Add Scheduler.from_pretrained and better scheduler changing (#1286)

* add conversion script for vae

* uP

* uP

* more changes

* push

* up

* finish again

* up

* up

* up

* up

* finish

* up

* uP

* up

* Apply suggestions from code review
Co-authored-by: Pedro Cuenca <pedro@huggingface.co>
Co-authored-by: Anton Lozhkov <anton@huggingface.co>
Co-authored-by: Suraj Patil <surajp815@gmail.com>

* up

* up
Co-authored-by: Pedro Cuenca <pedro@huggingface.co>
Co-authored-by: Anton Lozhkov <anton@huggingface.co>
Co-authored-by: Suraj Patil <surajp815@gmail.com>

README.md

@@ -152,15 +152,7 @@ it before the pipeline and pass it to `from_pretrained`.
 ```python
 from diffusers import LMSDiscreteScheduler
 
-lms = LMSDiscreteScheduler.from_config("CompVis/stable-diffusion-v1-4", subfolder="scheduler")
-
-pipe = StableDiffusionPipeline.from_pretrained(
-    "runwayml/stable-diffusion-v1-5", 
-    revision="fp16", 
-    torch_dtype=torch.float16,
-    scheduler=lms,
-)
-pipe = pipe.to("cuda")
+pipe.scheduler = LMSDiscreteScheduler.from_config(pipe.scheduler.config)
 
 prompt = "a photo of an astronaut riding a horse on mars"
 image = pipe(prompt).images[0]  

docs/source/_toctree.yml

@@ -10,6 +10,8 @@
   - sections:
     - local: using-diffusers/loading
       title: "Loading Pipelines, Models, and Schedulers"
+    - local: using-diffusers/schedulers
+      title: "Using different Schedulers"
     - local: using-diffusers/configuration
       title: "Configuring Pipelines, Models, and Schedulers"
     - local: using-diffusers/custom_pipeline_overview

docs/source/api/configuration.mdx

@@ -15,9 +15,9 @@ specific language governing permissions and limitations under the License.
 In Diffusers, schedulers of type [`schedulers.scheduling_utils.SchedulerMixin`], and models of type [`ModelMixin`] inherit from [`ConfigMixin`] which conveniently takes care of storing all parameters that are 
 passed to the respective `__init__` methods in a JSON-configuration file.
 
-TODO(PVP) - add example and better info here
-
 ## ConfigMixin
+
 [[autodoc]] ConfigMixin
+	- load_config
 	- from_config
 	- save_config

docs/source/api/pipelines/cycle_diffusion.mdx

@@ -39,7 +39,7 @@ from diffusers import CycleDiffusionPipeline, DDIMScheduler
 # load the pipeline
 # make sure you're logged in with `huggingface-cli login`
 model_id_or_path = "CompVis/stable-diffusion-v1-4"
-scheduler = DDIMScheduler.from_config(model_id_or_path, subfolder="scheduler")
+scheduler = DDIMScheduler.from_pretrained(model_id_or_path, subfolder="scheduler")
 pipe = CycleDiffusionPipeline.from_pretrained(model_id_or_path, scheduler=scheduler).to("cuda")
 
 # let's download an initial image

docs/source/api/pipelines/repaint.mdx

@@ -54,7 +54,7 @@ original_image = download_image(img_url).resize((256, 256))
 mask_image = download_image(mask_url).resize((256, 256))
 
 # Load the RePaint scheduler and pipeline based on a pretrained DDPM model
-scheduler = RePaintScheduler.from_config("google/ddpm-ema-celebahq-256")
+scheduler = RePaintScheduler.from_pretrained("google/ddpm-ema-celebahq-256")
 pipe = RePaintPipeline.from_pretrained("google/ddpm-ema-celebahq-256", scheduler=scheduler)
 pipe = pipe.to("cuda")
 

docs/source/api/pipelines/stable_diffusion.mdx

@@ -34,13 +34,17 @@ For more details about how Stable Diffusion works and how it differs from the ba
 ### How to load and use different schedulers.
 
 The stable diffusion pipeline uses [`PNDMScheduler`] scheduler by default. But `diffusers` provides many other schedulers that can be used with the stable diffusion pipeline such as [`DDIMScheduler`], [`LMSDiscreteScheduler`], [`EulerDiscreteScheduler`], [`EulerAncestralDiscreteScheduler`] etc.
-To use a different scheduler, you can pass the `scheduler` argument to `from_pretrained` method of the pipeline. For example, to use the [`EulerDiscreteScheduler`], you can do the following:
+To use a different scheduler, you can either change it via the [`ConfigMixin.from_config`] method or pass the `scheduler` argument to the `from_pretrained` method of the pipeline. For example, to use the [`EulerDiscreteScheduler`], you can do the following:
 
 ```python
-from diffusers import StableDiffusionPipeline, EulerDiscreteScheduler
+>>> from diffusers import StableDiffusionPipeline, EulerDiscreteScheduler
 
-euler_scheduler = EulerDiscreteScheduler.from_config("CompVis/stable-diffusion-v1-4", subfolder="scheduler")
-pipeline = StableDiffusionPipeline.from_pretrained("CompVis/stable-diffusion-v1-4", scheduler=euler_scheduler)
+>>> pipeline = StableDiffusionPipeline.from_pretrained("CompVis/stable-diffusion-v1-4")
+>>> pipeline.scheduler = EulerDiscreteScheduler.from_config(pipeline.scheduler.config)
+
+>>> # or
+>>> euler_scheduler = EulerDiscreteScheduler.from_pretrained("CompVis/stable-diffusion-v1-4", subfolder="scheduler")
+>>> pipeline = StableDiffusionPipeline.from_pretrained("CompVis/stable-diffusion-v1-4", scheduler=euler_scheduler)
 ```
 
 

docs/source/quicktour.mdx

@@ -41,7 +41,7 @@ In this guide though, you'll use [`DiffusionPipeline`] for text-to-image generat
 ```python
 >>> from diffusers import DiffusionPipeline
 
->>> generator = DiffusionPipeline.from_pretrained("CompVis/ldm-text2im-large-256")
+>>> pipeline = DiffusionPipeline.from_pretrained("CompVis/ldm-text2im-large-256")
 ```
 
 The [`DiffusionPipeline`] downloads and caches all modeling, tokenization, and scheduling components. 
@@ -49,13 +49,13 @@ Because the model consists of roughly 1.4 billion parameters, we strongly recomm
 You can move the generator object to GPU, just like you would in PyTorch.
 
 ```python
->>> generator.to("cuda")
+>>> pipeline.to("cuda")
 ```
 
-Now you can use the `generator` on your text prompt:
+Now you can use the `pipeline` on your text prompt:
 
 ```python
->>> image = generator("An image of a squirrel in Picasso style").images[0]
+>>> image = pipeline("An image of a squirrel in Picasso style").images[0]
 ```
 
 The output is by default wrapped into a [PIL Image object](https://pillow.readthedocs.io/en/stable/reference/Image.html?highlight=image#the-image-class).
@@ -82,7 +82,7 @@ just like we did before only that now you need to pass your `AUTH_TOKEN`:
 ```python
 >>> from diffusers import DiffusionPipeline
 
->>> generator = DiffusionPipeline.from_pretrained("runwayml/stable-diffusion-v1-5", use_auth_token=AUTH_TOKEN)
+>>> pipeline = DiffusionPipeline.from_pretrained("runwayml/stable-diffusion-v1-5", use_auth_token=AUTH_TOKEN)
 ```
 
 If you do not pass your authentication token you will see that the diffusion system will not be correctly 
@@ -102,7 +102,7 @@ token. Assuming that `"./stable-diffusion-v1-5"` is the local path to the cloned
 you can also load the pipeline as follows:
 
 ```python
->>> generator = DiffusionPipeline.from_pretrained("./stable-diffusion-v1-5")
+>>> pipeline = DiffusionPipeline.from_pretrained("./stable-diffusion-v1-5")
 ```
 
 Running the pipeline is then identical to the code above as it's the same model architecture.
@@ -115,19 +115,20 @@ Running the pipeline is then identical to the code above as it's the same model
 
 Diffusion systems can be used with multiple different [schedulers](./api/schedulers) each with their
 pros and cons. By default, Stable Diffusion runs with [`PNDMScheduler`], but it's very simple to 
-use a different scheduler. *E.g.* if you would instead like to use the [`LMSDiscreteScheduler`] scheduler,
+use a different scheduler. *E.g.* if you would instead like to use the [`EulerDiscreteScheduler`] scheduler,
 you could use it as follows:
 
 ```python
->>> from diffusers import LMSDiscreteScheduler
+>>> from diffusers import EulerDiscreteScheduler
 
->>> scheduler = LMSDiscreteScheduler.from_config("runwayml/stable-diffusion-v1-5", subfolder="scheduler")
+>>> pipeline = StableDiffusionPipeline.from_pretrained("runwayml/stable-diffusion-v1-5", use_auth_token=AUTH_TOKEN)
 
->>> generator = StableDiffusionPipeline.from_pretrained(
-...     "runwayml/stable-diffusion-v1-5", scheduler=scheduler, use_auth_token=AUTH_TOKEN
-... )
+>>> # change scheduler to Euler
+>>> pipeline.scheduler = EulerDiscreteScheduler.from_config(pipeline.scheduler.config)
 ```
 
+For more in-detail information on how to change between schedulers, please refer to the [Using Schedulers](./using-diffusers/schedulers) guide.
+
 [Stability AI's](https://stability.ai/) Stable Diffusion model is an impressive image generation model
 and can do much more than just generating images from text. We have dedicated a whole documentation page,
 just for Stable Diffusion [here](./conceptual/stable_diffusion).

docs/source/using-diffusers/loading.mdx

@@ -19,7 +19,7 @@ In the following we explain in-detail how to easily load:
 
 - *Complete Diffusion Pipelines* via the [`DiffusionPipeline.from_pretrained`]
 - *Diffusion Models* via [`ModelMixin.from_pretrained`]
-- *Schedulers* via [`ConfigMixin.from_config`]
+- *Schedulers* via [`SchedulerMixin.from_pretrained`]
 
 ## Loading pipelines
 
@@ -137,15 +137,15 @@ from diffusers import DiffusionPipeline, EulerDiscreteScheduler, DPMSolverMultis
 
 repo_id = "runwayml/stable-diffusion-v1-5"
 
-scheduler = EulerDiscreteScheduler.from_config(repo_id, subfolder="scheduler")
+scheduler = EulerDiscreteScheduler.from_pretrained(repo_id, subfolder="scheduler")
 # or
-# scheduler = DPMSolverMultistepScheduler.from_config(repo_id, subfolder="scheduler")
+# scheduler = DPMSolverMultistepScheduler.from_pretrained(repo_id, subfolder="scheduler")
 
 stable_diffusion = DiffusionPipeline.from_pretrained(repo_id, scheduler=scheduler)
 ```
 
 Three things are worth paying attention to here.
-- First, the scheduler is loaded with [`ConfigMixin.from_config`] since it only depends on a configuration file and not any parameterized weights
+- First, the scheduler is loaded with [`SchedulerMixin.from_pretrained`]
 - Second, the scheduler is loaded with a function argument, called `subfolder="scheduler"` as the configuration of stable diffusion's scheduling is defined in a [subfolder of the official pipeline repository](https://huggingface.co/runwayml/stable-diffusion-v1-5/tree/main/scheduler)
 - Third, the scheduler instance can simply be passed with the `scheduler` keyword argument to [`DiffusionPipeline.from_pretrained`]. This works because the [`StableDiffusionPipeline`] defines its scheduler with the `scheduler` attribute. It's not possible to use a different name, such as `sampler=scheduler` since `sampler` is not a defined keyword for [`StableDiffusionPipeline.__init__`]
 
@@ -337,8 +337,8 @@ model = UNet2DModel.from_pretrained(repo_id)
 
 ## Loading schedulers
 
-Schedulers cannot be loaded via a `from_pretrained` method, but instead rely on [`ConfigMixin.from_config`]. Schedulers are **not parameterized** or **trained**, but instead purely defined by a configuration file.
-Therefore the loading method was given a different name here.
+Schedulers rely on [`SchedulerMixin.from_pretrained`]. Schedulers are **not parameterized** or **trained**, but instead purely defined by a configuration file.
+For consistency, we use the same method name as we do for models or pipelines, but no weights are loaded in this case.
 
 In constrast to pipelines or models, loading schedulers does not consume any significant amount of memory and the same configuration file can often be used for a variety of different schedulers.
 For example, all of:
@@ -367,13 +367,13 @@ from diffusers import (
 
 repo_id = "runwayml/stable-diffusion-v1-5"
 
-ddpm = DDPMScheduler.from_config(repo_id, subfolder="scheduler")
-ddim = DDIMScheduler.from_config(repo_id, subfolder="scheduler")
-pndm = PNDMScheduler.from_config(repo_id, subfolder="scheduler")
-lms = LMSDiscreteScheduler.from_config(repo_id, subfolder="scheduler")
-euler_anc = EulerAncestralDiscreteScheduler.from_config(repo_id, subfolder="scheduler")
-euler = EulerDiscreteScheduler.from_config(repo_id, subfolder="scheduler")
-dpm = DPMSolverMultistepScheduler.from_config(repo_id, subfolder="scheduler")
+ddpm = DDPMScheduler.from_pretrained(repo_id, subfolder="scheduler")
+ddim = DDIMScheduler.from_pretrained(repo_id, subfolder="scheduler")
+pndm = PNDMScheduler.from_pretrained(repo_id, subfolder="scheduler")
+lms = LMSDiscreteScheduler.from_pretrained(repo_id, subfolder="scheduler")
+euler_anc = EulerAncestralDiscreteScheduler.from_pretrained(repo_id, subfolder="scheduler")
+euler = EulerDiscreteScheduler.from_pretrained(repo_id, subfolder="scheduler")
+dpm = DPMSolverMultistepScheduler.from_pretrained(repo_id, subfolder="scheduler")
 
 # replace `dpm` with any of `ddpm`, `ddim`, `pndm`, `lms`, `euler`, `euler_anc`
 pipeline = StableDiffusionPipeline.from_pretrained(repo_id, scheduler=dpm)

docs/source/using-diffusers/schedulers.mdx

+<!--Copyright 2022 The HuggingFace Team. All rights reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
+the License. You may obtain a copy of the License at
+
+http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
+an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
+specific language governing permissions and limitations under the License.
+-->
+
+# Schedulers
+
+Diffusion pipelines are inherently a collection of diffusion models and schedulers that are partly independent from each other. This means that one is able to switch out parts of the pipeline to better customize 
+a pipeline to one's use case. The best example of this are the [Schedulers](../api/schedulers.mdx).
+
+Whereas diffusion models usually simply define the forward pass from noise to a less noisy sample, 
+schedulers define the whole denoising process, *i.e.*:
+- How many denoising steps?
+- Stochastic or deterministic?
+- What algorithm to use to find the denoised sample
+
+They can be quite complex and often define a trade-off between **denoising speed** and **denoising quality**.
+It is extremely difficult to measure quantitatively which scheduler works best for a given diffusion pipeline, so it is often recommended to simply try out which works best.
+
+The following paragraphs shows how to do so with the 🧨 Diffusers library.
+
+## Load pipeline
+
+Let's start by loading the stable diffusion pipeline.
+Remember that you have to be a registered user on the 🤗 Hugging Face Hub, and have "click-accepted" the [license](https://huggingface.co/runwayml/stable-diffusion-v1-5) in order to use stable diffusion.
+
+```python
+from huggingface_hub import login
+from diffusers import DiffusionPipeline
+import torch
+
+# first we need to login with our access token
+login()
+
+# Now we can download the pipeline
+pipeline = DiffusionPipeline.from_pretrained("runwayml/stable-diffusion-v1-5", torch_dtype=torch.float16)
+```
+
+Next, we move it to GPU:
+
+```python
+pipeline.to("cuda")
+```
+
+## Access the scheduler
+
+The scheduler is always one of the components of the pipeline and is usually called `"scheduler"`.
+So it can be accessed via the `"scheduler"` property.
+
+```python
+pipeline.scheduler
+```
+
+**Output**:
+```
+PNDMScheduler {
+  "_class_name": "PNDMScheduler",
+  "_diffusers_version": "0.8.0.dev0",
+  "beta_end": 0.012,
+  "beta_schedule": "scaled_linear",
+  "beta_start": 0.00085,
+  "clip_sample": false,
+  "num_train_timesteps": 1000,
+  "set_alpha_to_one": false,
+  "skip_prk_steps": true,
+  "steps_offset": 1,
+  "trained_betas": null
+}
+```
+
+We can see that the scheduler is of type [`PNDMScheduler`]. 
+Cool, now let's compare the scheduler in its performance to other schedulers.
+First we define a prompt on which we will test all the different schedulers:
+
+```python
+prompt = "A photograph of an astronaut riding a horse on Mars, high resolution, high definition."
+```
+
+Next, we create a generator from a random seed that will ensure that we can generate similar images as well as run the pipeline:
+
+```python
+generator = torch.Generator(device="cuda").manual_seed(8)
+image = pipeline(prompt, generator=generator).images[0]
+image
+```
+
+<p align="center">
+    <br>
+    <img src="https://huggingface.co/datasets/patrickvonplaten/images/resolve/main/diffusers_docs/astronaut_pndm.png" width="400"/>
+    <br>
+</p>
+
+
+## Changing the scheduler
+
+Now we show how easy it is to change the scheduler of a pipeline. Every scheduler has a property [`SchedulerMixin.compatibles`] 
+which defines all compatible schedulers. You can take a look at all available, compatible schedulers for the Stable Diffusion pipeline as follows.
+
+```python
+pipeline.scheduler.compatibles
+```
+
+**Output**:
+```
+[diffusers.schedulers.scheduling_lms_discrete.LMSDiscreteScheduler,
+ diffusers.schedulers.scheduling_ddim.DDIMScheduler,
+ diffusers.schedulers.scheduling_dpmsolver_multistep.DPMSolverMultistepScheduler,
+ diffusers.schedulers.scheduling_euler_discrete.EulerDiscreteScheduler,
+ diffusers.schedulers.scheduling_pndm.PNDMScheduler,
+ diffusers.schedulers.scheduling_ddpm.DDPMScheduler,
+ diffusers.schedulers.scheduling_euler_ancestral_discrete.EulerAncestralDiscreteScheduler]
+```
+
+Cool, lots of schedulers to look at. Feel free to have a look at their respective class definitions: 
+
+- [`LMSDiscreteScheduler`], 
+- [`DDIMScheduler`], 
+- [`DPMSolverMultistepScheduler`], 
+- [`EulerDiscreteScheduler`], 
+- [`PNDMScheduler`], 
+- [`DDPMScheduler`], 
+- [`EulerAncestralDiscreteScheduler`].
+
+We will now compare the input prompt with all other schedulers. To change the scheduler of the pipeline you can make use of the 
+convenient [`ConfigMixin.config`] property in combination with the [`ConfigMixin.from_config`] function.
+
+```python
+pipeline.scheduler.config
+```
+
+returns a dictionary of the configuration of the scheduler:
+
+**Output**:
+```
+FrozenDict([('num_train_timesteps', 1000),
+            ('beta_start', 0.00085),
+            ('beta_end', 0.012),
+            ('beta_schedule', 'scaled_linear'),
+            ('trained_betas', None),
+            ('skip_prk_steps', True),
+            ('set_alpha_to_one', False),
+            ('steps_offset', 1),
+            ('_class_name', 'PNDMScheduler'),
+            ('_diffusers_version', '0.8.0.dev0'),
+            ('clip_sample', False)])
+```
+
+This configuration can then be used to instantiate a scheduler
+of a different class that is compatible with the pipeline. Here, 
+we change the scheduler to the [`DDIMScheduler`].
+
+```python
+from diffusers import DDIMScheduler
+
+pipeline.scheduler = DDIMScheduler.from_config(pipeline.scheduler.config)
+```
+
+Cool, now we can run the pipeline again to compare the generation quality.
+
+```python
+generator = torch.Generator(device="cuda").manual_seed(8)
+image = pipeline(prompt, generator=generator).images[0]
+image
+```
+
+<p align="center">
+    <br>
+    <img src="https://huggingface.co/datasets/patrickvonplaten/images/resolve/main/diffusers_docs/astronaut_ddim.png" width="400"/>
+    <br>
+</p>
+
+
+## Compare schedulers
+
+So far we have tried running the stable diffusion pipeline with two schedulers: [`PNDMScheduler`] and [`DDIMScheduler`]. 
+A number of better schedulers have been released that can be run with much fewer steps, let's compare them here:
+
+[`LMSDiscreteScheduler`] usually leads to better results:
+
+```python
+from diffusers import LMSDiscreteScheduler
+
+pipeline.scheduler = LMSDiscreteScheduler.from_config(pipeline.scheduler.config)
+
+generator = torch.Generator(device="cuda").manual_seed(8)
+image = pipeline(prompt, generator=generator).images[0]
+image
+```
+
+<p align="center">
+    <br>
+    <img src="https://huggingface.co/datasets/patrickvonplaten/images/resolve/main/diffusers_docs/astronaut_lms.png" width="400"/>
+    <br>
+</p>
+
+
+[`EulerDiscreteScheduler`] and [`EulerAncestralDiscreteScheduler`] can generate high quality results with as little as 30 steps.
+
+```python
+from diffusers import EulerDiscreteScheduler
+
+pipeline.scheduler = EulerDiscreteScheduler.from_config(pipeline.scheduler.config)
+
+generator = torch.Generator(device="cuda").manual_seed(8)
+image = pipeline(prompt, generator=generator, num_inference_steps=30).images[0]
+image
+```
+
+<p align="center">
+    <br>
+    <img src="https://huggingface.co/datasets/patrickvonplaten/images/resolve/main/diffusers_docs/astronaut_euler_discrete.png" width="400"/>
+    <br>
+</p>
+
+
+and:
+
+```python
+from diffusers import EulerAncestralDiscreteScheduler
+
+pipeline.scheduler = EulerAncestralDiscreteScheduler.from_config(pipeline.scheduler.config)
+
+generator = torch.Generator(device="cuda").manual_seed(8)
+image = pipeline(prompt, generator=generator, num_inference_steps=30).images[0]
+image
+```
+
+<p align="center">
+    <br>
+    <img src="https://huggingface.co/datasets/patrickvonplaten/images/resolve/main/diffusers_docs/astronaut_euler_ancestral.png" width="400"/>
+    <br>
+</p>
+
+
+At the time of writing this doc [`DPMSolverMultistepScheduler`] gives arguably the best speed/quality trade-off and can be run with as little 
+as 20 steps.
+
+```python
+from diffusers import DPMSolverMultistepScheduler
+
+pipeline.scheduler = DPMSolverMultistepScheduler.from_config(pipeline.scheduler.config)
+
+generator = torch.Generator(device="cuda").manual_seed(8)
+image = pipeline(prompt, generator=generator, num_inference_steps=20).images[0]
+image
+```
+
+<p align="center">
+    <br>
+    <img src="https://huggingface.co/datasets/patrickvonplaten/images/resolve/main/diffusers_docs/astronaut_dpm.png" width="400"/>
+    <br>
+</p>
+
+As you can see most images look very similar and are arguably of very similar quality. It often really depends on the specific use case which scheduler to choose. A good approach is always to run multiple different
+schedulers to compare results.

src/diffusers/configuration_utils.py

@@ -29,7 +29,7 @@ from huggingface_hub.utils import EntryNotFoundError, RepositoryNotFoundError, R
 from requests import HTTPError
 
 from . import __version__
-from .utils import DIFFUSERS_CACHE, HUGGINGFACE_CO_RESOLVE_ENDPOINT, logging
+from .utils import DIFFUSERS_CACHE, HUGGINGFACE_CO_RESOLVE_ENDPOINT, DummyObject, deprecate, logging
 
 
 logger = logging.get_logger(__name__)
@@ -37,6 +37,38 @@ logger = logging.get_logger(__name__)
 _re_configuration_file = re.compile(r"config\.(.*)\.json")
 
 
+class FrozenDict(OrderedDict):
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+
+        for key, value in self.items():
+            setattr(self, key, value)
+
+        self.__frozen = True
+
+    def __delitem__(self, *args, **kwargs):
+        raise Exception(f"You cannot use ``__delitem__`` on a {self.__class__.__name__} instance.")
+
+    def setdefault(self, *args, **kwargs):
+        raise Exception(f"You cannot use ``setdefault`` on a {self.__class__.__name__} instance.")
+
+    def pop(self, *args, **kwargs):
+        raise Exception(f"You cannot use ``pop`` on a {self.__class__.__name__} instance.")
+
+    def update(self, *args, **kwargs):
+        raise Exception(f"You cannot use ``update`` on a {self.__class__.__name__} instance.")
+
+    def __setattr__(self, name, value):
+        if hasattr(self, "__frozen") and self.__frozen:
+            raise Exception(f"You cannot use ``__setattr__`` on a {self.__class__.__name__} instance.")
+        super().__setattr__(name, value)
+
+    def __setitem__(self, name, value):
+        if hasattr(self, "__frozen") and self.__frozen:
+            raise Exception(f"You cannot use ``__setattr__`` on a {self.__class__.__name__} instance.")
+        super().__setitem__(name, value)
+
+
 class ConfigMixin:
     r"""
     Base class for all configuration classes. Stores all configuration parameters under `self.config` Also handles all
@@ -49,13 +81,12 @@ class ConfigMixin:
           [`~ConfigMixin.save_config`] (should be overridden by parent class).
         - **ignore_for_config** (`List[str]`) -- A list of attributes that should not be saved in the config (should be
           overridden by parent class).
-        - **_compatible_classes** (`List[str]`) -- A list of classes that are compatible with the parent class, so that
-          `from_config` can be used from a class different than the one used to save the config (should be overridden
-          by parent class).
+        - **has_compatibles** (`bool`) -- Whether the class has compatible classes (should be overridden by parent
+          class).
     """
     config_name = None
     ignore_for_config = []
-    _compatible_classes = []
+    has_compatibles = False
 
     def register_to_config(self, **kwargs):
         if self.config_name is None:
@@ -104,9 +135,98 @@ class ConfigMixin:
         logger.info(f"Configuration saved in {output_config_file}")
 
     @classmethod
-    def from_config(cls, pretrained_model_name_or_path: Union[str, os.PathLike], return_unused_kwargs=False, **kwargs):
+    def from_config(cls, config: Union[FrozenDict, Dict[str, Any]] = None, return_unused_kwargs=False, **kwargs):
+        r"""
+        Instantiate a Python class from a config dictionary
+
+        Parameters:
+            config (`Dict[str, Any]`):
+                A config dictionary from which the Python class will be instantiated. Make sure to only load
+                configuration files of compatible classes.
+            return_unused_kwargs (`bool`, *optional*, defaults to `False`):
+                Whether kwargs that are not consumed by the Python class should be returned or not.
+
+            kwargs (remaining dictionary of keyword arguments, *optional*):
+                Can be used to update the configuration object (after it being loaded) and initiate the Python class.
+                `**kwargs` will be directly passed to the underlying scheduler/model's `__init__` method and eventually
+                overwrite same named arguments of `config`.
+
+        Examples:
+
+        ```python
+        >>> from diffusers import DDPMScheduler, DDIMScheduler, PNDMScheduler
+
+        >>> # Download scheduler from huggingface.co and cache.
+        >>> scheduler = DDPMScheduler.from_pretrained("google/ddpm-cifar10-32")
+
+        >>> # Instantiate DDIM scheduler class with same config as DDPM
+        >>> scheduler = DDIMScheduler.from_config(scheduler.config)
+
+        >>> # Instantiate PNDM scheduler class with same config as DDPM
+        >>> scheduler = PNDMScheduler.from_config(scheduler.config)
+        ```
+        """
+        # <===== TO BE REMOVED WITH DEPRECATION
+        # TODO(Patrick) - make sure to remove the following lines when config=="model_path" is deprecated
+        if "pretrained_model_name_or_path" in kwargs:
+            config = kwargs.pop("pretrained_model_name_or_path")
+
+        if config is None:
+            raise ValueError("Please make sure to provide a config as the first positional argument.")
+        # ======>
+
+        if not isinstance(config, dict):
+            deprecation_message = "It is deprecated to pass a pretrained model name or path to `from_config`."
+            if "Scheduler" in cls.__name__:
+                deprecation_message += (
+                    f"If you were trying to load a scheduler, please use {cls}.from_pretrained(...) instead."
+                    " Otherwise, please make sure to pass a configuration dictionary instead. This functionality will"
+                    " be removed in v1.0.0."
+                )
+            elif "Model" in cls.__name__:
+                deprecation_message += (
+                    f"If you were trying to load a model, please use {cls}.load_config(...) followed by"
+                    f" {cls}.from_config(...) instead. Otherwise, please make sure to pass a configuration dictionary"
+                    " instead. This functionality will be removed in v1.0.0."
+                )
+            deprecate("config-passed-as-path", "1.0.0", deprecation_message, standard_warn=False)
+            config, kwargs = cls.load_config(pretrained_model_name_or_path=config, return_unused_kwargs=True, **kwargs)
+
+        init_dict, unused_kwargs, hidden_dict = cls.extract_init_dict(config, **kwargs)
+
+        # Allow dtype to be specified on initialization
+        if "dtype" in unused_kwargs:
+            init_dict["dtype"] = unused_kwargs.pop("dtype")
+
+        # Return model and optionally state and/or unused_kwargs
+        model = cls(**init_dict)
+
+        # make sure to also save config parameters that might be used for compatible classes
+        model.register_to_config(**hidden_dict)
+
+        # add hidden kwargs of compatible classes to unused_kwargs
+        unused_kwargs = {**unused_kwargs, **hidden_dict}
+
+        if return_unused_kwargs:
+            return (model, unused_kwargs)
+        else:
+            return model
+
+    @classmethod
+    def get_config_dict(cls, *args, **kwargs):
+        deprecation_message = (
+            f" The function get_config_dict is deprecated. Please use {cls}.load_config instead. This function will be"
+            " removed in version v1.0.0"
+        )
+        deprecate("get_config_dict", "1.0.0", deprecation_message, standard_warn=False)
+        return cls.load_config(*args, **kwargs)
+
+    @classmethod
+    def load_config(
+        cls, pretrained_model_name_or_path: Union[str, os.PathLike], return_unused_kwargs=False, **kwargs
+    ) -> Tuple[Dict[str, Any], Dict[str, Any]]:
         r"""
-        Instantiate a Python class from a pre-defined JSON-file.
+        Instantiate a Python class from a config dictionary
 
         Parameters:
             pretrained_model_name_or_path (`str` or `os.PathLike`, *optional*):
@@ -120,10 +240,6 @@ class ConfigMixin:
             cache_dir (`Union[str, os.PathLike]`, *optional*):
                 Path to a directory in which a downloaded pretrained model configuration should be cached if the
                 standard cache should not be used.
-            ignore_mismatched_sizes (`bool`, *optional*, defaults to `False`):
-                Whether or not to raise an error if some of the weights from the checkpoint do not have the same size
-                as the weights of the model (if for instance, you are instantiating a model with 10 labels from a
-                checkpoint with 3 labels).
             force_download (`bool`, *optional*, defaults to `False`):
                 Whether or not to force the (re-)download of the model weights and configuration files, overriding the
                 cached versions if they exist.
@@ -161,33 +277,7 @@ class ConfigMixin:
         use this method in a firewalled environment.
 
         </Tip>
-
         """
-        config_dict = cls.get_config_dict(pretrained_model_name_or_path=pretrained_model_name_or_path, **kwargs)
-        init_dict, unused_kwargs = cls.extract_init_dict(config_dict, **kwargs)
-
-        # Allow dtype to be specified on initialization
-        if "dtype" in unused_kwargs:
-            init_dict["dtype"] = unused_kwargs.pop("dtype")
-
-        # Return model and optionally state and/or unused_kwargs
-        model = cls(**init_dict)
-        return_tuple = (model,)
-
-        # Flax schedulers have a state, so return it.
-        if cls.__name__.startswith("Flax") and hasattr(model, "create_state") and getattr(model, "has_state", False):
-            state = model.create_state()
-            return_tuple += (state,)
-
-        if return_unused_kwargs:
-            return return_tuple + (unused_kwargs,)
-        else:
-            return return_tuple if len(return_tuple) > 1 else model
-
-    @classmethod
-    def get_config_dict(
-        cls, pretrained_model_name_or_path: Union[str, os.PathLike], **kwargs
-    ) -> Tuple[Dict[str, Any], Dict[str, Any]]:
         cache_dir = kwargs.pop("cache_dir", DIFFUSERS_CACHE)
         force_download = kwargs.pop("force_download", False)
         resume_download = kwargs.pop("resume_download", False)
@@ -283,6 +373,9 @@ class ConfigMixin:
         except (json.JSONDecodeError, UnicodeDecodeError):
             raise EnvironmentError(f"It looks like the config file at '{config_file}' is not a valid JSON file.")
 
+        if return_unused_kwargs:
+            return config_dict, kwargs
+
         return config_dict
 
     @staticmethod
@@ -291,6 +384,9 @@ class ConfigMixin:
 
     @classmethod
     def extract_init_dict(cls, config_dict, **kwargs):
+        # 0. Copy origin config dict
+        original_dict = {k: v for k, v in config_dict.items()}
+
         # 1. Retrieve expected config attributes from __init__ signature
         expected_keys = cls._get_init_keys(cls)
         expected_keys.remove("self")
@@ -310,10 +406,11 @@ class ConfigMixin:
         # load diffusers library to import compatible and original scheduler
         diffusers_library = importlib.import_module(__name__.split(".")[0])
 
-        # remove attributes from compatible classes that orig cannot expect
-        compatible_classes = [getattr(diffusers_library, c, None) for c in cls._compatible_classes]
-        # filter out None potentially undefined dummy classes
-        compatible_classes = [c for c in compatible_classes if c is not None]
+        if cls.has_compatibles:
+            compatible_classes = [c for c in cls._get_compatibles() if not isinstance(c, DummyObject)]
+        else:
+            compatible_classes = []
+
         expected_keys_comp_cls = set()
         for c in compatible_classes:
             expected_keys_c = cls._get_init_keys(c)
@@ -364,7 +461,10 @@ class ConfigMixin:
         # 6. Define unused keyword arguments
         unused_kwargs = {**config_dict, **kwargs}
 
-        return init_dict, unused_kwargs
+        # 7. Define "hidden" config parameters that were saved for compatible classes
+        hidden_config_dict = {k: v for k, v in original_dict.items() if k not in init_dict and not k.startswith("_")}
+
+        return init_dict, unused_kwargs, hidden_config_dict
 
     @classmethod
     def _dict_from_json_file(cls, json_file: Union[str, os.PathLike]):
@@ -377,6 +477,12 @@ class ConfigMixin:
 
     @property
     def config(self) -> Dict[str, Any]:
+        """
+        Returns the config of the class as a frozen dictionary
+
+        Returns:
+            `Dict[str, Any]`: Config of the class.
+        """
         return self._internal_dict
 
     def to_json_string(self) -> str:
@@ -401,38 +507,6 @@ class ConfigMixin:
             writer.write(self.to_json_string())
 
 
-class FrozenDict(OrderedDict):
-    def __init__(self, *args, **kwargs):
-        super().__init__(*args, **kwargs)
-
-        for key, value in self.items():
-            setattr(self, key, value)
-
-        self.__frozen = True
-
-    def __delitem__(self, *args, **kwargs):
-        raise Exception(f"You cannot use ``__delitem__`` on a {self.__class__.__name__} instance.")
-
-    def setdefault(self, *args, **kwargs):
-        raise Exception(f"You cannot use ``setdefault`` on a {self.__class__.__name__} instance.")
-
-    def pop(self, *args, **kwargs):
-        raise Exception(f"You cannot use ``pop`` on a {self.__class__.__name__} instance.")
-
-    def update(self, *args, **kwargs):
-        raise Exception(f"You cannot use ``update`` on a {self.__class__.__name__} instance.")
-
-    def __setattr__(self, name, value):
-        if hasattr(self, "__frozen") and self.__frozen:
-            raise Exception(f"You cannot use ``__setattr__`` on a {self.__class__.__name__} instance.")
-        super().__setattr__(name, value)
-
-    def __setitem__(self, name, value):
-        if hasattr(self, "__frozen") and self.__frozen:
-            raise Exception(f"You cannot use ``__setattr__`` on a {self.__class__.__name__} instance.")
-        super().__setitem__(name, value)
-
-
 def register_to_config(init):
     r"""
     Decorator to apply on the init of classes inheriting from [`ConfigMixin`] so that all the arguments are

src/diffusers/pipeline_flax_utils.py

@@ -47,7 +47,7 @@ logger = logging.get_logger(__name__)
 LOADABLE_CLASSES = {
     "diffusers": {
         "FlaxModelMixin": ["save_pretrained", "from_pretrained"],
-        "FlaxSchedulerMixin": ["save_config", "from_config"],
+        "FlaxSchedulerMixin": ["save_pretrained", "from_pretrained"],
         "FlaxDiffusionPipeline": ["save_pretrained", "from_pretrained"],
     },
     "transformers": {
@@ -280,7 +280,7 @@ class FlaxDiffusionPipeline(ConfigMixin):
         >>> from diffusers import FlaxDPMSolverMultistepScheduler
 
         >>> model_id = "runwayml/stable-diffusion-v1-5"
-        >>> sched, sched_state = FlaxDPMSolverMultistepScheduler.from_config(
+        >>> sched, sched_state = FlaxDPMSolverMultistepScheduler.from_pretrained(
         ...     model_id,
         ...     subfolder="scheduler",
         ... )
@@ -303,7 +303,7 @@ class FlaxDiffusionPipeline(ConfigMixin):
         # 1. Download the checkpoints and configs
         # use snapshot download here to get it working from from_pretrained
         if not os.path.isdir(pretrained_model_name_or_path):
-            config_dict = cls.get_config_dict(
+            config_dict = cls.load_config(
                 pretrained_model_name_or_path,
                 cache_dir=cache_dir,
                 resume_download=resume_download,
@@ -349,7 +349,7 @@ class FlaxDiffusionPipeline(ConfigMixin):
         else:
             cached_folder = pretrained_model_name_or_path
 
-        config_dict = cls.get_config_dict(cached_folder)
+        config_dict = cls.load_config(cached_folder)
 
         # 2. Load the pipeline class, if using custom module then load it from the hub
         # if we load from explicit class, let's use it
@@ -370,7 +370,7 @@ class FlaxDiffusionPipeline(ConfigMixin):
         expected_modules = set(inspect.signature(pipeline_class.__init__).parameters.keys())
         passed_class_obj = {k: kwargs.pop(k) for k in expected_modules if k in kwargs}
 
-        init_dict, _ = pipeline_class.extract_init_dict(config_dict, **kwargs)
+        init_dict, _, _ = pipeline_class.extract_init_dict(config_dict, **kwargs)
 
         init_kwargs = {}
 

src/diffusers/pipeline_utils.py

@@ -65,7 +65,7 @@ logger = logging.get_logger(__name__)
 LOADABLE_CLASSES = {
     "diffusers": {
         "ModelMixin": ["save_pretrained", "from_pretrained"],
-        "SchedulerMixin": ["save_config", "from_config"],
+        "SchedulerMixin": ["save_pretrained", "from_pretrained"],
         "DiffusionPipeline": ["save_pretrained", "from_pretrained"],
         "OnnxRuntimeModel": ["save_pretrained", "from_pretrained"],
     },
@@ -207,7 +207,7 @@ class DiffusionPipeline(ConfigMixin):
         if torch_device is None:
             return self
 
-        module_names, _ = self.extract_init_dict(dict(self.config))
+        module_names, _, _ = self.extract_init_dict(dict(self.config))
         for name in module_names.keys():
             module = getattr(self, name)
             if isinstance(module, torch.nn.Module):
@@ -228,7 +228,7 @@ class DiffusionPipeline(ConfigMixin):
         Returns:
             `torch.device`: The torch device on which the pipeline is located.
         """
-        module_names, _ = self.extract_init_dict(dict(self.config))
+        module_names, _, _ = self.extract_init_dict(dict(self.config))
         for name in module_names.keys():
             module = getattr(self, name)
             if isinstance(module, torch.nn.Module):
@@ -377,11 +377,11 @@ class DiffusionPipeline(ConfigMixin):
         >>> # of the documentation](https://huggingface.co/docs/hub/security-tokens)
         >>> pipeline = DiffusionPipeline.from_pretrained("runwayml/stable-diffusion-v1-5")
 
-        >>> # Download pipeline, but overwrite scheduler
+        >>> # Use a different scheduler
         >>> from diffusers import LMSDiscreteScheduler
 
-        >>> scheduler = LMSDiscreteScheduler.from_config("runwayml/stable-diffusion-v1-5", subfolder="scheduler")
-        >>> pipeline = DiffusionPipeline.from_pretrained("runwayml/stable-diffusion-v1-5", scheduler=scheduler)
+        >>> scheduler = LMSDiscreteScheduler.from_config(pipeline.scheduler.config)
+        >>> pipeline.scheduler = scheduler
         ```
         """
         cache_dir = kwargs.pop("cache_dir", DIFFUSERS_CACHE)
@@ -428,7 +428,7 @@ class DiffusionPipeline(ConfigMixin):
         # 1. Download the checkpoints and configs
         # use snapshot download here to get it working from from_pretrained
         if not os.path.isdir(pretrained_model_name_or_path):
-            config_dict = cls.get_config_dict(
+            config_dict = cls.load_config(
                 pretrained_model_name_or_path,
                 cache_dir=cache_dir,
                 resume_download=resume_download,
@@ -474,7 +474,7 @@ class DiffusionPipeline(ConfigMixin):
         else:
             cached_folder = pretrained_model_name_or_path
 
-        config_dict = cls.get_config_dict(cached_folder)
+        config_dict = cls.load_config(cached_folder)
 
         # 2. Load the pipeline class, if using custom module then load it from the hub
         # if we load from explicit class, let's use it
@@ -513,7 +513,7 @@ class DiffusionPipeline(ConfigMixin):
         expected_modules = set(inspect.signature(pipeline_class.__init__).parameters.keys()) - set(["self"])
         passed_class_obj = {k: kwargs.pop(k) for k in expected_modules if k in kwargs}
 
-        init_dict, unused_kwargs = pipeline_class.extract_init_dict(config_dict, **kwargs)
+        init_dict, unused_kwargs, _ = pipeline_class.extract_init_dict(config_dict, **kwargs)
 
         if len(unused_kwargs) > 0:
             logger.warning(f"Keyword arguments {unused_kwargs} not recognized.")

src/diffusers/pipelines/ddpm/pipeline_ddpm.py

@@ -71,7 +71,7 @@ class DDPMPipeline(DiffusionPipeline):
         """
         message = (
             "Please make sure to instantiate your scheduler with `predict_epsilon` instead. E.g. `scheduler ="
-            " DDPMScheduler.from_config(<model_id>, predict_epsilon=True)`."
+            " DDPMScheduler.from_pretrained(<model_id>, predict_epsilon=True)`."
         )
         predict_epsilon = deprecate("predict_epsilon", "0.10.0", message, take_from=kwargs)
 

src/diffusers/pipelines/stable_diffusion/README.md

@@ -72,7 +72,7 @@ image.save("astronaut_rides_horse.png")
 # make sure you're logged in with `huggingface-cli login`
 from diffusers import StableDiffusionPipeline, DDIMScheduler
 
-scheduler =  DDIMScheduler.from_config("CompVis/stable-diffusion-v1-4", subfolder="scheduler")
+scheduler =  DDIMScheduler.from_pretrained("CompVis/stable-diffusion-v1-4", subfolder="scheduler")
 
 pipe = StableDiffusionPipeline.from_pretrained(
     "runwayml/stable-diffusion-v1-5", 
@@ -91,7 +91,7 @@ image.save("astronaut_rides_horse.png")
 # make sure you're logged in with `huggingface-cli login`
 from diffusers import StableDiffusionPipeline, LMSDiscreteScheduler
 
-lms = LMSDiscreteScheduler.from_config("CompVis/stable-diffusion-v1-4", subfolder="scheduler")
+lms = LMSDiscreteScheduler.from_pretrained("CompVis/stable-diffusion-v1-4", subfolder="scheduler")
 
 pipe = StableDiffusionPipeline.from_pretrained(
     "runwayml/stable-diffusion-v1-5", 
@@ -120,7 +120,7 @@ from diffusers import CycleDiffusionPipeline, DDIMScheduler
 # load the pipeline
 # make sure you're logged in with `huggingface-cli login`
 model_id_or_path = "CompVis/stable-diffusion-v1-4"
-scheduler = DDIMScheduler.from_config(model_id_or_path, subfolder="scheduler")
+scheduler = DDIMScheduler.from_pretrained(model_id_or_path, subfolder="scheduler")
 pipe = CycleDiffusionPipeline.from_pretrained(model_id_or_path, scheduler=scheduler).to("cuda")
 
 # let's download an initial image

src/diffusers/schedulers/scheduling_ddim.py

@@ -23,7 +23,7 @@ import numpy as np
 import torch
 
 from ..configuration_utils import ConfigMixin, register_to_config
-from ..utils import BaseOutput
+from ..utils import _COMPATIBLE_STABLE_DIFFUSION_SCHEDULERS, BaseOutput
 from .scheduling_utils import SchedulerMixin
 
 
@@ -82,8 +82,8 @@ class DDIMScheduler(SchedulerMixin, ConfigMixin):
 
     [`~ConfigMixin`] takes care of storing all config attributes that are passed in the scheduler's `__init__`
     function, such as `num_train_timesteps`. They can be accessed via `scheduler.config.num_train_timesteps`.
-    [`~ConfigMixin`] also provides general loading and saving functionality via the [`~ConfigMixin.save_config`] and
-    [`~ConfigMixin.from_config`] functions.
+    [`SchedulerMixin`] provides general loading and saving functionality via the [`SchedulerMixin.save_pretrained`] and
+    [`~SchedulerMixin.from_pretrained`] functions.
 
     For more details, see the original paper: https://arxiv.org/abs/2010.02502
 
@@ -109,14 +109,7 @@ class DDIMScheduler(SchedulerMixin, ConfigMixin):
 
     """
 
-    _compatible_classes = [
-        "PNDMScheduler",
-        "DDPMScheduler",
-        "LMSDiscreteScheduler",
-        "EulerDiscreteScheduler",
-        "EulerAncestralDiscreteScheduler",
-        "DPMSolverMultistepScheduler",
-    ]
+    _compatibles = _COMPATIBLE_STABLE_DIFFUSION_SCHEDULERS.copy()
 
     @register_to_config
     def __init__(

src/diffusers/schedulers/scheduling_ddim_flax.py

@@ -23,7 +23,12 @@ import flax
 import jax.numpy as jnp
 
 from ..configuration_utils import ConfigMixin, register_to_config
-from .scheduling_utils_flax import FlaxSchedulerMixin, FlaxSchedulerOutput, broadcast_to_shape_from_left
+from .scheduling_utils_flax import (
+    _FLAX_COMPATIBLE_STABLE_DIFFUSION_SCHEDULERS,
+    FlaxSchedulerMixin,
+    FlaxSchedulerOutput,
+    broadcast_to_shape_from_left,
+)
 
 
 def betas_for_alpha_bar(num_diffusion_timesteps, max_beta=0.999) -> jnp.ndarray:
@@ -79,8 +84,8 @@ class FlaxDDIMScheduler(FlaxSchedulerMixin, ConfigMixin):
 
     [`~ConfigMixin`] takes care of storing all config attributes that are passed in the scheduler's `__init__`
     function, such as `num_train_timesteps`. They can be accessed via `scheduler.config.num_train_timesteps`.
-    [`~ConfigMixin`] also provides general loading and saving functionality via the [`~ConfigMixin.save_config`] and
-    [`~ConfigMixin.from_config`] functions.
+    [`SchedulerMixin`] provides general loading and saving functionality via the [`SchedulerMixin.save_pretrained`] and
+    [`~SchedulerMixin.from_pretrained`] functions.
 
     For more details, see the original paper: https://arxiv.org/abs/2010.02502
 
@@ -105,6 +110,8 @@ class FlaxDDIMScheduler(FlaxSchedulerMixin, ConfigMixin):
             stable diffusion.
     """
 
+    _compatibles = _FLAX_COMPATIBLE_STABLE_DIFFUSION_SCHEDULERS.copy()
+
     @property
     def has_state(self):
         return True

src/diffusers/schedulers/scheduling_ddpm.py

@@ -22,7 +22,7 @@ import numpy as np
 import torch
 
 from ..configuration_utils import ConfigMixin, FrozenDict, register_to_config
-from ..utils import BaseOutput, deprecate
+from ..utils import _COMPATIBLE_STABLE_DIFFUSION_SCHEDULERS, BaseOutput, deprecate
 from .scheduling_utils import SchedulerMixin
 
 
@@ -80,8 +80,8 @@ class DDPMScheduler(SchedulerMixin, ConfigMixin):
 
     [`~ConfigMixin`] takes care of storing all config attributes that are passed in the scheduler's `__init__`
     function, such as `num_train_timesteps`. They can be accessed via `scheduler.config.num_train_timesteps`.
-    [`~ConfigMixin`] also provides general loading and saving functionality via the [`~ConfigMixin.save_config`] and
-    [`~ConfigMixin.from_config`] functions.
+    [`SchedulerMixin`] provides general loading and saving functionality via the [`SchedulerMixin.save_pretrained`] and
+    [`~SchedulerMixin.from_pretrained`] functions.
 
     For more details, see the original paper: https://arxiv.org/abs/2006.11239
 
@@ -104,14 +104,7 @@ class DDPMScheduler(SchedulerMixin, ConfigMixin):
 
     """
 
-    _compatible_classes = [
-        "DDIMScheduler",
-        "PNDMScheduler",
-        "LMSDiscreteScheduler",
-        "EulerDiscreteScheduler",
-        "EulerAncestralDiscreteScheduler",
-        "DPMSolverMultistepScheduler",
-    ]
+    _compatibles = _COMPATIBLE_STABLE_DIFFUSION_SCHEDULERS.copy()
 
     @register_to_config
     def __init__(
@@ -249,7 +242,7 @@ class DDPMScheduler(SchedulerMixin, ConfigMixin):
         """
         message = (
             "Please make sure to instantiate your scheduler with `predict_epsilon` instead. E.g. `scheduler ="
-            " DDPMScheduler.from_config(<model_id>, predict_epsilon=True)`."
+            " DDPMScheduler.from_pretrained(<model_id>, predict_epsilon=True)`."
         )
         predict_epsilon = deprecate("predict_epsilon", "0.10.0", message, take_from=kwargs)
         if predict_epsilon is not None and predict_epsilon != self.config.predict_epsilon:

src/diffusers/schedulers/scheduling_ddpm_flax.py

@@ -24,7 +24,12 @@ from jax import random
 
 from ..configuration_utils import ConfigMixin, FrozenDict, register_to_config
 from ..utils import deprecate
-from .scheduling_utils_flax import FlaxSchedulerMixin, FlaxSchedulerOutput, broadcast_to_shape_from_left
+from .scheduling_utils_flax import (
+    _FLAX_COMPATIBLE_STABLE_DIFFUSION_SCHEDULERS,
+    FlaxSchedulerMixin,
+    FlaxSchedulerOutput,
+    broadcast_to_shape_from_left,
+)
 
 
 def betas_for_alpha_bar(num_diffusion_timesteps, max_beta=0.999) -> jnp.ndarray:
@@ -79,8 +84,8 @@ class FlaxDDPMScheduler(FlaxSchedulerMixin, ConfigMixin):
 
     [`~ConfigMixin`] takes care of storing all config attributes that are passed in the scheduler's `__init__`
     function, such as `num_train_timesteps`. They can be accessed via `scheduler.config.num_train_timesteps`.
-    [`~ConfigMixin`] also provides general loading and saving functionality via the [`~ConfigMixin.save_config`] and
-    [`~ConfigMixin.from_config`] functions.
+    [`SchedulerMixin`] provides general loading and saving functionality via the [`SchedulerMixin.save_pretrained`] and
+    [`~SchedulerMixin.from_pretrained`] functions.
 
     For more details, see the original paper: https://arxiv.org/abs/2006.11239
 
@@ -103,6 +108,8 @@ class FlaxDDPMScheduler(FlaxSchedulerMixin, ConfigMixin):
 
     """
 
+    _compatibles = _FLAX_COMPATIBLE_STABLE_DIFFUSION_SCHEDULERS.copy()
+
     @property
     def has_state(self):
         return True
@@ -221,7 +228,7 @@ class FlaxDDPMScheduler(FlaxSchedulerMixin, ConfigMixin):
         """
         message = (
             "Please make sure to instantiate your scheduler with `predict_epsilon` instead. E.g. `scheduler ="
-            " DDPMScheduler.from_config(<model_id>, predict_epsilon=True)`."
+            " DDPMScheduler.from_pretrained(<model_id>, predict_epsilon=True)`."
         )
         predict_epsilon = deprecate("predict_epsilon", "0.10.0", message, take_from=kwargs)
         if predict_epsilon is not None and predict_epsilon != self.config.predict_epsilon:

src/diffusers/schedulers/scheduling_dpmsolver_multistep.py

@@ -21,6 +21,7 @@ import numpy as np
 import torch
 
 from ..configuration_utils import ConfigMixin, register_to_config
+from ..utils import _COMPATIBLE_STABLE_DIFFUSION_SCHEDULERS
 from .scheduling_utils import SchedulerMixin, SchedulerOutput
 
 
@@ -71,8 +72,8 @@ class DPMSolverMultistepScheduler(SchedulerMixin, ConfigMixin):
 
     [`~ConfigMixin`] takes care of storing all config attributes that are passed in the scheduler's `__init__`
     function, such as `num_train_timesteps`. They can be accessed via `scheduler.config.num_train_timesteps`.
-    [`~ConfigMixin`] also provides general loading and saving functionality via the [`~ConfigMixin.save_config`] and
-    [`~ConfigMixin.from_config`] functions.
+    [`SchedulerMixin`] provides general loading and saving functionality via the [`SchedulerMixin.save_pretrained`] and
+    [`~SchedulerMixin.from_pretrained`] functions.
 
     Args:
         num_train_timesteps (`int`): number of diffusion steps used to train the model.
@@ -116,14 +117,7 @@ class DPMSolverMultistepScheduler(SchedulerMixin, ConfigMixin):
 
     """
 
-    _compatible_classes = [
-        "DDIMScheduler",
-        "DDPMScheduler",
-        "PNDMScheduler",
-        "LMSDiscreteScheduler",
-        "EulerDiscreteScheduler",
-        "EulerAncestralDiscreteScheduler",
-    ]
+    _compatibles = _COMPATIBLE_STABLE_DIFFUSION_SCHEDULERS.copy()
 
     @register_to_config
     def __init__(

src/diffusers/schedulers/scheduling_dpmsolver_multistep_flax.py

@@ -23,7 +23,12 @@ import jax
 import jax.numpy as jnp
 
 from ..configuration_utils import ConfigMixin, register_to_config
-from .scheduling_utils_flax import FlaxSchedulerMixin, FlaxSchedulerOutput, broadcast_to_shape_from_left
+from .scheduling_utils_flax import (
+    _FLAX_COMPATIBLE_STABLE_DIFFUSION_SCHEDULERS,
+    FlaxSchedulerMixin,
+    FlaxSchedulerOutput,
+    broadcast_to_shape_from_left,
+)
 
 
 def betas_for_alpha_bar(num_diffusion_timesteps: int, max_beta=0.999) -> jnp.ndarray:
@@ -96,8 +101,8 @@ class FlaxDPMSolverMultistepScheduler(FlaxSchedulerMixin, ConfigMixin):
 
     [`~ConfigMixin`] takes care of storing all config attributes that are passed in the scheduler's `__init__`
     function, such as `num_train_timesteps`. They can be accessed via `scheduler.config.num_train_timesteps`.
-    [`~ConfigMixin`] also provides general loading and saving functionality via the [`~ConfigMixin.save_config`] and
-    [`~ConfigMixin.from_config`] functions.
+    [`SchedulerMixin`] provides general loading and saving functionality via the [`SchedulerMixin.save_pretrained`] and
+    [`~SchedulerMixin.from_pretrained`] functions.
 
     For more details, see the original paper: https://arxiv.org/abs/2206.00927 and https://arxiv.org/abs/2211.01095
 
@@ -143,6 +148,8 @@ class FlaxDPMSolverMultistepScheduler(FlaxSchedulerMixin, ConfigMixin):
 
     """
 
+    _compatibles = _FLAX_COMPATIBLE_STABLE_DIFFUSION_SCHEDULERS.copy()
+
     @property
     def has_state(self):
         return True