-        if args.validation_prompt is None:
-            del vae
-            if torch.cuda.is_available():
-                torch.cuda.empty_cache()
-
     # Scheduler and math around the number of training steps.
     overrode_max_train_steps = False
     num_update_steps_per_epoch = math.ceil(len(train_dataloader) / args.gradient_accumulation_steps)
@@ -1740,7 +1715,9 @@ def main(args):
         unet.train()
         for step, batch in enumerate(train_dataloader):
             with accelerator.accumulate(unet):
+                pixel_values = batch["pixel_values"].to(dtype=vae.dtype)
                 prompts = batch["prompts"]
+                # print(prompts)
                 # encode batch prompts when custom prompts are provided for each image -
                 if train_dataset.custom_instance_prompts:
                     if freeze_text_encoder:
@@ -1752,13 +1729,9 @@ def main(args):
                         tokens_one = tokenize_prompt(tokenizer_one, prompts, add_special_tokens)
                         tokens_two = tokenize_prompt(tokenizer_two, prompts, add_special_tokens)
 
-                if args.cache_latents:
-                    model_input = latents_cache[step].sample()
-                else:
-                    pixel_values = batch["pixel_values"].to(dtype=vae.dtype)
-                    model_input = vae.encode(pixel_values).latent_dist.sample()
-
-                model_input = model_input * vae_scaling_factor
+                # Convert images to latent space
+                model_input = vae.encode(pixel_values).latent_dist.sample()
+                model_input = model_input * vae.config.scaling_factor
                 if args.pretrained_vae_model_name_or_path is None:
                     model_input = model_input.to(weight_dtype)
 

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)
 

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/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/README.md

@@ -1,6 +1,6 @@
 # Latent Consistency Distillation Example:
 
-[Latent Consistency Models (LCMs)](https://arxiv.org/abs/2310.04378) is a method to distill a latent diffusion model to enable swift inference with minimal steps. This example demonstrates how to use latent consistency distillation to distill stable-diffusion-v1.5 for inference with few timesteps.
+[Latent Consistency Models (LCMs)](https://arxiv.org/abs/2310.04378) is method to distill latent diffusion model to enable swift inference with minimal steps. This example demonstrates how to use the latent consistency distillation to distill stable-diffusion-v1.5 for less timestep inference.
 
 ## Full model distillation
 
@@ -24,7 +24,7 @@ Then cd in the example folder and run
 pip install -r requirements.txt
 ```
 
-And initialize an [🤗 Accelerate](https://github.com/huggingface/accelerate/) environment with:
+And initialize an [🤗Accelerate](https://github.com/huggingface/accelerate/) environment with:
 
 ```bash
 accelerate config
@@ -46,16 +46,12 @@ write_basic_config()
 When running `accelerate config`, if we specify torch compile mode to True there can be dramatic speedups.
 
 
-#### Example
-
-The following uses the [Conceptual Captions 12M (CC12M) dataset](https://github.com/google-research-datasets/conceptual-12m) as an example, and for illustrative purposes only. For best results you may consider large and high-quality text-image datasets such as [LAION](https://laion.ai/blog/laion-400-open-dataset/). You may also need to search the hyperparameter space according to the dataset you use.
+#### Example with LAION-A6+ dataset
 
 ```bash
-export MODEL_NAME="runwayml/stable-diffusion-v1-5"
-export OUTPUT_DIR="path/to/saved/model"
-
-accelerate launch train_lcm_distill_sd_wds.py \
-    --pretrained_teacher_model=$MODEL_NAME \
+runwayml/stable-diffusion-v1-5
+PROGRAM="train_lcm_distill_sd_wds.py \
+    --pretrained_teacher_model=$MODEL_DIR \
     --output_dir=$OUTPUT_DIR \
     --mixed_precision=fp16 \
     --resolution=512 \
@@ -63,7 +59,7 @@ accelerate launch train_lcm_distill_sd_wds.py \
     --max_train_steps=1000 \
     --max_train_samples=4000000 \
     --dataloader_num_workers=8 \
-    --train_shards_path_or_url="pipe:curl -L -s https://huggingface.co/datasets/laion/conceptual-captions-12m-webdataset/resolve/main/data/{00000..01099}.tar?download=true" \
+    --train_shards_path_or_url='pipe:aws s3 cp s3://muse-datasets/laion-aesthetic6plus-min512-data/{00000..01210}.tar -' \
     --validation_steps=200 \
     --checkpointing_steps=200 --checkpoints_total_limit=10 \
     --train_batch_size=12 \
@@ -73,23 +69,19 @@ accelerate launch train_lcm_distill_sd_wds.py \
     --resume_from_checkpoint=latest \
     --report_to=wandb \
     --seed=453645634 \
-    --push_to_hub
+    --push_to_hub \
 ```
 
 ## LCM-LoRA
 
 Instead of fine-tuning the full model, we can also just train a LoRA that can be injected into any SDXL model.
 
-### Example
-
-The following uses the [Conceptual Captions 12M (CC12M) dataset](https://github.com/google-research-datasets/conceptual-12m) as an example. For best results you may consider large and high-quality text-image datasets such as [LAION](https://laion.ai/blog/laion-400-open-dataset/).
-
+### Example with LAION-A6+ dataset
+    
 ```bash
-export MODEL_NAME="runwayml/stable-diffusion-v1-5"
-export OUTPUT_DIR="path/to/saved/model"
-
-accelerate launch train_lcm_distill_lora_sd_wds.py \
-    --pretrained_teacher_model=$MODEL_NAME \
+runwayml/stable-diffusion-v1-5
+PROGRAM="train_lcm_distill_lora_sd_wds.py \
+    --pretrained_teacher_model=$MODEL_DIR \
     --output_dir=$OUTPUT_DIR \
     --mixed_precision=fp16 \
     --resolution=512 \
@@ -98,7 +90,7 @@ accelerate launch train_lcm_distill_lora_sd_wds.py \
     --max_train_steps=1000 \
     --max_train_samples=4000000 \
     --dataloader_num_workers=8 \
-    --train_shards_path_or_url="pipe:curl -L -s https://huggingface.co/datasets/laion/conceptual-captions-12m-webdataset/resolve/main/data/{00000..01099}.tar?download=true" \
+    --train_shards_path_or_url='pipe:aws s3 cp s3://muse-datasets/laion-aesthetic6plus-min512-data/{00000..01210}.tar -' \
     --validation_steps=200 \
     --checkpointing_steps=200 --checkpoints_total_limit=10 \
     --train_batch_size=12 \

examples/consistency_distillation/README_sdxl.md

@@ -1,6 +1,6 @@
 # Latent Consistency Distillation Example:
 
-[Latent Consistency Models (LCMs)](https://arxiv.org/abs/2310.04378) is a method to distill a latent diffusion model to enable swift inference with minimal steps. This example demonstrates how to use latent consistency distillation to distill SDXL for inference with few timesteps.
+[Latent Consistency Models (LCMs)](https://arxiv.org/abs/2310.04378) is method to distill latent diffusion model to enable swift inference with minimal steps. This example demonstrates how to use the latent consistency distillation to distill SDXL for less timestep inference.
 
 ## Full model distillation
 
@@ -24,7 +24,7 @@ Then cd in the example folder and run
 pip install -r requirements.txt
 ```
 
-And initialize an [🤗 Accelerate](https://github.com/huggingface/accelerate/) environment with:
+And initialize an [🤗Accelerate](https://github.com/huggingface/accelerate/) environment with:
 
 ```bash
 accelerate config
@@ -46,16 +46,12 @@ write_basic_config()
 When running `accelerate config`, if we specify torch compile mode to True there can be dramatic speedups.
 
 
-#### Example
-
-The following uses the [Conceptual Captions 12M (CC12M) dataset](https://github.com/google-research-datasets/conceptual-12m) as an example, and for illustrative purposes only. For best results you may consider large and high-quality text-image datasets such as [LAION](https://laion.ai/blog/laion-400-open-dataset/). You may also need to search the hyperparameter space according to the dataset you use.
+#### Example with LAION-A6+ dataset
 
 ```bash
-export MODEL_NAME="stabilityai/stable-diffusion-xl-base-1.0"
-export OUTPUT_DIR="path/to/saved/model"
-
-accelerate launch train_lcm_distill_sdxl_wds.py \
-    --pretrained_teacher_model=$MODEL_NAME \
+export MODEL_DIR="stabilityai/stable-diffusion-xl-base-1.0"
+PROGRAM="train_lcm_distill_sdxl_wds.py \
+    --pretrained_teacher_model=$MODEL_DIR \
     --pretrained_vae_model_name_or_path=madebyollin/sdxl-vae-fp16-fix \
     --output_dir=$OUTPUT_DIR \
     --mixed_precision=fp16 \
@@ -64,7 +60,7 @@ accelerate launch train_lcm_distill_sdxl_wds.py \
     --max_train_steps=1000 \
     --max_train_samples=4000000 \
     --dataloader_num_workers=8 \
-    --train_shards_path_or_url="pipe:curl -L -s https://huggingface.co/datasets/laion/conceptual-captions-12m-webdataset/resolve/main/data/{00000..01099}.tar?download=true" \
+    --train_shards_path_or_url='pipe:aws s3 cp s3://muse-datasets/laion-aesthetic6plus-min512-data/{00000..01210}.tar -' \
     --validation_steps=200 \
     --checkpointing_steps=200 --checkpoints_total_limit=10 \
     --train_batch_size=12 \
@@ -81,15 +77,11 @@ accelerate launch train_lcm_distill_sdxl_wds.py \
 
 Instead of fine-tuning the full model, we can also just train a LoRA that can be injected into any SDXL model.
 
-### Example
-
-The following uses the [Conceptual Captions 12M (CC12M) dataset](https://github.com/google-research-datasets/conceptual-12m) as an example. For best results you may consider large and high-quality text-image datasets such as [LAION](https://laion.ai/blog/laion-400-open-dataset/).
-
+### Example with LAION-A6+ dataset
+    
 ```bash
-export MODEL_NAME="stabilityai/stable-diffusion-xl-base-1.0"
-export OUTPUT_DIR="path/to/saved/model"
-
-accelerate launch train_lcm_distill_lora_sdxl_wds.py \
+export MODEL_DIR="stabilityai/stable-diffusion-xl-base-1.0"
+PROGRAM="train_lcm_distill_lora_sdxl_wds.py \
     --pretrained_teacher_model=$MODEL_DIR \
     --pretrained_vae_model_name_or_path=madebyollin/sdxl-vae-fp16-fix \
     --output_dir=$OUTPUT_DIR \
@@ -100,7 +92,7 @@ accelerate launch train_lcm_distill_lora_sdxl_wds.py \
     --max_train_steps=1000 \
     --max_train_samples=4000000 \
     --dataloader_num_workers=8 \
-    --train_shards_path_or_url="pipe:curl -L -s https://huggingface.co/datasets/laion/conceptual-captions-12m-webdataset/resolve/main/data/{00000..01099}.tar?download=true" \
+    --train_shards_path_or_url='pipe:aws s3 cp s3://muse-datasets/laion-aesthetic6plus-min512-data/{00000..01210}.tar -' \
     --validation_steps=200 \
     --checkpointing_steps=200 --checkpoints_total_limit=10 \
     --train_batch_size=12 \

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]
 
@@ -1123,7 +1123,7 @@ def main(args):
     for epoch in range(first_epoch, args.num_train_epochs):
         for step, batch in enumerate(train_dataloader):
             with accelerator.accumulate(unet):
-                image, text = batch
+                image, text, _, _ = batch
 
                 image = image.to(accelerator.device, non_blocking=True)
                 encoded_text = compute_embeddings_fn(text)

examples/consistency_distillation/train_lcm_distill_lora_sdxl_wds.py

@@ -68,11 +68,6 @@ from diffusers.utils.import_utils import is_xformers_available
 
 MAX_SEQ_LENGTH = 77
 
-# Adjust for your dataset
-WDS_JSON_WIDTH = "width"  # original_width for LAION
-WDS_JSON_HEIGHT = "height"  # original_height for LAION
-MIN_SIZE = 700  # ~960 for LAION, ideal: 1024 if the dataset contains large images
-
 if is_wandb_available():
     import wandb
 
@@ -151,10 +146,10 @@ class WebdatasetFilter:
         try:
             if "json" in x:
                 x_json = json.loads(x["json"])
-                filter_size = (x_json.get(WDS_JSON_WIDTH, 0.0) or 0.0) >= self.min_size and x_json.get(
-                    WDS_JSON_HEIGHT, 0
+                filter_size = (x_json.get("original_width", 0.0) or 0.0) >= self.min_size and x_json.get(
+                    "original_height", 0
                 ) >= self.min_size
-                filter_watermark = (x_json.get("pwatermark", 0.0) or 0.0) <= self.max_pwatermark
+                filter_watermark = (x_json.get("pwatermark", 1.0) or 1.0) <= self.max_pwatermark
                 return filter_size and filter_watermark
             else:
                 return False
@@ -185,7 +180,7 @@ class Text2ImageDataset:
             if use_fix_crop_and_size:
                 return (resolution, resolution)
             else:
-                return (int(json.get(WDS_JSON_WIDTH, 0.0)), int(json.get(WDS_JSON_HEIGHT, 0.0)))
+                return (int(json.get("original_width", 0.0)), int(json.get("original_height", 0.0)))
 
         def transform(example):
             # resize image
@@ -217,7 +212,7 @@ class Text2ImageDataset:
         pipeline = [
             wds.ResampledShards(train_shards_path_or_url),
             tarfile_to_samples_nothrow,
-            wds.select(WebdatasetFilter(min_size=MIN_SIZE)),
+            wds.select(WebdatasetFilter(min_size=960)),
             wds.shuffle(shuffle_buffer_size),
             *processing_pipeline,
             wds.batched(per_gpu_batch_size, partial=False, collation_fn=default_collate),
@@ -397,7 +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]
 
@@ -1106,7 +1106,7 @@ def main(args):
     for epoch in range(first_epoch, args.num_train_epochs):
         for step, batch in enumerate(train_dataloader):
             with accelerator.accumulate(unet):
-                image, text = batch
+                image, text, _, _ = batch
 
                 image = image.to(accelerator.device, non_blocking=True)
                 encoded_text = compute_embeddings_fn(text)

examples/consistency_distillation/train_lcm_distill_sdxl_wds.py

@@ -67,11 +67,6 @@ from diffusers.utils.import_utils import is_xformers_available
 
 MAX_SEQ_LENGTH = 77
 
-# Adjust for your dataset
-WDS_JSON_WIDTH = "width"  # original_width for LAION
-WDS_JSON_HEIGHT = "height"  # original_height for LAION
-MIN_SIZE = 700  # ~960 for LAION, ideal: 1024 if the dataset contains large images
-
 if is_wandb_available():
     import wandb
 
@@ -133,10 +128,10 @@ class WebdatasetFilter:
         try:
             if "json" in x:
                 x_json = json.loads(x["json"])
-                filter_size = (x_json.get(WDS_JSON_WIDTH, 0.0) or 0.0) >= self.min_size and x_json.get(
-                    WDS_JSON_HEIGHT, 0
+                filter_size = (x_json.get("original_width", 0.0) or 0.0) >= self.min_size and x_json.get(
+                    "original_height", 0
                 ) >= self.min_size
-                filter_watermark = (x_json.get("pwatermark", 0.0) or 0.0) <= self.max_pwatermark
+                filter_watermark = (x_json.get("pwatermark", 1.0) or 1.0) <= self.max_pwatermark
                 return filter_size and filter_watermark
             else:
                 return False
@@ -167,7 +162,7 @@ class Text2ImageDataset:
             if use_fix_crop_and_size:
                 return (resolution, resolution)
             else:
-                return (int(json.get(WDS_JSON_WIDTH, 0.0)), int(json.get(WDS_JSON_HEIGHT, 0.0)))
+                return (int(json.get("original_width", 0.0)), int(json.get("original_height", 0.0)))
 
         def transform(example):
             # resize image
@@ -199,7 +194,7 @@ class Text2ImageDataset:
         pipeline = [
             wds.ResampledShards(train_shards_path_or_url),
             tarfile_to_samples_nothrow,
-            wds.select(WebdatasetFilter(min_size=MIN_SIZE)),
+            wds.select(WebdatasetFilter(min_size=960)),
             wds.shuffle(shuffle_buffer_size),
             *processing_pipeline,
             wds.batched(per_gpu_batch_size, partial=False, collation_fn=default_collate),
@@ -419,7 +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,
             )
@@ -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/attention_processor.py

@@ -113,14 +113,12 @@ class Attention(nn.Module):
     ):
         super().__init__()
         self.inner_dim = out_dim if out_dim is not None else dim_head * heads
-        self.query_dim = query_dim
         self.cross_attention_dim = cross_attention_dim if cross_attention_dim is not None else query_dim
         self.upcast_attention = upcast_attention
         self.upcast_softmax = upcast_softmax
         self.rescale_output_factor = rescale_output_factor
         self.residual_connection = residual_connection
         self.dropout = dropout
-        self.fused_projections = False
         self.out_dim = out_dim if out_dim is not None else query_dim
 
         # we make use of this private variable to know whether this class is loaded
@@ -182,7 +180,6 @@ class Attention(nn.Module):
         else:
             linear_cls = LoRACompatibleLinear
 
-        self.linear_cls = linear_cls
         self.to_q = linear_cls(query_dim, self.inner_dim, bias=bias)
 
         if not self.only_cross_attention:
@@ -695,32 +692,6 @@ class Attention(nn.Module):
 
         return encoder_hidden_states
 
-    @torch.no_grad()
-    def fuse_projections(self, fuse=True):
-        is_cross_attention = self.cross_attention_dim != self.query_dim
-        device = self.to_q.weight.data.device
-        dtype = self.to_q.weight.data.dtype
-
-        if not is_cross_attention:
-            # fetch weight matrices.
-            concatenated_weights = torch.cat([self.to_q.weight.data, self.to_k.weight.data, self.to_v.weight.data])
-            in_features = concatenated_weights.shape[1]
-            out_features = concatenated_weights.shape[0]
-
-            # create a new single projection layer and copy over the weights.
-            self.to_qkv = self.linear_cls(in_features, out_features, bias=False, device=device, dtype=dtype)
-            self.to_qkv.weight.copy_(concatenated_weights)
-
-        else:
-            concatenated_weights = torch.cat([self.to_k.weight.data, self.to_v.weight.data])
-            in_features = concatenated_weights.shape[1]
-            out_features = concatenated_weights.shape[0]
-
-            self.to_kv = self.linear_cls(in_features, out_features, bias=False, device=device, dtype=dtype)
-            self.to_kv.weight.copy_(concatenated_weights)
-
-        self.fused_projections = fuse
-
 
 class AttnProcessor:
     r"""
@@ -1213,6 +1184,9 @@ class AttnProcessor2_0:
         scale: float = 1.0,
     ) -> torch.FloatTensor:
         residual = hidden_states
+
+        args = () if USE_PEFT_BACKEND else (scale,)
+
         if attn.spatial_norm is not None:
             hidden_states = attn.spatial_norm(hidden_states, temb)
 
@@ -1279,103 +1253,6 @@ class AttnProcessor2_0:
         return hidden_states
 
 
-class FusedAttnProcessor2_0:
-    r"""
-    Processor for implementing scaled dot-product attention (enabled by default if you're using PyTorch 2.0).
-    It uses fused projection layers. For self-attention modules, all projection matrices (i.e., query,
-    key, value) are fused. For cross-attention modules, key and value projection matrices are fused.
-
-    <Tip warning={true}>
-
-    This API is currently 🧪 experimental in nature and can change in future.
-
-    </Tip>
-    """
-
-    def __init__(self):
-        if not hasattr(F, "scaled_dot_product_attention"):
-            raise ImportError(
-                "FusedAttnProcessor2_0 requires at least PyTorch 2.0, to use it. Please upgrade PyTorch to > 2.0."
-            )
-
-    def __call__(
-        self,
-        attn: Attention,
-        hidden_states: torch.FloatTensor,
-        encoder_hidden_states: Optional[torch.FloatTensor] = None,
-        attention_mask: Optional[torch.FloatTensor] = None,
-        temb: Optional[torch.FloatTensor] = None,
-        scale: float = 1.0,
-    ) -> torch.FloatTensor:
-        residual = hidden_states
-        if attn.spatial_norm is not None:
-            hidden_states = attn.spatial_norm(hidden_states, temb)
-
-        input_ndim = hidden_states.ndim
-
-        if input_ndim == 4:
-            batch_size, channel, height, width = hidden_states.shape
-            hidden_states = hidden_states.view(batch_size, channel, height * width).transpose(1, 2)
-
-        batch_size, sequence_length, _ = (
-            hidden_states.shape if encoder_hidden_states is None else encoder_hidden_states.shape
-        )
-
-        if attention_mask is not None:
-            attention_mask = attn.prepare_attention_mask(attention_mask, sequence_length, batch_size)
-            # scaled_dot_product_attention expects attention_mask shape to be
-            # (batch, heads, source_length, target_length)
-            attention_mask = attention_mask.view(batch_size, attn.heads, -1, attention_mask.shape[-1])
-
-        if attn.group_norm is not None:
-            hidden_states = attn.group_norm(hidden_states.transpose(1, 2)).transpose(1, 2)
-
-        args = () if USE_PEFT_BACKEND else (scale,)
-        if encoder_hidden_states is None:
-            qkv = attn.to_qkv(hidden_states, *args)
-            split_size = qkv.shape[-1] // 3
-            query, key, value = torch.split(qkv, split_size, dim=-1)
-        else:
-            if attn.norm_cross:
-                encoder_hidden_states = attn.norm_encoder_hidden_states(encoder_hidden_states)
-            query = attn.to_q(hidden_states, *args)
-
-            kv = attn.to_kv(encoder_hidden_states, *args)
-            split_size = kv.shape[-1] // 2
-            key, value = torch.split(kv, split_size, dim=-1)
-
-        inner_dim = key.shape[-1]
-        head_dim = inner_dim // attn.heads
-
-        query = query.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
-        key = key.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
-        value = value.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
-
-        # the output of sdp = (batch, num_heads, seq_len, head_dim)
-        # TODO: add support for attn.scale when we move to Torch 2.1
-        hidden_states = F.scaled_dot_product_attention(
-            query, key, value, attn_mask=attention_mask, dropout_p=0.0, is_causal=False
-        )
-
-        hidden_states = hidden_states.transpose(1, 2).reshape(batch_size, -1, attn.heads * head_dim)
-        hidden_states = hidden_states.to(query.dtype)
-
-        # linear proj
-        hidden_states = attn.to_out[0](hidden_states, *args)
-        # dropout
-        hidden_states = attn.to_out[1](hidden_states)
-
-        if input_ndim == 4:
-            hidden_states = hidden_states.transpose(-1, -2).reshape(batch_size, channel, height, width)
-
-        if attn.residual_connection:
-            hidden_states = hidden_states + residual
-
-        hidden_states = hidden_states / attn.rescale_output_factor
-
-        return hidden_states
-
-
 class CustomDiffusionXFormersAttnProcessor(nn.Module):
     r"""
     Processor for implementing memory efficient attention using xFormers for the Custom Diffusion method.
@@ -2374,7 +2251,6 @@ CROSS_ATTENTION_PROCESSORS = (
 AttentionProcessor = Union[
     AttnProcessor,
     AttnProcessor2_0,
-    FusedAttnProcessor2_0,
     XFormersAttnProcessor,
     SlicedAttnProcessor,
     AttnAddedKVProcessor,

src/diffusers/models/autoencoder_kl.py

@@ -22,7 +22,6 @@ from ..utils.accelerate_utils import apply_forward_hook
 from .attention_processor import (
     ADDED_KV_ATTENTION_PROCESSORS,
     CROSS_ATTENTION_PROCESSORS,
-    Attention,
     AttentionProcessor,
     AttnAddedKVProcessor,
     AttnProcessor,
@@ -449,41 +448,3 @@ class AutoencoderKL(ModelMixin, ConfigMixin, FromOriginalVAEMixin):
             return (dec,)
 
         return DecoderOutput(sample=dec)
-
-    # Copied from diffusers.models.unet_2d_condition.UNet2DConditionModel.fuse_qkv_projections
-    def fuse_qkv_projections(self):
-        """
-        Enables fused QKV projections. For self-attention modules, all projection matrices (i.e., query,
-        key, value) are fused. For cross-attention modules, key and value projection matrices are fused.
-
-        <Tip warning={true}>
-
-        This API is 🧪 experimental.
-
-        </Tip>
-        """
-        self.original_attn_processors = None
-
-        for _, attn_processor in self.attn_processors.items():
-            if "Added" in str(attn_processor.__class__.__name__):
-                raise ValueError("`fuse_qkv_projections()` is not supported for models having added KV projections.")
-
-        self.original_attn_processors = self.attn_processors
-
-        for module in self.modules():
-            if isinstance(module, Attention):
-                module.fuse_projections(fuse=True)
-
-    # Copied from diffusers.models.unet_2d_condition.UNet2DConditionModel.unfuse_qkv_projections
-    def unfuse_qkv_projections(self):
-        """Disables the fused QKV projection if enabled.
-
-        <Tip warning={true}>
-
-        This API is 🧪 experimental.
-
-        </Tip>
-
-        """
-        if self.original_attn_processors is not None:
-            self.set_attn_processor(self.original_attn_processors)

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/models/unet_2d_condition.py

@@ -25,7 +25,6 @@ from .activations import get_activation
 from .attention_processor import (
     ADDED_KV_ATTENTION_PROCESSORS,
     CROSS_ATTENTION_PROCESSORS,
-    Attention,
     AttentionProcessor,
     AttnAddedKVProcessor,
     AttnProcessor,
@@ -795,42 +794,6 @@ class UNet2DConditionModel(ModelMixin, ConfigMixin, UNet2DConditionLoadersMixin)
                 if hasattr(upsample_block, k) or getattr(upsample_block, k, None) is not None:
                     setattr(upsample_block, k, None)
 
-    def fuse_qkv_projections(self):
-        """
-        Enables fused QKV projections. For self-attention modules, all projection matrices (i.e., query,
-        key, value) are fused. For cross-attention modules, key and value projection matrices are fused.
-
-        <Tip warning={true}>
-
-        This API is 🧪 experimental.
-
-        </Tip>
-        """
-        self.original_attn_processors = None
-
-        for _, attn_processor in self.attn_processors.items():
-            if "Added" in str(attn_processor.__class__.__name__):
-                raise ValueError("`fuse_qkv_projections()` is not supported for models having added KV projections.")
-
-        self.original_attn_processors = self.attn_processors
-
-        for module in self.modules():
-            if isinstance(module, Attention):
-                module.fuse_projections(fuse=True)
-
-    def unfuse_qkv_projections(self):
-        """Disables the fused QKV projection if enabled.
-
-        <Tip warning={true}>
-
-        This API is 🧪 experimental.
-
-        </Tip>
-
-        """
-        if self.original_attn_processors is not None:
-            self.set_attn_processor(self.original_attn_processors)
-
     def forward(
         self,
         sample: torch.FloatTensor,

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/pipelines/stable_diffusion/convert_from_ckpt.py

@@ -446,9 +446,8 @@ def convert_ldm_unet_checkpoint(
         new_checkpoint["add_embedding.linear_2.bias"] = unet_state_dict["label_emb.0.2.bias"]
 
     # Relevant to StableDiffusionUpscalePipeline
-    if "num_class_embeds" in config:
-        if (config["num_class_embeds"] is not None) and ("label_emb.weight" in unet_state_dict):
-            new_checkpoint["class_embedding.weight"] = unet_state_dict["label_emb.weight"]
+    if (config["num_class_embeds"] is not None) and ("label_emb.weight" in unet_state_dict):
+        new_checkpoint["class_embedding.weight"] = unet_state_dict["label_emb.weight"]
 
     new_checkpoint["conv_in.weight"] = unet_state_dict["input_blocks.0.0.weight"]
     new_checkpoint["conv_in.bias"] = unet_state_dict["input_blocks.0.0.bias"]

src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl.py

@@ -34,7 +34,6 @@ from ...loaders import (
 from ...models import AutoencoderKL, ImageProjection, UNet2DConditionModel
 from ...models.attention_processor import (
     AttnProcessor2_0,
-    FusedAttnProcessor2_0,
     LoRAAttnProcessor2_0,
     LoRAXFormersAttnProcessor,
     XFormersAttnProcessor,
@@ -682,6 +681,7 @@ class StableDiffusionXLPipeline(
         add_time_ids = torch.tensor([add_time_ids], dtype=dtype)
         return add_time_ids
 
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_upscale.StableDiffusionUpscalePipeline.upcast_vae
     def upcast_vae(self):
         dtype = self.vae.dtype
         self.vae.to(dtype=torch.float32)
@@ -692,7 +692,6 @@ class StableDiffusionXLPipeline(
                 XFormersAttnProcessor,
                 LoRAXFormersAttnProcessor,
                 LoRAAttnProcessor2_0,
-                FusedAttnProcessor2_0,
             ),
         )
         # if xformers or torch_2_0 is used attention block does not need
@@ -730,65 +729,6 @@ class StableDiffusionXLPipeline(
         """Disables the FreeU mechanism if enabled."""
         self.unet.disable_freeu()
 
-    def fuse_qkv_projections(self, unet: bool = True, vae: bool = True):
-        """
-        Enables fused QKV projections. For self-attention modules, all projection matrices (i.e., query,
-        key, value) are fused. For cross-attention modules, key and value projection matrices are fused.
-
-        <Tip warning={true}>
-
-        This API is 🧪 experimental.
-
-        </Tip>
-
-        Args:
-            unet (`bool`, defaults to `True`): To apply fusion on the UNet.
-            vae (`bool`, defaults to `True`): To apply fusion on the VAE.
-        """
-        self.fusing_unet = False
-        self.fusing_vae = False
-
-        if unet:
-            self.fusing_unet = True
-            self.unet.fuse_qkv_projections()
-            self.unet.set_attn_processor(FusedAttnProcessor2_0())
-
-        if vae:
-            if not isinstance(self.vae, AutoencoderKL):
-                raise ValueError("`fuse_qkv_projections()` is only supported for the VAE of type `AutoencoderKL`.")
-
-            self.fusing_vae = True
-            self.vae.fuse_qkv_projections()
-            self.vae.set_attn_processor(FusedAttnProcessor2_0())
-
-    def unfuse_qkv_projections(self, unet: bool = True, vae: bool = True):
-        """Disable QKV projection fusion if enabled.
-
-        <Tip warning={true}>
-
-        This API is 🧪 experimental.
-
-        </Tip>
-
-        Args:
-            unet (`bool`, defaults to `True`): To apply fusion on the UNet.
-            vae (`bool`, defaults to `True`): To apply fusion on the VAE.
-
-        """
-        if unet:
-            if not self.fusing_unet:
-                logger.warning("The UNet was not initially fused for QKV projections. Doing nothing.")
-            else:
-                self.unet.unfuse_qkv_projections()
-                self.fusing_unet = False
-
-        if vae:
-            if not self.fusing_vae:
-                logger.warning("The VAE was not initially fused for QKV projections. Doing nothing.")
-            else:
-                self.vae.unfuse_qkv_projections()
-                self.fusing_vae = False
-
     # Copied from diffusers.pipelines.latent_consistency_models.pipeline_latent_consistency_text2img.LatentConsistencyModelPipeline.get_guidance_scale_embedding
     def get_guidance_scale_embedding(self, w, embedding_dim=512, dtype=torch.float32):
         """

src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_instruct_pix2pix.py

@@ -24,7 +24,6 @@ from ...loaders import FromSingleFileMixin, StableDiffusionXLLoraLoaderMixin, Te
 from ...models import AutoencoderKL, UNet2DConditionModel
 from ...models.attention_processor import (
     AttnProcessor2_0,
-    FusedAttnProcessor2_0,
     LoRAAttnProcessor2_0,
     LoRAXFormersAttnProcessor,
     XFormersAttnProcessor,
@@ -611,7 +610,6 @@ class StableDiffusionXLInstructPix2PixPipeline(
                 XFormersAttnProcessor,
                 LoRAXFormersAttnProcessor,
                 LoRAAttnProcessor2_0,
-                FusedAttnProcessor2_0,
             ),
         )
         # if xformers or torch_2_0 is used attention block does not need

src/diffusers/pipelines/versatile_diffusion/modeling_text_unet.py

@@ -10,10 +10,10 @@ from diffusers.utils import deprecate
 from ...configuration_utils import ConfigMixin, register_to_config
 from ...models import ModelMixin
 from ...models.activations import get_activation
+from ...models.attention import Attention
 from ...models.attention_processor import (
     ADDED_KV_ATTENTION_PROCESSORS,
     CROSS_ATTENTION_PROCESSORS,
-    Attention,
     AttentionProcessor,
     AttnAddedKVProcessor,
     AttnAddedKVProcessor2_0,
@@ -1000,42 +1000,6 @@ class UNetFlatConditionModel(ModelMixin, ConfigMixin):
                 if hasattr(upsample_block, k) or getattr(upsample_block, k, None) is not None:
                     setattr(upsample_block, k, None)
 
-    def fuse_qkv_projections(self):
-        """
-        Enables fused QKV projections. For self-attention modules, all projection matrices (i.e., query,
-        key, value) are fused. For cross-attention modules, key and value projection matrices are fused.
-
-        <Tip warning={true}>
-
-        This API is 🧪 experimental.
-
-        </Tip>
-        """
-        self.original_attn_processors = None
-
-        for _, attn_processor in self.attn_processors.items():
-            if "Added" in str(attn_processor.__class__.__name__):
-                raise ValueError("`fuse_qkv_projections()` is not supported for models having added KV projections.")
-
-        self.original_attn_processors = self.attn_processors
-
-        for module in self.modules():
-            if isinstance(module, Attention):
-                module.fuse_projections(fuse=True)
-
-    def unfuse_qkv_projections(self):
-        """Disables the fused QKV projection if enabled.
-
-        <Tip warning={true}>
-
-        This API is 🧪 experimental.
-
-        </Tip>
-
-        """
-        if self.original_attn_processors is not None:
-            self.set_attn_processor(self.original_attn_processors)
-
     def forward(
         self,
         sample: torch.FloatTensor,

src/diffusers/schedulers/scheduling_euler_discrete.py

@@ -191,11 +191,10 @@ class EulerDiscreteScheduler(SchedulerMixin, ConfigMixin):
     @property
     def init_noise_sigma(self):
         # standard deviation of the initial noise distribution
-        max_sigma = max(self.sigmas) if isinstance(self.sigmas, list) else self.sigmas.max()
         if self.config.timestep_spacing in ["linspace", "trailing"]:
-            return max_sigma
+            return self.sigmas.max()
 
-        return (max_sigma**2 + 1) ** 0.5
+        return (self.sigmas.max() ** 2 + 1) ** 0.5
 
     @property
     def step_index(self):

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:

tests/pipelines/stable_diffusion_xl/test_stable_diffusion_xl.py

@@ -938,37 +938,6 @@ class StableDiffusionXLPipelineFastTests(
 
         assert np.abs(image_slices[0] - image_slices[1]).max() < 1e-3
 
-    def test_stable_diffusion_xl_with_fused_qkv_projections(self):
-        device = "cpu"  # ensure determinism for the device-dependent torch.Generator
-        components = self.get_dummy_components()
-        sd_pipe = StableDiffusionXLPipeline(**components)
-        sd_pipe = sd_pipe.to(device)
-        sd_pipe.set_progress_bar_config(disable=None)
-
-        inputs = self.get_dummy_inputs(device)
-        image = sd_pipe(**inputs).images
-        original_image_slice = image[0, -3:, -3:, -1]
-
-        sd_pipe.fuse_qkv_projections()
-        inputs = self.get_dummy_inputs(device)
-        image = sd_pipe(**inputs).images
-        image_slice_fused = image[0, -3:, -3:, -1]
-
-        sd_pipe.unfuse_qkv_projections()
-        inputs = self.get_dummy_inputs(device)
-        image = sd_pipe(**inputs).images
-        image_slice_disabled = image[0, -3:, -3:, -1]
-
-        assert np.allclose(
-            original_image_slice, image_slice_fused, atol=1e-2, rtol=1e-2
-        ), "Fusion of QKV projections shouldn't affect the outputs."
-        assert np.allclose(
-            image_slice_fused, image_slice_disabled, atol=1e-2, rtol=1e-2
-        ), "Outputs, with QKV projection fusion enabled, shouldn't change when fused QKV projections are disabled."
-        assert np.allclose(
-            original_image_slice, image_slice_disabled, atol=1e-2, rtol=1e-2
-        ), "Original outputs should match when fused QKV projections are disabled."
-
 
 @slow
 class StableDiffusionXLPipelineIntegrationTests(unittest.TestCase):