push

* add output_type=mesh

* update img2img

* make style

* add doc

* make style

* Apply suggestions from code review

Co-authored-by: Pedro Cuenca <pedro@huggingface.co>

* add docstring for output_type

* add a section in doc about hub mesh visualization/ rotation

* update conversion script so default background is white

* Apply suggestions from code review

Co-authored-by: Sayak Paul <spsayakpaul@gmail.com>

* Update src/diffusers/pipelines/shap_e/pipeline_shap_e_img2img.py

Co-authored-by: Sayak Paul <spsayakpaul@gmail.com>

* renderer -> shap_e_renderer

* img2img renderer -> shap_e_renderer

* fix tests

---------

Co-authored-by: yiyixuxu <yixu310@gmail,com>
Co-authored-by: Pedro Cuenca <pedro@huggingface.co>
Co-authored-by: Sayak Paul <spsayakpaul@gmail.com>

.github/workflows/pr_tests.yml

@@ -4,6 +4,9 @@ on:
   pull_request:
     branches:
       - main
+  push:
+    branches:
+      - ci-*
 
 concurrency:
   group: ${{ github.workflow }}-${{ github.head_ref || github.run_id }}

CONTRIBUTING.md

@@ -297,7 +297,7 @@ if you don't know yet what specific component you would like to add:
 - [Model or pipeline](https://github.com/huggingface/diffusers/issues?q=is%3Aopen+is%3Aissue+label%3A%22New+pipeline%2Fmodel%22)
 - [Scheduler](https://github.com/huggingface/diffusers/issues?q=is%3Aopen+is%3Aissue+label%3A%22New+scheduler%22)
 
-Before adding any of the three components, it is strongly recommended that you give the [Philosophy guide](https://github.com/huggingface/diffusers/issues?q=is%3Aopen+is%3Aissue+label%3A%22Good+second+issue%22) a read to better understand the design of any of the three components. Please be aware that
+Before adding any of the three components, it is strongly recommended that you give the [Philosophy guide](https://github.com/huggingface/diffusers/blob/main/PHILOSOPHY.md) a read to better understand the design of any of the three components. Please be aware that
 we cannot merge model, scheduler, or pipeline additions that strongly diverge from our design philosophy
 as it will lead to API inconsistencies. If you fundamentally disagree with a design choice, please
 open a [Feedback issue](https://github.com/huggingface/diffusers/issues/new?assignees=&labels=&template=feedback.md&title=) instead so that it can be discussed whether a certain design

README.md

@@ -143,9 +143,14 @@ just hang out ☕.
   </tr>
   <tr>
     <td>Text-to-Image</td>
-    <td><a href="https://huggingface.co/docs/diffusers/api/pipelines/if">if</a></td>
+    <td><a href="https://huggingface.co/docs/diffusers/api/pipelines/if">DeepFloyd IF</a></td>
       <td><a href="https://huggingface.co/DeepFloyd/IF-I-XL-v1.0"> DeepFloyd/IF-I-XL-v1.0 </a></td>
   </tr>
+  <tr>
+    <td>Text-to-Image</td>
+    <td><a href="https://huggingface.co/docs/diffusers/api/pipelines/kandinsky">Kandinsky</a></td>
+      <td><a href="https://huggingface.co/kandinsky-community/kandinsky-2-2-decoder"> kandinsky-community/kandinsky-2-2-decoder </a></td>
+  </tr>
   <tr style="border-top: 2px solid black">
     <td>Text-guided Image-to-Image</td>
     <td><a href="https://huggingface.co/docs/diffusers/api/pipelines/stable_diffusion/controlnet">Controlnet</a></td>
@@ -153,7 +158,7 @@ just hang out ☕.
   </tr>
   <tr>
     <td>Text-guided Image-to-Image</td>
-    <td><a href="https://huggingface.co/docs/diffusers/api/pipelines/stable_diffusion/pix2pix">Instruct Pix2Pix</a></td>
+    <td><a href="https://huggingface.co/docs/diffusers/api/pipelines/pix2pix">Instruct Pix2Pix</a></td>
       <td><a href="https://huggingface.co/timbrooks/instruct-pix2pix"> timbrooks/instruct-pix2pix </a></td>
   </tr>
   <tr>

docs/source/en/_toctree.yml

@@ -117,6 +117,8 @@
     title: Habana Gaudi
   - local: optimization/tome
     title: Token Merging
+  - local: optimization/bentoml
+    title: BentoML Integration
   title: Optimization/Special Hardware
 - sections:
   - local: conceptual/philosophy
@@ -164,6 +166,8 @@
       title: VQModel
     - local: api/models/autoencoderkl
       title: AutoencoderKL
+    - local: api/models/asymmetricautoencoderkl
+      title: AsymmetricAutoencoderKL
     - local: api/models/transformer2d
       title: Transformer2D
     - local: api/models/transformer_temporal
@@ -196,12 +200,12 @@
       title: DDIM
     - local: api/pipelines/ddpm
       title: DDPM
+    - local: api/pipelines/deepfloyd_if
+      title: DeepFloyd IF
     - local: api/pipelines/diffedit
       title: DiffEdit
     - local: api/pipelines/dit
       title: DiT
-    - local: api/pipelines/if
-      title: IF
     - local: api/pipelines/pix2pix
       title: InstructPix2Pix
     - local: api/pipelines/kandinsky
@@ -255,6 +259,8 @@
         title: Super-Resolution
       - local: api/pipelines/stable_diffusion/ldm3d_diffusion
         title: LDM3D Text-to-(RGB, Depth)
+      - local: api/pipelines/stable_diffusion/adapter
+        title: Stable Diffusion T2I-adapter
       title: Stable Diffusion
     - local: api/pipelines/stable_unclip
       title: Stable unCLIP

docs/source/en/api/loaders.mdx

@@ -35,3 +35,11 @@ Adapters (textual inversion, LoRA, hypernetworks) allow you to modify a diffusio
 ## FromSingleFileMixin
 
 [[autodoc]] loaders.FromSingleFileMixin
+
+## FromOriginalControlnetMixin
+
+[[autodoc]] loaders.FromOriginalControlnetMixin
+
+## FromOriginalVAEMixin
+
+[[autodoc]] loaders.FromOriginalVAEMixin

docs/source/en/api/models/asymmetricautoencoderkl.mdx

@@ -0,0 +1,55 @@
+# AsymmetricAutoencoderKL
+
+Improved larger variational autoencoder (VAE) model with KL loss for inpainting task: [Designing a Better Asymmetric VQGAN for StableDiffusion](https://arxiv.org/abs/2306.04632) by Zixin Zhu, Xuelu Feng, Dongdong Chen, Jianmin Bao, Le Wang, Yinpeng Chen, Lu Yuan, Gang Hua.
+
+The abstract from the paper is:
+
+*StableDiffusion is a revolutionary text-to-image generator that is causing a stir in the world of image generation and editing. Unlike traditional methods that learn a diffusion model in pixel space, StableDiffusion learns a diffusion model in the latent space via a VQGAN, ensuring both efficiency and quality. It not only supports image generation tasks, but also enables image editing for real images, such as image inpainting and local editing. However, we have observed that the vanilla VQGAN used in StableDiffusion leads to significant information loss, causing distortion artifacts even in non-edited image regions. To this end, we propose a new asymmetric VQGAN with two simple designs. Firstly, in addition to the input from the encoder, the decoder contains a conditional branch that incorporates information from task-specific priors, such as the unmasked image region in inpainting. Secondly, the decoder is much heavier than the encoder, allowing for more detailed recovery while only slightly increasing the total inference cost. The training cost of our asymmetric VQGAN is cheap, and we only need to retrain a new asymmetric decoder while keeping the vanilla VQGAN encoder and StableDiffusion unchanged. Our asymmetric VQGAN can be widely used in StableDiffusion-based inpainting and local editing methods. Extensive experiments demonstrate that it can significantly improve the inpainting and editing performance, while maintaining the original text-to-image capability. The code is available at https://github.com/buxiangzhiren/Asymmetric_VQGAN*
+
+Evaluation results can be found in section 4.1 of the original paper. 
+
+## Available checkpoints
+
+* [https://huggingface.co/cross-attention/asymmetric-autoencoder-kl-x-1-5](https://huggingface.co/cross-attention/asymmetric-autoencoder-kl-x-1-5)
+* [https://huggingface.co/cross-attention/asymmetric-autoencoder-kl-x-2](https://huggingface.co/cross-attention/asymmetric-autoencoder-kl-x-2)
+
+## Example Usage
+
+```python
+from io import BytesIO
+from PIL import Image
+import requests
+from diffusers import AsymmetricAutoencoderKL, StableDiffusionInpaintPipeline
+
+
+def download_image(url: str) -> Image.Image:
+    response = requests.get(url)
+    return Image.open(BytesIO(response.content)).convert("RGB")
+
+
+prompt = "a photo of a person"
+img_url = "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/repaint/celeba_hq_256.png"
+mask_url = "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/repaint/mask_256.png"
+
+image = download_image(img_url).resize((256, 256))
+mask_image = download_image(mask_url).resize((256, 256))
+
+pipe = StableDiffusionInpaintPipeline.from_pretrained("runwayml/stable-diffusion-inpainting")
+pipe.vae = AsymmetricAutoencoderKL.from_pretrained("cross-attention/asymmetric-autoencoder-kl-x-1-5")
+pipe.to("cuda")
+
+image = pipe(prompt=prompt, image=image, mask_image=mask_image).images[0]
+image.save("image.jpeg")
+```
+
+## AsymmetricAutoencoderKL
+
+[[autodoc]] models.autoencoder_asym_kl.AsymmetricAutoencoderKL
+
+## AutoencoderKLOutput
+
+[[autodoc]] models.autoencoder_kl.AutoencoderKLOutput
+
+## DecoderOutput
+
+[[autodoc]] models.vae.DecoderOutput

docs/source/en/api/models/autoencoderkl.mdx

@@ -6,6 +6,18 @@ The abstract from the paper is:
 
 *How can we perform efficient inference and learning in directed probabilistic models, in the presence of continuous latent variables with intractable posterior distributions, and large datasets? We introduce a stochastic variational inference and learning algorithm that scales to large datasets and, under some mild differentiability conditions, even works in the intractable case. Our contributions are two-fold. First, we show that a reparameterization of the variational lower bound yields a lower bound estimator that can be straightforwardly optimized using standard stochastic gradient methods. Second, we show that for i.i.d. datasets with continuous latent variables per datapoint, posterior inference can be made especially efficient by fitting an approximate inference model (also called a recognition model) to the intractable posterior using the proposed lower bound estimator. Theoretical advantages are reflected in experimental results.*
 
+## Loading from the original format
+
+By default the [`AutoencoderKL`] should be loaded with [`~ModelMixin.from_pretrained`], but it can also be loaded
+from the original format using [`FromOriginalVAEMixin.from_single_file`] as follows:
+
+```py
+from diffusers import AutoencoderKL
+
+url = "https://huggingface.co/stabilityai/sd-vae-ft-mse-original/blob/main/vae-ft-mse-840000-ema-pruned.safetensors"  # can also be local file
+model = AutoencoderKL.from_single_file(url)
+```
+
 ## AutoencoderKL
 
 [[autodoc]] AutoencoderKL
@@ -28,4 +40,4 @@ The abstract from the paper is:
 
 ## FlaxDecoderOutput
 
-[[autodoc]] models.vae_flax.FlaxDecoderOutput
+[[autodoc]] models.vae_flax.FlaxDecoderOutput

docs/source/en/api/models/controlnet.mdx

@@ -6,6 +6,21 @@ The abstract from the paper is:
 
 *We present a neural network structure, ControlNet, to control pretrained large diffusion models to support additional input conditions. The ControlNet learns task-specific conditions in an end-to-end way, and the learning is robust even when the training dataset is small (< 50k). Moreover, training a ControlNet is as fast as fine-tuning a diffusion model, and the model can be trained on a personal devices. Alternatively, if powerful computation clusters are available, the model can scale to large amounts (millions to billions) of data. We report that large diffusion models like Stable Diffusion can be augmented with ControlNets to enable conditional inputs like edge maps, segmentation maps, keypoints, etc. This may enrich the methods to control large diffusion models and further facilitate related applications.*
 
+## Loading from the original format
+
+By default the [`ControlNetModel`] should be loaded with [`~ModelMixin.from_pretrained`], but it can also be loaded
+from the original format using [`FromOriginalControlnetMixin.from_single_file`] as follows:
+
+```py
+from diffusers import StableDiffusionControlnetPipeline, ControlNetModel
+
+url = "https://huggingface.co/lllyasviel/ControlNet-v1-1/blob/main/control_v11p_sd15_canny.pth"  # can also be a local path
+controlnet = ControlNetModel.from_single_file(url)
+
+url = "https://huggingface.co/runwayml/stable-diffusion-v1-5/blob/main/v1-5-pruned.safetensors"  # can also be a local path
+pipe = StableDiffusionControlnetPipeline.from_single_file(url, controlnet=controlnet)
+```
+
 ## ControlNetModel
 
 [[autodoc]] ControlNetModel
@@ -20,4 +35,4 @@ The abstract from the paper is:
 
 ## FlaxControlNetOutput
 
-[[autodoc]] models.controlnet_flax.FlaxControlNetOutput
+[[autodoc]] models.controlnet_flax.FlaxControlNetOutput

docs/source/en/api/pipelines/deepfloyd_if.mdx

@@ -10,7 +10,7 @@ an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express o
 specific language governing permissions and limitations under the License.
 -->
 
-# IF
+# DeepFloyd IF 
 
 ## Overview
 

docs/source/en/api/pipelines/diffedit.mdx

@@ -71,7 +71,7 @@ First, let's load our pipeline:
 
 ```py
 import torch
-from diffusers import DDIMScheduler, DDIMInverseScheduler, StableDiffusionPix2PixZeroPipeline
+from diffusers import DDIMScheduler, DDIMInverseScheduler, StableDiffusionDiffEditPipeline
 
 sd_model_ckpt = "stabilityai/stable-diffusion-2-1"
 pipeline = StableDiffusionDiffEditPipeline.from_pretrained(
@@ -357,4 +357,4 @@ images[0].save("edited_image.png")
     - all
     - generate_mask
     - invert
-	- __call__
+	- __call__

docs/source/en/api/pipelines/overview.mdx

@@ -66,6 +66,7 @@ available a colab notebook to directly try them out.
 | [score_sde_ve](./score_sde_ve) | [**Score-Based Generative Modeling through Stochastic Differential Equations**](https://openreview.net/forum?id=PxTIG12RRHS) | Unconditional Image Generation | 
 | [score_sde_vp](./score_sde_vp) | [**Score-Based Generative Modeling through Stochastic Differential Equations**](https://openreview.net/forum?id=PxTIG12RRHS) | Unconditional Image Generation | 
 | [semantic_stable_diffusion](./semantic_stable_diffusion) | [**SEGA: Instructing Diffusion using Semantic Dimensions**](https://arxiv.org/abs/2301.12247) | Text-to-Image Generation |
+| [stable_diffusion_adapter](./stable_diffusion/adapter) | [**T2I-Adapter**](https://arxiv.org/abs/2302.08453) | Image-to-Image Text-Guided Generation with Adapters | -
 | [stable_diffusion_text2img](./stable_diffusion/text2img) | [**Stable Diffusion**](https://stability.ai/blog/stable-diffusion-public-release) | Text-to-Image Generation | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/huggingface/notebooks/blob/main/diffusers/training_example.ipynb)
 | [stable_diffusion_img2img](./stable_diffusion/img2img) | [**Stable Diffusion**](https://stability.ai/blog/stable-diffusion-public-release) | Image-to-Image Text-Guided Generation | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/huggingface/notebooks/blob/main/diffusers/image_2_image_using_diffusers.ipynb)
 | [stable_diffusion_inpaint](./stable_diffusion/inpaint) | [**Stable Diffusion**](https://stability.ai/blog/stable-diffusion-public-release) | Text-Guided Image Inpainting | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/huggingface/notebooks/blob/main/diffusers/in_painting_with_stable_diffusion_using_diffusers.ipynb)

docs/source/en/api/pipelines/panorama.mdx

@@ -60,6 +60,25 @@ and increase the VRAM usage.
 
 </Tip>
 
+<Tip>
+
+Circular padding is applied to ensure there are no stitching artifacts when working with 
+panoramas that needs to seamlessly transition from the rightmost part to the leftmost part. 
+By enabling circular padding (set `circular_padding=True`), the operation applies additional 
+crops after the rightmost point of the image, allowing the model to "see” the transition 
+from the rightmost part to the leftmost part. This helps maintain visual consistency in 
+a 360-degree sense and creates a proper “panorama” that can be viewed using 360-degree 
+panorama viewers. When decoding latents in StableDiffusion, circular padding is applied 
+to ensure that the decoded latents match in the RGB space.
+
+Without circular padding, there is a stitching artifact (default):
+![img](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/indoor_%20no_circular_padding.png)
+
+With circular padding, the right and the left parts are matching (`circular_padding=True`):
+![img](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/indoor_%20circular_padding.png)
+
+</Tip>
+
 ## StableDiffusionPanoramaPipeline
 [[autodoc]] StableDiffusionPanoramaPipeline
 	- __call__

docs/source/en/api/pipelines/shap_e.mdx

@@ -128,6 +128,63 @@ gif_path = export_to_gif(images[0], "burger_3d.gif")
 ```
 ![img](https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/shap_e/burger_out.gif)
 
+### Generate mesh
+
+For both [`ShapEPipeline`] and [`ShapEImg2ImgPipeline`], you can generate mesh output by passing `output_type` as `mesh` to the pipeline, and then use the [`ShapEPipeline.export_to_ply`] utility function to save the output as a `ply` file. We also provide a [`ShapEPipeline.export_to_obj`] function that you can use to save mesh outputs as `obj` files.
+
+```python
+import torch
+
+from diffusers import DiffusionPipeline
+from diffusers.utils import export_to_ply
+
+device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+
+repo = "openai/shap-e"
+pipe = DiffusionPipeline.from_pretrained(repo, torch_dtype=torch.float16, variant="fp16")
+pipe = pipe.to(device)
+
+guidance_scale = 15.0
+prompt = "A birthday cupcake"
+
+images = pipe(prompt, guidance_scale=guidance_scale, num_inference_steps=64, frame_size=256, output_type="mesh").images
+
+ply_path = export_to_ply(images[0], "3d_cake.ply")
+print(f"saved to folder: {ply_path}")
+```
+
+Huggingface Datasets supports mesh visualization for mesh files in `glb` format. Below we will show you how to convert your mesh file into `glb` format so that you can use the Dataset viewer to render 3D objects. 
+
+We need to install `trimesh` library.
+
+```
+pip install trimesh
+```
+
+To convert the mesh file into `glb` format, 
+
+```python
+import trimesh
+
+mesh = trimesh.load("3d_cake.ply")
+mesh.export("3d_cake.glb", file_type="glb")
+```
+
+By default, the mesh output of Shap-E is from the bottom viewpoint; you can change the default viewpoint by applying a rotation transformation
+
+```python
+import trimesh
+import numpy as np
+
+mesh = trimesh.load("3d_cake.ply")
+rot = trimesh.transformations.rotation_matrix(-np.pi / 2, [1, 0, 0])
+mesh = mesh.apply_transform(rot)
+mesh.export("3d_cake.glb", file_type="glb")
+```
+
+Now you can upload your mesh file to your dataset and visualize it! Here is the link to the 3D cake we just generated
+https://huggingface.co/datasets/hf-internal-testing/diffusers-images/blob/main/shap_e/3d_cake.glb
+
 ## ShapEPipeline
 [[autodoc]] ShapEPipeline
 	- all

docs/source/en/api/pipelines/stable_diffusion/adapter.mdx

@@ -0,0 +1,187 @@
+<!--Copyright 2023 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.
+-->
+
+# Text-to-Image Generation with Adapter Conditioning
+
+## Overview
+
+[T2I-Adapter: Learning Adapters to Dig out More Controllable Ability for Text-to-Image Diffusion Models](https://arxiv.org/abs/2302.08453) by Chong Mou, Xintao Wang, Liangbin Xie, Jian Zhang, Zhongang Qi, Ying Shan, Xiaohu Qie.
+
+Using the pretrained models we can provide control images (for example, a depth map) to control Stable Diffusion text-to-image generation so that it follows the structure of the depth image and fills in the details.
+
+The abstract of the paper is the following:
+
+*The incredible generative ability of large-scale text-to-image (T2I) models has demonstrated strong power of learning complex structures and meaningful semantics. However, relying solely on text prompts cannot fully take advantage of the knowledge learned by the model, especially when flexible and accurate structure control is needed. In this paper, we aim to ``dig out" the capabilities that T2I models have implicitly learned, and then explicitly use them to control the generation more granularly. Specifically, we propose to learn simple and small T2I-Adapters to align internal knowledge in T2I models with external control signals, while freezing the original large T2I models. In this way, we can train various adapters according to different conditions, and achieve rich control and editing effects. Further, the proposed T2I-Adapters have attractive properties of practical value, such as composability and generalization ability. Extensive experiments demonstrate that our T2I-Adapter has promising generation quality and a wide range of applications.*
+
+This model was contributed by the community contributor [HimariO](https://github.com/HimariO) ❤️ .
+
+## Available Pipelines:
+
+| Pipeline | Tasks | Demo
+|---|---|:---:|
+| [StableDiffusionAdapterPipeline](https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_adapter.py) | *Text-to-Image Generation with T2I-Adapter Conditioning* | -
+
+## Usage example
+
+In the following we give a simple example of how to use a *T2IAdapter* checkpoint with Diffusers for inference.
+All adapters use the same pipeline.
+
+ 1. Images are first converted into the appropriate *control image* format.
+ 2. The *control image* and *prompt* are passed to the [`StableDiffusionAdapterPipeline`].
+
+Let's have a look at a simple example using the [Color Adapter](https://huggingface.co/TencentARC/t2iadapter_color_sd14v1).
+
+```python
+from diffusers.utils import load_image
+
+image = load_image("https://huggingface.co/datasets/diffusers/docs-images/resolve/main/t2i-adapter/color_ref.png")
+```
+
+![img](https://huggingface.co/datasets/diffusers/docs-images/resolve/main/t2i-adapter/color_ref.png)
+
+
+Then we can create our color palette by simply resizing it to 8 by 8 pixels and then scaling it back to original size.
+
+```python
+from PIL import Image
+
+color_palette = image.resize((8, 8))
+color_palette = color_palette.resize((512, 512), resample=Image.Resampling.NEAREST)
+```
+
+Let's take a look at the processed image.
+
+![img](https://huggingface.co/datasets/diffusers/docs-images/resolve/main/t2i-adapter/color_palette.png)
+
+
+Next, create the adapter pipeline
+
+```py
+import torch
+from diffusers import StableDiffusionAdapterPipeline, T2IAdapter
+
+adapter = T2IAdapter.from_pretrained("TencentARC/t2iadapter_color_sd14v1")
+pipe = StableDiffusionAdapterPipeline.from_pretrained(
+    "CompVis/stable-diffusion-v1-4",
+    adapter=adapter,
+    torch_dtype=torch.float16,
+)
+pipe.to("cuda")
+```
+
+Finally, pass the prompt and control image to the pipeline
+
+```py
+# fix the random seed, so you will get the same result as the example
+generator = torch.manual_seed(7)
+
+out_image = pipe(
+    "At night, glowing cubes in front of the beach",
+    image=color_palette,
+    generator=generator,
+).images[0]
+```
+
+![img](https://huggingface.co/datasets/diffusers/docs-images/resolve/main/t2i-adapter/color_output.png)
+
+
+## Available checkpoints
+
+Non-diffusers checkpoints can be found under [TencentARC/T2I-Adapter](https://huggingface.co/TencentARC/T2I-Adapter/tree/main/models).
+
+### T2I-Adapter with Stable Diffusion 1.4
+
+| Model Name | Control Image Overview| Control Image Example | Generated Image Example |
+|---|---|---|---|
+|[TencentARC/t2iadapter_color_sd14v1](https://huggingface.co/TencentARC/t2iadapter_color_sd14v1)<br/> *Trained with spatial color palette* | A image with 8x8 color palette.|<a href="https://huggingface.co/datasets/diffusers/docs-images/resolve/main/t2i-adapter/color_sample_input.png"><img width="64" style="margin:0;padding:0;" src="https://huggingface.co/datasets/diffusers/docs-images/resolve/main/t2i-adapter/color_sample_input.png"/></a>|<a href="https://huggingface.co/datasets/diffusers/docs-images/resolve/main/t2i-adapter/color_sample_output.png"><img width="64" src="https://huggingface.co/datasets/diffusers/docs-images/resolve/main/t2i-adapter/color_sample_output.png"/></a>|
+|[TencentARC/t2iadapter_canny_sd14v1](https://huggingface.co/TencentARC/t2iadapter_canny_sd14v1)<br/> *Trained with canny edge detection* | A monochrome image with white edges on a black background.|<a href="https://huggingface.co/datasets/diffusers/docs-images/resolve/main/t2i-adapter/canny_sample_input.png"><img width="64" style="margin:0;padding:0;" src="https://huggingface.co/datasets/diffusers/docs-images/resolve/main/t2i-adapter/canny_sample_input.png"/></a>|<a href="https://huggingface.co/datasets/diffusers/docs-images/resolve/main/t2i-adapter/canny_sample_output.png"><img width="64" src="https://huggingface.co/datasets/diffusers/docs-images/resolve/main/t2i-adapter/canny_sample_output.png"/></a>|
+|[TencentARC/t2iadapter_sketch_sd14v1](https://huggingface.co/TencentARC/t2iadapter_sketch_sd14v1)<br/> *Trained with [PidiNet](https://github.com/zhuoinoulu/pidinet) edge detection* | A hand-drawn monochrome image with white outlines on a black background.|<a href="https://huggingface.co/datasets/diffusers/docs-images/resolve/main/t2i-adapter/sketch_sample_input.png"><img width="64" style="margin:0;padding:0;" src="https://huggingface.co/datasets/diffusers/docs-images/resolve/main/t2i-adapter/sketch_sample_input.png"/></a>|<a href="https://huggingface.co/datasets/diffusers/docs-images/resolve/main/t2i-adapter/sketch_sample_output.png"><img width="64" src="https://huggingface.co/datasets/diffusers/docs-images/resolve/main/t2i-adapter/sketch_sample_output.png"/></a>|
+|[TencentARC/t2iadapter_depth_sd14v1](https://huggingface.co/TencentARC/t2iadapter_depth_sd14v1)<br/> *Trained with Midas depth estimation*  | A grayscale image with black representing deep areas and white representing shallow areas.|<a href="https://huggingface.co/datasets/diffusers/docs-images/resolve/main/t2i-adapter/depth_sample_input.png"><img width="64" src="https://huggingface.co/datasets/diffusers/docs-images/resolve/main/t2i-adapter/depth_sample_input.png"/></a>|<a href="https://huggingface.co/datasets/diffusers/docs-images/resolve/main/t2i-adapter/depth_sample_output.png"><img width="64" src="https://huggingface.co/datasets/diffusers/docs-images/resolve/main/t2i-adapter/depth_sample_output.png"/></a>|
+|[TencentARC/t2iadapter_openpose_sd14v1](https://huggingface.co/TencentARC/t2iadapter_openpose_sd14v1)<br/> *Trained with OpenPose bone image*  | A [OpenPose bone](https://github.com/CMU-Perceptual-Computing-Lab/openpose) image.|<a href="https://huggingface.co/datasets/diffusers/docs-images/resolve/main/t2i-adapter/openpose_sample_input.png"><img width="64" src="https://huggingface.co/datasets/diffusers/docs-images/resolve/main/t2i-adapter/openpose_sample_input.png"/></a>|<a href="https://huggingface.co/datasets/diffusers/docs-images/resolve/main/t2i-adapter/openpose_sample_output.png"><img width="64" src="https://huggingface.co/datasets/diffusers/docs-images/resolve/main/t2i-adapter/openpose_sample_output.png"/></a>|
+|[TencentARC/t2iadapter_keypose_sd14v1](https://huggingface.co/TencentARC/t2iadapter_keypose_sd14v1)<br/> *Trained with mmpose skeleton image*  | A [mmpose skeleton](https://github.com/open-mmlab/mmpose) image.|<a href="https://huggingface.co/datasets/diffusers/docs-images/resolve/main/t2i-adapter/keypose_sample_input.png"><img width="64" src="https://huggingface.co/datasets/diffusers/docs-images/resolve/main/t2i-adapter/keypose_sample_input.png"/></a>|<a href="https://huggingface.co/datasets/diffusers/docs-images/resolve/main/t2i-adapter/keypose_sample_output.png"><img width="64" src="https://huggingface.co/datasets/diffusers/docs-images/resolve/main/t2i-adapter/keypose_sample_output.png"/></a>|
+|[TencentARC/t2iadapter_seg_sd14v1](https://huggingface.co/TencentARC/t2iadapter_seg_sd14v1)<br/>*Trained with semantic segmentation*  | An [custom](https://github.com/TencentARC/T2I-Adapter/discussions/25) segmentation protocol image.|<a href="https://huggingface.co/datasets/diffusers/docs-images/resolve/main/t2i-adapter/seg_sample_input.png"><img width="64" src="https://huggingface.co/datasets/diffusers/docs-images/resolve/main/t2i-adapter/seg_sample_input.png"/></a>|<a href="https://huggingface.co/datasets/diffusers/docs-images/resolve/main/t2i-adapter/seg_sample_output.png"><img width="64" src="https://huggingface.co/datasets/diffusers/docs-images/resolve/main/t2i-adapter/seg_sample_output.png"/></a> |
+|[TencentARC/t2iadapter_canny_sd15v2](https://huggingface.co/TencentARC/t2iadapter_canny_sd15v2)||
+|[TencentARC/t2iadapter_depth_sd15v2](https://huggingface.co/TencentARC/t2iadapter_depth_sd15v2)||
+|[TencentARC/t2iadapter_sketch_sd15v2](https://huggingface.co/TencentARC/t2iadapter_sketch_sd15v2)||
+|[TencentARC/t2iadapter_zoedepth_sd15v1](https://huggingface.co/TencentARC/t2iadapter_zoedepth_sd15v1)||
+
+## Combining multiple adapters
+
+[`MultiAdapter`] can be used for applying multiple conditionings at once.
+
+Here we use the keypose adapter for the character posture and the depth adapter for creating the scene.
+
+```py
+import torch
+from PIL import Image
+from diffusers.utils import load_image
+
+cond_keypose = load_image(
+    "https://huggingface.co/datasets/diffusers/docs-images/resolve/main/t2i-adapter/keypose_sample_input.png"
+)
+cond_depth = load_image(
+    "https://huggingface.co/datasets/diffusers/docs-images/resolve/main/t2i-adapter/depth_sample_input.png"
+)
+cond = [[cond_keypose, cond_depth]]
+
+prompt = ["A man walking in an office room with a nice view"]
+```
+
+The two control images look as such:
+
+![img](https://huggingface.co/datasets/diffusers/docs-images/resolve/main/t2i-adapter/keypose_sample_input.png)
+![img](https://huggingface.co/datasets/diffusers/docs-images/resolve/main/t2i-adapter/depth_sample_input.png)
+
+
+`MultiAdapter` combines keypose and depth adapters. 
+
+`adapter_conditioning_scale` balances the relative influence of the different adapters.
+
+```py
+from diffusers import StableDiffusionAdapterPipeline, MultiAdapter
+
+adapters = MultiAdapter(
+    [
+        T2IAdapter.from_pretrained("TencentARC/t2iadapter_keypose_sd14v1"),
+        T2IAdapter.from_pretrained("TencentARC/t2iadapter_depth_sd14v1"),
+    ]
+)
+adapters = adapters.to(torch.float16)
+
+pipe = StableDiffusionAdapterPipeline.from_pretrained(
+    "CompVis/stable-diffusion-v1-4",
+    torch_dtype=torch.float16,
+    adapter=adapters,
+)
+
+images = pipe(prompt, cond, adapter_conditioning_scale=[0.8, 0.8])
+```
+
+![img](https://huggingface.co/datasets/diffusers/docs-images/resolve/main/t2i-adapter/keypose_depth_sample_output.png)
+
+
+## T2I Adapter vs ControlNet
+
+T2I-Adapter is similar to [ControlNet](https://huggingface.co/docs/diffusers/main/en/api/pipelines/controlnet). 
+T2i-Adapter uses a smaller auxiliary network which is only run once for the entire diffusion process. 
+However, T2I-Adapter performs slightly worse than ControlNet. 
+
+## StableDiffusionAdapterPipeline
+[[autodoc]] StableDiffusionAdapterPipeline
+	- all
+	- __call__
+	- enable_attention_slicing
+	- disable_attention_slicing
+	- enable_vae_slicing
+	- disable_vae_slicing
+	- enable_xformers_memory_efficient_attention
+	- disable_xformers_memory_efficient_attention

docs/source/en/api/pipelines/stable_diffusion/stable_diffusion_xl.mdx

@@ -21,7 +21,7 @@ The abstract of the paper is the following:
 ## Tips
 
 - Stable Diffusion XL works especially well with images between 768 and 1024.
-- Stable Diffusion XL output image can be improved by making use of a refiner as shown below
+- Stable Diffusion XL output image can be improved by making use of a refiner as shown below.
 
 ### Available checkpoints:
 
@@ -37,10 +37,10 @@ You can install the libraries as follows:
 pip install transformers
 pip install accelerate
 pip install safetensors
-pip install invisible-watermark>=2.0
+pip install invisible-watermark>=0.2.0
 ```
 
-### *Text-to-Image*
+### Text-to-Image
 
 You can use SDXL as follows for *text-to-image*:
 
@@ -57,22 +57,163 @@ prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k"
 image = pipe(prompt=prompt).images[0]
 ```
 
-### Refining the image output
+### Image-to-image 
 
-The image can be refined by making use of [stabilityai/stable-diffusion-xl-refiner-0.9](https://huggingface.co/stabilityai/stable-diffusion-xl-refiner-0.9).
-In this case, you only have to output the `latents` from the base model.
+You can use SDXL as follows for *image-to-image*:
 
-```py
-from diffusers import StableDiffusionXLPipeline, StableDiffusionXLImg2ImgPipeline
+```py 
 import torch
+from diffusers import StableDiffusionXLImg2ImgPipeline
+from diffusers.utils import load_image
 
-pipe = StableDiffusionXLPipeline.from_pretrained(
+pipe = StableDiffusionXLImg2ImgPipeline.from_pretrained(
+    "stabilityai/stable-diffusion-xl-refiner-0.9", torch_dtype=torch.float16, variant="fp16", use_safetensors=True
+)
+pipe = pipe.to("cuda")
+url = "https://huggingface.co/datasets/patrickvonplaten/images/resolve/main/aa_xl/000000009.png"
+
+init_image = load_image(url).convert("RGB")
+prompt = "a photo of an astronaut riding a horse on mars"
+image = pipe(prompt, image=init_image).images[0]
+```
+
+### Inpainting
+
+You can use SDXL as follows for *inpainting*
+
+```py 
+import torch
+from diffusers import StableDiffusionXLInpaintPipeline
+from diffusers.utils import load_image
+
+pipe = StableDiffusionXLInpaintPipeline.from_pretrained(
     "stabilityai/stable-diffusion-xl-base-0.9", torch_dtype=torch.float16, variant="fp16", use_safetensors=True
 )
 pipe.to("cuda")
 
-refiner = StableDiffusionXLImg2ImgPipeline.from_pretrained(
-    "stabilityai/stable-diffusion-xl-refiner-0.9", torch_dtype=torch.float16, use_safetensors=True, variant="fp16"
+img_url = "https://raw.githubusercontent.com/CompVis/latent-diffusion/main/data/inpainting_examples/overture-creations-5sI6fQgYIuo.png"
+mask_url = "https://raw.githubusercontent.com/CompVis/latent-diffusion/main/data/inpainting_examples/overture-creations-5sI6fQgYIuo_mask.png"
+
+init_image = load_image(img_url).convert("RGB")
+mask_image = load_image(mask_url).convert("RGB")
+
+prompt = "A majestic tiger sitting on a bench"
+image = pipe(prompt=prompt, image=init_image, mask_image=mask_image, num_inference_steps=50, strength=0.80).images[0]
+```
+
+### Refining the image output
+
+In addition to the [base model checkpoint](https://huggingface.co/stabilityai/stable-diffusion-xl-base-0.9), 
+StableDiffusion-XL also includes a [refiner checkpoint](huggingface.co/stabilityai/stable-diffusion-xl-refiner-0.9)
+that is specialized in denoising low-noise stage images to generate images of improved high-frequency quality.
+This refiner checkpoint can be used as a "second-step" pipeline after having run the base checkpoint to improve
+image quality.
+
+When using the refiner, one can easily 
+- 1.) employ the base model and refiner as an *Ensemble of Expert Denoisers* as first proposed in [eDiff-I](https://research.nvidia.com/labs/dir/eDiff-I/) or
+- 2.) simply run the refiner in [SDEdit](https://arxiv.org/abs/2108.01073) fashion after the base model.
+
+**Note**: The idea of using SD-XL base & refiner as an ensemble of experts was first brought forward by 
+a couple community contributors which also helped shape the following `diffusers` implementation, namely:
+- [SytanSD](https://github.com/SytanSD)
+- [bghira](https://github.com/bghira)
+- [Birch-san](https://github.com/Birch-san)
+
+#### 1.) Ensemble of Expert Denoisers
+
+When using the base and refiner model as an ensemble of expert of denoisers, the base model should serve as the 
+expert for the high-noise diffusion stage and the refiner serves as the expert for the low-noise diffusion stage.
+
+The advantage of 1.) over 2.) is that it requires less overall denoising steps and therefore should be significantly
+faster. The drawback is that one cannot really inspect the output of the base model; it will still be heavily denoised.
+
+To use the base model and refiner as an ensemble of expert denoisers, make sure to define the fraction
+of timesteps which should be run through the high-noise denoising stage (*i.e.* the base model) and the low-noise
+denoising stage (*i.e.* the refiner model) respectively. This fraction should be set as the [`denoising_end`](https://huggingface.co/docs/diffusers/main/en/api/pipelines/stable_diffusion/stable_diffusion_xl#diffusers.StableDiffusionXLPipeline.__call__.denoising_end) of the base model 
+and as the [`denoising_start`](https://huggingface.co/docs/diffusers/main/en/api/pipelines/stable_diffusion/stable_diffusion_xl#diffusers.StableDiffusionXLImg2ImgPipeline.__call__.denoising_start) of the refiner model.
+
+Let's look at an example.
+First, we import the two pipelines. Since the text encoders and variational autoencoder are the same
+you don't have to load those again for the refiner.
+
+```py
+from diffusers import DiffusionPipeline
+import torch
+
+base = DiffusionPipeline.from_pretrained(
+    "stabilityai/stable-diffusion-xl-base-0.9", torch_dtype=torch.float16, variant="fp16", use_safetensors=True
+)
+pipe.to("cuda")
+
+refiner = DiffusionPipeline.from_pretrained(
+    "stabilityai/stable-diffusion-xl-refiner-0.9",
+    text_encoder_2=base.text_encoder_2,
+    vae=base.vae,
+    torch_dtype=torch.float16,
+    use_safetensors=True,
+    variant="fp16",
+)
+refiner.to("cuda")
+```
+
+Now we define the number of inference steps and the fraction at which the model shall be run through the 
+high-noise denoising stage (*i.e.* the base model).
+
+```py
+n_steps = 40
+high_noise_frac = 0.7
+```
+
+A fraction of 0.7 means that 70% of the 40 inference steps (28 steps) are run through the base model
+and the remaining 12 steps are run through the refiner. Let's run the two pipelines now.
+Make sure to set `denoising_end` and `denoising_start` to the same values and keep `num_inference_steps`
+constant. Also remember that the output of the base model should be in latent space:
+
+```py
+prompt = "A majestic lion jumping from a big stone at night"
+
+image = base(prompt=prompt, num_inference_steps=n_steps, denoising_end=high_noise_frac, output_type="latent").images
+image = refiner(prompt=prompt, num_inference_steps=n_steps, denoising_start=high_noise_frac, image=image).images[0]
+```
+
+Let's have a look at the image
+
+| Original Image | Ensemble of Denoisers Experts |
+|---|---|
+| ![lion_base](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/lion_base.png) | ![lion_ref](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/lion_refined.png)
+
+If we would have just run the base model on the same 40 steps, the image would have been arguably less detailed (e.g. the lion eyes and nose):
+
+<Tip>
+
+The ensemble-of-experts method works well on all available schedulers!
+
+</Tip>
+
+#### 2.) Refining the image output from fully denoised base image
+
+In standard [`StableDiffusionImg2ImgPipeline`]-fashion, the fully-denoised image generated of the base model 
+can be further improved using the [refiner checkpoint](huggingface.co/stabilityai/stable-diffusion-xl-refiner-0.9).
+
+For this, you simply run the refiner as a normal image-to-image pipeline after the "base" text-to-image 
+pipeline. You can leave the outputs of the base model in latent space.
+
+```py
+from diffusers import DiffusionPipeline
+import torch
+
+pipe = DiffusionPipeline.from_pretrained(
+    "stabilityai/stable-diffusion-xl-base-0.9", torch_dtype=torch.float16, variant="fp16", use_safetensors=True
+)
+pipe.to("cuda")
+
+refiner = DiffusionPipeline.from_pretrained(
+    "stabilityai/stable-diffusion-xl-refiner-0.9",
+    text_encoder_2=pipe.text_encoder_2,
+    vae=pipe.vae,
+    torch_dtype=torch.float16,
+    use_safetensors=True,
+    variant="fp16",
 )
 refiner.to("cuda")
 
@@ -82,7 +223,70 @@ image = pipe(prompt=prompt, output_type="latent" if use_refiner else "pil").imag
 image = refiner(prompt=prompt, image=image[None, :]).images[0]
 ```
 
-### Loading single file checkpoitns / original file format
+| Original Image | Refined Image |
+|---|---|
+| ![](https://huggingface.co/datasets/diffusers/docs-images/resolve/main/sd_xl/init_image.png) | ![](https://huggingface.co/datasets/diffusers/docs-images/resolve/main/sd_xl/refined_image.png) |
+
+<Tip>
+
+The refiner can also very well be used in an in-painting setting. To do so just make
+  sure you use the [`StableDiffusionXLInpaintPipeline`] classes as shown below
+
+</Tip>
+
+To use the refiner for inpainting in the Ensemble of Expert Denoisers setting you can do the following:
+
+```py
+from diffusers import StableDiffusionXLInpaintPipeline
+from diffusers.utils import load_image
+
+pipe = StableDiffusionXLInpaintPipeline.from_pretrained(
+    "stabilityai/stable-diffusion-xl-base-0.9", torch_dtype=torch.float16, variant="fp16", use_safetensors=True
+)
+pipe.to("cuda")
+
+refiner = StableDiffusionXLInpaintPipeline.from_pretrained(
+    "stabilityai/stable-diffusion-xl-refiner-0.9",
+    text_encoder_2=pipe.text_encoder_2,
+    vae=pipe.vae,
+    torch_dtype=torch.float16,
+    use_safetensors=True,
+    variant="fp16",
+)
+refiner.to("cuda")
+
+img_url = "https://raw.githubusercontent.com/CompVis/latent-diffusion/main/data/inpainting_examples/overture-creations-5sI6fQgYIuo.png"
+mask_url = "https://raw.githubusercontent.com/CompVis/latent-diffusion/main/data/inpainting_examples/overture-creations-5sI6fQgYIuo_mask.png"
+
+init_image = load_image(img_url).convert("RGB")
+mask_image = load_image(mask_url).convert("RGB")
+
+prompt = "A majestic tiger sitting on a bench"
+num_inference_steps = 75
+high_noise_frac = 0.7
+
+image = pipe(
+    prompt=prompt,
+    image=init_image,
+    mask_image=mask_image,
+    num_inference_steps=num_inference_steps,
+    strength=0.80,
+    denoising_start=high_noise_frac,
+    output_type="latent",
+).images
+image = refiner(
+    prompt=prompt,
+    image=image,
+    mask_image=mask_image,
+    num_inference_steps=num_inference_steps,
+    denoising_start=high_noise_frac,
+).images[0]
+```
+
+To use the refiner for inpainting in the standard SDE-style setting, simply remove `denoising_end` and `denoising_start` and choose a smaller
+number of inference steps for the refiner.
+
+### Loading single file checkpoints / original file format
 
 By making use of [`~diffusers.loaders.FromSingleFileMixin.from_single_file`] you can also load the 
 original file format into `diffusers`:
@@ -91,13 +295,13 @@ original file format into `diffusers`:
 from diffusers import StableDiffusionXLPipeline, StableDiffusionXLImg2ImgPipeline
 import torch
 
-pipe = StableDiffusionXLPipeline.from_pretrained(
-    "stabilityai/stable-diffusion-xl-base-0.9", torch_dtype=torch.float16, variant="fp16", use_safetensors=True
+pipe = StableDiffusionXLPipeline.from_single_file(
+    "./sd_xl_base_0.9.safetensors", torch_dtype=torch.float16, variant="fp16", use_safetensors=True
 )
 pipe.to("cuda")
 
-refiner = StableDiffusionXLImg2ImgPipeline.from_pretrained(
-    "stabilityai/stable-diffusion-xl-refiner-0.9", torch_dtype=torch.float16, use_safetensors=True, variant="fp16"
+refiner = StableDiffusionXLImg2ImgPipeline.from_single_file(
+    "./sd_xl_refiner_0.9.safetensors", torch_dtype=torch.float16, use_safetensors=True, variant="fp16"
 )
 refiner.to("cuda")
 ```
@@ -127,7 +331,7 @@ You can speed up inference by making use of `torch.compile`. This should give yo
 + refiner.unet = torch.compile(refiner.unet, mode="reduce-overhead", fullgraph=True)
 ```
 
-### Running with `torch` < 2.0
+### Running with `torch < 2.0`
 
 **Note** that if you want to run Stable Diffusion XL with `torch` < 2.0, please make sure to enable xformers 
 attention:
@@ -152,3 +356,9 @@ pip install xformers
 [[autodoc]] StableDiffusionXLImg2ImgPipeline
 	- all
 	- __call__
+
+## StableDiffusionXLInpaintPipeline
+
+[[autodoc]] StableDiffusionXLInpaintPipeline
+	- all
+	- __call__

docs/source/en/index.mdx

@@ -40,7 +40,7 @@ The library has three main components:
       ><div class="w-full text-center bg-gradient-to-br from-pink-400 to-pink-500 rounded-lg py-1.5 font-semibold mb-5 text-white text-lg leading-relaxed">Conceptual guides</div>
       <p class="text-gray-700">Understand why the library was designed the way it was, and learn more about the ethical guidelines and safety implementations for using the library.</p>
    </a>
-    <a class="!no-underline border dark:border-gray-700 p-5 rounded-lg shadow hover:shadow-lg" href="./api/models"
+    <a class="!no-underline border dark:border-gray-700 p-5 rounded-lg shadow hover:shadow-lg" href="./api/models/overview"
       ><div class="w-full text-center bg-gradient-to-br from-purple-400 to-purple-500 rounded-lg py-1.5 font-semibold mb-5 text-white text-lg leading-relaxed">Reference</div>
       <p class="text-gray-700">Technical descriptions of how 🤗 Diffusers classes and methods work.</p>
     </a>
@@ -69,6 +69,7 @@ The library has three main components:
 | [score_sde_ve](./api/pipelines/score_sde_ve) | [Score-Based Generative Modeling through Stochastic Differential Equations](https://openreview.net/forum?id=PxTIG12RRHS) | Unconditional Image Generation |
 | [score_sde_vp](./api/pipelines/score_sde_vp) | [Score-Based Generative Modeling through Stochastic Differential Equations](https://openreview.net/forum?id=PxTIG12RRHS) | Unconditional Image Generation |
 | [semantic_stable_diffusion](./api/pipelines/semantic_stable_diffusion) | [Semantic Guidance](https://arxiv.org/abs/2301.12247) | Text-Guided Generation |
+| [stable_diffusion_adapter](./api/pipelines/stable_diffusion/adapter) | [**T2I-Adapter**](https://arxiv.org/abs/2302.08453) | Image-to-Image Text-Guided Generation | -
 | [stable_diffusion_text2img](./api/pipelines/stable_diffusion/text2img) | [Stable Diffusion](https://stability.ai/blog/stable-diffusion-public-release) | Text-to-Image Generation |
 | [stable_diffusion_img2img](./api/pipelines/stable_diffusion/img2img) | [Stable Diffusion](https://stability.ai/blog/stable-diffusion-public-release) | Image-to-Image Text-Guided Generation |
 | [stable_diffusion_inpaint](./api/pipelines/stable_diffusion/inpaint) | [Stable Diffusion](https://stability.ai/blog/stable-diffusion-public-release) | Text-Guided Image Inpainting |

docs/source/en/optimization/bentoml.mdx

@@ -0,0 +1,200 @@
+<!--Copyright 2023 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.
+-->
+
+# BentoML Integration Guide
+
+[[open-in-colab]]
+
+[BentoML](https://github.com/bentoml/BentoML/) is an open-source framework designed for building,
+shipping, and scaling AI applications. It allows users to easily package and serve diffusion models
+for production, ensuring reliable and efficient deployments. It features out-of-the-box operational
+management tools like monitoring and tracing, and facilitates the deployment to various cloud platforms
+with ease. BentoML's distributed architecture and the separation of API server logic from
+model inference logic enable efficient scaling of deployments, even with budget constraints.
+As a result, integrating it with Diffusers provides a valuable tool for real-world deployments.
+
+This tutorial demonstrates how to integrate BentoML with Diffusers.
+
+## Prerequisites
+
+- Install [Diffusers](https://huggingface.co/docs/diffusers/installation).
+- Install BentoML by running `pip install bentoml`. For more information, see the [BentoML documentation](https://docs.bentoml.com).
+
+## Import a diffusion model
+
+First, you need to prepare the model. BentoML has its own [Model Store](https://docs.bentoml.com/en/latest/concepts/model.html)
+for model management. Create a `download_model.py` file as below to import a diffusion model into BentoML's Model
+Store:
+
+```py
+import bentoml
+
+bentoml.diffusers.import_model(
+    "sd2.1",  # Model tag in the BentoML Model Store
+    "stabilityai/stable-diffusion-2-1",  # Hugging Face model identifier
+)
+```
+
+This code snippet downloads the Stable Diffusion 2.1 model (using it's repo id
+`stabilityai/stable-diffusion-2-1`) from the Hugging Face Hub (or use the cached download
+files if the model is already downloaded) and imports it into the BentoML Model
+Store with the name `sd2.1`.
+
+For models already fine-tuned and stored on disk, you can provide the path instead of
+the repo id.
+
+```py
+import bentoml
+
+bentoml.diffusers.import_model(
+    "sd2.1-local",
+    "./local_stable_diffusion_2.1/",
+)
+```
+
+You can view the model in the Model Store:
+
+```
+bentoml models list
+
+Tag                                                                 Module                              Size       Creation Time       
+sd2.1:ysrlmubascajwnry                                              bentoml.diffusers                   33.85 GiB  2023-07-12 16:47:44 
+```
+
+## Turn a diffusion model into a RESTful service with BentoML
+
+Once the diffusion model is in BentoML's Model Store, you can implement a text-to-image
+service with it. The Stable Diffusion model accepts various arguments
+in addition to the required prompt to guide the image generation process.
+To validate these input arguments, use BentoML's [pydantic](https://github.com/pydantic/pydantic) integration.
+Create a `sdargs.py` file with an example pydantic model:
+
+```py
+import typing as t
+
+from pydantic import BaseModel
+
+
+class SDArgs(BaseModel):
+    prompt: str
+    negative_prompt: t.Optional[str] = None
+    height: t.Optional[int] = 512
+    width: t.Optional[int] = 512
+
+    class Config:
+        extra = "allow"
+```
+
+This pydantic model requires a string field `prompt` and three optional fields: `height`, `width`, and `negative_prompt`,
+each with corresponding types. The `extra = "allow"` line supports adding additional fields not defined in the `SDArgs` class.
+In a real-world scenario, you may define all the desired fields and not allow extra ones.
+
+Next, create a BentoML Service file that defines a Stable Diffusion service:
+
+```py
+import bentoml
+from bentoml.io import Image, JSON
+
+from sdargs import SDArgs
+
+bento_model = bentoml.diffusers.get("sd2.1:latest")
+sd21_runner = bento_model.to_runner(name="sd21-runner")
+
+svc = bentoml.Service("stable-diffusion-21", runners=[sd21_runner])
+
+
+@svc.api(input=JSON(pydantic_model=SDArgs), output=Image())
+async def txt2img(input_data):
+    kwargs = input_data.dict()
+    res = await sd21_runner.async_run(**kwargs)
+    images = res[0]
+    return images[0]
+```
+
+Save the file as `service.py`, and spin up a BentoML Service endpoint using:
+
+```
+bentoml serve service:svc
+```
+
+An HTTP server with `/txt2img` endpoint that accepts a JSON dictionary should be up at
+port 3000. Go to <http://127.0.0.1:3000> in your web browser to access the Swagger UI.
+
+You can also test the text-to-image generation using `curl` and write the returned image to
+`output.jpg`.
+
+```
+curl -X POST http://127.0.0.1:3000/txt2img \
+     -H 'Content-Type: application/json' \
+     -d "{\"prompt\":\"a black cat\", \"height\":768, \"width\":768}" \
+     --output output.jpg
+```
+
+## Package a BentoML Service for cloud deployment
+
+To deploy a BentoML Service, you need to pack it into a BentoML
+[Bento](https://docs.bentoml.com/en/latest/concepts/bento.html), a file archive with all the source code,
+models, data files, and dependencies. This can be done by providing a `bentofile.yaml` file as follows:
+
+```yaml
+service: "service.py:svc"
+include:
+  - "service.py"
+python:
+  packages:
+    - torch
+    - transformers
+    - accelerate
+    - diffusers
+    - triton
+    - xformers
+    - pydantic
+docker:
+    distro: debian
+    cuda_version: "11.6"
+```
+
+The `bentofile.yaml` file contains [Bento build
+options](https://docs.bentoml.com/en/latest/concepts/bento.html#bento-build-options),
+such as package dependencies and Docker options.
+
+Then you build a Bento using:
+
+```
+bentoml build
+```
+
+The output looks like:
+
+```
+Successfully built Bento(tag="stable-diffusion-21:crkuh7a7rw5bcasc").
+
+Possible next steps:
+
+ * Containerize your Bento with `bentoml containerize`:
+    $ bentoml containerize stable-diffusion-21:crkuh7a7rw5bcasc
+
+ * Push to BentoCloud with `bentoml push`:
+    $ bentoml push stable-diffusion-21:crkuh7a7rw5bcasc
+```
+
+You can create a Docker image based on the Bento by running the following command and deploy it to a cloud provider.
+
+```
+bentoml containerize stable-diffusion-21:crkuh7a7rw5bcasc
+```
+
+If you want an end-to-end solution for deploying and managing models, you can push the Bento to [Yatai](https://github.com/bentoml/Yatai) or
+[BentoCloud](https://bentoml.com/cloud) for a distributed deployment.
+
+For more information about BentoML's integration with Diffusers, see the [BentoML Diffusers
+Guide](https://docs.bentoml.com/en/latest/frameworks/diffusers.html).

docs/source/en/training/dreambooth.mdx

@@ -701,3 +701,7 @@ accelerate launch train_dreambooth.py \
   --class_labels_conditioning timesteps \
   --push_to_hub
 ```
+
+## Stable Diffusion XL
+
+We support fine-tuning of the UNet shipped in [Stable Diffusion XL](https://huggingface.co/papers/2307.01952) with DreamBooth and LoRA via the `train_dreambooth_lora_sdxl.py` script. Please refer to the docs [here](https://github.com/huggingface/diffusers/blob/main/examples/dreambooth/README_sdxl.md). 

docs/source/en/training/lora.mdx

@@ -280,6 +280,10 @@ Note that the use of [`~diffusers.loaders.LoraLoaderMixin.load_lora_weights`] is
 **Note** that it is possible to provide a local directory path to [`~diffusers.loaders.LoraLoaderMixin.load_lora_weights`] as well as [`~diffusers.loaders.UNet2DConditionLoadersMixin.load_attn_procs`]. To know about the supported inputs,
 refer to the respective docstrings.
 
+## Unloading LoRA parameters
+
+You can call [`~diffusers.loaders.LoraLoaderMixin.unload_lora_weights`] on a pipeline to unload the LoRA parameters.
+
 ## Supporting A1111 themed LoRA checkpoints from Diffusers
 
 To provide seamless interoperability with A1111 to our users, we support loading A1111 formatted

docs/source/en/using-diffusers/controlling_generation.mdx

@@ -59,6 +59,7 @@ For convenience, we provide a table to denote which methods are inference-only a
 | [Custom Diffusion](#custom-diffusion) | ❌ | ✅ |  |
 | [Model Editing](#model-editing) | ✅ | ❌ |  |
 | [DiffEdit](#diffedit) | ✅ | ❌ |  |
+| [T2I-Adapter](#t2i-adapter) | ✅ | ❌ |  |
 
 ## Instruct Pix2Pix
 
@@ -215,4 +216,13 @@ To know more details, check out the [official doc](../api/pipelines/stable_diffu
 [DiffEdit](../api/pipelines/stable_diffusion/diffedit) allows for semantic editing of input images along with 
 input prompts while preserving the original input images as much as possible. 
 
-To know more details, check out the [official doc](../api/pipelines/stable_diffusion/model_editing).
+To know more details, check out the [official doc](../api/pipelines/stable_diffusion/model_editing).
+## T2I-Adapter
+
+[Paper](https://arxiv.org/abs/2302.08453)
+
+[T2I-Adapter](../api/pipelines/stable_diffusion/adapter) is an auxiliary network which adds an extra condition.
+There are 8 canonical pre-trained adapters trained on different conditionings such as edge detection, sketch, 
+depth maps, and semantic segmentations.
+
+See [here](../api/pipelines/stable_diffusion/adapter) for more information on how to use it.

docs/source/ko/_toctree.yml

@@ -8,14 +8,69 @@
   - local: installation
     title: "설치"
   title: "시작하기"
-
+- sections:
+  - local: tutorials/tutorial_overview
+    title: 개요
+  - local: using-diffusers/write_own_pipeline
+    title: 모델과 스케줄러 이해하기
+  - local: tutorials/basic_training
+    title: Diffusion 모델 학습하기
+  title: Tutorials
 - sections:
   - sections:
     - local: in_translation
       title: 개요
-    - local: in_translation
+    - local: using-diffusers/loading
+      title: 파이프라인, 모델, 스케줄러 불러오기
+    - local: using-diffusers/schedulers
+      title: 다른 스케줄러들을 가져오고 비교하기
+    - local: using-diffusers/custom_pipeline_overview
+      title: 커뮤니티 파이프라인 불러오기
+    - local: using-diffusers/using_safetensors
+      title: 세이프텐서 불러오기
+    - local: using-diffusers/other-formats
+      title: 다른 형식의 Stable Diffusion 불러오기
+    title: 불러오기 & 허브
+  - sections:
+    - local: using-diffusers/pipeline_overview
+      title: 개요
+    - local: using-diffusers/unconditional_image_generation
       title: Unconditional 이미지 생성
     - local: in_translation
+      title: Text-to-image 생성
+    - local: using-diffusers/img2img
+      title: Text-guided image-to-image
+    - local: using-diffusers/inpaint
+      title: Text-guided 이미지 인페인팅
+    - local: using-diffusers/depth2img
+      title: Text-guided depth-to-image
+    - local: in_translation
+      title: Textual inversion
+    - local: in_translation
+      title: 여러 GPU를 사용한 분산 추론
+    - local: using-diffusers/reusing_seeds
+      title: Deterministic 생성으로 이미지 퀄리티 높이기
+    - local: in_translation
+      title: 재현 가능한 파이프라인 생성하기
+    - local: using-diffusers/custom_pipeline_examples
+      title: 커뮤니티 파이프라인들
+    - local: in_translation
+      title: 커뮤티니 파이프라인에 기여하는 방법
+    - local: in_translation
+      title: JAX/Flax에서의 Stable Diffusion
+    - local: in_translation
+      title: Weighting Prompts
+    title: 추론을 위한 파이프라인
+  - sections:
+    - local: training/overview
+      title: 개요
+    - local: in_translation
+      title: 학습을 위한 데이터셋 생성하기
+    - local: training/adapt_a_model
+      title: 새로운 태스크에 모델 적용하기
+    - local: training/unconditional_training
+      title: Unconditional 이미지 생성
+    - local: training/text_inversion
       title: Textual Inversion
     - local: training/dreambooth
       title: DreamBooth
@@ -27,13 +82,16 @@
       title: ControlNet
     - local: in_translation
       title: InstructPix2Pix 학습
-    title: 학습
+    - local: in_translation
+      title: Custom Diffusion
+    title: Training
+  title: Diffusers 사용하기
 - sections:
-  - local: in_translation
+  - local: optimization/opt_overview
     title: 개요
   - local: optimization/fp16
     title: 메모리와 속도
-  - local: in_translation
+  - local: optimization/torch2.0
     title: Torch2.0 지원
   - local: optimization/xformers
     title: xFormers
@@ -41,8 +99,12 @@
     title: ONNX
   - local: optimization/open_vino
     title: OpenVINO
+  - local: in_translation
+    title: Core ML
   - local: optimization/mps
     title: MPS
   - local: optimization/habana
     title: Habana Gaudi
+  - local: in_translation
+    title: Token Merging
   title: 최적화/특수 하드웨어

docs/source/ko/optimization/fp16.mdx

@@ -59,7 +59,7 @@ torch.backends.cuda.matmul.allow_tf32 = True
 
 ## 반정밀도 가중치
 
-더 많은 GPU 메모리를 절약하고 더 빠른 속도를 얻기 위해 모델 가중치를 반정밀도(half precision)로 직접 로드하고 실행할 수 있습니다. 
+더 많은 GPU 메모리를 절약하고 더 빠른 속도를 얻기 위해 모델 가중치를 반정밀도(half precision)로 직접 불러오고 실행할 수 있습니다. 
 여기에는 `fp16`이라는 브랜치에 저장된 float16 버전의 가중치를 불러오고, 그 때 `float16` 유형을 사용하도록 PyTorch에 지시하는 작업이 포함됩니다.
 
 ```Python

docs/source/ko/optimization/opt_overview.mdx

@@ -0,0 +1,17 @@
+<!--Copyright 2023 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.
+-->
+
+# 개요
+
+노이즈가 많은 출력에서 적은 출력으로 만드는 과정으로 고품질 생성 모델의 출력을 만드는 각각의 반복되는 스텝은 많은 계산이 필요합니다. 🧨 Diffuser의 목표 중 하나는 모든 사람이 이 기술을 널리 이용할 수 있도록 하는 것이며, 여기에는 소비자 및 특수 하드웨어에서 빠른 추론을 가능하게 하는 것을 포함합니다. 
+
+이 섹션에서는 추론 속도를 최적화하고 메모리 소비를 줄이기 위한 반정밀(half-precision) 가중치 및 sliced attention과 같은 팁과 요령을 다룹니다. 또한 [`torch.compile`](https://pytorch.org/tutorials/intermediate/torch_compile_tutorial.html) 또는 [ONNX Runtime](https://onnxruntime.ai/docs/)을 사용하여 PyTorch 코드의 속도를 높이고, [xFormers](https://facebookresearch.github.io/xformers/)를 사용하여 memory-efficient attention을 활성화하는 방법을 배울 수 있습니다. Apple Silicon, Intel 또는 Habana 프로세서와 같은 특정 하드웨어에서 추론을 실행하기 위한 가이드도 있습니다.

docs/source/ko/optimization/torch2.0.mdx

@@ -0,0 +1,445 @@
+<!--Copyright 2023 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.
+-->
+
+# Diffusers에서의 PyTorch 2.0 가속화 지원
+
+`0.13.0` 버전부터 Diffusers는 [PyTorch 2.0](https://pytorch.org/get-started/pytorch-2.0/)에서의 최신 최적화를 지원합니다. 이는 다음을 포함됩니다.
+1. momory-efficient attention을 사용한 가속화된 트랜스포머 지원 - `xformers`같은 추가적인 dependencies 필요 없음
+2. 추가 성능 향상을 위한 개별 모델에 대한 컴파일 기능 [torch.compile](https://pytorch.org/tutorials/intermediate/torch_compile_tutorial.html) 지원
+
+
+## 설치
+가속화된 어텐션 구현과 및 `torch.compile()`을 사용하기 위해, pip에서 최신 버전의 PyTorch 2.0을 설치되어 있고 diffusers 0.13.0. 버전 이상인지 확인하세요. 아래 설명된 바와 같이, PyTorch 2.0이 활성화되어 있을 때 diffusers는 최적화된 어텐션 프로세서([`AttnProcessor2_0`](https://github.com/huggingface/diffusers/blob/1a5797c6d4491a879ea5285c4efc377664e0332d/src/diffusers/models/attention_processor.py#L798))를 사용합니다.
+
+```bash
+pip install --upgrade torch diffusers
+```
+
+## 가속화된 트랜스포머와 `torch.compile` 사용하기.
+
+
+1. **가속화된 트랜스포머 구현**
+
+   PyTorch 2.0에는 [`torch.nn.functional.scaled_dot_product_attention`](https://pytorch.org/docs/master/generated/torch.nn.functional.scaled_dot_product_attention) 함수를 통해 최적화된 memory-efficient attention의 구현이 포함되어 있습니다. 이는 입력 및 GPU 유형에 따라 여러 최적화를 자동으로 활성화합니다. 이는 [xFormers](https://github.com/facebookresearch/xformers)의 `memory_efficient_attention`과 유사하지만 기본적으로 PyTorch에 내장되어 있습니다.
+   
+   이러한 최적화는 PyTorch 2.0이 설치되어 있고 `torch.nn.functional.scaled_dot_product_attention`을 사용할 수 있는 경우 Diffusers에서 기본적으로 활성화됩니다. 이를 사용하려면 `torch 2.0`을 설치하고 파이프라인을 사용하기만 하면 됩니다. 예를 들어:
+
+    ```Python
+    import torch
+    from diffusers import DiffusionPipeline
+
+    pipe = DiffusionPipeline.from_pretrained("runwayml/stable-diffusion-v1-5", torch_dtype=torch.float16)
+    pipe = pipe.to("cuda")
+
+    prompt = "a photo of an astronaut riding a horse on mars"
+    image = pipe(prompt).images[0]
+    ```
+
+    이를 명시적으로 활성화하려면(필수는 아님) 아래와 같이 수행할 수 있습니다.
+
+    ```diff
+    import torch
+    from diffusers import DiffusionPipeline
+    + from diffusers.models.attention_processor import AttnProcessor2_0
+
+    pipe = DiffusionPipeline.from_pretrained("runwayml/stable-diffusion-v1-5", torch_dtype=torch.float16).to("cuda")
+    + pipe.unet.set_attn_processor(AttnProcessor2_0())
+
+    prompt = "a photo of an astronaut riding a horse on mars"
+    image = pipe(prompt).images[0]
+    ```
+
+    이 실행 과정은 `xFormers`만큼 빠르고 메모리적으로 효율적이어야 합니다. 자세한 내용은 [벤치마크](#benchmark)에서 확인하세요.
+
+    파이프라인을 보다 deterministic으로 만들거나 파인 튜닝된 모델을 [Core ML](https://huggingface.co/docs/diffusers/v0.16.0/en/optimization/coreml#how-to-run-stable-diffusion-with-core-ml)과 같은 다른 형식으로 변환해야 하는 경우 바닐라 어텐션 프로세서 ([`AttnProcessor`](https://github.com/huggingface/diffusers/blob/1a5797c6d4491a879ea5285c4efc377664e0332d/src/diffusers/models/attention_processor.py#L402))로 되돌릴 수 있습니다. 일반 어텐션 프로세서를 사용하려면 [`~diffusers.UNet2DConditionModel.set_default_attn_processor`] 함수를 사용할 수 있습니다:
+
+    ```Python
+    import torch
+    from diffusers import DiffusionPipeline
+    from diffusers.models.attention_processor import AttnProcessor
+
+    pipe = DiffusionPipeline.from_pretrained("runwayml/stable-diffusion-v1-5", torch_dtype=torch.float16).to("cuda")
+    pipe.unet.set_default_attn_processor()
+
+    prompt = "a photo of an astronaut riding a horse on mars"
+    image = pipe(prompt).images[0]
+    ```
+
+2. **torch.compile**
+
+    추가적인 속도 향상을 위해 새로운 `torch.compile` 기능을 사용할 수 있습니다. 파이프라인의 UNet은 일반적으로 계산 비용이 가장 크기 때문에 나머지 하위 모델(텍스트 인코더와 VAE)은 그대로 두고 `unet`을 `torch.compile`로 래핑합니다. 자세한 내용과 다른 옵션은 [torch 컴파일 문서](https://pytorch.org/tutorials/intermediate/torch_compile_tutorial.html)를 참조하세요.
+
+    ```python
+    pipe.unet = torch.compile(pipe.unet, mode="reduce-overhead", fullgraph=True)
+    images = pipe(prompt, num_inference_steps=steps, num_images_per_prompt=batch_size).images
+    ```
+
+    GPU 유형에 따라 `compile()`은 가속화된 트랜스포머 최적화를 통해 **5% - 300%**의 _추가 성능 향상_을 얻을 수 있습니다. 그러나 컴파일은 Ampere(A100, 3090), Ada(4090) 및 Hopper(H100)와 같은 최신 GPU 아키텍처에서 더 많은 성능 향상을 가져올 수 있음을 참고하세요.
+    
+    컴파일은 완료하는 데 약간의 시간이 걸리므로, 파이프라인을 한 번 준비한 다음 동일한 유형의 추론 작업을 여러 번 수행해야 하는 상황에 가장 적합합니다. 다른 이미지 크기에서 컴파일된 파이프라인을 호출하면 시간적 비용이 많이 들 수 있는 컴파일 작업이 다시 트리거됩니다.
+
+
+## 벤치마크
+
+PyTorch 2.0의 효율적인 어텐션 구현과 `torch.compile`을 사용하여 가장 많이 사용되는 5개의 파이프라인에 대해 다양한 GPU와 배치 크기에 걸쳐 포괄적인 벤치마크를 수행했습니다. 여기서는 [`torch.compile()`이 최적으로 활용되도록 하는](https://github.com/huggingface/diffusers/pull/3313) `diffusers 0.17.0.dev0`을 사용했습니다.
+
+### 벤치마킹 코드
+
+#### Stable Diffusion text-to-image 
+
+```python 
+from diffusers import DiffusionPipeline
+import torch
+
+path = "runwayml/stable-diffusion-v1-5"
+
+run_compile = True  # Set True / False
+
+pipe = DiffusionPipeline.from_pretrained(path, torch_dtype=torch.float16)
+pipe = pipe.to("cuda")
+pipe.unet.to(memory_format=torch.channels_last)
+
+if run_compile:
+    print("Run torch compile")
+    pipe.unet = torch.compile(pipe.unet, mode="reduce-overhead", fullgraph=True)
+
+prompt = "ghibli style, a fantasy landscape with castles"
+
+for _ in range(3):
+    images = pipe(prompt=prompt).images
+```
+
+#### Stable Diffusion image-to-image 
+
+```python 
+from diffusers import StableDiffusionImg2ImgPipeline
+import requests
+import torch
+from PIL import Image
+from io import BytesIO
+
+url = "https://raw.githubusercontent.com/CompVis/stable-diffusion/main/assets/stable-samples/img2img/sketch-mountains-input.jpg"
+
+response = requests.get(url)
+init_image = Image.open(BytesIO(response.content)).convert("RGB")
+init_image = init_image.resize((512, 512))
+
+path = "runwayml/stable-diffusion-v1-5"
+
+run_compile = True  # Set True / False
+
+pipe = StableDiffusionImg2ImgPipeline.from_pretrained(path, torch_dtype=torch.float16)
+pipe = pipe.to("cuda")
+pipe.unet.to(memory_format=torch.channels_last)
+
+if run_compile:
+    print("Run torch compile")
+    pipe.unet = torch.compile(pipe.unet, mode="reduce-overhead", fullgraph=True)
+
+prompt = "ghibli style, a fantasy landscape with castles"
+
+for _ in range(3):
+    image = pipe(prompt=prompt, image=init_image).images[0]
+```
+
+#### Stable Diffusion - inpainting
+
+```python 
+from diffusers import StableDiffusionInpaintPipeline
+import requests
+import torch
+from PIL import Image
+from io import BytesIO
+
+url = "https://raw.githubusercontent.com/CompVis/stable-diffusion/main/assets/stable-samples/img2img/sketch-mountains-input.jpg"
+
+def download_image(url):
+    response = requests.get(url)
+    return Image.open(BytesIO(response.content)).convert("RGB")
+
+
+img_url = "https://raw.githubusercontent.com/CompVis/latent-diffusion/main/data/inpainting_examples/overture-creations-5sI6fQgYIuo.png"
+mask_url = "https://raw.githubusercontent.com/CompVis/latent-diffusion/main/data/inpainting_examples/overture-creations-5sI6fQgYIuo_mask.png"
+
+init_image = download_image(img_url).resize((512, 512))
+mask_image = download_image(mask_url).resize((512, 512))
+
+path = "runwayml/stable-diffusion-inpainting"
+
+run_compile = True  # Set True / False
+
+pipe = StableDiffusionInpaintPipeline.from_pretrained(path, torch_dtype=torch.float16)
+pipe = pipe.to("cuda")
+pipe.unet.to(memory_format=torch.channels_last)
+
+if run_compile:
+    print("Run torch compile")
+    pipe.unet = torch.compile(pipe.unet, mode="reduce-overhead", fullgraph=True)
+
+prompt = "ghibli style, a fantasy landscape with castles"
+
+for _ in range(3):
+    image = pipe(prompt=prompt, image=init_image, mask_image=mask_image).images[0]
+```
+
+#### ControlNet 
+
+```python 
+from diffusers import StableDiffusionControlNetPipeline, ControlNetModel
+import requests
+import torch
+from PIL import Image
+from io import BytesIO
+
+url = "https://raw.githubusercontent.com/CompVis/stable-diffusion/main/assets/stable-samples/img2img/sketch-mountains-input.jpg"
+
+response = requests.get(url)
+init_image = Image.open(BytesIO(response.content)).convert("RGB")
+init_image = init_image.resize((512, 512))
+
+path = "runwayml/stable-diffusion-v1-5"
+
+run_compile = True  # Set True / False
+controlnet = ControlNetModel.from_pretrained("lllyasviel/sd-controlnet-canny", torch_dtype=torch.float16)
+pipe = StableDiffusionControlNetPipeline.from_pretrained(
+    path, controlnet=controlnet, torch_dtype=torch.float16
+)
+
+pipe = pipe.to("cuda")
+pipe.unet.to(memory_format=torch.channels_last)
+pipe.controlnet.to(memory_format=torch.channels_last)
+
+if run_compile:
+    print("Run torch compile")
+    pipe.unet = torch.compile(pipe.unet, mode="reduce-overhead", fullgraph=True)
+    pipe.controlnet = torch.compile(pipe.controlnet, mode="reduce-overhead", fullgraph=True)
+
+prompt = "ghibli style, a fantasy landscape with castles"
+
+for _ in range(3):
+    image = pipe(prompt=prompt, image=init_image).images[0]
+```
+
+#### IF text-to-image + upscaling
+
+```python 
+from diffusers import DiffusionPipeline
+import torch
+
+run_compile = True  # Set True / False
+
+pipe = DiffusionPipeline.from_pretrained("DeepFloyd/IF-I-M-v1.0", variant="fp16", text_encoder=None, torch_dtype=torch.float16)
+pipe.to("cuda")
+pipe_2 = DiffusionPipeline.from_pretrained("DeepFloyd/IF-II-M-v1.0", variant="fp16", text_encoder=None, torch_dtype=torch.float16)
+pipe_2.to("cuda")
+pipe_3 = DiffusionPipeline.from_pretrained("stabilityai/stable-diffusion-x4-upscaler", torch_dtype=torch.float16)
+pipe_3.to("cuda")
+
+
+pipe.unet.to(memory_format=torch.channels_last)
+pipe_2.unet.to(memory_format=torch.channels_last)
+pipe_3.unet.to(memory_format=torch.channels_last)
+
+if run_compile:
+    pipe.unet = torch.compile(pipe.unet, mode="reduce-overhead", fullgraph=True)
+    pipe_2.unet = torch.compile(pipe_2.unet, mode="reduce-overhead", fullgraph=True)
+    pipe_3.unet = torch.compile(pipe_3.unet, mode="reduce-overhead", fullgraph=True)
+
+prompt = "the blue hulk"
+
+prompt_embeds = torch.randn((1, 2, 4096), dtype=torch.float16)
+neg_prompt_embeds = torch.randn((1, 2, 4096), dtype=torch.float16)
+
+for _ in range(3):
+    image = pipe(prompt_embeds=prompt_embeds, negative_prompt_embeds=neg_prompt_embeds, output_type="pt").images
+    image_2 = pipe_2(image=image, prompt_embeds=prompt_embeds, negative_prompt_embeds=neg_prompt_embeds, output_type="pt").images
+    image_3 = pipe_3(prompt=prompt, image=image, noise_level=100).images
+```
+
+PyTorch 2.0 및 `torch.compile()`로 얻을 수 있는 가능한 속도 향상에 대해, [Stable Diffusion text-to-image pipeline](StableDiffusionPipeline)에 대한 상대적인 속도 향상을 보여주는 차트를 5개의 서로 다른 GPU 제품군(배치 크기 4)에 대해 나타냅니다:
+
+![t2i_speedup](https://huggingface.co/datasets/diffusers/docs-images/resolve/main/pt2_benchmarks/t2i_speedup.png)
+
+To give you an even better idea of how this speed-up holds for the other pipelines presented above, consider the following 
+plot that shows the benchmarking numbers from an A100 across three different batch sizes
+(with PyTorch 2.0 nightly and `torch.compile()`):
+이 속도 향상이 위에 제시된 다른 파이프라인에 대해서도 어떻게 유지되는지 더 잘 이해하기 위해, 세 가지의 다른 배치 크기에 걸쳐 A100의 벤치마킹(PyTorch 2.0 nightly 및 `torch.compile() 사용) 수치를 보여주는 차트를 보입니다:
+
+![a100_numbers](https://huggingface.co/datasets/diffusers/docs-images/resolve/main/pt2_benchmarks/a100_numbers.png)
+
+_(위 차트의 벤치마크 메트릭은 **초당 iteration 수(iterations/second)**입니다)_
+
+그러나 투명성을 위해 모든 벤치마킹 수치를 공개합니다!
+
+다음 표들에서는, **_초당 처리되는 iteration_** 수 측면에서의 결과를 보여줍니다.
+
+### A100 (batch size: 1)
+
+| **Pipeline** | **torch 2.0 - <br>no compile** | **torch nightly - <br>no compile** | **torch 2.0 - <br>compile** | **torch nightly - <br>compile** |
+|:---:|:---:|:---:|:---:|:---:|
+| SD - txt2img | 21.66 | 23.13 | 44.03 | 49.74 |
+| SD - img2img | 21.81 | 22.40 | 43.92 | 46.32 |
+| SD - inpaint | 22.24 | 23.23 | 43.76 | 49.25 |
+| SD - controlnet | 15.02 | 15.82 | 32.13 | 36.08 |
+| IF | 20.21 / <br>13.84 / <br>24.00 | 20.12 / <br>13.70 / <br>24.03 | ❌ | 97.34 / <br>27.23 / <br>111.66 |
+
+### A100 (batch size: 4)
+
+| **Pipeline** | **torch 2.0 - <br>no compile** | **torch nightly - <br>no compile** | **torch 2.0 - <br>compile** | **torch nightly - <br>compile** |
+|:---:|:---:|:---:|:---:|:---:|
+| SD - txt2img | 11.6 | 13.12 | 14.62 | 17.27 |
+| SD - img2img | 11.47 | 13.06 | 14.66 | 17.25 |
+| SD - inpaint | 11.67 | 13.31 | 14.88 | 17.48 |
+| SD - controlnet | 8.28 | 9.38 | 10.51 | 12.41 |
+| IF | 25.02 | 18.04 | ❌ | 48.47 |
+
+### A100 (batch size: 16)
+
+| **Pipeline** | **torch 2.0 - <br>no compile** | **torch nightly - <br>no compile** | **torch 2.0 - <br>compile** | **torch nightly - <br>compile** |
+|:---:|:---:|:---:|:---:|:---:|
+| SD - txt2img | 3.04 | 3.6 | 3.83 | 4.68 |
+| SD - img2img | 2.98 | 3.58 | 3.83 | 4.67 |
+| SD - inpaint | 3.04 | 3.66 | 3.9 | 4.76 |
+| SD - controlnet | 2.15 | 2.58 | 2.74 | 3.35 |
+| IF | 8.78 | 9.82 | ❌ | 16.77 |
+
+### V100 (batch size: 1)
+
+| **Pipeline** | **torch 2.0 - <br>no compile** | **torch nightly - <br>no compile** | **torch 2.0 - <br>compile** | **torch nightly - <br>compile** |
+|:---:|:---:|:---:|:---:|:---:|
+| SD - txt2img | 18.99 | 19.14 | 20.95 | 22.17 |
+| SD - img2img | 18.56 | 19.18 | 20.95 | 22.11 |
+| SD - inpaint | 19.14 | 19.06 | 21.08 | 22.20 |
+| SD - controlnet | 13.48 | 13.93 | 15.18 | 15.88 |
+| IF |  20.01 / <br>9.08 / <br>23.34 | 19.79 / <br>8.98 / <br>24.10 | ❌ | 55.75 / <br>11.57 / <br>57.67 |
+
+### V100 (batch size: 4)
+
+| **Pipeline** | **torch 2.0 - <br>no compile** | **torch nightly - <br>no compile** | **torch 2.0 - <br>compile** | **torch nightly - <br>compile** |
+|:---:|:---:|:---:|:---:|:---:|
+| SD - txt2img | 5.96 | 5.89 | 6.83 | 6.86 |
+| SD - img2img | 5.90 | 5.91 | 6.81 | 6.82 |
+| SD - inpaint | 5.99 | 6.03 | 6.93 | 6.95 |
+| SD - controlnet | 4.26 | 4.29 | 4.92 | 4.93 |
+| IF | 15.41 | 14.76 | ❌ | 22.95 |
+
+### V100 (batch size: 16)
+
+| **Pipeline** | **torch 2.0 - <br>no compile** | **torch nightly - <br>no compile** | **torch 2.0 - <br>compile** | **torch nightly - <br>compile** |
+|:---:|:---:|:---:|:---:|:---:|
+| SD - txt2img | 1.66 | 1.66 | 1.92 | 1.90 |
+| SD - img2img | 1.65 | 1.65 | 1.91 | 1.89 |
+| SD - inpaint | 1.69 | 1.69 | 1.95 | 1.93 |
+| SD - controlnet | 1.19 | 1.19 | OOM after warmup | 1.36 |
+| IF | 5.43 | 5.29 | ❌ | 7.06 |
+
+### T4 (batch size: 1)
+
+| **Pipeline** | **torch 2.0 - <br>no compile** | **torch nightly - <br>no compile** | **torch 2.0 - <br>compile** | **torch nightly - <br>compile** |
+|:---:|:---:|:---:|:---:|:---:|
+| SD - txt2img | 6.9 | 6.95 | 7.3 | 7.56 |
+| SD - img2img | 6.84 | 6.99 | 7.04 | 7.55 |
+| SD - inpaint | 6.91 | 6.7 | 7.01 | 7.37 |
+| SD - controlnet | 4.89 | 4.86 | 5.35 | 5.48 |
+| IF | 17.42 / <br>2.47 / <br>18.52 | 16.96 / <br>2.45 / <br>18.69 | ❌ | 24.63 / <br>2.47 / <br>23.39 |
+
+### T4 (batch size: 4)
+
+| **Pipeline** | **torch 2.0 - <br>no compile** | **torch nightly - <br>no compile** | **torch 2.0 - <br>compile** | **torch nightly - <br>compile** |
+|:---:|:---:|:---:|:---:|:---:|
+| SD - txt2img | 1.79 | 1.79 | 2.03 | 1.99 |
+| SD - img2img | 1.77 | 1.77 | 2.05 | 2.04 |
+| SD - inpaint | 1.81 | 1.82 | 2.09 | 2.09 |
+| SD - controlnet | 1.34 | 1.27 | 1.47 | 1.46 |
+| IF | 5.79 |  5.61 | ❌ | 7.39 |
+
+### T4 (batch size: 16)
+
+| **Pipeline** | **torch 2.0 - <br>no compile** | **torch nightly - <br>no compile** | **torch 2.0 - <br>compile** | **torch nightly - <br>compile** |
+|:---:|:---:|:---:|:---:|:---:|
+| SD - txt2img | 2.34s | 2.30s | OOM after 2nd iteration | 1.99s |
+| SD - img2img | 2.35s | 2.31s | OOM after warmup | 2.00s |
+| SD - inpaint | 2.30s | 2.26s | OOM after 2nd iteration | 1.95s |
+| SD - controlnet | OOM after 2nd iteration | OOM after 2nd iteration | OOM after warmup | OOM after warmup |
+| IF * | 1.44 | 1.44 | ❌ | 1.94 |
+
+### RTX 3090 (batch size: 1)
+
+| **Pipeline** | **torch 2.0 - <br>no compile** | **torch nightly - <br>no compile** | **torch 2.0 - <br>compile** | **torch nightly - <br>compile** |
+|:---:|:---:|:---:|:---:|:---:|
+| SD - txt2img | 22.56 | 22.84 | 23.84 | 25.69 |
+| SD - img2img | 22.25 | 22.61 | 24.1 | 25.83 |
+| SD - inpaint | 22.22 | 22.54 | 24.26 | 26.02 |
+| SD - controlnet | 16.03 | 16.33 | 17.38 | 18.56 |
+| IF | 27.08 / <br>9.07 / <br>31.23 | 26.75 / <br>8.92 / <br>31.47 | ❌ | 68.08 / <br>11.16 / <br>65.29 |
+
+### RTX 3090 (batch size: 4)
+
+| **Pipeline** | **torch 2.0 - <br>no compile** | **torch nightly - <br>no compile** | **torch 2.0 - <br>compile** | **torch nightly - <br>compile** |
+|:---:|:---:|:---:|:---:|:---:|
+| SD - txt2img | 6.46 | 6.35 | 7.29 | 7.3 |
+| SD - img2img | 6.33 | 6.27 | 7.31 | 7.26 |
+| SD - inpaint | 6.47 | 6.4 | 7.44 | 7.39 |
+| SD - controlnet | 4.59 | 4.54 | 5.27 | 5.26 |
+| IF | 16.81 | 16.62 | ❌ | 21.57 |
+
+### RTX 3090 (batch size: 16)
+
+| **Pipeline** | **torch 2.0 - <br>no compile** | **torch nightly - <br>no compile** | **torch 2.0 - <br>compile** | **torch nightly - <br>compile** |
+|:---:|:---:|:---:|:---:|:---:|
+| SD - txt2img | 1.7 | 1.69 | 1.93 | 1.91 |
+| SD - img2img | 1.68 | 1.67 | 1.93 | 1.9 |
+| SD - inpaint | 1.72 | 1.71 | 1.97 | 1.94 |
+| SD - controlnet | 1.23 | 1.22 | 1.4 | 1.38 |
+| IF | 5.01 | 5.00 | ❌ | 6.33 |
+
+### RTX 4090 (batch size: 1)
+
+| **Pipeline** | **torch 2.0 - <br>no compile** | **torch nightly - <br>no compile** | **torch 2.0 - <br>compile** | **torch nightly - <br>compile** |
+|:---:|:---:|:---:|:---:|:---:|
+| SD - txt2img | 40.5 | 41.89 | 44.65 | 49.81 |
+| SD - img2img | 40.39 | 41.95 | 44.46 | 49.8 |
+| SD - inpaint | 40.51 | 41.88 | 44.58 | 49.72 |
+| SD - controlnet | 29.27 | 30.29 | 32.26 | 36.03 |
+| IF | 69.71 / <br>18.78 / <br>85.49 | 69.13 / <br>18.80 / <br>85.56 | ❌ | 124.60 / <br>26.37 / <br>138.79 |
+
+### RTX 4090 (batch size: 4)
+
+| **Pipeline** | **torch 2.0 - <br>no compile** | **torch nightly - <br>no compile** | **torch 2.0 - <br>compile** | **torch nightly - <br>compile** |
+|:---:|:---:|:---:|:---:|:---:|
+| SD - txt2img | 12.62 | 12.84 | 15.32 | 15.59 |
+| SD - img2img | 12.61 | 12,.79 | 15.35 | 15.66 |
+| SD - inpaint | 12.65 | 12.81 | 15.3 | 15.58 |
+| SD - controlnet | 9.1 | 9.25 | 11.03 | 11.22 |
+| IF | 31.88 | 31.14 | ❌ | 43.92 |
+
+### RTX 4090 (batch size: 16)
+
+| **Pipeline** | **torch 2.0 - <br>no compile** | **torch nightly - <br>no compile** | **torch 2.0 - <br>compile** | **torch nightly - <br>compile** |
+|:---:|:---:|:---:|:---:|:---:|
+| SD - txt2img | 3.17 | 3.2 | 3.84 | 3.85 |
+| SD - img2img | 3.16 | 3.2 | 3.84 | 3.85 |
+| SD - inpaint | 3.17 | 3.2 | 3.85 | 3.85 |
+| SD - controlnet | 2.23 | 2.3 | 2.7 | 2.75 |
+| IF | 9.26 | 9.2 | ❌ | 13.31 |
+
+## 참고
+
+* Follow [this PR](https://github.com/huggingface/diffusers/pull/3313) for more details on the environment used for conducting the benchmarks. 
+* For the IF pipeline and batch sizes > 1, we only used a batch size of >1 in the first IF pipeline for text-to-image generation and NOT for upscaling. So, that means the two upscaling pipelines received a batch size of 1. 
+
+*Thanks to [Horace He](https://github.com/Chillee) from the PyTorch team for their support in improving our support of `torch.compile()` in Diffusers.*
+
+* 벤치마크 수행에 사용된 환경에 대한 자세한 내용은 [이 PR](https://github.com/huggingface/diffusers/pull/3313)을 참조하세요.
+* IF 파이프라인와 배치 크기 > 1의 경우 첫 번째 IF 파이프라인에서 text-to-image 생성을 위한 배치 크기 > 1만 사용했으며 업스케일링에는 사용하지 않았습니다. 즉, 두 개의 업스케일링 파이프라인이 배치 크기 1임을 의미합니다.
+
+*Diffusers에서 `torch.compile()` 지원을 개선하는 데 도움을 준 PyTorch 팀의 [Horace He](https://github.com/Chillee)에게 감사드립니다.*

docs/source/ko/training/adapt_a_model.mdx

@@ -0,0 +1,54 @@
+<!--Copyright 2023 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.
+-->
+
+# 새로운 작업에 대한 모델을 적용하기
+
+많은 diffusion 시스템은 같은 구성 요소들을 공유하므로 한 작업에 대해 사전학습된 모델을 완전히 다른 작업에 적용할 수 있습니다.
+
+이 인페인팅을 위한 가이드는 사전학습된 [`UNet2DConditionModel`]의 아키텍처를 초기화하고 수정하여 사전학습된 text-to-image 모델을 어떻게 인페인팅에 적용하는지를 알려줄 것입니다.
+
+## UNet2DConditionModel 파라미터 구성
+
+[`UNet2DConditionModel`]은 [input sample](https://huggingface.co/docs/diffusers/v0.16.0/en/api/models#diffusers.UNet2DConditionModel.in_channels)에서 4개의 채널을 기본적으로 허용합니다. 예를 들어,  [`runwayml/stable-diffusion-v1-5`](https://huggingface.co/runwayml/stable-diffusion-v1-5)와 같은 사전학습된 text-to-image 모델을 불러오고 `in_channels`의 수를 확인합니다:
+
+```py
+from diffusers import StableDiffusionPipeline
+
+pipeline = StableDiffusionPipeline.from_pretrained("runwayml/stable-diffusion-v1-5")
+pipeline.unet.config["in_channels"]
+4
+```
+
+인페인팅은 입력 샘플에 9개의 채널이 필요합니다. [`runwayml/stable-diffusion-inpainting`](https://huggingface.co/runwayml/stable-diffusion-inpainting)와 같은 사전학습된 인페인팅 모델에서 이 값을 확인할 수 있습니다:
+
+```py
+from diffusers import StableDiffusionPipeline
+
+pipeline = StableDiffusionPipeline.from_pretrained("runwayml/stable-diffusion-inpainting")
+pipeline.unet.config["in_channels"]
+9
+```
+
+인페인팅에 대한 text-to-image 모델을 적용하기 위해, `in_channels` 수를 4에서 9로 수정해야 할 것입니다.
+
+사전학습된 text-to-image 모델의 가중치와 [`UNet2DConditionModel`]을 초기화하고 `in_channels`를 9로 수정해 주세요. `in_channels`의 수를 수정하면 크기가 달라지기 때문에 크기가 안 맞는 오류를 피하기 위해 `ignore_mismatched_sizes=True` 및 `low_cpu_mem_usage=False`를 설정해야 합니다.
+
+```py
+from diffusers import UNet2DConditionModel
+
+model_id = "runwayml/stable-diffusion-v1-5"
+unet = UNet2DConditionModel.from_pretrained(
+    model_id, subfolder="unet", in_channels=9, low_cpu_mem_usage=False, ignore_mismatched_sizes=True
+)
+```
+
+Text-to-image 모델로부터 다른 구성 요소의 사전학습된 가중치는 체크포인트로부터 초기화되지만 `unet`의 입력 채널 가중치 (`conv_in.weight`)는 랜덤하게 초기화됩니다. 그렇지 않으면 모델이 노이즈를 리턴하기 때문에 인페인팅의 모델을 파인튜닝 할 때 중요합니다.

docs/source/ko/training/dreambooth.mdx

@@ -273,7 +273,7 @@ from diffusers import DiffusionPipeline, UNet2DConditionModel
 from transformers import CLIPTextModel
 import torch
 
-# 학습에 사용된 것과 동일한 인수(model, revision)로 파이프라인을 로드합니다.
+# 학습에 사용된 것과 동일한 인수(model, revision)로 파이프라인을 불러옵니다.
 model_id = "CompVis/stable-diffusion-v1-4"
 
 unet = UNet2DConditionModel.from_pretrained("/sddata/dreambooth/daruma-v2-1/checkpoint-100/unet")
@@ -294,7 +294,7 @@ If you have **`"accelerate<0.16.0"`** installed, you need to convert it to an in
 from accelerate import Accelerator
 from diffusers import DiffusionPipeline
 
-# 학습에 사용된 것과 동일한 인수(model, revision)로 파이프라인을 로드합니다.
+# 학습에 사용된 것과 동일한 인수(model, revision)로 파이프라인을 불러옵니다.
 model_id = "CompVis/stable-diffusion-v1-4"
 pipeline = DiffusionPipeline.from_pretrained(model_id)
 

docs/source/ko/training/lora.mdx

@@ -102,7 +102,7 @@ accelerate launch train_dreambooth_lora.py \
 >>> pipe = StableDiffusionPipeline.from_pretrained(model_base, torch_dtype=torch.float16)
 ```
 
-*기본 모델의 가중치 위에* 파인튜닝된 DreamBooth 모델에서 LoRA 가중치를 로드한 다음, 더 빠른 추론을 위해 파이프라인을 GPU로 이동합니다. LoRA 가중치를 프리징된 사전 훈련된 모델 가중치와 병합할 때, 선택적으로 'scale' 매개변수로 어느 정도의 가중치를 병합할 지 조절할 수 있습니다:
+*기본 모델의 가중치 위에* 파인튜닝된 DreamBooth 모델에서 LoRA 가중치를 불러온 다음, 더 빠른 추론을 위해 파이프라인을 GPU로 이동합니다. LoRA 가중치를 프리징된 사전 훈련된 모델 가중치와 병합할 때, 선택적으로 'scale' 매개변수로 어느 정도의 가중치를 병합할 지 조절할 수 있습니다:
 
 <Tip>
 

docs/source/ko/training/overview.mdx

@@ -0,0 +1,73 @@
+<!--Copyright 2023 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.
+-->
+
+# 🧨 Diffusers 학습 예시
+
+이번 챕터에서는 다양한 유즈케이스들에 대한 예제 코드들을 통해 어떻게하면 효과적으로 `diffusers` 라이브러리를 사용할 수 있을까에 대해 알아보도록 하겠습니다. 
+
+**Note**: 혹시 오피셜한 예시코드를 찾고 있다면, [여기](https://github.com/huggingface/diffusers/tree/main/src/diffusers/pipelines)를 참고해보세요!
+
+여기서 다룰 예시들은 다음을 지향합니다.
+
+- **손쉬운 디펜던시 설치** (Self-contained) : 여기서 사용될 예시 코드들의 디펜던시 패키지들은 전부 `pip install` 명령어를 통해 설치 가능한 패키지들입니다. 또한 친절하게 `requirements.txt` 파일에 해당 패키지들이 명시되어 있어, `pip install -r requirements.txt`로 간편하게 해당 디펜던시들을 설치할 수 있습니다. 예시: [train_unconditional.py](https://github.com/huggingface/diffusers/blob/main/examples/unconditional_image_generation/train_unconditional.py), [requirements.txt](https://github.com/huggingface/diffusers/blob/main/examples/unconditional_image_generation/requirements.txt) 
+- **손쉬운 수정** (Easy-to-tweak) : 저희는 가능하면 많은 유즈 케이스들을 제공하고자 합니다. 하지만 예시는 결국 그저 예시라는 점들 기억해주세요. 여기서 제공되는 예시코드들을 그저 단순히 복사-붙혀넣기하는 식으로는 여러분이 마주한 문제들을 손쉽게 해결할 순 없을 것입니다. 다시 말해 어느 정도는 여러분의 상황과 니즈에 맞춰 코드를 일정 부분 고쳐나가야 할 것입니다. 따라서 대부분의 학습 예시들은 데이터의 전처리 과정과 학습 과정에 대한 코드들을 함께 제공함으로써, 사용자가 니즈에 맞게 손쉬운 수정할 수 있도록 돕고 있습니다.
+- **입문자 친화적인** (Beginner-friendly) : 이번 챕터는 diffusion 모델과 `diffusers` 라이브러리에 대한 전반적인 이해를 돕기 위해 작성되었습니다. 따라서 diffusion 모델에 대한 최신 SOTA (state-of-the-art) 방법론들 가운데서도, 입문자에게는 많이 어려울 수 있다고 판단되면, 해당 방법론들은 여기서 다루지 않으려고 합니다.
+- **하나의 태스크만 포함할 것**(One-purpose-only): 여기서 다룰 예시들은 하나의 태스크만 포함하고 있어야 합니다. 물론 이미지 초해상화(super-resolution)와 이미지 보정(modification)과 같은 유사한 모델링 프로세스를 갖는 태스크들이 존재하겠지만, 하나의 예제에 하나의 태스크만을 담는 것이 더 이해하기 용이하다고 판단했기 때문입니다.
+
+
+
+저희는 diffusion 모델의 대표적인 태스크들을 다루는 공식 예제를 제공하고 있습니다. *공식* 예제는 현재 진행형으로 `diffusers` 관리자들(maintainers)에 의해 관리되고 있습니다. 또한 저희는 앞서 정의한 저희의 철학을 엄격하게 따르고자 노력하고 있습니다. 혹시 여러분께서 이러한 예시가 반드시 필요하다고 생각되신다면, 언제든지 [Feature Request](https://github.com/huggingface/diffusers/issues/new?assignees=&labels=&template=feature_request.md&title=) 혹은 직접 [Pull Request](https://github.com/huggingface/diffusers/compare)를 주시기 바랍니다. 저희는 언제나 환영입니다!
+
+학습 예시들은 다양한 태스크들에 대해 diffusion 모델을 사전학습(pretrain)하거나 파인튜닝(fine-tuning)하는 법을 보여줍니다. 현재 다음과 같은 예제들을 지원하고 있습니다.
+
+- [Unconditional Training](./unconditional_training)
+- [Text-to-Image Training](./text2image)
+- [Text Inversion](./text_inversion)
+- [Dreambooth](./dreambooth)
+
+memory-efficient attention 연산을 수행하기 위해, 가능하면 [xFormers](../optimization/xformers)를 설치해주시기 바랍니다. 이를 통해 학습 속도를 늘리고 메모리에 대한 부담을 줄일 수 있습니다.
+
+| Task | 🤗 Accelerate | 🤗 Datasets | Colab
+|---|---|:---:|:---:|
+| [**Unconditional Image Generation**](./unconditional_training) | ✅ | ✅ | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/huggingface/notebooks/blob/main/diffusers/training_example.ipynb)
+| [**Text-to-Image fine-tuning**](./text2image) | ✅ | ✅ | 
+| [**Textual Inversion**](./text_inversion) | ✅ | - | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/huggingface/notebooks/blob/main/diffusers/sd_textual_inversion_training.ipynb)
+| [**Dreambooth**](./dreambooth) | ✅ | - | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/huggingface/notebooks/blob/main/diffusers/sd_dreambooth_training.ipynb)
+| [**Training with LoRA**](./lora) | ✅ | - | - |
+| [**ControlNet**](./controlnet) | ✅ | ✅ | - |
+| [**InstructPix2Pix**](./instructpix2pix) | ✅ | ✅ | - |
+| [**Custom Diffusion**](./custom_diffusion) | ✅ | ✅ | - |
+
+
+## 커뮤니티
+
+공식 예제 외에도 **커뮤니티 예제** 역시 제공하고 있습니다. 해당 예제들은 우리의 커뮤니티에 의해 관리됩니다. 커뮤니티 예쩨는 학습 예시나 추론 파이프라인으로 구성될 수 있습니다. 이러한 커뮤니티 예시들의 경우,  앞서 정의했던 철학들을 좀 더 관대하게 적용하고 있습니다. 또한 이러한 커뮤니티 예시들의 경우, 모든 이슈들에 대한 유지보수를 보장할 수는 없습니다.
+
+유용하긴 하지만, 아직은 대중적이지 못하거나 저희의 철학에 부합하지 않는 예제들은 [community examples](https://github.com/huggingface/diffusers/tree/main/examples/community) 폴더에 담기게 됩니다.
+
+**Note**: 커뮤니티 예제는 `diffusers`에 기여(contribution)를 희망하는 분들에게 [아주 좋은 기여 수단](https://github.com/huggingface/diffusers/issues?q=is%3Aopen+is%3Aissue+label%3A%22good+first+issue%22)이 될 수 있습니다.
+
+## 주목할 사항들
+
+최신 버전의 예시 코드들의 성공적인 구동을 보장하기 위해서는, 반드시 **소스코드를 통해 `diffusers`를 설치해야 하며,** 해당 예시 코드들이 요구하는 디펜던시들 역시 설치해야 합니다. 이를 위해 새로운 가상 환경을 구축하고 다음의 명령어를 실행해야 합니다.
+
+```bash
+git clone https://github.com/huggingface/diffusers
+cd diffusers
+pip install .
+```
+
+그 다음 `cd` 명령어를 통해 해당 예제 디렉토리에 접근해서 다음 명령어를 실행하면 됩니다.
+
+```bash
+pip install -r requirements.txt
+```

docs/source/ko/training/text_inversion.mdx

@@ -0,0 +1,275 @@
+ <!--Copyright 2023 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.
+-->
+
+
+
+# Textual-Inversion
+
+[[open-in-colab]]
+
+[textual-inversion](https://arxiv.org/abs/2208.01618)은 소수의 예시 이미지에서 새로운 콘셉트를 포착하는 기법입니다. 이 기술은 원래 [Latent Diffusion](https://github.com/CompVis/latent-diffusion)에서 시연되었지만, 이후 [Stable Diffusion](https://huggingface.co/docs/diffusers/main/en/conceptual/stable_diffusion)과 같은 유사한 다른 모델에도 적용되었습니다. 학습된 콘셉트는 text-to-image 파이프라인에서 생성된 이미지를 더 잘 제어하는 데 사용할 수 있습니다. 이 모델은 텍스트 인코더의 임베딩 공간에서 새로운 '단어'를 학습하여 개인화된 이미지 생성을 위한 텍스트 프롬프트 내에서 사용됩니다.
+
+![Textual Inversion example](https://textual-inversion.github.io/static/images/editing/colorful_teapot.JPG)
+<small>By using just 3-5 images you can teach new concepts to a model such as Stable Diffusion for personalized image generation <a href="https://github.com/rinongal/textual_inversion">(image source)</a>.</small>
+
+이 가이드에서는 textual-inversion으로 [`runwayml/stable-diffusion-v1-5`](https://huggingface.co/runwayml/stable-diffusion-v1-5) 모델을 학습하는 방법을 설명합니다. 이 가이드에서 사용된 모든 textual-inversion 학습 스크립트는 [여기](https://github.com/huggingface/diffusers/tree/main/examples/textual_inversion)에서 확인할 수 있습니다. 내부적으로 어떻게 작동하는지 자세히 살펴보고 싶으시다면 해당 링크를 참조해주시기 바랍니다.
+
+<Tip>
+
+[Stable Diffusion Textual Inversion Concepts Library](https://huggingface.co/sd-concepts-library)에는 커뮤니티에서 제작한 학습된 textual-inversion 모델들이 있습니다. 시간이 지남에 따라 더 많은 콘셉트들이 추가되어 유용한 리소스로 성장할 것입니다!
+
+</Tip>
+
+시작하기 전에 학습을 위한 의존성 라이브러리들을 설치해야 합니다:
+
+```bash
+pip install diffusers accelerate transformers
+```
+
+의존성 라이브러리들의 설치가 완료되면, [🤗Accelerate](https://github.com/huggingface/accelerate/) 환경을 초기화시킵니다.
+
+```bash
+accelerate config
+```
+
+별도의 설정없이, 기본 🤗Accelerate 환경을 설정하려면 다음과 같이 하세요:
+
+```bash
+accelerate config default
+```
+
+또는 사용 중인 환경이 노트북과 같은 대화형 셸을 지원하지 않는다면, 다음과 같이 사용할 수 있습니다:
+
+```py
+from accelerate.utils import write_basic_config
+
+write_basic_config()
+```
+
+마지막으로, Memory-Efficient Attention을 통해 메모리 사용량을 줄이기 위해 [xFormers](https://huggingface.co/docs/diffusers/main/en/training/optimization/xformers)를 설치합니다. xFormers를 설치한 후, 학습 스크립트에 `--enable_xformers_memory_efficient_attention` 인자를 추가합니다. xFormers는 Flax에서 지원되지 않습니다.
+
+## 허브에 모델 업로드하기
+
+모델을 허브에 저장하려면, 학습 스크립트에 다음 인자를 추가해야 합니다.
+
+```bash
+--push_to_hub
+```
+
+## 체크포인트 저장 및 불러오기
+
+학습중에 모델의 체크포인트를 정기적으로 저장하는 것이 좋습니다. 이렇게 하면 어떤 이유로든 학습이 중단된 경우 저장된 체크포인트에서 학습을 다시 시작할 수 있습니다. 학습 스크립트에 다음 인자를 전달하면 500단계마다 전체 학습 상태가 `output_dir`의 하위 폴더에 체크포인트로서 저장됩니다.
+
+```bash
+--checkpointing_steps=500
+```
+
+저장된 체크포인트에서 학습을 재개하려면, 학습 스크립트와 재개할 특정 체크포인트에 다음 인자를 전달하세요.
+
+```bash
+--resume_from_checkpoint="checkpoint-1500"
+```
+
+## 파인 튜닝
+
+학습용 데이터셋으로 [고양이 장난감 데이터셋](https://huggingface.co/datasets/diffusers/cat_toy_example)을 다운로드하여 디렉토리에 저장하세요. 여러분만의 고유한 데이터셋을 사용하고자 한다면, [학습용 데이터셋 만들기](https://huggingface.co/docs/diffusers/training/create_dataset) 가이드를 살펴보시기 바랍니다.
+
+```py
+from huggingface_hub import snapshot_download
+
+local_dir = "./cat"
+snapshot_download(
+    "diffusers/cat_toy_example", local_dir=local_dir, repo_type="dataset", ignore_patterns=".gitattributes"
+)
+```
+
+모델의 리포지토리 ID(또는 모델 가중치가 포함된 디렉터리 경로)를 `MODEL_NAME` 환경 변수에 할당하고, 해당 값을 [`pretrained_model_name_or_path`](https://huggingface.co/docs/diffusers/en/api/diffusion_pipeline#diffusers.DiffusionPipeline.from_pretrained.pretrained_model_name_or_path) 인자에 전달합니다. 그리고 이미지가 포함된 디렉터리 경로를 `DATA_DIR` 환경 변수에 할당합니다.
+
+이제 [학습 스크립트](https://github.com/huggingface/diffusers/blob/main/examples/textual_inversion/textual_inversion.py)를 실행할 수 있습니다. 스크립트는 다음 파일을 생성하고 리포지토리에 저장합니다.
+
+- `learned_embeds.bin` 
+- `token_identifier.txt`
+- `type_of_concept.txt`.
+
+<Tip>
+
+💡V100 GPU 1개를 기준으로 전체 학습에는 최대 1시간이 걸립니다. 학습이 완료되기를 기다리는 동안 궁금한 점이 있으면 아래 섹션에서 [textual-inversion이 어떻게 작동하는지](https://huggingface.co/docs/diffusers/training/text_inversion#how-it-works) 자유롭게 확인하세요 ! 
+
+</Tip>
+
+<frameworkcontent>
+<pt>
+```bash
+export MODEL_NAME="runwayml/stable-diffusion-v1-5"
+export DATA_DIR="./cat"
+
+accelerate launch textual_inversion.py \
+  --pretrained_model_name_or_path=$MODEL_NAME \
+  --train_data_dir=$DATA_DIR \
+  --learnable_property="object" \
+  --placeholder_token="<cat-toy>" --initializer_token="toy" \
+  --resolution=512 \
+  --train_batch_size=1 \
+  --gradient_accumulation_steps=4 \
+  --max_train_steps=3000 \
+  --learning_rate=5.0e-04 --scale_lr \
+  --lr_scheduler="constant" \
+  --lr_warmup_steps=0 \
+  --output_dir="textual_inversion_cat" \
+  --push_to_hub
+```
+
+<Tip>
+
+💡학습 성능을 올리기 위해, 플레이스홀더 토큰(`<cat-toy>`)을 (단일한 임베딩 벡터가 아닌) 복수의 임베딩 벡터로 표현하는 것 역시 고려할 있습니다.  이러한 트릭이 모델이 보다 복잡한 이미지의 스타일(앞서 말한 콘셉트)을 더 잘 캡처하는 데 도움이 될 수 있습니다. 복수의 임베딩 벡터 학습을 활성화하려면 다음 옵션을 전달하십시오.
+
+```bash
+--num_vectors=5
+```
+
+</Tip>
+</pt>
+<jax>
+
+TPU에 액세스할 수 있는 경우, [Flax 학습 스크립트](https://github.com/huggingface/diffusers/blob/main/examples/textual_inversion/textual_inversion_flax.py)를 사용하여 더 빠르게 모델을 학습시켜보세요. (물론 GPU에서도 작동합니다.) 동일한 설정에서 Flax 학습 스크립트는 PyTorch 학습 스크립트보다 최소 70% 더 빨라야 합니다! ⚡️
+
+시작하기 앞서 Flax에 대한 의존성 라이브러리들을 설치해야 합니다.
+
+```bash
+pip install -U -r requirements_flax.txt
+```
+
+모델의 리포지토리 ID(또는 모델 가중치가 포함된 디렉터리 경로)를 `MODEL_NAME` 환경 변수에 할당하고, 해당 값을 [`pretrained_model_name_or_path`](https://huggingface.co/docs/diffusers/en/api/diffusion_pipeline#diffusers.DiffusionPipeline.from_pretrained.pretrained_model_name_or_path) 인자에 전달합니다.
+
+그런 다음 [학습 스크립트](https://github.com/huggingface/diffusers/blob/main/examples/textual_inversion/textual_inversion_flax.py)를 시작할 수 있습니다.
+
+```bash
+export MODEL_NAME="duongna/stable-diffusion-v1-4-flax"
+export DATA_DIR="./cat"
+
+python textual_inversion_flax.py \
+  --pretrained_model_name_or_path=$MODEL_NAME \
+  --train_data_dir=$DATA_DIR \
+  --learnable_property="object" \
+  --placeholder_token="<cat-toy>" --initializer_token="toy" \
+  --resolution=512 \
+  --train_batch_size=1 \
+  --max_train_steps=3000 \
+  --learning_rate=5.0e-04 --scale_lr \
+  --output_dir="textual_inversion_cat" \
+  --push_to_hub
+```
+</jax>
+</frameworkcontent>
+
+### 중간 로깅
+
+모델의 학습 진행 상황을 추적하는 데 관심이 있는 경우, 학습 과정에서 생성된 이미지를 저장할 수 있습니다. 학습 스크립트에 다음 인수를 추가하여 중간 로깅을 활성화합니다.
+
+- `validation_prompt` : 샘플을 생성하는 데 사용되는 프롬프트(기본값은 `None`으로 설정되며, 이 때 중간 로깅은 비활성화됨)
+- `num_validation_images` : 생성할 샘플 이미지 수
+- `validation_steps` : `validation_prompt`로부터 샘플 이미지를 생성하기 전 스텝의 수
+
+```bash
+--validation_prompt="A <cat-toy> backpack"
+--num_validation_images=4
+--validation_steps=100
+```
+
+## 추론
+
+모델을 학습한 후에는, 해당 모델을 [`StableDiffusionPipeline`]을 사용하여 추론에 사용할 수 있습니다.
+
+textual-inversion 스크립트는 기본적으로 textual-inversion을 통해 얻어진 임베딩 벡터만을 저장합니다. 해당 임베딩 벡터들은 텍스트 인코더의 임베딩 행렬에 추가되어 있습습니다.
+
+<frameworkcontent>
+<pt>
+<Tip>
+
+💡 커뮤니티는 [sd-concepts-library](https://huggingface.co/sd-concepts-library) 라는 대규모의 textual-inversion 임베딩 벡터 라이브러리를 만들었습니다. textual-inversion 임베딩을 밑바닥부터 학습하는 대신, 해당 라이브러리에 본인이 찾는 textual-inversion 임베딩이 이미 추가되어 있지 않은지를 확인하는 것도 좋은 방법이 될 것 같습니다.
+
+</Tip>
+
+textual-inversion 임베딩 벡터을 불러오기 위해서는, 먼저 해당 임베딩 벡터를 학습할 때 사용한 모델을 불러와야 합니다. 여기서는  [`runwayml/stable-diffusion-v1-5`](https://huggingface.co/docs/diffusers/training/runwayml/stable-diffusion-v1-5) 모델이 사용되었다고 가정하고 불러오겠습니다.
+
+```python
+from diffusers import StableDiffusionPipeline
+import torch
+
+model_id = "runwayml/stable-diffusion-v1-5"
+pipe = StableDiffusionPipeline.from_pretrained(model_id, torch_dtype=torch.float16).to("cuda")
+```
+
+다음으로 `TextualInversionLoaderMixin.load_textual_inversion` 함수를 통해, textual-inversion 임베딩 벡터를 불러와야 합니다. 여기서 우리는 이전의 `<cat-toy>` 예제의 임베딩을 불러올 것입니다.
+
+```python
+pipe.load_textual_inversion("sd-concepts-library/cat-toy")
+```
+
+이제 플레이스홀더 토큰(`<cat-toy>`)이 잘 동작하는지를 확인하는 파이프라인을 실행할 수 있습니다.
+
+```python
+prompt = "A <cat-toy> backpack"
+
+image = pipe(prompt, num_inference_steps=50).images[0]
+image.save("cat-backpack.png")
+```
+
+`TextualInversionLoaderMixin.load_textual_inversion`은 Diffusers 형식으로 저장된 텍스트 임베딩 벡터를 로드할 수 있을 뿐만 아니라, [Automatic1111](https://github.com/AUTOMATIC1111/stable-diffusion-webui) 형식으로 저장된 임베딩 벡터도 로드할 수 있습니다. 이렇게 하려면, 먼저 [civitAI](https://civitai.com/models/3036?modelVersionId=8387)에서 임베딩 벡터를 다운로드한 다음 로컬에서 불러와야 합니다.
+
+```python
+pipe.load_textual_inversion("./charturnerv2.pt")
+```
+</pt>
+<jax>
+
+현재 Flax에 대한 `load_textual_inversion` 함수는 없습니다. 따라서 학습 후 textual-inversion 임베딩 벡터가 모델의 일부로서 저장되었는지를 확인해야 합니다. 그런 다음은 다른 Flax 모델과 마찬가지로 실행할 수 있습니다.
+
+```python
+import jax
+import numpy as np
+from flax.jax_utils import replicate
+from flax.training.common_utils import shard
+from diffusers import FlaxStableDiffusionPipeline
+
+model_path = "path-to-your-trained-model"
+pipeline, params = FlaxStableDiffusionPipeline.from_pretrained(model_path, dtype=jax.numpy.bfloat16)
+
+prompt = "A <cat-toy> backpack"
+prng_seed = jax.random.PRNGKey(0)
+num_inference_steps = 50
+
+num_samples = jax.device_count()
+prompt = num_samples * [prompt]
+prompt_ids = pipeline.prepare_inputs(prompt)
+
+# shard inputs and rng
+params = replicate(params)
+prng_seed = jax.random.split(prng_seed, jax.device_count())
+prompt_ids = shard(prompt_ids)
+
+images = pipeline(prompt_ids, params, prng_seed, num_inference_steps, jit=True).images
+images = pipeline.numpy_to_pil(np.asarray(images.reshape((num_samples,) + images.shape[-3:])))
+image.save("cat-backpack.png")
+```
+</jax>
+</frameworkcontent>
+
+## 작동 방식
+
+![Diagram from the paper showing overview](https://textual-inversion.github.io/static/images/training/training.JPG)
+<small>Architecture overview from the Textual Inversion <a href="https://textual-inversion.github.io/">blog post.</a></small>
+
+일반적으로 텍스트 프롬프트는 모델에 전달되기 전에 임베딩으로 토큰화됩니다. textual-inversion은 비슷한 작업을 수행하지만, 위 다이어그램의 특수 토큰 `S*`로부터 새로운 토큰 임베딩 `v*`를 학습합니다. 모델의 아웃풋은 디퓨전 모델을 조정하는 데 사용되며, 디퓨전 모델이 단 몇 개의 예제 이미지에서 신속하고 새로운 콘셉트를 이해하는 데 도움을 줍니다.
+
+이를 위해 textual-inversion은 제너레이터 모델과 학습용 이미지의 노이즈 버전을 사용합니다. 제너레이터는 노이즈가 적은 버전의 이미지를 예측하려고 시도하며 토큰 임베딩 `v*`은 제너레이터의 성능에 따라 최적화됩니다. 토큰 임베딩이 새로운 콘셉트를 성공적으로 포착하면 디퓨전 모델에 더 유용한 정보를 제공하고 노이즈가 적은 더 선명한 이미지를 생성하는 데 도움이 됩니다. 이러한 최적화 프로세스는 일반적으로 다양한 프롬프트와 이미지에 수천 번에 노출됨으로써 이루어집니다.
+

docs/source/ko/training/unconditional_training.mdx

@@ -0,0 +1,144 @@
+<!--Copyright 2023 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.
+-->
+
+# Unconditional 이미지 생성
+
+unconditional 이미지 생성은 text-to-image 또는 image-to-image 모델과 달리 텍스트나 이미지에 대한 조건이 없이 학습 데이터 분포와 유사한 이미지만을 생성합니다.
+
+<iframe
+	src="https://stevhliu-ddpm-butterflies-128.hf.space"
+	frameborder="0"
+	width="850"
+	height="550"
+></iframe>
+
+
+이 가이드에서는 기존에 존재하던 데이터셋과 자신만의 커스텀 데이터셋에 대해 unconditional image generation 모델을 훈련하는 방법을 설명합니다. 훈련 세부 사항에 대해 더 자세히 알고 싶다면 unconditional image generation을 위한 모든 학습 스크립트를 [여기](https://github.com/huggingface/diffusers/tree/main/examples/unconditional_image_generation)에서 확인할 수 있습니다.
+
+스크립트를 실행하기 전, 먼저 의존성 라이브러리들을 설치해야 합니다.
+
+```bash
+pip install diffusers[training] accelerate datasets
+```
+
+그 다음 🤗 [Accelerate](https://github.com/huggingface/accelerate/) 환경을 초기화합니다.
+
+```bash
+accelerate config
+```
+
+별도의 설정 없이 기본 설정으로 🤗 [Accelerate](https://github.com/huggingface/accelerate/) 환경을 초기화해봅시다.
+
+```bash
+accelerate config default
+```
+
+노트북과 같은 대화형 쉘을 지원하지 않는 환경의 경우, 다음과 같이 사용해볼 수도 있습니다.
+
+```py
+from accelerate.utils import write_basic_config
+
+write_basic_config()
+```
+
+## 모델을 허브에 업로드하기
+
+학습 스크립트에 다음 인자를 추가하여 허브에 모델을 업로드할 수 있습니다.
+
+```bash
+--push_to_hub
+```
+
+## 체크포인트 저장하고 불러오기
+
+훈련 중 문제가 발생할 경우를 대비하여 체크포인트를 정기적으로 저장하는 것이 좋습니다. 체크포인트를 저장하려면 학습 스크립트에 다음 인자를 전달합니다:
+
+```bash
+--checkpointing_steps=500
+```
+
+전체 훈련 상태는 500스텝마다 `output_dir`의 하위 폴더에 저장되며, 학습 스크립트에 `--resume_from_checkpoint` 인자를 전달함으로써 체크포인트를 불러오고 훈련을 재개할 수 있습니다.
+
+```bash
+--resume_from_checkpoint="checkpoint-1500"
+```
+
+## 파인튜닝
+
+이제 학습 스크립트를 시작할 준비가 되었습니다! `--dataset_name` 인자에 파인튜닝할 데이터셋 이름을 지정한 다음, `--output_dir` 인자에 지정된 경로로 저장합니다. 본인만의 데이터셋를 사용하려면, [학습용 데이터셋 만들기](create_dataset) 가이드를 참조하세요.
+
+학습 스크립트는 `diffusion_pytorch_model.bin` 파일을 생성하고, 그것을 당신의 리포지토리에 저장합니다.
+
+<Tip>
+
+💡 전체 학습은 V100 GPU 4개를 사용할 경우, 2시간이 소요됩니다.
+
+</Tip>
+
+예를 들어, [Oxford Flowers](https://huggingface.co/datasets/huggan/flowers-102-categories) 데이터셋을 사용해 파인튜닝할 경우:
+
+```bash
+accelerate launch train_unconditional.py \
+  --dataset_name="huggan/flowers-102-categories" \
+  --resolution=64 \
+  --output_dir="ddpm-ema-flowers-64" \
+  --train_batch_size=16 \
+  --num_epochs=100 \
+  --gradient_accumulation_steps=1 \
+  --learning_rate=1e-4 \
+  --lr_warmup_steps=500 \
+  --mixed_precision=no \
+  --push_to_hub
+```
+
+<div class="flex justify-center">
+    <img src="https://user-images.githubusercontent.com/26864830/180248660-a0b143d0-b89a-42c5-8656-2ebf6ece7e52.png"/>
+</div>
+[Pokemon](https://huggingface.co/datasets/huggan/pokemon) 데이터셋을 사용할 경우:
+
+```bash
+accelerate launch train_unconditional.py \
+  --dataset_name="huggan/pokemon" \
+  --resolution=64 \
+  --output_dir="ddpm-ema-pokemon-64" \
+  --train_batch_size=16 \
+  --num_epochs=100 \
+  --gradient_accumulation_steps=1 \
+  --learning_rate=1e-4 \
+  --lr_warmup_steps=500 \
+  --mixed_precision=no \
+  --push_to_hub
+```
+
+<div class="flex justify-center">
+    <img src="https://user-images.githubusercontent.com/26864830/180248200-928953b4-db38-48db-b0c6-8b740fe6786f.png"/>
+</div>
+
+### 여러개의 GPU로 훈련하기
+
+`accelerate`을 사용하면 원활한 다중 GPU 훈련이 가능합니다. `accelerate`을 사용하여 분산 훈련을 실행하려면 [여기](https://huggingface.co/docs/accelerate/basic_tutorials/launch) 지침을 따르세요. 다음은 명령어 예제입니다.
+
+```bash
+accelerate launch --mixed_precision="fp16" --multi_gpu train_unconditional.py \
+  --dataset_name="huggan/pokemon" \
+  --resolution=64 --center_crop --random_flip \
+  --output_dir="ddpm-ema-pokemon-64" \
+  --train_batch_size=16 \
+  --num_epochs=100 \
+  --gradient_accumulation_steps=1 \
+  --use_ema \
+  --learning_rate=1e-4 \
+  --lr_warmup_steps=500 \
+  --mixed_precision="fp16" \
+  --logger="wandb" \
+  --push_to_hub
+```

docs/source/ko/tutorials/basic_training.mdx

@@ -0,0 +1,405 @@
+<!--Copyright 2023 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.
+-->
+
+[[open-in-colab]]
+
+
+# Diffusion 모델을 학습하기
+
+Unconditional 이미지 생성은 학습에 사용된 데이터셋과 유사한 이미지를 생성하는 diffusion 모델에서 인기 있는 어플리케이션입니다. 일반적으로, 가장 좋은 결과는 특정 데이터셋에 사전 훈련된 모델을 파인튜닝하는 것으로 얻을 수 있습니다. 이 [허브](https://huggingface.co/search/full-text?q=unconditional-image-generation&type=model)에서 이러한 많은 체크포인트를 찾을 수 있지만, 만약 마음에 드는 체크포인트를 찾지 못했다면, 언제든지 스스로 학습할 수 있습니다!
+
+이 튜토리얼은 나만의 🦋 나비 🦋를 생성하기 위해 [Smithsonian Butterflies](https://huggingface.co/datasets/huggan/smithsonian_butterflies_subset) 데이터셋의 하위 집합에서 [`UNet2DModel`] 모델을 학습하는 방법을 가르쳐줄 것입니다.
+
+<Tip>
+
+💡 이 학습 튜토리얼은 [Training with 🧨 Diffusers](https://colab.research.google.com/github/huggingface/notebooks/blob/main/diffusers/training_example.ipynb) 노트북 기반으로 합니다. Diffusion 모델의 작동 방식 및 자세한 내용은 노트북을 확인하세요!
+
+</Tip>
+
+시작 전에, 🤗 Datasets을 불러오고 전처리하기 위해 데이터셋이 설치되어 있는지 다수 GPU에서 학습을 간소화하기 위해 🤗 Accelerate 가 설치되어 있는지 확인하세요. 그 후 학습 메트릭을 시각화하기 위해 [TensorBoard](https://www.tensorflow.org/tensorboard)를 또한 설치하세요. (또한 학습 추적을 위해 [Weights & Biases](https://docs.wandb.ai/)를 사용할 수 있습니다.)
+
+```bash
+!pip install diffusers[training]
+```
+
+커뮤니티에 모델을 공유할 것을 권장하며, 이를 위해서 Hugging Face 계정에 로그인을 해야 합니다. (계정이 없다면 [여기](https://hf.co/join)에서 만들 수 있습니다.) 노트북에서 로그인할 수 있으며 메시지가 표시되면 토큰을 입력할 수 있습니다.
+
+```py
+>>> from huggingface_hub import notebook_login
+
+>>> notebook_login()
+```
+
+또는 터미널로 로그인할 수 있습니다:
+
+```bash
+huggingface-cli login
+```
+
+모델 체크포인트가 상당히 크기 때문에 [Git-LFS](https://git-lfs.com/)에서 대용량 파일의 버전 관리를 할 수 있습니다.
+
+```bash
+!sudo apt -qq install git-lfs
+!git config --global credential.helper store
+```
+
+
+## 학습 구성
+
+편의를 위해 학습 파라미터들을 포함한 `TrainingConfig` 클래스를 생성합니다 (자유롭게 조정 가능):
+
+```py
+>>> from dataclasses import dataclass
+
+
+>>> @dataclass
+... class TrainingConfig:
+...     image_size = 128  # 생성되는 이미지 해상도
+...     train_batch_size = 16
+...     eval_batch_size = 16  # 평가 동안에 샘플링할 이미지 수
+...     num_epochs = 50
+...     gradient_accumulation_steps = 1
+...     learning_rate = 1e-4
+...     lr_warmup_steps = 500
+...     save_image_epochs = 10
+...     save_model_epochs = 30
+...     mixed_precision = "fp16"  # `no`는 float32, 자동 혼합 정밀도를 위한 `fp16`
+...     output_dir = "ddpm-butterflies-128"  # 로컬 및 HF Hub에 저장되는 모델명
+
+...     push_to_hub = True  # 저장된 모델을 HF Hub에 업로드할지 여부
+...     hub_private_repo = False
+...     overwrite_output_dir = True  # 노트북을 다시 실행할 때 이전 모델에 덮어씌울지
+...     seed = 0
+
+
+>>> config = TrainingConfig()
+```
+
+
+## 데이터셋 불러오기
+
+🤗 Datasets 라이브러리와 [Smithsonian Butterflies](https://huggingface.co/datasets/huggan/smithsonian_butterflies_subset) 데이터셋을 쉽게 불러올 수 있습니다.
+
+```py
+>>> from datasets import load_dataset
+
+>>> config.dataset_name = "huggan/smithsonian_butterflies_subset"
+>>> dataset = load_dataset(config.dataset_name, split="train")
+```
+
+💡[HugGan Community Event](https://huggingface.co/huggan) 에서 추가의 데이터셋을 찾거나 로컬의 [`ImageFolder`](https://huggingface.co/docs/datasets/image_dataset#imagefolder)를 만듦으로써 나만의 데이터셋을 사용할 수 있습니다. HugGan Community Event 에 가져온 데이터셋의 경우 레포지토리의 id로 `config.dataset_name` 을 설정하고, 나만의 이미지를 사용하는 경우 `imagefolder` 를 설정합니다.
+
+🤗 Datasets은 [`~datasets.Image`] 기능을 사용해 자동으로 이미지 데이터를 디코딩하고 [`PIL.Image`](https://pillow.readthedocs.io/en/stable/reference/Image.html)로 불러옵니다. 이를 시각화 해보면:
+
+```py
+>>> import matplotlib.pyplot as plt
+
+>>> fig, axs = plt.subplots(1, 4, figsize=(16, 4))
+>>> for i, image in enumerate(dataset[:4]["image"]):
+...     axs[i].imshow(image)
+...     axs[i].set_axis_off()
+>>> fig.show()
+```
+
+![](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/butterflies_ds.png)
+
+이미지는 모두 다른 사이즈이기 때문에, 우선 전처리가 필요합니다:
+
+-   `Resize` 는 `config.image_size` 에 정의된 이미지 사이즈로 변경합니다.
+-   `RandomHorizontalFlip` 은 랜덤적으로 이미지를 미러링하여 데이터셋을 보강합니다.
+-   `Normalize` 는 모델이 예상하는 [-1, 1] 범위로 픽셀 값을 재조정 하는데 중요합니다.
+
+```py
+>>> from torchvision import transforms
+
+>>> preprocess = transforms.Compose(
+...     [
+...         transforms.Resize((config.image_size, config.image_size)),
+...         transforms.RandomHorizontalFlip(),
+...         transforms.ToTensor(),
+...         transforms.Normalize([0.5], [0.5]),
+...     ]
+... )
+```
+
+ 학습 도중에 `preprocess` 함수를 적용하려면 🤗 Datasets의 [`~datasets.Dataset.set_transform`] 방법이 사용됩니다.
+
+```py
+>>> def transform(examples):
+...     images = [preprocess(image.convert("RGB")) for image in examples["image"]]
+...     return {"images": images}
+
+
+>>> dataset.set_transform(transform)
+```
+
+이미지의 크기가 조정되었는지 확인하기 위해 이미지를 다시 시각화해보세요. 이제 [DataLoader](https://pytorch.org/docs/stable/data#torch.utils.data.DataLoader)에 데이터셋을 포함해 학습할 준비가 되었습니다!
+
+```py
+>>> import torch
+
+>>> train_dataloader = torch.utils.data.DataLoader(dataset, batch_size=config.train_batch_size, shuffle=True)
+```
+
+
+## UNet2DModel 생성하기
+
+🧨 Diffusers에 사전학습된 모델들은 모델 클래스에서 원하는 파라미터로 쉽게 생성할 수 있습니다. 예를 들어, [`UNet2DModel`]를 생성하려면:
+
+```py
+>>> from diffusers import UNet2DModel
+
+>>> model = UNet2DModel(
+...     sample_size=config.image_size,  # 타겟 이미지 해상도
+...     in_channels=3,  # 입력 채널 수, RGB 이미지에서 3
+...     out_channels=3,  # 출력 채널 수
+...     layers_per_block=2,  # UNet 블럭당 몇 개의 ResNet 레이어가 사용되는지
+...     block_out_channels=(128, 128, 256, 256, 512, 512),  # 각 UNet 블럭을 위한 출력 채널 수
+...     down_block_types=(
+...         "DownBlock2D",  # 일반적인 ResNet 다운샘플링 블럭
+...         "DownBlock2D",
+...         "DownBlock2D",
+...         "DownBlock2D",
+...         "AttnDownBlock2D",  # spatial self-attention이 포함된 일반적인 ResNet 다운샘플링 블럭
+...         "DownBlock2D",
+...     ),
+...     up_block_types=(
+...         "UpBlock2D",  # 일반적인 ResNet 업샘플링 블럭
+...         "AttnUpBlock2D",  # spatial self-attention이 포함된 일반적인 ResNet 업샘플링 블럭
+...         "UpBlock2D",
+...         "UpBlock2D",
+...         "UpBlock2D",
+...         "UpBlock2D",
+...     ),
+... )
+```
+
+샘플의 이미지 크기와 모델 출력 크기가 맞는지 빠르게 확인하기 위한 좋은 아이디어가 있습니다:
+
+```py
+>>> sample_image = dataset[0]["images"].unsqueeze(0)
+>>> print("Input shape:", sample_image.shape)
+Input shape: torch.Size([1, 3, 128, 128])
+
+>>> print("Output shape:", model(sample_image, timestep=0).sample.shape)
+Output shape: torch.Size([1, 3, 128, 128])
+```
+
+훌륭해요! 다음, 이미지에 약간의 노이즈를 더하기 위해 스케줄러가 필요합니다.
+
+
+## 스케줄러 생성하기
+
+스케줄러는 모델을 학습 또는 추론에 사용하는지에 따라 다르게 작동합니다. 추론시에, 스케줄러는 노이즈로부터 이미지를 생성합니다. 학습시 스케줄러는 diffusion 과정에서의 특정 포인트로부터 모델의 출력 또는 샘플을 가져와 *노이즈 스케줄* 과 *업데이트 규칙*에 따라 이미지에 노이즈를 적용합니다.
+
+`DDPMScheduler`를 보면 이전으로부터 `sample_image`에 랜덤한 노이즈를 더하는 `add_noise` 메서드를 사용합니다:
+
+```py
+>>> import torch
+>>> from PIL import Image
+>>> from diffusers import DDPMScheduler
+
+>>> noise_scheduler = DDPMScheduler(num_train_timesteps=1000)
+>>> noise = torch.randn(sample_image.shape)
+>>> timesteps = torch.LongTensor([50])
+>>> noisy_image = noise_scheduler.add_noise(sample_image, noise, timesteps)
+
+>>> Image.fromarray(((noisy_image.permute(0, 2, 3, 1) + 1.0) * 127.5).type(torch.uint8).numpy()[0])
+```
+
+![](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/noisy_butterfly.png)
+
+모델의 학습 목적은 이미지에 더해진 노이즈를 예측하는 것입니다. 이 단계에서 손실은 다음과 같이 계산될 수 있습니다:
+
+```py
+>>> import torch.nn.functional as F
+
+>>> noise_pred = model(noisy_image, timesteps).sample
+>>> loss = F.mse_loss(noise_pred, noise)
+```
+
+## 모델 학습하기
+
+지금까지, 모델 학습을 시작하기 위해 많은 부분을 갖추었으며 이제 남은 것은 모든 것을 조합하는 것입니다.
+
+우선 옵티마이저(optimizer)와 학습률 스케줄러(learning rate scheduler)가 필요할 것입니다:
+
+```py
+>>> from diffusers.optimization import get_cosine_schedule_with_warmup
+
+>>> optimizer = torch.optim.AdamW(model.parameters(), lr=config.learning_rate)
+>>> lr_scheduler = get_cosine_schedule_with_warmup(
+...     optimizer=optimizer,
+...     num_warmup_steps=config.lr_warmup_steps,
+...     num_training_steps=(len(train_dataloader) * config.num_epochs),
+... )
+```
+
+그 후, 모델을 평가하는 방법이 필요합니다. 평가를 위해, `DDPMPipeline`을 사용해 배치의 이미지 샘플들을 생성하고 그리드 형태로 저장할 수 있습니다:
+
+```py
+>>> from diffusers import DDPMPipeline
+>>> import math
+>>> import os
+
+
+>>> def make_grid(images, rows, cols):
+...     w, h = images[0].size
+...     grid = Image.new("RGB", size=(cols * w, rows * h))
+...     for i, image in enumerate(images):
+...         grid.paste(image, box=(i % cols * w, i // cols * h))
+...     return grid
+
+
+>>> def evaluate(config, epoch, pipeline):
+...     # 랜덤한 노이즈로 부터 이미지를 추출합니다.(이는 역전파 diffusion 과정입니다.)
+...     # 기본 파이프라인 출력 형태는 `List[PIL.Image]` 입니다.
+...     images = pipeline(
+...         batch_size=config.eval_batch_size,
+...         generator=torch.manual_seed(config.seed),
+...     ).images
+
+...     # 이미지들을 그리드로 만들어줍니다.
+...     image_grid = make_grid(images, rows=4, cols=4)
+
+...     # 이미지들을 저장합니다.
+...     test_dir = os.path.join(config.output_dir, "samples")
+...     os.makedirs(test_dir, exist_ok=True)
+...     image_grid.save(f"{test_dir}/{epoch:04d}.png")
+```
+
+TensorBoard에 로깅, 그래디언트 누적 및 혼합 정밀도 학습을 쉽게 수행하기 위해 🤗 Accelerate를 학습 루프에 함께 앞서 말한 모든 구성 정보들을 묶어 진행할 수 있습니다. 허브에 모델을 업로드 하기 위해 레포지토리 이름 및 정보를 가져오기 위한 함수를 작성하고 허브에 업로드할 수 있습니다.
+
+💡아래의 학습 루프는 어렵고 길어 보일 수 있지만, 나중에 한 줄의 코드로 학습을 한다면 그만한 가치가 있을 것입니다! 만약 기다리지 못하고 이미지를 생성하고 싶다면, 아래 코드를 자유롭게 붙여넣고 작동시키면 됩니다. 🤗
+
+```py
+>>> from accelerate import Accelerator
+>>> from huggingface_hub import HfFolder, Repository, whoami
+>>> from tqdm.auto import tqdm
+>>> from pathlib import Path
+>>> import os
+
+
+>>> def get_full_repo_name(model_id: str, organization: str = None, token: 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 train_loop(config, model, noise_scheduler, optimizer, train_dataloader, lr_scheduler):
+...     # accelerator와 tensorboard 로깅 초기화
+...     accelerator = Accelerator(
+...         mixed_precision=config.mixed_precision,
+...         gradient_accumulation_steps=config.gradient_accumulation_steps,
+...         log_with="tensorboard",
+...         logging_dir=os.path.join(config.output_dir, "logs"),
+...     )
+...     if accelerator.is_main_process:
+...         if config.push_to_hub:
+...             repo_name = get_full_repo_name(Path(config.output_dir).name)
+...             repo = Repository(config.output_dir, clone_from=repo_name)
+...         elif config.output_dir is not None:
+...             os.makedirs(config.output_dir, exist_ok=True)
+...         accelerator.init_trackers("train_example")
+
+...     # 모든 것이 준비되었습니다.
+...     # 기억해야 할 특정한 순서는 없으며 준비한 방법에 제공한 것과 동일한 순서로 객체의 압축을 풀면 됩니다.
+...     model, optimizer, train_dataloader, lr_scheduler = accelerator.prepare(
+...         model, optimizer, train_dataloader, lr_scheduler
+...     )
+
+...     global_step = 0
+
+...     # 이제 모델을 학습합니다.
+...     for epoch in range(config.num_epochs):
+...         progress_bar = tqdm(total=len(train_dataloader), disable=not accelerator.is_local_main_process)
+...         progress_bar.set_description(f"Epoch {epoch}")
+
+...         for step, batch in enumerate(train_dataloader):
+...             clean_images = batch["images"]
+...             # 이미지에 더할 노이즈를 샘플링합니다.
+...             noise = torch.randn(clean_images.shape).to(clean_images.device)
+...             bs = clean_images.shape[0]
+
+...             # 각 이미지를 위한 랜덤한 타임스텝(timestep)을 샘플링합니다.
+...             timesteps = torch.randint(
+...                 0, noise_scheduler.config.num_train_timesteps, (bs,), device=clean_images.device
+...             ).long()
+
+...             # 각 타임스텝의 노이즈 크기에 따라 깨끗한 이미지에 노이즈를 추가합니다.
+...             # (이는 foward diffusion 과정입니다.)
+...             noisy_images = noise_scheduler.add_noise(clean_images, noise, timesteps)
+
+...             with accelerator.accumulate(model):
+...                 # 노이즈를 반복적으로 예측합니다.
+...                 noise_pred = model(noisy_images, timesteps, return_dict=False)[0]
+...                 loss = F.mse_loss(noise_pred, noise)
+...                 accelerator.backward(loss)
+
+...                 accelerator.clip_grad_norm_(model.parameters(), 1.0)
+...                 optimizer.step()
+...                 lr_scheduler.step()
+...                 optimizer.zero_grad()
+
+...             progress_bar.update(1)
+...             logs = {"loss": loss.detach().item(), "lr": lr_scheduler.get_last_lr()[0], "step": global_step}
+...             progress_bar.set_postfix(**logs)
+...             accelerator.log(logs, step=global_step)
+...             global_step += 1
+
+...         # 각 에포크가 끝난 후 evaluate()와 몇 가지 데모 이미지를 선택적으로 샘플링하고 모델을 저장합니다.
+...         if accelerator.is_main_process:
+...             pipeline = DDPMPipeline(unet=accelerator.unwrap_model(model), scheduler=noise_scheduler)
+
+...             if (epoch + 1) % config.save_image_epochs == 0 or epoch == config.num_epochs - 1:
+...                 evaluate(config, epoch, pipeline)
+
+...             if (epoch + 1) % config.save_model_epochs == 0 or epoch == config.num_epochs - 1:
+...                 if config.push_to_hub:
+...                     repo.push_to_hub(commit_message=f"Epoch {epoch}", blocking=True)
+...                 else:
+...                     pipeline.save_pretrained(config.output_dir)
+```
+
+휴, 코드가 꽤 많았네요! 하지만 🤗 Accelerate의 [`~accelerate.notebook_launcher`] 함수와 학습을 시작할 준비가 되었습니다. 함수에 학습 루프, 모든 학습 인수, 학습에 사용할 프로세스 수(사용 가능한 GPU의 수를 변경할 수 있음)를 전달합니다:
+
+```py
+>>> from accelerate import notebook_launcher
+
+>>> args = (config, model, noise_scheduler, optimizer, train_dataloader, lr_scheduler)
+
+>>> notebook_launcher(train_loop, args, num_processes=1)
+```
+
+한번 학습이 완료되면, diffusion 모델로 생성된 최종 🦋이미지🦋를 확인해보길 바랍니다!
+
+```py
+>>> import glob
+
+>>> sample_images = sorted(glob.glob(f"{config.output_dir}/samples/*.png"))
+>>> Image.open(sample_images[-1])
+```
+
+![](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/butterflies_final.png)
+
+## 다음 단계
+
+Unconditional 이미지 생성은 학습될 수 있는 작업 중 하나의 예시입니다. 다른 작업과 학습 방법은 [🧨 Diffusers 학습 예시](../training/overview) 페이지에서 확인할 수 있습니다. 다음은 학습할 수 있는 몇 가지 예시입니다:
+
+-   [Textual Inversion](../training/text_inversion), 특정 시각적 개념을 학습시켜 생성된 이미지에 통합시키는 알고리즘입니다.
+-   [DreamBooth](../training/dreambooth), 주제에 대한 몇 가지 입력 이미지들이 주어지면 주제에 대한 개인화된 이미지를 생성하기 위한 기술입니다.
+-   [Guide](../training/text2image) 데이터셋에 Stable Diffusion 모델을 파인튜닝하는 방법입니다.
+-   [Guide](../training/lora)  LoRA를 사용해 매우 큰 모델을 빠르게 파인튜닝하기 위한 메모리 효율적인 기술입니다.

docs/source/ko/tutorials/tutorial_overview.mdx

@@ -0,0 +1,23 @@
+<!--Copyright 2023 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.
+-->
+
+# Overview
+
+🧨 Diffusers에 오신 걸 환영합니다! 여러분이 diffusion 모델과 생성 AI를 처음 접하고, 더 많은 걸 배우고 싶으셨다면 제대로 찾아오셨습니다. 이 튜토리얼은 diffusion model을 여러분에게 젠틀하게 소개하고, 라이브러리의 기본 사항(핵심 구성요소와 🧨 Diffusers 사용법)을 이해하는 데 도움이 되도록 설계되었습니다.
+
+여러분은 이 튜토리얼을 통해 빠르게 생성하기 위해선 추론 파이프라인을 어떻게 사용해야 하는지, 그리고 라이브러리를 modular toolbox처럼 이용해서 여러분만의 diffusion system을 구축할 수 있도록 파이프라인을 분해하는 법을 배울 수 있습니다. 다음 단원에서는 여러분이 원하는 것을 생성하기 위해 자신만의 diffusion model을 학습하는 방법을 배우게 됩니다.
+
+튜토리얼을 완료한다면 여러분은 라이브러리를 직접 탐색하고, 자신의 프로젝트와 애플리케이션에 적용할 스킬들을 습득할 수 있을 겁니다. 
+
+[Discord](https://discord.com/invite/JfAtkvEtRb)나 [포럼](https://discuss.huggingface.co/c/discussion-related-to-httpsgithubcomhuggingfacediffusers/63) 커뮤니티에 자유롭게 참여해서 다른 사용자와 개발자들과 교류하고 협업해 보세요!
+
+자 지금부터 diffusing을 시작해 보겠습니다! 🧨

docs/source/ko/using-diffusers/custom_pipeline_examples.mdx

@@ -0,0 +1,275 @@
+<!--Copyright 2023 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.
+-->
+
+# 커뮤니티 파이프라인
+
+> **커뮤니티 파이프라인에 대한 자세한 내용은 [이 이슈](https://github.com/huggingface/diffusers/issues/841)를 참조하세요.
+
+**커뮤니티** 예제는 커뮤니티에서 추가한 추론 및 훈련 예제로 구성되어 있습니다.
+다음 표를 참조하여 모든 커뮤니티 예제에 대한 개요를 확인하시기 바랍니다. **코드 예제**를 클릭하면 복사하여 붙여넣기할 수 있는 코드 예제를 확인할 수 있습니다.
+커뮤니티가 예상대로 작동하지 않는 경우 이슈를 개설하고 작성자에게 핑을 보내주세요.
+
+| 예                                     | 설명                                                                                                                                                                                                                                                                                                                                                                                                                                                                                             | 코드 예제                                                                         | 콜랩                                                                                                                                                                                                              |저자                                                         |
+|:---------------------------------------|:------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|:---------------------------------------------------------------------------------|:-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|-----------------------------------------------------------:|
+| CLIP Guided Stable Diffusion           | CLIP 가이드 기반의 Stable Diffusion으로 텍스트에서 이미지로 생성하기                                                                                                                                                                                                                                                                                                                                                                                                                                | [CLIP Guided Stable Diffusion](#clip-guided-stable-diffusion)                    | [![콜랩에서 열기](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/huggingface/notebooks/blob/main/diffusers/CLIP_Guided_Stable_diffusion_with_diffusers.ipynb)  | [Suraj Patil](https://github.com/patil-suraj/)             |
+| One Step U-Net (Dummy)                 | 커뮤니티 파이프라인을 어떻게 사용해야 하는지에 대한 예시(참고 https://github.com/huggingface/diffusers/issues/841)                                                                                                                                                                                                                                                                                                                                                                                   | [One Step U-Net](#one-step-unet)                                                 | -                                                                                                                                                                                                                  | [Patrick von Platen](https://github.com/patrickvonplaten/) |
+| Stable Diffusion Interpolation         | 서로 다른 프롬프트/시드 간 Stable Diffusion의 latent space 보간                                                                                                                                                                                                                                                                                                                                                                                                                                    | [Stable Diffusion Interpolation](#stable-diffusion-interpolation)                | -                                                                                                                                                                                                                  | [Nate Raw](https://github.com/nateraw/)                    |
+| Stable Diffusion Mega                  | 모든 기능을 갖춘 **하나의** Stable Diffusion 파이프라인 [Text2Image](https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion.py), [Image2Image](https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_img2img.py) and [Inpainting](https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_inpaint.py) | [Stable Diffusion Mega](#stable-diffusion-mega)                                  | -                                                                                                                                                                                                                  | [Patrick von Platen](https://github.com/patrickvonplaten/) |
+| Long Prompt Weighting Stable Diffusion | 토큰 길이 제한이 없고 프롬프트에서 파싱 가중치 지원을 하는 **하나의** Stable Diffusion 파이프라인,                                                                                                                                                                                                                                                                                                                                                                                                    | [Long Prompt Weighting Stable Diffusion](#long-prompt-weighting-stable-diffusion) |-                                                                                                                                                                                                                  | [SkyTNT](https://github.com/SkyTNT)                        |
+| Speech to Image                        | 자동 음성 인식을 사용하여 텍스트를 작성하고 Stable Diffusion을 사용하여 이미지를 생성합니다.                                                                                                                                                                                                                                                                                                                                                                                                          | [Speech to Image](#speech-to-image)                                               | -                                                                                                                                                                                                                 | [Mikail Duzenli](https://github.com/MikailINTech)          |
+
+커스텀 파이프라인을 불러오려면 `diffusers/examples/community`에 있는 파일 중 하나로서 `custom_pipeline` 인수를 `DiffusionPipeline`에 전달하기만 하면 됩니다. 자신만의 파이프라인이 있는 PR을 보내주시면 빠르게 병합해드리겠습니다.
+```py
+pipe = DiffusionPipeline.from_pretrained(
+    "CompVis/stable-diffusion-v1-4", custom_pipeline="filename_in_the_community_folder"
+)
+```
+
+## 사용 예시
+
+### CLIP 가이드 기반의 Stable Diffusion
+
+모든 노이즈 제거 단계에서 추가 CLIP 모델을 통해 Stable Diffusion을 가이드함으로써 CLIP 모델 기반의 Stable Diffusion은 보다 더 사실적인 이미지를 생성을 할 수 있습니다.
+
+다음 코드는 약 12GB의 GPU RAM이 필요합니다.
+
+```python
+from diffusers import DiffusionPipeline
+from transformers import CLIPImageProcessor, CLIPModel
+import torch
+
+
+feature_extractor = CLIPImageProcessor.from_pretrained("laion/CLIP-ViT-B-32-laion2B-s34B-b79K")
+clip_model = CLIPModel.from_pretrained("laion/CLIP-ViT-B-32-laion2B-s34B-b79K", torch_dtype=torch.float16)
+
+
+guided_pipeline = DiffusionPipeline.from_pretrained(
+    "CompVis/stable-diffusion-v1-4",
+    custom_pipeline="clip_guided_stable_diffusion",
+    clip_model=clip_model,
+    feature_extractor=feature_extractor,
+    torch_dtype=torch.float16,
+)
+guided_pipeline.enable_attention_slicing()
+guided_pipeline = guided_pipeline.to("cuda")
+
+prompt = "fantasy book cover, full moon, fantasy forest landscape, golden vector elements, fantasy magic, dark light night, intricate, elegant, sharp focus, illustration, highly detailed, digital painting, concept art, matte, art by WLOP and Artgerm and Albert Bierstadt, masterpiece"
+
+generator = torch.Generator(device="cuda").manual_seed(0)
+images = []
+for i in range(4):
+    image = guided_pipeline(
+        prompt,
+        num_inference_steps=50,
+        guidance_scale=7.5,
+        clip_guidance_scale=100,
+        num_cutouts=4,
+        use_cutouts=False,
+        generator=generator,
+    ).images[0]
+    images.append(image)
+
+# 이미지 로컬에 저장하기
+for i, img in enumerate(images):
+    img.save(f"./clip_guided_sd/image_{i}.png")
+```
+
+이미지` 목록에는 로컬에 저장하거나 구글 콜랩에 직접 표시할 수 있는 PIL 이미지 목록이 포함되어 있습니다. 생성된 이미지는 기본적으로 안정적인 확산을 사용하는 것보다 품질이 높은 경향이 있습니다. 예를 들어 위의 스크립트는 다음과 같은 이미지를 생성합니다:
+
+![clip_guidance](https://huggingface.co/datasets/patrickvonplaten/images/resolve/main/clip_guidance/merged_clip_guidance.jpg).
+
+### One Step Unet
+
+예시 "one-step-unet"는 다음과 같이 실행할 수 있습니다.
+
+```python
+from diffusers import DiffusionPipeline
+
+pipe = DiffusionPipeline.from_pretrained("google/ddpm-cifar10-32", custom_pipeline="one_step_unet")
+pipe()
+```
+
+**참고**: 이 커뮤니티 파이프라인은 기능으로 유용하지 않으며 커뮤니티 파이프라인을 추가할 수 있는 방법의 예시일 뿐입니다(https://github.com/huggingface/diffusers/issues/841 참조).
+
+### Stable Diffusion Interpolation
+
+다음 코드는 최소 8GB VRAM의 GPU에서 실행할 수 있으며 약 5분 정도 소요됩니다.
+
+```python
+from diffusers import DiffusionPipeline
+import torch
+
+pipe = DiffusionPipeline.from_pretrained(
+    "CompVis/stable-diffusion-v1-4",
+    torch_dtype=torch.float16,
+    safety_checker=None,  # Very important for videos...lots of false positives while interpolating
+    custom_pipeline="interpolate_stable_diffusion",
+).to("cuda")
+pipe.enable_attention_slicing()
+
+frame_filepaths = pipe.walk(
+    prompts=["a dog", "a cat", "a horse"],
+    seeds=[42, 1337, 1234],
+    num_interpolation_steps=16,
+    output_dir="./dreams",
+    batch_size=4,
+    height=512,
+    width=512,
+    guidance_scale=8.5,
+    num_inference_steps=50,
+)
+```
+
+walk(...)` 함수의 출력은 `output_dir`에 정의된 대로 폴더에 저장된 이미지 목록을 반환합니다. 이 이미지를 사용하여 안정적으로 확산되는 동영상을 만들 수 있습니다.
+
+> 안정된 확산을 이용한 동영상 제작 방법과 더 많은 기능에 대한 자세한 내용은 https://github.com/nateraw/stable-diffusion-videos 에서 확인하시기 바랍니다.
+
+### Stable Diffusion Mega
+
+The Stable Diffusion Mega 파이프라인을 사용하면 Stable Diffusion 파이프라인의 주요 사용 사례를 단일 클래스에서 사용할 수 있습니다.
+```python
+#!/usr/bin/env python3
+from diffusers import DiffusionPipeline
+import PIL
+import requests
+from io import BytesIO
+import torch
+
+
+def download_image(url):
+    response = requests.get(url)
+    return PIL.Image.open(BytesIO(response.content)).convert("RGB")
+
+
+pipe = DiffusionPipeline.from_pretrained(
+    "CompVis/stable-diffusion-v1-4",
+    custom_pipeline="stable_diffusion_mega",
+    torch_dtype=torch.float16,
+)
+pipe.to("cuda")
+pipe.enable_attention_slicing()
+
+
+### Text-to-Image
+
+images = pipe.text2img("An astronaut riding a horse").images
+
+### Image-to-Image
+
+init_image = download_image(
+    "https://raw.githubusercontent.com/CompVis/stable-diffusion/main/assets/stable-samples/img2img/sketch-mountains-input.jpg"
+)
+
+prompt = "A fantasy landscape, trending on artstation"
+
+images = pipe.img2img(prompt=prompt, image=init_image, strength=0.75, guidance_scale=7.5).images
+
+### Inpainting
+
+img_url = "https://raw.githubusercontent.com/CompVis/latent-diffusion/main/data/inpainting_examples/overture-creations-5sI6fQgYIuo.png"
+mask_url = "https://raw.githubusercontent.com/CompVis/latent-diffusion/main/data/inpainting_examples/overture-creations-5sI6fQgYIuo_mask.png"
+init_image = download_image(img_url).resize((512, 512))
+mask_image = download_image(mask_url).resize((512, 512))
+
+prompt = "a cat sitting on a bench"
+images = pipe.inpaint(prompt=prompt, image=init_image, mask_image=mask_image, strength=0.75).images
+```
+
+위에 표시된 것처럼 하나의 파이프라인에서 '텍스트-이미지 변환', '이미지-이미지 변환', '인페인팅'을 모두 실행할 수 있습니다.
+
+### Long Prompt Weighting Stable Diffusion
+
+파이프라인을 사용하면 77개의 토큰 길이 제한 없이 프롬프트를 입력할 수 있습니다. 또한 "()"를 사용하여 단어 가중치를 높이거나 "[]"를 사용하여 단어 가중치를 낮출 수 있습니다.
+또한 파이프라인을 사용하면 단일 클래스에서 Stable Diffusion 파이프라인의 주요 사용 사례를 사용할 수 있습니다.
+
+#### pytorch
+
+```python
+from diffusers import DiffusionPipeline
+import torch
+
+pipe = DiffusionPipeline.from_pretrained(
+    "hakurei/waifu-diffusion", custom_pipeline="lpw_stable_diffusion", torch_dtype=torch.float16
+)
+pipe = pipe.to("cuda")
+
+prompt = "best_quality (1girl:1.3) bow bride brown_hair closed_mouth frilled_bow frilled_hair_tubes frills (full_body:1.3) fox_ear hair_bow hair_tubes happy hood japanese_clothes kimono long_sleeves red_bow smile solo tabi uchikake white_kimono wide_sleeves cherry_blossoms"
+neg_prompt = "lowres, bad_anatomy, error_body, error_hair, error_arm, error_hands, bad_hands, error_fingers, bad_fingers, missing_fingers, error_legs, bad_legs, multiple_legs, missing_legs, error_lighting, error_shadow, error_reflection, text, error, extra_digit, fewer_digits, cropped, worst_quality, low_quality, normal_quality, jpeg_artifacts, signature, watermark, username, blurry"
+
+pipe.text2img(prompt, negative_prompt=neg_prompt, width=512, height=512, max_embeddings_multiples=3).images[0]
+```
+
+#### onnxruntime
+
+```python
+from diffusers import DiffusionPipeline
+import torch
+
+pipe = DiffusionPipeline.from_pretrained(
+    "CompVis/stable-diffusion-v1-4",
+    custom_pipeline="lpw_stable_diffusion_onnx",
+    revision="onnx",
+    provider="CUDAExecutionProvider",
+)
+
+prompt = "a photo of an astronaut riding a horse on mars, best quality"
+neg_prompt = "lowres, bad anatomy, error body, error hair, error arm, error hands, bad hands, error fingers, bad fingers, missing fingers, error legs, bad legs, multiple legs, missing legs, error lighting, error shadow, error reflection, text, error, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry"
+
+pipe.text2img(prompt, negative_prompt=neg_prompt, width=512, height=512, max_embeddings_multiples=3).images[0]
+```
+
+토큰 인덱스 시퀀스 길이가 이 모델에 지정된 최대 시퀀스 길이보다 길면(*** > 77). 이 시퀀스를 모델에서 실행하면 인덱싱 오류가 발생합니다`. 정상적인 현상이니 걱정하지 마세요.
+### Speech to Image
+
+다음 코드는 사전학습된 OpenAI whisper-small과 Stable Diffusion을 사용하여 오디오 샘플에서 이미지를 생성할 수 있습니다.
+```Python
+import torch
+
+import matplotlib.pyplot as plt
+from datasets import load_dataset
+from diffusers import DiffusionPipeline
+from transformers import (
+    WhisperForConditionalGeneration,
+    WhisperProcessor,
+)
+
+
+device = "cuda" if torch.cuda.is_available() else "cpu"
+
+ds = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation")
+
+audio_sample = ds[3]
+
+text = audio_sample["text"].lower()
+speech_data = audio_sample["audio"]["array"]
+
+model = WhisperForConditionalGeneration.from_pretrained("openai/whisper-small").to(device)
+processor = WhisperProcessor.from_pretrained("openai/whisper-small")
+
+diffuser_pipeline = DiffusionPipeline.from_pretrained(
+    "CompVis/stable-diffusion-v1-4",
+    custom_pipeline="speech_to_image_diffusion",
+    speech_model=model,
+    speech_processor=processor,
+    
+    torch_dtype=torch.float16,
+)
+
+diffuser_pipeline.enable_attention_slicing()
+diffuser_pipeline = diffuser_pipeline.to(device)
+
+output = diffuser_pipeline(speech_data)
+plt.imshow(output.images[0])
+```
+위 예시는 다음의 결과 이미지를 보입니다.
+
+![image](https://user-images.githubusercontent.com/45072645/196901736-77d9c6fc-63ee-4072-90b0-dc8b903d63e3.png)

docs/source/ko/using-diffusers/custom_pipeline_overview.mdx

@@ -0,0 +1,56 @@
+<!--Copyright 2023 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.
+-->
+
+# 커스텀 파이프라인 불러오기
+
+[[open-in-colab]]
+
+커뮤니티 파이프라인은 논문에 명시된 원래의 구현체와 다른 형태로 구현된 모든 [`DiffusionPipeline`] 클래스를 의미합니다. (예를 들어, [`StableDiffusionControlNetPipeline`]는 ["Text-to-Image Generation with ControlNet Conditioning"](https://arxiv.org/abs/2302.05543) 해당) 이들은 추가 기능을 제공하거나 파이프라인의 원래 구현을 확장합니다.
+
+[Speech to Image](https://github.com/huggingface/diffusers/tree/main/examples/community#speech-to-image) 또는 [Composable Stable Diffusion](https://github.com/huggingface/diffusers/tree/main/examples/community#composable-stable-diffusion) 과 같은 멋진 커뮤니티 파이프라인이 많이 있으며 [여기에서](https://github.com/huggingface/diffusers/tree/main/examples/community) 모든 공식 커뮤니티 파이프라인을 찾을 수 있습니다.
+
+허브에서 커뮤니티 파이프라인을 로드하려면, 커뮤니티 파이프라인의 리포지토리 ID와 (파이프라인 가중치 및 구성 요소를 로드하려는) 모델의 리포지토리 ID를 인자로 전달해야 합니다. 예를 들어, 아래 예시에서는 `hf-internal-testing/diffusers-dummy-pipeline`에서 더미 파이프라인을 불러오고, `google/ddpm-cifar10-32`에서 파이프라인의 가중치와 컴포넌트들을 로드합니다.
+
+<Tip warning={true}>
+
+🔒 허깅 페이스 허브에서 커뮤니티 파이프라인을 불러오는 것은 곧 해당 코드가 안전하다고 신뢰하는 것입니다. 코드를 자동으로 불러오고 실행하기 앞서 반드시 온라인으로 해당 코드의 신뢰성을 검사하세요!
+
+</Tip>
+
+```py
+from diffusers import DiffusionPipeline
+
+pipeline = DiffusionPipeline.from_pretrained(
+    "google/ddpm-cifar10-32", custom_pipeline="hf-internal-testing/diffusers-dummy-pipeline"
+)
+```
+
+공식 커뮤니티 파이프라인을 불러오는 것은 비슷하지만, 공식 리포지토리 ID에서 가중치를 불러오는 것과 더불어 해당 파이프라인 내의 컴포넌트를 직접 지정하는 것 역시 가능합니다. 아래 예제를 보면 커뮤니티 [CLIP Guided Stable Diffusion](https://github.com/huggingface/diffusers/tree/main/examples/community#clip-guided-stable-diffusion) 파이프라인을 로드할 때, 해당 파이프라인에서 사용할 `clip_model` 컴포넌트와 `feature_extractor` 컴포넌트를 직접 설정하는 것을 확인할 수 있습니다.
+
+```py
+from diffusers import DiffusionPipeline
+from transformers import CLIPImageProcessor, CLIPModel
+
+clip_model_id = "laion/CLIP-ViT-B-32-laion2B-s34B-b79K"
+
+feature_extractor = CLIPImageProcessor.from_pretrained(clip_model_id)
+clip_model = CLIPModel.from_pretrained(clip_model_id)
+
+pipeline = DiffusionPipeline.from_pretrained(
+    "runwayml/stable-diffusion-v1-5",
+    custom_pipeline="clip_guided_stable_diffusion",
+    clip_model=clip_model,
+    feature_extractor=feature_extractor,
+)
+```
+
+커뮤니티 파이프라인에 대한 자세한 내용은 [커뮤니티 파이프라인](https://github.com/huggingface/diffusers/blob/main/docs/source/en/using-diffusers/custom_pipeline_examples) 가이드를 살펴보세요. 커뮤니티 파이프라인 등록에 관심이 있는 경우 [커뮤니티 파이프라인에 기여하는 방법](https://github.com/huggingface/diffusers/blob/main/docs/source/en/using-diffusers/contribute_pipeline)에 대한 가이드를 확인하세요 !

docs/source/ko/using-diffusers/depth2img.mdx

@@ -0,0 +1,57 @@
+<!--Copyright 2023 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.
+-->
+
+# Text-guided depth-to-image 생성
+
+[[open-in-colab]]
+
+[`StableDiffusionDepth2ImgPipeline`]을 사용하면 텍스트 프롬프트와 초기 이미지를 전달하여 새 이미지의 생성을 조절할 수 있습니다. 또한 이미지 구조를 보존하기 위해 `depth_map`을 전달할 수도 있습니다. `depth_map`이 제공되지 않으면 파이프라인은 통합된 [depth-estimation model](https://github.com/isl-org/MiDaS)을 통해 자동으로 깊이를 예측합니다.
+
+
+먼저 [`StableDiffusionDepth2ImgPipeline`]의 인스턴스를 생성합니다:
+
+```python
+import torch
+import requests
+from PIL import Image
+
+from diffusers import StableDiffusionDepth2ImgPipeline
+
+pipe = StableDiffusionDepth2ImgPipeline.from_pretrained(
+    "stabilityai/stable-diffusion-2-depth",
+    torch_dtype=torch.float16,
+).to("cuda")
+```
+
+이제 프롬프트를 파이프라인에 전달합니다. 특정 단어가 이미지 생성을 가이드 하는것을 방지하기 위해 `negative_prompt`를 전달할 수도 있습니다:
+
+```python
+url = "http://images.cocodataset.org/val2017/000000039769.jpg"
+init_image = Image.open(requests.get(url, stream=True).raw)
+prompt = "two tigers"
+n_prompt = "bad, deformed, ugly, bad anatomy"
+image = pipe(prompt=prompt, image=init_image, negative_prompt=n_prompt, strength=0.7).images[0]
+image
+```
+
+| Input                                                                           | Output                                                                                                                                |
+|---------------------------------------------------------------------------------|---------------------------------------------------------------------------------------------------------------------------------------|
+| <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/coco-cats.png" width="500"/> | <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/depth2img-tigers.png" width="500"/> |
+
+아래의 Spaces를 가지고 놀며 depth map이 있는 이미지와 없는 이미지의 차이가 있는지 확인해 보세요!
+
+<iframe
+	src="https://radames-stable-diffusion-depth2img.hf.space"
+	frameborder="0"
+	width="850"
+	height="500"
+></iframe>

docs/source/ko/using-diffusers/img2img.mdx

@@ -0,0 +1,100 @@
+<!--Copyright 2023 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.
+-->
+
+# 텍스트 기반 image-to-image 생성
+
+[[Colab에서 열기]]
+
+[`StableDiffusionImg2ImgPipeline`]을 사용하면 텍스트 프롬프트와 시작 이미지를 전달하여 새 이미지 생성의 조건을 지정할 수 있습니다.
+
+시작하기 전에 필요한 라이브러리가 모두 설치되어 있는지 확인하세요:
+
+```bash
+!pip install diffusers transformers ftfy accelerate
+```
+
+[`nitrosocke/Ghibli-Diffusion`](https://huggingface.co/nitrosocke/Ghibli-Diffusion)과 같은 사전학습된 stable diffusion 모델로 [`StableDiffusionImg2ImgPipeline`]을 생성하여 시작하세요.
+
+
+```python
+import torch
+import requests
+from PIL import Image
+from io import BytesIO
+from diffusers import StableDiffusionImg2ImgPipeline
+
+device = "cuda"
+pipe = StableDiffusionImg2ImgPipeline.from_pretrained("nitrosocke/Ghibli-Diffusion", torch_dtype=torch.float16).to(
+    device
+)
+```
+
+초기 이미지를 다운로드하고 사전 처리하여 파이프라인에 전달할 수 있습니다:
+
+```python
+url = "https://raw.githubusercontent.com/CompVis/stable-diffusion/main/assets/stable-samples/img2img/sketch-mountains-input.jpg"
+
+response = requests.get(url)
+init_image = Image.open(BytesIO(response.content)).convert("RGB")
+init_image.thumbnail((768, 768))
+init_image
+```
+
+<div class="flex justify-center">
+    <img src="https://huggingface.co/datasets/YiYiXu/test-doc-assets/resolve/main/image_2_image_using_diffusers_cell_8_output_0.jpeg"/>
+</div>
+
+<Tip>
+
+💡 `strength`는 입력 이미지에 추가되는 노이즈의 양을 제어하는 0.0에서 1.0 사이의 값입니다. 1.0에 가까운 값은 다양한 변형을 허용하지만 입력 이미지와 의미적으로 일치하지 않는 이미지를 생성합니다.
+
+</Tip>
+
+프롬프트를 정의하고(지브리 스타일(Ghibli-style)에 맞게 조정된 이 체크포인트의 경우 프롬프트 앞에 `ghibli style` 토큰을 붙여야 합니다) 파이프라인을 실행합니다:
+
+```python
+prompt = "ghibli style, a fantasy landscape with castles"
+generator = torch.Generator(device=device).manual_seed(1024)
+image = pipe(prompt=prompt, image=init_image, strength=0.75, guidance_scale=7.5, generator=generator).images[0]
+image
+```
+
+<div class="flex justify-center">
+    <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/ghibli-castles.png"/>
+</div>
+
+다른 스케줄러로 실험하여 출력에 어떤 영향을 미치는지 확인할 수도 있습니다:
+
+```python
+from diffusers import LMSDiscreteScheduler
+
+lms = LMSDiscreteScheduler.from_config(pipe.scheduler.config)
+pipe.scheduler = lms
+generator = torch.Generator(device=device).manual_seed(1024)
+image = pipe(prompt=prompt, image=init_image, strength=0.75, guidance_scale=7.5, generator=generator).images[0]
+image
+```
+
+<div class="flex justify-center">
+    <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/lms-ghibli.png"/>
+</div>
+
+아래 공백을 확인하고 `strength` 값을 다르게 설정하여 이미지를 생성해 보세요. `strength`를 낮게 설정하면 원본 이미지와 더 유사한 이미지가 생성되는 것을 확인할 수 있습니다.
+
+자유롭게 스케줄러를 [`LMSDiscreteScheduler`]로 전환하여 출력에 어떤 영향을 미치는지 확인해 보세요.
+
+<iframe
+	src="https://stevhliu-ghibli-img2img.hf.space"
+	frameborder="0"
+	width="850"
+	height="500"
+></iframe>

docs/source/ko/using-diffusers/inpaint.mdx

@@ -0,0 +1,75 @@
+<!--Copyright 2023 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.
+-->
+
+# Text-guided 이미지 인페인팅(inpainting)
+
+[[코랩에서 열기]]
+
+[`StableDiffusionInpaintPipeline`]은 마스크와 텍스트 프롬프트를 제공하여 이미지의 특정 부분을 편집할 수 있도록 합니다. 이 기능은 인페인팅 작업을 위해 특별히 훈련된 [`runwayml/stable-diffusion-inpainting`](https://huggingface.co/runwayml/stable-diffusion-inpainting)과 같은 Stable Diffusion 버전을 사용합니다.
+
+먼저 [`StableDiffusionInpaintPipeline`] 인스턴스를 불러옵니다:
+
+```python
+import PIL
+import requests
+import torch
+from io import BytesIO
+
+from diffusers import StableDiffusionInpaintPipeline
+
+pipeline = StableDiffusionInpaintPipeline.from_pretrained(
+    "runwayml/stable-diffusion-inpainting",
+    torch_dtype=torch.float16,
+)
+pipeline = pipeline.to("cuda")
+```
+
+나중에 교체할 강아지 이미지와 마스크를 다운로드하세요:
+
+```python
+def download_image(url):
+    response = requests.get(url)
+    return PIL.Image.open(BytesIO(response.content)).convert("RGB")
+
+
+img_url = "https://raw.githubusercontent.com/CompVis/latent-diffusion/main/data/inpainting_examples/overture-creations-5sI6fQgYIuo.png"
+mask_url = "https://raw.githubusercontent.com/CompVis/latent-diffusion/main/data/inpainting_examples/overture-creations-5sI6fQgYIuo_mask.png"
+
+init_image = download_image(img_url).resize((512, 512))
+mask_image = download_image(mask_url).resize((512, 512))
+```
+
+이제 마스크를 다른 것으로 교체하라는 프롬프트를 만들 수 있습니다:
+
+```python
+prompt = "Face of a yellow cat, high resolution, sitting on a park bench"
+image = pipe(prompt=prompt, image=init_image, mask_image=mask_image).images[0]
+```
+
+`image`          | `mask_image` | `prompt` | output |
+:-------------------------:|:-------------------------:|:-------------------------:|-------------------------:|
+<img src="https://raw.githubusercontent.com/CompVis/latent-diffusion/main/data/inpainting_examples/overture-creations-5sI6fQgYIuo.png" alt="drawing" width="250"/> | <img src="https://raw.githubusercontent.com/CompVis/latent-diffusion/main/data/inpainting_examples/overture-creations-5sI6fQgYIuo_mask.png" alt="drawing" width="250"/> | ***Face of a yellow cat, high resolution, sitting on a park bench*** | <img src="https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/in_paint/yellow_cat_sitting_on_a_park_bench.png" alt="drawing" width="250"/> |
+
+<Tip warning={true}>
+
+이전의 실험적인 인페인팅 구현에서는 품질이 낮은 다른 프로세스를 사용했습니다. 이전 버전과의 호환성을 보장하기 위해 새 모델이 포함되지 않은 사전학습된 파이프라인을 불러오면 이전 인페인팅 방법이 계속 적용됩니다.
+
+</Tip>
+
+아래 Space에서 이미지 인페인팅을 직접 해보세요!
+
+<iframe
+  src="https://runwayml-stable-diffusion-inpainting.hf.space"
+  frameborder="0"
+  width="850"
+  height="500"
+></iframe>

docs/source/ko/using-diffusers/loading.mdx

@@ -0,0 +1,442 @@
+<!--Copyright 2023 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.
+-->
+
+
+
+# 파이프라인, 모델, 스케줄러 불러오기
+
+기본적으로 diffusion 모델은 다양한 컴포넌트들(모델, 토크나이저, 스케줄러) 간의 복잡한 상호작용을 기반으로 동작합니다. 디퓨저스(Diffusers)는 이러한 diffusion 모델을 보다 쉽고 간편한 API로 제공하는 것을 목표로 설계되었습니다. [`DiffusionPipeline`]은 diffusion 모델이 갖는 복잡성을 하나의 파이프라인 API로 통합하고, 동시에 이를 구성하는 각각의 컴포넌트들을 태스크에 맞춰 유연하게 커스터마이징할 수 있도록 지원하고 있습니다.
+
+diffusion 모델의 훈련과 추론에 필요한 모든 것은 [`DiffusionPipeline.from_pretrained`] 메서드를 통해 접근할 수 있습니다. (이 말의 의미는 다음 단락에서 보다 자세하게 다뤄보도록 하겠습니다.)
+
+이 문서에서는 설명할 내용은 다음과 같습니다.
+
+* 허브를 통해 혹은 로컬로 파이프라인을 불러오는 법
+
+* 파이프라인에 다른 컴포넌트들을 적용하는 법
+* 오리지널 체크포인트가 아닌 variant를 불러오는 법  (variant란 기본으로 설정된 `fp32`가 아닌 다른  부동 소수점 타입(예: `fp16`)을 사용하거나 Non-EMA 가중치를 사용하는 체크포인트들을 의미합니다.)
+* 모델과 스케줄러를 불러오는 법
+
+
+
+## Diffusion 파이프라인
+
+<Tip>
+
+💡 [`DiffusionPipeline`] 클래스가 동작하는 방식에 보다 자세한 내용이 궁금하다면,  [DiffusionPipeline explained](#diffusionpipeline에-대해-알아보기) 섹션을 확인해보세요.
+
+</Tip>
+
+[`DiffusionPipeline`] 클래스는 diffusion 모델을 [허브](https://huggingface.co/models?library=diffusers)로부터 불러오는 가장 심플하면서 보편적인 방식입니다. [`DiffusionPipeline.from_pretrained`] 메서드는 적합한 파이프라인 클래스를 자동으로 탐지하고, 필요한 구성요소(configuration)와 가중치(weight) 파일들을 다운로드하고 캐싱한 다음, 해당 파이프라인 인스턴스를 반환합니다.
+
+```python
+from diffusers import DiffusionPipeline
+
+repo_id = "runwayml/stable-diffusion-v1-5"
+pipe = DiffusionPipeline.from_pretrained(repo_id)
+```
+
+물론 [`DiffusionPipeline`] 클래스를 사용하지 않고, 명시적으로 직접 해당 파이프라인 클래스를 불러오는 것도 가능합니다. 아래 예시 코드는 위 예시와 동일한 인스턴스를 반환합니다. 
+
+```python
+from diffusers import StableDiffusionPipeline
+
+repo_id = "runwayml/stable-diffusion-v1-5"
+pipe = StableDiffusionPipeline.from_pretrained(repo_id)
+```
+
+[CompVis/stable-diffusion-v1-4](https://huggingface.co/CompVis/stable-diffusion-v1-4)이나 [runwayml/stable-diffusion-v1-5](https://huggingface.co/runwayml/stable-diffusion-v1-5) 같은 체크포인트들의 경우, 하나 이상의 다양한 태스크에 활용될 수 있습니다. (예를 들어 위의 두 체크포인트의 경우, text-to-image와 image-to-image에 모두 활용될 수 있습니다.)  만약 이러한 체크포인트들을 기본 설정 태스크가 아닌 다른 태스크에 활용하고자 한다면, 해당 태스크에 대응되는 파이프라인(task-specific pipeline)을 사용해야 합니다.
+
+```python
+from diffusers import StableDiffusionImg2ImgPipeline
+
+repo_id = "runwayml/stable-diffusion-v1-5"
+pipe = StableDiffusionImg2ImgPipeline.from_pretrained(repo_id)
+```
+
+
+
+### 로컬 파이프라인
+
+파이프라인을 로컬로 불러오고자 한다면, `git-lfs`를 사용하여 직접 체크포인트를 로컬 디스크에 다운로드 받아야 합니다. 아래의 명령어를 실행하면 `./stable-diffusion-v1-5`란 이름으로 폴더가 로컬디스크에 생성됩니다.
+
+```bash
+git lfs install
+git clone https://huggingface.co/runwayml/stable-diffusion-v1-5
+```
+
+그런 다음 해당 로컬 경로를 [`~DiffusionPipeline.from_pretrained`] 메서드에 전달합니다.
+
+```python
+from diffusers import DiffusionPipeline
+
+repo_id = "./stable-diffusion-v1-5"
+stable_diffusion = DiffusionPipeline.from_pretrained(repo_id)
+```
+
+위의 예시코드처럼 만약 `repo_id`가 로컬 패스(local path)라면, [`~DiffusionPipeline.from_pretrained`] 메서드는 이를 자동으로 감지하여 허브에서 파일을 다운로드하지 않습니다. 만약 로컬 디스크에 저장된 파이프라인 체크포인트가 최신 버전이 아닐 경우에도, 최신 버전을 다운로드하지 않고 기존 로컬 디스크에 저장된 체크포인트를 사용한다는 것을 의미합니다.
+
+
+
+### 파이프라인 내부의 컴포넌트 교체하기
+
+파이프라인 내부의 컴포넌트들은 호환 가능한 다른 컴포넌트로 교체될 수 있습니다. 이와 같은 컴포넌트 교체가 중요한 이유는 다음과 같습니다. 
+
+- 어떤 스케줄러를 사용할 것인가는 생성속도와 생성품질 간의 트레이드오프를 정의하는 중요한 요소입니다.
+- diffusion 모델 내부의 컴포넌트들은 일반적으로 각각 독립적으로 훈련되기 때문에, 더 좋은 성능을 보여주는 컴포넌트가 있다면 그걸로 교체하는 식으로 성능을 향상시킬 수 있습니다.
+- 파인 튜닝 단계에서는 일반적으로 UNet 혹은 텍스트 인코더와 같은 일부 컴포넌트들만 훈련하게 됩니다.
+
+어떤 스케줄러들이 호환가능한지는 `compatibles` 속성을 통해 확인할 수 있습니다.
+
+```python
+from diffusers import DiffusionPipeline
+
+repo_id = "runwayml/stable-diffusion-v1-5"
+stable_diffusion = DiffusionPipeline.from_pretrained(repo_id)
+stable_diffusion.scheduler.compatibles
+```
+
+이번에는 [`SchedulerMixin.from_pretrained`] 메서드를 사용해서, 기존 기본 스케줄러였던 [`PNDMScheduler`]를 보다 우수한 성능의 [`EulerDiscreteScheduler`]로 바꿔봅시다. 스케줄러를 로드할 때는 `subfolder` 인자를 통해, 해당 파이프라인의 레포지토리에서 [스케줄러에 관한 하위폴더](https://huggingface.co/runwayml/stable-diffusion-v1-5/tree/main/scheduler)를  명시해주어야 합니다. 
+
+그 다음 새롭게 생성한 [`EulerDiscreteScheduler`] 인스턴스를 [`DiffusionPipeline`]의 `scheduler` 인자에 전달합니다.
+
+```python
+from diffusers import DiffusionPipeline, EulerDiscreteScheduler, DPMSolverMultistepScheduler
+
+repo_id = "runwayml/stable-diffusion-v1-5"
+
+scheduler = EulerDiscreteScheduler.from_pretrained(repo_id, subfolder="scheduler")
+
+stable_diffusion = DiffusionPipeline.from_pretrained(repo_id, scheduler=scheduler)
+```
+
+### 세이프티 체커
+
+스테이블 diffusion과 같은 diffusion 모델들은 유해한 이미지를 생성할 수도 있습니다. 이를 예방하기 위해 디퓨저스는 생성된 이미지의 유해성을 판단하는 [세이프티 체커(safety checker)](https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/stable_diffusion/safety_checker.py) 기능을 지원하고 있습니다. 만약 세이프티 체커의 사용을 원하지 않는다면, `safety_checker` 인자에 `None`을 전달해주시면 됩니다.
+
+```python
+from diffusers import DiffusionPipeline
+
+repo_id = "runwayml/stable-diffusion-v1-5"
+stable_diffusion = DiffusionPipeline.from_pretrained(repo_id, safety_checker=None)
+```
+
+### 컴포넌트 재사용
+
+복수의 파이프라인에 동일한 모델이 반복적으로 사용한다면, 굳이 해당 모델의 동일한 가중치를 중복으로 RAM에 불러올 필요는 없을 것입니다.  [`~DiffusionPipeline.components`] 속성을 통해 파이프라인 내부의 컴포넌트들을 참조할 수 있는데, 이번 단락에서는 이를 통해 동일한 모델 가중치를 RAM에 중복으로 불러오는 것을 방지하는 법에 대해 알아보겠습니다.
+
+```python
+from diffusers import StableDiffusionPipeline, StableDiffusionImg2ImgPipeline
+
+model_id = "runwayml/stable-diffusion-v1-5"
+stable_diffusion_txt2img = StableDiffusionPipeline.from_pretrained(model_id)
+
+components = stable_diffusion_txt2img.components
+```
+
+그 다음 위 예시 코드에서 선언한 `components` 변수를 다른 파이프라인에 전달함으로써, 모델의 가중치를 중복으로 RAM에 로딩하지 않고, 동일한 컴포넌트를 재사용할 수 있습니다.
+
+```python
+stable_diffusion_img2img = StableDiffusionImg2ImgPipeline(**components)
+```
+
+물론 각각의 컴포넌트들을 따로 따로 파이프라인에 전달할 수도 있습니다.  예를 들어 `stable_diffusion_txt2img` 파이프라인 안의 컴포넌트들 가운데서 세이프티 체커(`safety_checker`)와 피쳐 익스트랙터(`feature_extractor`)를 제외한 컴포넌트들만 `stable_diffusion_img2img` 파이프라인에서 재사용하는 방식 역시 가능합니다. 
+
+```python
+from diffusers import StableDiffusionPipeline, StableDiffusionImg2ImgPipeline
+
+model_id = "runwayml/stable-diffusion-v1-5"
+stable_diffusion_txt2img = StableDiffusionPipeline.from_pretrained(model_id)
+stable_diffusion_img2img = StableDiffusionImg2ImgPipeline(
+    vae=stable_diffusion_txt2img.vae,
+    text_encoder=stable_diffusion_txt2img.text_encoder,
+    tokenizer=stable_diffusion_txt2img.tokenizer,
+    unet=stable_diffusion_txt2img.unet,
+    scheduler=stable_diffusion_txt2img.scheduler,
+    safety_checker=None,
+    feature_extractor=None,
+    requires_safety_checker=False,
+)
+```
+
+## Checkpoint variants
+
+Variant란 일반적으로 다음과 같은 체크포인트들을 의미합니다.
+
+-  `torch.float16`과 같이 정밀도는 더 낮지만, 용량 역시 더 작은 부동소수점 타입의 가중치를 사용하는 체크포인트. *(다만 이와 같은 variant의 경우, 추가적인 훈련과 CPU환경에서의 구동이 불가능합니다.)* 
+- Non-EMA 가중치를 사용하는 체크포인트. *(Non-EMA 가중치의 경우, 파인 튜닝 단계에서 사용하는 것이 권장되는데, 추론 단계에선 사용하지 않는 것이 권장됩니다.)*
+
+<Tip>
+
+💡 모델 구조는 동일하지만 서로 다른 학습 환경에서 서로 다른 데이터셋으로 학습된 체크포인트들이 있을 경우, 해당 체크포인트들은 variant 단계가 아닌 레포지토리 단계에서 분리되어 관리되어야 합니다. (즉, 해당 체크포인트들은 서로 다른 레포지토리에서 따로 관리되어야 합니다. 예시: [`stable-diffusion-v1-4`], [`stable-diffusion-v1-5`]). 
+
+</Tip>
+
+| **checkpoint type** | **weight name**                     | **argument for loading weights** |
+| ------------------- | ----------------------------------- | -------------------------------- |
+| original            | diffusion_pytorch_model.bin         |                                  |
+| floating point      | diffusion_pytorch_model.fp16.bin    | `variant`, `torch_dtype`         |
+| non-EMA             | diffusion_pytorch_model.non_ema.bin | `variant`                        |
+
+variant를 로드할 때 2개의 중요한 argument가 있습니다.
+
+* `torch_dtype`은 불러올 체크포인트의 부동소수점을 정의합니다. 예를 들어 `torch_dtype=torch.float16`을 명시함으로써 가중치의 부동소수점 타입을 `fl16`으로 변환할 수 있습니다. (만약 따로 설정하지 않을 경우, 기본값으로 `fp32` 타입의 가중치가 로딩됩니다.) 또한 `variant` 인자를 명시하지 않은 채로 체크포인트를 불러온 다음, 해당 체크포인트를 `torch_dtype=torch.float16` 인자를 통해 `fp16` 타입으로 변환하는 것 역시 가능합니다. 이 경우 기본으로 설정된 `fp32` 가중치가 먼저 다운로드되고, 해당 가중치들을 불러온 다음 `fp16` 타입으로 변환하게 됩니다.
+* `variant` 인자는 레포지토리에서 어떤 variant를 불러올 것인가를 정의합니다. 가령  [`diffusers/stable-diffusion-variants`](https://huggingface.co/diffusers/stable-diffusion-variants/tree/main/unet) 레포지토리로부터 `non_ema` 체크포인트를 불러오고자 한다면, `variant="non_ema"` 인자를 전달해야 합니다.
+
+```python
+from diffusers import DiffusionPipeline
+
+# load fp16 variant
+stable_diffusion = DiffusionPipeline.from_pretrained(
+    "runwayml/stable-diffusion-v1-5", variant="fp16", torch_dtype=torch.float16
+)
+# load non_ema variant
+stable_diffusion = DiffusionPipeline.from_pretrained("runwayml/stable-diffusion-v1-5", variant="non_ema")
+```
+
+다른 부동소수점 타입의 가중치 혹은 non-EMA 가중치를 사용하는 체크포인트를 저장하기 위해서는, [`DiffusionPipeline.save_pretrained`] 메서드를 사용해야 하며, 이 때 `variant` 인자를 명시해줘야 합니다. 원래의 체크포인트와 동일한 폴더에 variant를 저장해야 하며, 이렇게 하면 동일한 폴더에서 오리지널 체크포인트과 variant를 모두 불러올 수 있습니다.
+
+```python
+from diffusers import DiffusionPipeline
+
+# save as fp16 variant
+stable_diffusion.save_pretrained("runwayml/stable-diffusion-v1-5", variant="fp16")
+# save as non-ema variant
+stable_diffusion.save_pretrained("runwayml/stable-diffusion-v1-5", variant="non_ema")
+```
+
+만약 variant를 기존 폴더에 저장하지 않을 경우, `variant` 인자를 반드시 명시해야 합니다. 그렇게 하지 않을 경우 원래의 오리지널 체크포인트를 찾을 수 없게 되기 때문에 에러가 발생합니다.
+
+```python
+# 👎 this won't work
+stable_diffusion = DiffusionPipeline.from_pretrained("./stable-diffusion-v1-5", torch_dtype=torch.float16)
+# 👍 this works
+stable_diffusion = DiffusionPipeline.from_pretrained(
+    "./stable-diffusion-v1-5", variant="fp16", torch_dtype=torch.float16
+)
+```
+
+### 모델 불러오기
+
+모델들은 [`ModelMixin.from_pretrained`] 메서드를 통해 불러올 수 있습니다. 해당 메서드는 최신 버전의 모델 가중치 파일과 설정 파일(configurations)을 다운로드하고 캐싱합니다. 만약 이러한 파일들이 최신 버전으로 로컬 캐시에 저장되어 있다면, [`ModelMixin.from_pretrained`]는 굳이 해당 파일들을 다시 다운로드하지 않으며, 그저 캐시에 있는 최신 파일들을 재사용합니다.
+
+모델은 `subfolder` 인자에 명시된 하위 폴더로부터 로드됩니다. 예를 들어 `runwayml/stable-diffusion-v1-5`의 UNet 모델의 가중치는 [`unet`](https://huggingface.co/runwayml/stable-diffusion-v1-5/tree/main/unet) 폴더에 저장되어 있습니다.
+
+```python
+from diffusers import UNet2DConditionModel
+
+repo_id = "runwayml/stable-diffusion-v1-5"
+model = UNet2DConditionModel.from_pretrained(repo_id, subfolder="unet")
+```
+
+혹은 [해당 모델의 레포지토리](https://huggingface.co/google/ddpm-cifar10-32/tree/main)로부터 다이렉트로 가져오는 것 역시 가능합니다. 
+
+```python
+from diffusers import UNet2DModel
+
+repo_id = "google/ddpm-cifar10-32"
+model = UNet2DModel.from_pretrained(repo_id)
+```
+
+또한 앞서 봤던 `variant` 인자를 명시함으로써, Non-EMA나 `fp16`의 가중치를 가져오는 것 역시 가능합니다. 
+
+```python
+from diffusers import UNet2DConditionModel
+
+model = UNet2DConditionModel.from_pretrained("runwayml/stable-diffusion-v1-5", subfolder="unet", variant="non-ema")
+model.save_pretrained("./local-unet", variant="non-ema")
+```
+
+### 스케줄러
+
+스케줄러들은 [`SchedulerMixin.from_pretrained`] 메서드를 통해 불러올 수 있습니다. 모델과 달리 스케줄러는 별도의 가중치를 갖지 않으며, 따라서 당연히 별도의 학습과정을 요구하지 않습니다. 이러한 스케줄러들은 (해당 스케줄러 하위폴더의) configration 파일을 통해 정의됩니다. 
+
+여러개의 스케줄러를 불러온다고 해서 많은 메모리를 소모하는 것은 아니며, 다양한 스케줄러들에 동일한 스케줄러 configration을  적용하는 것 역시 가능합니다. 다음 예시 코드에서 불러오는 스케줄러들은 모두 [`StableDiffusionPipeline`]과 호환되는데, 이는 곧 해당 스케줄러들에 동일한 스케줄러 configration 파일을 적용할 수 있음을 의미합니다.
+
+```python
+from diffusers import StableDiffusionPipeline
+from diffusers import (
+    DDPMScheduler,
+    DDIMScheduler,
+    PNDMScheduler,
+    LMSDiscreteScheduler,
+    EulerDiscreteScheduler,
+    EulerAncestralDiscreteScheduler,
+    DPMSolverMultistepScheduler,
+)
+
+repo_id = "runwayml/stable-diffusion-v1-5"
+
+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_anc`, `euler`
+pipeline = StableDiffusionPipeline.from_pretrained(repo_id, scheduler=dpm)
+```
+
+### DiffusionPipeline에 대해 알아보기
+
+클래스 메서드로서  [`DiffusionPipeline.from_pretrained`]은 2가지를 담당합니다.
+
+- 첫째로, `from_pretrained` 메서드는 최신 버전의 파이프라인을 다운로드하고, 캐시에 저장합니다. 이미 로컬 캐시에 최신 버전의 파이프라인이 저장되어 있다면, [`DiffusionPipeline.from_pretrained`]은 해당 파일들을 다시 다운로드하지 않고, 로컬 캐시에 저장되어 있는 파이프라인을 불러옵니다.
+-  `model_index.json` 파일을 통해 체크포인트에 대응되는 적합한 파이프라인 클래스로 불러옵니다.
+
+파이프라인의 폴더 구조는 해당 파이프라인 클래스의 구조와 직접적으로 일치합니다. 예를 들어 [`StableDiffusionPipeline`] 클래스는 [`runwayml/stable-diffusion-v1-5`](https://huggingface.co/runwayml/stable-diffusion-v1-5) 레포지토리와 대응되는 구조를 갖습니다.
+
+```python
+from diffusers import DiffusionPipeline
+
+repo_id = "runwayml/stable-diffusion-v1-5"
+pipeline = DiffusionPipeline.from_pretrained(repo_id)
+print(pipeline)
+```
+
+위의 코드 출력 결과를 확인해보면, `pipeline`은 [`StableDiffusionPipeline`]의 인스턴스이며, 다음과 같이 총 7개의 컴포넌트로 구성된다는 것을 알 수 있습니다.
+
+- `"feature_extractor"`: [`~transformers.CLIPFeatureExtractor`]의 인스턴스
+- `"safety_checker"`: 유해한 컨텐츠를 스크리닝하기 위한 [컴포넌트](https://github.com/huggingface/diffusers/blob/e55687e1e15407f60f32242027b7bb8170e58266/src/diffusers/pipelines/stable_diffusion/safety_checker.py#L32)
+- `"scheduler"`: [`PNDMScheduler`]의 인스턴스
+- `"text_encoder"`: [`~transformers.CLIPTextModel`]의 인스턴스
+- `"tokenizer"`: a [`~transformers.CLIPTokenizer`]의 인스턴스
+- `"unet"`: [`UNet2DConditionModel`]의 인스턴스
+- `"vae"` [`AutoencoderKL`]의 인스턴스
+
+```json
+StableDiffusionPipeline {
+  "feature_extractor": [
+    "transformers",
+    "CLIPImageProcessor"
+  ],
+  "safety_checker": [
+    "stable_diffusion",
+    "StableDiffusionSafetyChecker"
+  ],
+  "scheduler": [
+    "diffusers",
+    "PNDMScheduler"
+  ],
+  "text_encoder": [
+    "transformers",
+    "CLIPTextModel"
+  ],
+  "tokenizer": [
+    "transformers",
+    "CLIPTokenizer"
+  ],
+  "unet": [
+    "diffusers",
+    "UNet2DConditionModel"
+  ],
+  "vae": [
+    "diffusers",
+    "AutoencoderKL"
+  ]
+}
+```
+
+파이프라인 인스턴스의 컴포넌트들을  [`runwayml/stable-diffusion-v1-5`](https://huggingface.co/runwayml/stable-diffusion-v1-5)의 폴더 구조와 비교해볼 경우, 각각의 컴포넌트마다 별도의 폴더가 있음을 확인할 수 있습니다.
+
+```
+.
+├── feature_extractor
+│   └── preprocessor_config.json
+├── model_index.json
+├── safety_checker
+│   ├── config.json
+│   └── pytorch_model.bin
+├── scheduler
+│   └── scheduler_config.json
+├── text_encoder
+│   ├── config.json
+│   └── pytorch_model.bin
+├── tokenizer
+│   ├── merges.txt
+│   ├── special_tokens_map.json
+│   ├── tokenizer_config.json
+│   └── vocab.json
+├── unet
+│   ├── config.json
+│   ├── diffusion_pytorch_model.bin
+└── vae
+    ├── config.json
+    ├── diffusion_pytorch_model.bin
+```
+
+또한 각각의 컴포넌트들을 파이프라인 인스턴스의 속성으로써 참조할 수 있습니다. 
+
+```py
+pipeline.tokenizer
+```
+
+```python
+CLIPTokenizer(
+    name_or_path="/root/.cache/huggingface/hub/models--runwayml--stable-diffusion-v1-5/snapshots/39593d5650112b4cc580433f6b0435385882d819/tokenizer",
+    vocab_size=49408,
+    model_max_length=77,
+    is_fast=False,
+    padding_side="right",
+    truncation_side="right",
+    special_tokens={
+        "bos_token": AddedToken("<|startoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=True),
+        "eos_token": AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=True),
+        "unk_token": AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=True),
+        "pad_token": "<|endoftext|>",
+    },
+)
+```
+
+모든 파이프라인은 `model_index.json` 파일을 통해 [`DiffusionPipeline`]에 다음과 같은 정보를 전달합니다.
+
+- `_class_name` 는 어떤 파이프라인 클래스를 사용해야 하는지에 대해 알려줍니다.
+- `_diffusers_version`는 어떤 버전의 디퓨저스로 파이프라인 안의 모델들이 만들어졌는지를 알려줍니다.
+- 그 다음은 각각의 컴포넌트들이 어떤 라이브러리의 어떤 클래스로 만들어졌는지에 대해 알려줍니다. (아래 예시에서 `"feature_extractor" : ["transformers", "CLIPImageProcessor"]`의 경우, `feature_extractor` 컴포넌트는 `transformers` 라이브러리의 `CLIPImageProcessor` 클래스를 통해 만들어졌다는 것을 의미합니다.)
+
+```json
+{
+  "_class_name": "StableDiffusionPipeline",
+  "_diffusers_version": "0.6.0",
+  "feature_extractor": [
+    "transformers",
+    "CLIPImageProcessor"
+  ],
+  "safety_checker": [
+    "stable_diffusion",
+    "StableDiffusionSafetyChecker"
+  ],
+  "scheduler": [
+    "diffusers",
+    "PNDMScheduler"
+  ],
+  "text_encoder": [
+    "transformers",
+    "CLIPTextModel"
+  ],
+  "tokenizer": [
+    "transformers",
+    "CLIPTokenizer"
+  ],
+  "unet": [
+    "diffusers",
+    "UNet2DConditionModel"
+  ],
+  "vae": [
+    "diffusers",
+    "AutoencoderKL"
+  ]
+}
+```
+

docs/source/ko/using-diffusers/other-formats.mdx

@@ -0,0 +1,191 @@
+<!--Copyright 2023 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.
+-->
+
+# 다양한 Stable Diffusion 포맷 불러오기
+
+Stable Diffusion 모델들은 학습 및 저장된 프레임워크와 다운로드 위치에 따라 다양한 형식으로 제공됩니다. 이러한 형식을 🤗 Diffusers에서 사용할 수 있도록 변환하면 추론을 위한 [다양한 스케줄러 사용](schedulers), 사용자 지정 파이프라인 구축, 추론 속도 최적화를 위한 다양한 기법과 방법 등 라이브러리에서 지원하는 모든 기능을 사용할 수 있습니다.
+
+<Tip>
+
+우리는 `.safetensors` 형식을 추천합니다. 왜냐하면 기존의 pickled 파일은 취약하고 머신에서 코드를 실행할 때 악용될 수 있는 것에 비해 훨씬 더 안전합니다. (safetensors 불러오기 가이드에서 자세히 알아보세요.)
+
+</Tip>
+
+이 가이드에서는 다른 Stable Diffusion 형식을 🤗 Diffusers와 호환되도록 변환하는 방법을 설명합니다.
+
+## PyTorch .ckpt
+
+체크포인트 또는 `.ckpt` 형식은 일반적으로 모델을 저장하는 데 사용됩니다. `.ckpt` 파일은 전체 모델을 포함하며 일반적으로 크기가 몇 GB입니다. `.ckpt` 파일을 [~StableDiffusionPipeline.from_ckpt] 메서드를 사용하여 직접 불러와서 사용할 수도 있지만, 일반적으로 두 가지 형식을 모두 사용할 수 있도록 `.ckpt` 파일을 🤗 Diffusers로 변환하는 것이 더 좋습니다.
+
+`.ckpt` 파일을 변환하는 두 가지 옵션이 있습니다. Space를 사용하여 체크포인트를 변환하거나 스크립트를 사용하여 `.ckpt` 파일을 변환합니다.
+
+### Space로 변환하기
+
+`.ckpt` 파일을 변환하는 가장 쉽고 편리한 방법은 SD에서 Diffusers로 스페이스를 사용하는 것입니다. Space의 지침에 따라 .ckpt 파일을 변환 할 수 있습니다.
+
+이 접근 방식은 기본 모델에서는 잘 작동하지만 더 많은 사용자 정의 모델에서는 어려움을 겪을 수 있습니다. 빈 pull request나 오류를 반환하면 Space가 실패한 것입니다.
+이 경우 스크립트를 사용하여 `.ckpt` 파일을 변환해 볼 수 있습니다.
+
+### 스크립트로 변환하기
+
+🤗 Diffusers는 `.ckpt`  파일 변환을 위한 변환 스크립트를 제공합니다. 이 접근 방식은 위의 Space보다 더 안정적입니다.
+
+시작하기 전에 스크립트를 실행할 🤗 Diffusers의 로컬 클론(clone)이 있는지 확인하고 Hugging Face 계정에 로그인하여 pull request를 열고 변환된 모델을 허브에 푸시할 수 있도록 하세요.
+
+```bash
+huggingface-cli login
+```
+
+스크립트를 사용하려면:
+
+1. 변환하려는 `.ckpt`  파일이 포함된 리포지토리를 Git으로 클론(clone)합니다.
+
+이 예제에서는 TemporalNet .ckpt 파일을 변환해 보겠습니다:
+
+```bash
+git lfs install
+git clone https://huggingface.co/CiaraRowles/TemporalNet
+```
+
+2. 체크포인트를 변환할 리포지토리에서 pull request를 엽니다:
+
+```bash
+cd TemporalNet && git fetch origin refs/pr/13:pr/13
+git checkout pr/13
+```
+
+3. 변환 스크립트에서 구성할 입력 인수는 여러 가지가 있지만 가장 중요한 인수는 다음과 같습니다:
+
+- `checkpoint_path`: 변환할 `.ckpt` 파일의 경로를 입력합니다.
+- `original_config_file`: 원래 아키텍처의 구성을 정의하는 YAML 파일입니다. 이 파일을 찾을 수 없는 경우 `.ckpt` 파일을 찾은 GitHub 리포지토리에서 YAML 파일을 검색해 보세요.
+- `dump_path`: 변환된 모델의 경로
+
+예를 들어, TemporalNet 모델은 Stable Diffusion v1.5 및 ControlNet 모델이기 때문에 ControlNet 리포지토리에서 cldm_v15.yaml 파일을 가져올 수 있습니다.
+
+4. 이제 스크립트를 실행하여 .ckpt 파일을 변환할 수 있습니다:
+
+```bash
+python ../diffusers/scripts/convert_original_stable_diffusion_to_diffusers.py --checkpoint_path temporalnetv3.ckpt --original_config_file cldm_v15.yaml --dump_path ./ --controlnet
+```
+
+5. 변환이 완료되면 변환된 모델을 업로드하고 결과물을 pull request [pull request](https://huggingface.co/CiaraRowles/TemporalNet/discussions/13)를 테스트하세요!
+
+```bash
+git push origin pr/13:refs/pr/13
+```
+
+## **Keras .pb or .h5**
+
+🧪 이 기능은 실험적인 기능입니다. 현재로서는 Stable Diffusion v1 체크포인트만 변환 KerasCV Space에서 지원됩니다.
+
+[KerasCV](https://keras.io/keras_cv/)는 [Stable Diffusion](https://github.com/keras-team/keras-cv/blob/master/keras_cv/models/stable_diffusion)  v1 및 v2에 대한 학습을 지원합니다. 그러나 추론 및 배포를 위한 Stable Diffusion 모델 실험을 제한적으로 지원하는 반면, 🤗 Diffusers는 다양한 [noise schedulers](https://huggingface.co/docs/diffusers/using-diffusers/schedulers), [flash attention](https://huggingface.co/docs/diffusers/optimization/xformers), and [other optimization techniques](https://huggingface.co/docs/diffusers/optimization/fp16) 등 이러한 목적을 위한 보다 완벽한 기능을 갖추고 있습니다.
+
+[Convert KerasCV](https://huggingface.co/spaces/sayakpaul/convert-kerascv-sd-diffusers) Space 변환은 `.pb` 또는 `.h5`을 PyTorch로 변환한 다음, 추론할 수 있도록 [`StableDiffusionPipeline`] 으로 감싸서 준비합니다. 변환된 체크포인트는 Hugging Face Hub의 리포지토리에 저장됩니다.
+
+예제로, textual-inversion으로 학습된 `[sayakpaul/textual-inversion-kerasio](https://huggingface.co/sayakpaul/textual-inversion-kerasio/tree/main)` 체크포인트를 변환해 보겠습니다. 이것은 특수 토큰  `<my-funny-cat>`을 사용하여 고양이로 이미지를 개인화합니다.
+
+KerasCV Space 변환에서는 다음을 입력할 수 있습니다:
+
+- Hugging Face 토큰.
+- UNet 과 텍스트 인코더(text encoder) 가중치를 다운로드하는 경로입니다. 모델을 어떻게 학습할지 방식에 따라, UNet과 텍스트 인코더의 경로를 모두 제공할 필요는 없습니다. 예를 들어, textual-inversion에는 텍스트 인코더의 임베딩만 필요하고 텍스트-이미지(text-to-image) 모델 변환에는 UNet 가중치만 필요합니다.
+- Placeholder 토큰은 textual-inversion 모델에만 적용됩니다.
+- `output_repo_prefix`는 변환된 모델이 저장되는 리포지토리의 이름입니다.
+
+**Submit** (제출) 버튼을 클릭하면 KerasCV 체크포인트가 자동으로 변환됩니다! 체크포인트가 성공적으로 변환되면, 변환된 체크포인트가 포함된 새 리포지토리로 연결되는 링크가 표시됩니다. 새 리포지토리로 연결되는 링크를 따라가면 변환된 모델을 사용해 볼 수 있는 추론 위젯이 포함된 모델 카드가 생성된 KerasCV Space 변환을 확인할 수 있습니다.
+
+코드를 사용하여 추론을 실행하려면 모델 카드의 오른쪽 상단 모서리에 있는 **Use in Diffusers**  버튼을 클릭하여 예시 코드를 복사하여 붙여넣습니다:
+
+```py
+from diffusers import DiffusionPipeline
+
+pipeline = DiffusionPipeline.from_pretrained("sayakpaul/textual-inversion-cat-kerascv_sd_diffusers_pipeline")
+```
+
+그러면 다음과 같은 이미지를 생성할 수 있습니다:
+
+```py
+from diffusers import DiffusionPipeline
+
+pipeline = DiffusionPipeline.from_pretrained("sayakpaul/textual-inversion-cat-kerascv_sd_diffusers_pipeline")
+pipeline.to("cuda")
+
+placeholder_token = "<my-funny-cat-token>"
+prompt = f"two {placeholder_token} getting married, photorealistic, high quality"
+image = pipeline(prompt, num_inference_steps=50).images[0]
+```
+
+## **A1111 LoRA files**
+
+[Automatic1111](https://github.com/AUTOMATIC1111/stable-diffusion-webui) (A1111)은 Stable Diffusion을 위해 널리 사용되는 웹 UI로, [Civitai](https://civitai.com/) 와 같은 모델 공유 플랫폼을 지원합니다. 특히 LoRA 기법으로 학습된 모델은 학습 속도가 빠르고 완전히 파인튜닝된 모델보다 파일 크기가 훨씬 작기 때문에 인기가 높습니다.
+
+🤗 Diffusers는 [`~loaders.LoraLoaderMixin.load_lora_weights`]:를 사용하여 A1111 LoRA 체크포인트 불러오기를 지원합니다:
+
+```py
+from diffusers import DiffusionPipeline, UniPCMultistepScheduler
+import torch
+
+pipeline = DiffusionPipeline.from_pretrained(
+    "andite/anything-v4.0", torch_dtype=torch.float16, safety_checker=None
+).to("cuda")
+pipeline.scheduler = UniPCMultistepScheduler.from_config(pipeline.scheduler.config)
+```
+
+Civitai에서 LoRA 체크포인트를 다운로드하세요; 이 예제에서는  [Howls Moving Castle,Interior/Scenery LoRA (Ghibli Stlye)](https://civitai.com/models/14605?modelVersionId=19998) 체크포인트를 사용했지만, 어떤 LoRA 체크포인트든 자유롭게 사용해 보세요!
+
+```bash
+!wget https://civitai.com/api/download/models/19998 -O howls_moving_castle.safetensors
+```
+
+메서드를 사용하여 파이프라인에 LoRA 체크포인트를 불러옵니다:
+
+```py
+pipeline.load_lora_weights(".", weight_name="howls_moving_castle.safetensors")
+```
+
+이제 파이프라인을 사용하여 이미지를 생성할 수 있습니다:
+
+```py
+prompt = "masterpiece, illustration, ultra-detailed, cityscape, san francisco, golden gate bridge, california, bay area, in the snow, beautiful detailed starry sky"
+negative_prompt = "lowres, cropped, worst quality, low quality, normal quality, artifacts, signature, watermark, username, blurry, more than one bridge, bad architecture"
+
+images = pipeline(
+    prompt=prompt,
+    negative_prompt=negative_prompt,
+    width=512,
+    height=512,
+    num_inference_steps=25,
+    num_images_per_prompt=4,
+    generator=torch.manual_seed(0),
+).images
+```
+
+마지막으로, 디스플레이에 이미지를 표시하는 헬퍼 함수를 만듭니다:
+
+```py
+from PIL import Image
+
+
+def image_grid(imgs, rows=2, cols=2):
+    w, h = imgs[0].size
+    grid = Image.new("RGB", size=(cols * w, rows * h))
+
+    for i, img in enumerate(imgs):
+        grid.paste(img, box=(i % cols * w, i // cols * h))
+    return grid
+
+
+image_grid(images)
+```
+
+<div class="flex justify-center">
+  <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/a1111-lora-sf.png" />
+</div>

docs/source/ko/using-diffusers/pipeline_overview.mdx

@@ -0,0 +1,17 @@
+<!--Copyright 2023 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.
+-->
+
+# Overview
+
+파이프라인은 독립적으로 훈련된 모델과 스케줄러를 함께 모아서 추론을 위해 diffusion 시스템을 빠르고 쉽게 사용할 수 있는 방법을 제공하는 end-to-end 클래스입니다. 모델과 스케줄러의 특정 조합은 특수한 기능과 함께 [`StableDiffusionPipeline`] 또는 [`StableDiffusionControlNetPipeline`]과 같은 특정 파이프라인 유형을 정의합니다. 모든 파이프라인 유형은 기본 [`DiffusionPipeline`] 클래스에서 상속됩니다. 어느 체크포인트를 전달하면, 파이프라인 유형을 자동으로 감지하고 필요한 구성 요소들을 불러옵니다.
+
+이 섹션에서는 unconditional 이미지 생성, text-to-image 생성의 다양한 테크닉과 변화를 파이프라인에서 지원하는 작업들을 소개합니다. 프롬프트에 있는 특정 단어가 출력에 영향을 미치는 것을 조정하기 위해 재현성을 위한 시드 설정과 프롬프트에 가중치를 부여하는 것으로 생성 프로세스를 더 잘 제어하는 방법에 대해 배울 수 있습니다. 마지막으로 음성에서부터 이미지 생성과 같은 커스텀 작업을 위한 커뮤니티 파이프라인을 만드는 방법을 알 수 있습니다.

docs/source/ko/using-diffusers/reusing_seeds.mdx

@@ -0,0 +1,63 @@
+<!--Copyright 2023 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.
+-->
+
+# Deterministic(결정적) 생성을 통한 이미지 품질 개선
+
+생성된 이미지의 품질을 개선하는 일반적인 방법은 *결정적 batch(배치) 생성*을 사용하는 것입니다. 이 방법은 이미지 batch(배치)를 생성하고 두 번째 추론 라운드에서 더 자세한 프롬프트와 함께 개선할 이미지 하나를 선택하는 것입니다. 핵심은 일괄 이미지 생성을 위해 파이프라인에 [`torch.Generator`](https://pytorch.org/docs/stable/generated/torch.Generator.html#generator) 목록을 전달하고, 각 `Generator`를 시드에 연결하여 이미지에 재사용할 수 있도록 하는 것입니다.
+
+예를 들어 [`runwayml/stable-diffusion-v1-5`](runwayml/stable-diffusion-v1-5)를 사용하여 다음 프롬프트의 여러 버전을 생성해 봅시다.
+
+```py
+prompt = "Labrador in the style of Vermeer"
+```
+
+(가능하다면) 파이프라인을 [`DiffusionPipeline.from_pretrained`]로 인스턴스화하여 GPU에 배치합니다.
+
+```python
+>>> from diffusers import DiffusionPipeline
+
+>>> pipe = DiffusionPipeline.from_pretrained("runwayml/stable-diffusion-v1-5", torch_dtype=torch.float16)
+>>> pipe = pipe.to("cuda")
+```
+
+이제 네 개의 서로 다른 `Generator`를 정의하고 각 `Generator`에 시드(`0` ~ `3`)를 할당하여 나중에 특정 이미지에 대해 `Generator`를 재사용할 수 있도록 합니다.
+
+```python
+>>> import torch
+
+>>> generator = [torch.Generator(device="cuda").manual_seed(i) for i in range(4)]
+```
+
+이미지를 생성하고 살펴봅니다.
+
+```python
+>>> images = pipe(prompt, generator=generator, num_images_per_prompt=4).images
+>>> images
+```
+
+![img](https://huggingface.co/datasets/diffusers/diffusers-images-docs/resolve/main/reusabe_seeds.jpg)
+
+이 예제에서는 첫 번째 이미지를 개선했지만 실제로는 원하는 모든 이미지를 사용할 수 있습니다(심지어 두 개의 눈이 있는 이미지도!). 첫 번째 이미지에서는 시드가 '0'인 '생성기'를 사용했기 때문에 두 번째 추론 라운드에서는 이 '생성기'를 재사용할 것입니다. 이미지의 품질을 개선하려면 프롬프트에 몇 가지 텍스트를 추가합니다:
+
+```python
+prompt = [prompt + t for t in [", highly realistic", ", artsy", ", trending", ", colorful"]]
+generator = [torch.Generator(device="cuda").manual_seed(0) for i in range(4)]
+```
+
+시드가 `0`인 제너레이터 4개를 생성하고, 이전 라운드의 첫 번째 이미지처럼 보이는 다른 이미지 batch(배치)를 생성합니다!
+
+```python
+>>> images = pipe(prompt, generator=generator).images
+>>> images
+```
+
+![img](https://huggingface.co/datasets/diffusers/diffusers-images-docs/resolve/main/reusabe_seeds_2.jpg)

docs/source/ko/using-diffusers/schedulers.mdx

@@ -0,0 +1,329 @@
+<!--Copyright 2023 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.
+-->
+
+# 스케줄러
+
+diffusion 파이프라인은 diffusion 모델, 스케줄러 등의 컴포넌트들로 구성됩니다. 그리고 파이프라인 안의 일부 컴포넌트를 다른 컴포넌트로 교체하는 식의 커스터마이징 역시 가능합니다.  이와 같은 컴포넌트 커스터마이징의 가장 대표적인 예시가 바로 [스케줄러](../api/schedulers/overview.mdx)를 교체하는 것입니다.
+
+
+
+스케쥴러는 다음과 같이 diffusion 시스템의 전반적인 디노이징 프로세스를 정의합니다.
+
+- 디노이징 스텝을 얼마나 가져가야 할까?
+- 확률적으로(stochastic) 혹은 확정적으로(deterministic)?
+- 디노이징 된 샘플을 찾아내기 위해 어떤 알고리즘을 사용해야 할까?
+
+이러한 프로세스는 다소 난해하고, 디노이징 속도와 디노이징 퀄리티 사이의 트레이드 오프를 정의해야 하는 문제가 될 수 있습니다. 주어진 파이프라인에 어떤 스케줄러가 가장 적합한지를 정량적으로 판단하는 것은 매우 어려운 일입니다. 이로 인해 일단 해당 스케줄러를 직접 사용하여, 생성되는 이미지를 직접 눈으로 보며, 정성적으로 성능을 판단해보는 것이 추천되곤 합니다.
+
+
+
+
+
+## 파이프라인 불러오기
+
+먼저 스테이블 diffusion 파이프라인을 불러오도록 해보겠습니다. 물론 스테이블 diffusion을 사용하기 위해서는, 허깅페이스 허브에 등록된 사용자여야 하며, 관련 [라이센스](https://huggingface.co/runwayml/stable-diffusion-v1-5)에 동의해야 한다는 점을 잊지 말아주세요. 
+
+*역자 주: 다만, 현재 신규로 생성한 허깅페이스 계정에 대해서는 라이센스 동의를 요구하지 않는 것으로 보입니다!*
+
+```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)
+```
+
+다음으로, GPU로 이동합니다.
+
+```python
+pipeline.to("cuda")
+```
+
+
+
+
+
+## 스케줄러 액세스
+
+스케줄러는 언제나 파이프라인의 컴포넌트로서 존재하며, 일반적으로 파이프라인 인스턴스 내에 `scheduler`라는 이름의 속성(property)으로 정의되어 있습니다. 
+
+```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
+}
+```
+
+출력 결과를 통해, 우리는 해당 스케줄러가 [`PNDMScheduler`]의 인스턴스라는 것을 알 수 있습니다. 이제 [`PNDMScheduler`]와 다른 스케줄러들의 성능을 비교해보도록 하겠습니다. 먼저 테스트에 사용할 프롬프트를 다음과 같이 정의해보도록 하겠습니다. 
+
+```python
+prompt = "A photograph of an astronaut riding a horse on Mars, high resolution, high definition."
+```
+
+다음으로 유사한 이미지 생성을 보장하기 위해서, 다음과 같이 랜덤시드를 고정해주도록 하겠습니다. 
+
+```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>
+
+
+
+
+## 스케줄러 교체하기
+
+다음으로 파이프라인의 스케줄러를 다른 스케줄러로 교체하는 방법에 대해 알아보겠습니다. 모든 스케줄러는 [`SchedulerMixin.compatibles`]라는 속성(property)을 갖고 있습니다. 해당 속성은 **호환 가능한** 스케줄러들에 대한 정보를 담고 있습니다. 
+
+```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]
+```
+
+호환되는 스케줄러들을 살펴보면 아래와 같습니다.
+
+- [`LMSDiscreteScheduler`], 
+- [`DDIMScheduler`], 
+- [`DPMSolverMultistepScheduler`], 
+- [`EulerDiscreteScheduler`], 
+- [`PNDMScheduler`], 
+- [`DDPMScheduler`], 
+- [`EulerAncestralDiscreteScheduler`].
+
+앞서 정의했던 프롬프트를 사용해서 각각의 스케줄러들을 비교해보도록 하겠습니다.
+
+먼저 파이프라인 안의 스케줄러를 바꾸기 위해 [`ConfigMixin.config`] 속성과 [`ConfigMixin.from_config`] 메서드를 활용해보려고 합니다.
+
+
+
+```python
+pipeline.scheduler.config
+```
+
+**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)])
+```
+
+기존 스케줄러의 config를 호환 가능한 다른 스케줄러에 이식하는 것 역시 가능합니다. 
+
+다음 예시는 기존 스케줄러(`pipeline.scheduler`)를 다른 종류의 스케줄러(`DDIMScheduler`)로 바꾸는 코드입니다. 기존 스케줄러가 갖고 있던 config를 `.from_config` 메서드의 인자로 전달하는 것을 확인할 수 있습니다.
+
+```python
+from diffusers import DDIMScheduler
+
+pipeline.scheduler = DDIMScheduler.from_config(pipeline.scheduler.config)
+```
+
+
+
+이제 파이프라인을 실행해서 두 스케줄러 사이의 생성된 이미지의 퀄리티를 비교해봅시다.
+
+```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>
+
+
+
+
+## 스케줄러들 비교해보기
+
+지금까지는 [`PNDMScheduler`]와 [`DDIMScheduler`] 스케줄러를 실행해보았습니다. 아직 비교해볼 스케줄러들이 더 많이 남아있으니 계속 비교해보도록 하겠습니다.
+
+
+
+[`LMSDiscreteScheduler`]을 일반적으로 더 좋은 결과를 보여줍니다.
+
+```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`]와 [`EulerAncestralDiscreteScheduler`] 고작 30번의 inference step만으로도 높은 퀄리티의 이미지를 생성하는 것을 알 수 있습니다.
+
+```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>
+
+
+```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>
+
+
+지금 이 문서를 작성하는 현시점 기준에선, [`DPMSolverMultistepScheduler`]가 시간 대비 가장 좋은 품질의 이미지를 생성하는 것 같습니다. 20번 정도의 스텝만으로도 실행될 수 있습니다.
+
+
+
+```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>
+
+
+보시다시피 생성된 이미지들은 매우 비슷하고, 비슷한 퀄리티를 보이는 것 같습니다. 실제로 어떤 스케줄러를 선택할 것인가는 종종 특정 이용 사례에 기반해서 결정되곤 합니다. 결국 여러 종류의 스케줄러를 직접 실행시켜보고 눈으로 직접 비교해서 판단하는 게 좋은 선택일 것 같습니다.
+
+
+
+## Flax에서 스케줄러 교체하기
+
+JAX/Flax 사용자인 경우 기본 파이프라인 스케줄러를 변경할 수도 있습니다. 다음은 Flax Stable Diffusion 파이프라인과 초고속 [DDPM-Solver++ 스케줄러를](../api/schedulers/multistep_dpm_solver) 사용하여 추론을 실행하는 방법에 대한 예시입니다 .
+
+```Python
+import jax
+import numpy as np
+from flax.jax_utils import replicate
+from flax.training.common_utils import shard
+
+from diffusers import FlaxStableDiffusionPipeline, FlaxDPMSolverMultistepScheduler
+
+model_id = "runwayml/stable-diffusion-v1-5"
+scheduler, scheduler_state = FlaxDPMSolverMultistepScheduler.from_pretrained(
+    model_id,
+    subfolder="scheduler"
+)
+pipeline, params = FlaxStableDiffusionPipeline.from_pretrained(
+    model_id,
+    scheduler=scheduler,
+    revision="bf16",
+    dtype=jax.numpy.bfloat16,
+)
+params["scheduler"] = scheduler_state
+
+# Generate 1 image per parallel device (8 on TPUv2-8 or TPUv3-8)
+prompt = "a photo of an astronaut riding a horse on mars"
+num_samples = jax.device_count()
+prompt_ids = pipeline.prepare_inputs([prompt] * num_samples)
+
+prng_seed = jax.random.PRNGKey(0)
+num_inference_steps = 25
+
+# shard inputs and rng
+params = replicate(params)
+prng_seed = jax.random.split(prng_seed, jax.device_count())
+prompt_ids = shard(prompt_ids)
+
+images = pipeline(prompt_ids, params, prng_seed, num_inference_steps, jit=True).images
+images = pipeline.numpy_to_pil(np.asarray(images.reshape((num_samples,) + images.shape[-3:])))
+```
+
+<Tip warning={true}>
+
+다음 Flax 스케줄러는 *아직* Flax Stable Diffusion 파이프라인과 호환되지 않습니다.
+
+- `FlaxLMSDiscreteScheduler`
+- `FlaxDDPMScheduler`
+
+</Tip>
+

docs/source/ko/using-diffusers/unconditional_image_generation.mdx

@@ -0,0 +1,54 @@
+<!--Copyright 2023 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.
+-->
+
+# Unconditional 이미지 생성
+
+[[Colab에서 열기]]
+
+Unconditional 이미지 생성은 비교적 간단한 작업입니다. 모델이 텍스트나 이미지와 같은 추가 조건 없이 이미 학습된 학습 데이터와 유사한 이미지만 생성합니다.
+
+['DiffusionPipeline']은 추론을 위해 미리 학습된 diffusion 시스템을 사용하는 가장 쉬운 방법입니다.
+
+먼저 ['DiffusionPipeline']의 인스턴스를 생성하고 다운로드할 파이프라인의 [체크포인트](https://huggingface.co/models?library=diffusers&sort=downloads)를 지정합니다. 허브의 🧨 diffusion 체크포인트 중 하나를 사용할 수 있습니다(사용할 체크포인트는 나비 이미지를 생성합니다).
+
+<Tip>
+
+💡 나만의 unconditional 이미지 생성 모델을 학습시키고 싶으신가요? 학습 가이드를 살펴보고 나만의 이미지를 생성하는 방법을 알아보세요.
+
+</Tip>
+
+
+이 가이드에서는 unconditional 이미지 생성에 ['DiffusionPipeline']과 [DDPM](https://arxiv.org/abs/2006.11239)을 사용합니다:
+
+	```python
+ >>> from diffusers import DiffusionPipeline
+
+ >>> generator = DiffusionPipeline.from_pretrained("anton-l/ddpm-butterflies-128")
+	```
+[diffusion 파이프라인]은 모든 모델링, 토큰화, 스케줄링 구성 요소를 다운로드하고 캐시합니다. 이 모델은 약 14억 개의 파라미터로 구성되어 있기 때문에 GPU에서 실행할 것을 강력히 권장합니다. PyTorch에서와 마찬가지로 제너레이터 객체를 GPU로 옮길 수 있습니다:
+	```python
+ >>> generator.to("cuda")
+	```
+이제 제너레이터를 사용하여 이미지를 생성할 수 있습니다:
+	```python
+ >>> image = generator().images[0]
+	```
+출력은 기본적으로 [PIL.Image](https://pillow.readthedocs.io/en/stable/reference/Image.html?highlight=image#the-image-class) 객체로 감싸집니다.
+
+다음을 호출하여 이미지를 저장할 수 있습니다:
+	```python
+ >>> image.save("generated_image.png")
+	```
+	
+아래 스페이스(데모 링크)를 이용해 보고, 추론 단계의 매개변수를 자유롭게 조절하여 이미지 품질에 어떤 영향을 미치는지 확인해 보세요!
+
+<iframe src="https://stevhliu-ddpm-butterflies-128.hf.space" frameborder="0" width="850" height="500"></iframe>

docs/source/ko/using-diffusers/using_safetensors.mdx

@@ -0,0 +1,14 @@
+# 세이프센서란 무엇인가요? 
+
+[세이프텐서](https://github.com/huggingface/safetensors)는 피클을 사용하는 파이토치를 사용하는 기존의 '.bin'과는 다른 형식입니다.
+
+피클은 악의적인 파일이 임의의 코드를 실행할 수 있는 안전하지 않은 것으로 악명이 높습니다.
+허브 자체에서 문제를 방지하기 위해 노력하고 있지만 만병통치약은 아닙니다.
+
+세이프텐서의 가장 중요한 목표는 컴퓨터를 탈취할 수 없다는 의미에서 머신 러닝 모델 로딩을 *안전하게* 만드는 것입니다.
+
+# 왜 세이프센서를 사용하나요?
+
+**잘 알려지지 않은 모델을 사용하려는 경우, 그리고 파일의 출처가 확실하지 않은 경우 "안전성"이 하나의 이유가 될 수 있습니다.
+
+그리고 두 번째 이유는 **로딩 속도**입니다. 세이프센서는 일반 피클 파일보다 훨씬 빠르게 모델을 훨씬 빠르게 로드할 수 있습니다. 모델을 전환하는 데 많은 시간을 소비하는 경우, 이는 엄청난 시간 절약이 가능합니다.

docs/source/ko/using-diffusers/write_own_pipeline.mdx

@@ -0,0 +1,290 @@
+<!--Copyright 2023 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.
+-->
+
+# 파이프라인, 모델 및 스케줄러 이해하기
+
+[[colab에서 열기]]
+
+🧨 Diffusers는 사용자 친화적이며 유연한 도구 상자로, 사용사례에 맞게 diffusion 시스템을 구축 할 수 있도록 설계되었습니다. 이 도구 상자의 핵심은 모델과 스케줄러입니다. [`DiffusionPipeline`]은 편의를 위해 이러한 구성 요소를 번들로 제공하지만, 파이프라인을 분리하고 모델과 스케줄러를 개별적으로 사용해 새로운 diffusion 시스템을 만들 수도 있습니다. 
+
+이 튜토리얼에서는 기본 파이프라인부터 시작해 Stable Diffusion 파이프라인까지 진행하며 모델과 스케줄러를 사용해 추론을 위한 diffusion 시스템을 조립하는 방법을 배웁니다.
+
+## 기본 파이프라인 해체하기
+
+파이프라인은 추론을 위해 모델을 실행하는 빠르고 쉬운 방법으로, 이미지를 생성하는 데 코드가 4줄 이상 필요하지 않습니다:
+
+```py
+>>> from diffusers import DDPMPipeline
+
+>>> ddpm = DDPMPipeline.from_pretrained("google/ddpm-cat-256").to("cuda")
+>>> image = ddpm(num_inference_steps=25).images[0]
+>>> image
+```
+
+<div class="flex justify-center">
+    <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/ddpm-cat.png" alt="Image of cat created from DDPMPipeline"/>
+</div>
+
+정말 쉽습니다. 그런데 파이프라인은 어떻게 이렇게 할 수 있었을까요? 파이프라인을 세분화하여 내부에서 어떤 일이 일어나고 있는지 살펴보겠습니다.
+
+위 예시에서 파이프라인에는 [`UNet2DModel`] 모델과 [`DDPMScheduler`]가 포함되어 있습니다. 파이프라인은 원하는 출력 크기의 랜덤 노이즈를 받아 모델을 여러번 통과시켜 이미지의 노이즈를 제거합니다. 각 timestep에서 모델은 *noise residual*을 예측하고 스케줄러는 이를 사용하여 노이즈가 적은 이미지를 예측합니다. 파이프라인은 지정된 추론 스텝수에 도달할 때까지 이 과정을 반복합니다. 
+
+모델과 스케줄러를 별도로 사용하여 파이프라인을 다시 생성하기 위해 자체적인 노이즈 제거 프로세스를 작성해 보겠습니다.
+
+1. 모델과 스케줄러를 불러옵니다:
+
+    ```py
+    >>> from diffusers import DDPMScheduler, UNet2DModel
+
+    >>> scheduler = DDPMScheduler.from_pretrained("google/ddpm-cat-256")
+    >>> model = UNet2DModel.from_pretrained("google/ddpm-cat-256").to("cuda")
+    ```
+
+2. 노이즈 제거 프로세스를 실행할 timestep 수를 설정합니다:
+
+    ```py
+    >>> scheduler.set_timesteps(50)
+    ```
+
+3. 스케줄러의 timestep을 설정하면 균등한 간격의 구성 요소를 가진 텐서가 생성됩니다.(이 예시에서는 50개) 각 요소는 모델이 이미지의 노이즈를 제거하는 시간 간격에 해당합니다. 나중에 노이즈 제거 루프를 만들 때 이 텐서를 반복하여 이미지의 노이즈를 제거합니다:
+
+    ```py
+    >>> scheduler.timesteps
+    tensor([980, 960, 940, 920, 900, 880, 860, 840, 820, 800, 780, 760, 740, 720,
+        700, 680, 660, 640, 620, 600, 580, 560, 540, 520, 500, 480, 460, 440,
+        420, 400, 380, 360, 340, 320, 300, 280, 260, 240, 220, 200, 180, 160,
+        140, 120, 100,  80,  60,  40,  20,   0])
+    ```
+
+4. 원하는 출력과 같은 모양을 가진 랜덤 노이즈를 생성합니다:
+
+    ```py
+    >>> import torch
+
+    >>> sample_size = model.config.sample_size
+    >>> noise = torch.randn((1, 3, sample_size, sample_size)).to("cuda")
+    ```
+
+5. 이제 timestep을 반복하는 루프를 작성합니다. 각 timestep에서 모델은 [`UNet2DModel.forward`]를 통해 noisy residual을 반환합니다. 스케줄러의 [`~DDPMScheduler.step`] 메서드는 noisy residual, timestep, 그리고 입력을 받아 이전 timestep에서 이미지를 예측합니다. 이 출력은 노이즈 제거 루프의 모델에 대한 다음 입력이 되며, `timesteps` 배열의 끝에 도달할 때까지 반복됩니다.
+
+    ```py
+    >>> input = noise
+
+    >>> for t in scheduler.timesteps:
+    ...     with torch.no_grad():
+    ...         noisy_residual = model(input, t).sample
+    ...     previous_noisy_sample = scheduler.step(noisy_residual, t, input).prev_sample
+    ...     input = previous_noisy_sample
+    ```
+
+    이것이 전체 노이즈 제거 프로세스이며, 동일한 패턴을 사용해 모든 diffusion 시스템을 작성할 수 있습니다.
+
+6. 마지막 단계는 노이즈가 제거된 출력을 이미지로 변환하는 것입니다:
+
+    ```py
+    >>> from PIL import Image
+    >>> import numpy as np
+
+    >>> image = (input / 2 + 0.5).clamp(0, 1)
+    >>> image = image.cpu().permute(0, 2, 3, 1).numpy()[0]
+    >>> image = Image.fromarray((image * 255).round().astype("uint8"))
+    >>> image
+    ```
+
+다음 섹션에서는 여러분의 기술을 시험해보고 좀 더 복잡한 Stable Diffusion 파이프라인을 분석해 보겠습니다. 방법은 거의 동일합니다. 필요한 구성요소들을 초기화하고 timestep수를 설정하여 `timestep` 배열을 생성합니다. 노이즈 제거 루프에서 `timestep` 배열이 사용되며, 이 배열의 각 요소에 대해 모델은 노이즈가 적은 이미지를 예측합니다. 노이즈 제거 루프는 `timestep`을 반복하고 각 timestep에서 noise residual을 출력하고 스케줄러는 이를 사용하여 이전 timestep에서 노이즈가 덜한 이미지를 예측합니다. 이 프로세스는 `timestep` 배열의 끝에 도달할 때까지 반복됩니다.
+
+한번 사용해 봅시다!
+
+## Stable Diffusion 파이프라인 해체하기
+
+Stable Diffusion 은 text-to-image *latent diffusion* 모델입니다. latent diffusion 모델이라고 불리는 이유는 실제 픽셀 공간 대신 이미지의 저차원의 표현으로 작업하기 때문이고, 메모리 효율이 더 높습니다. 인코더는 이미지를 더 작은 표현으로 압축하고, 디코더는 압축된 표현을 다시 이미지로 변환합니다. text-to-image 모델의 경우 텍스트 임베딩을 생성하기 위해 tokenizer와 인코더가 필요합니다. 이전 예제에서 이미 UNet 모델과 스케줄러가 필요하다는 것은 알고 계셨을 것입니다.
+
+보시다시피, 이것은 UNet 모델만 포함된 DDPM 파이프라인보다 더 복잡합니다. Stable Diffusion 모델에는 세 개의 개별 사전학습된 모델이 있습니다.
+
+<Tip>
+
+💡 VAE, UNet 및 텍스트 인코더 모델의 작동방식에 대한 자세한 내용은 [How does Stable Diffusion work?](https://huggingface.co/blog/stable_diffusion#how-does-stable-diffusion-work) 블로그를 참조하세요.
+
+</Tip>
+
+이제 Stable Diffusion 파이프라인에 필요한 구성요소들이 무엇인지 알았으니, [`~ModelMixin.from_pretrained`] 메서드를 사용해 모든 구성요소를 불러옵니다. 사전학습된 체크포인트 [`runwayml/stable-diffusion-v1-5`](https://huggingface.co/runwayml/stable-diffusion-v1-5)에서 찾을 수 있으며, 각 구성요소들은 별도의 하위 폴더에 저장되어 있습니다:
+
+```py
+>>> from PIL import Image
+>>> import torch
+>>> from transformers import CLIPTextModel, CLIPTokenizer
+>>> from diffusers import AutoencoderKL, UNet2DConditionModel, PNDMScheduler
+
+>>> vae = AutoencoderKL.from_pretrained("CompVis/stable-diffusion-v1-4", subfolder="vae")
+>>> tokenizer = CLIPTokenizer.from_pretrained("CompVis/stable-diffusion-v1-4", subfolder="tokenizer")
+>>> text_encoder = CLIPTextModel.from_pretrained("CompVis/stable-diffusion-v1-4", subfolder="text_encoder")
+>>> unet = UNet2DConditionModel.from_pretrained("CompVis/stable-diffusion-v1-4", subfolder="unet")
+```
+
+기본 [`PNDMScheduler`] 대신, [`UniPCMultistepScheduler`]로 교체하여 다른 스케줄러를 얼마나 쉽게 연결할 수 있는지 확인합니다:
+
+```py
+>>> from diffusers import UniPCMultistepScheduler
+
+>>> scheduler = UniPCMultistepScheduler.from_pretrained("CompVis/stable-diffusion-v1-4", subfolder="scheduler")
+```
+
+추론 속도를 높이려면 스케줄러와 달리 학습 가능한 가중치가 있으므로 모델을 GPU로 옮기세요:
+
+```py
+>>> torch_device = "cuda"
+>>> vae.to(torch_device)
+>>> text_encoder.to(torch_device)
+>>> unet.to(torch_device)
+```
+
+### 텍스트 임베딩 생성하기
+
+다음 단계는 임베딩을 생성하기 위해 텍스트를 토큰화하는 것입니다. 이 텍스트는 UNet 모델에서 condition으로 사용되고 입력 프롬프트와 유사한 방향으로 diffusion 프로세스를 조정하는 데 사용됩니다.
+
+<Tip>
+
+💡 `guidance_scale` 매개변수는 이미지를 생성할 때 프롬프트에 얼마나 많은 가중치를 부여할지 결정합니다.
+
+</Tip>
+
+다른 프롬프트를 생성하고 싶다면 원하는 프롬프트를 자유롭게 선택하세요!
+
+```py
+>>> prompt = ["a photograph of an astronaut riding a horse"]
+>>> height = 512  # Stable Diffusion의 기본 높이
+>>> width = 512  # Stable Diffusion의 기본 너비
+>>> num_inference_steps = 25  # 노이즈 제거 스텝 수
+>>> guidance_scale = 7.5  # classifier-free guidance를 위한 scale
+>>> generator = torch.manual_seed(0)  # 초기 잠재 노이즈를 생성하는 seed generator
+>>> batch_size = len(prompt)
+```
+
+텍스트를 토큰화하고 프롬프트에서 임베딩을 생성합니다:
+
+```py
+>>> text_input = tokenizer(
+...     prompt, padding="max_length", max_length=tokenizer.model_max_length, truncation=True, return_tensors="pt"
+... )
+
+>>> with torch.no_grad():
+...     text_embeddings = text_encoder(text_input.input_ids.to(torch_device))[0]
+```
+
+또한 패딩 토큰의 임베딩인 *unconditional 텍스트 임베딩*을 생성해야 합니다. 이 임베딩은 조건부 `text_embeddings`과 동일한 shape(`batch_size` 그리고 `seq_length`)을 가져야 합니다:
+
+```py
+>>> max_length = text_input.input_ids.shape[-1]
+>>> uncond_input = tokenizer([""] * batch_size, padding="max_length", max_length=max_length, return_tensors="pt")
+>>> uncond_embeddings = text_encoder(uncond_input.input_ids.to(torch_device))[0]
+```
+
+두번의 forward pass를 피하기 위해 conditional 임베딩과 unconditional 임베딩을 배치(batch)로 연결하겠습니다:
+
+```py
+>>> text_embeddings = torch.cat([uncond_embeddings, text_embeddings])
+```
+
+### 랜덤 노이즈 생성
+
+그다음 diffusion 프로세스의 시작점으로 초기 랜덤 노이즈를 생성합니다. 이것이 이미지의 잠재적 표현이며 점차적으로 노이즈가 제거됩니다. 이 시점에서 `latent` 이미지는 최종 이미지 크기보다 작지만 나중에 모델이 이를 512x512 이미지 크기로 변환하므로 괜찮습니다.
+
+<Tip>
+
+💡 `vae` 모델에는 3개의 다운 샘플링 레이어가 있기 때문에 높이와 너비가 8로 나뉩니다. 다음을 실행하여 확인할 수 있습니다:
+
+```py
+2 ** (len(vae.config.block_out_channels) - 1) == 8
+```
+
+</Tip>
+
+```py
+>>> latents = torch.randn(
+...     (batch_size, unet.in_channels, height // 8, width // 8),
+...     generator=generator,
+... )
+>>> latents = latents.to(torch_device)
+```
+
+### 이미지 노이즈 제거
+
+먼저 [`UniPCMultistepScheduler`]와 같은 향상된 스케줄러에 필요한 노이즈 스케일 값인 초기 노이즈 분포 *sigma* 로 입력을 스케일링 하는 것부터 시작합니다:
+
+```py
+>>> latents = latents * scheduler.init_noise_sigma
+```
+
+마지막 단계는 `latent`의 순수한 노이즈를 점진적으로 프롬프트에 설명된 이미지로 변환하는 노이즈 제거 루프를 생성하는 것입니다. 노이즈 제거 루프는 세 가지 작업을 수행해야 한다는 점을 기억하세요:
+
+1. 노이즈 제거 중에 사용할 스케줄러의 timesteps를 설정합니다.
+2. timestep을 따라 반복합니다.
+3. 각 timestep에서 UNet 모델을 호출하여 noise residual을 예측하고 스케줄러에 전달하여 이전 노이즈 샘플을 계산합니다.
+
+```py
+>>> from tqdm.auto import tqdm
+
+>>> scheduler.set_timesteps(num_inference_steps)
+
+>>> for t in tqdm(scheduler.timesteps):
+...     # classifier-free guidance를 수행하는 경우 두번의 forward pass를 수행하지 않도록 latent를 확장.
+...     latent_model_input = torch.cat([latents] * 2)
+
+...     latent_model_input = scheduler.scale_model_input(latent_model_input, timestep=t)
+
+...     # noise residual 예측
+...     with torch.no_grad():
+...         noise_pred = unet(latent_model_input, t, encoder_hidden_states=text_embeddings).sample
+
+...     # guidance 수행
+...     noise_pred_uncond, noise_pred_text = noise_pred.chunk(2)
+...     noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond)
+
+...     # 이전 노이즈 샘플을 계산 x_t -> x_t-1
+...     latents = scheduler.step(noise_pred, t, latents).prev_sample
+```
+
+### 이미지 디코딩
+
+마지막 단계는 `vae`를 이용하여 잠재 표현을 이미지로 디코딩하고 `sample`과 함께 디코딩된 출력을 얻는 것입니다:
+
+```py
+# latent를 스케일링하고 vae로 이미지 디코딩
+latents = 1 / 0.18215 * latents
+with torch.no_grad():
+    image = vae.decode(latents).sample
+```
+
+마지막으로 이미지를 `PIL.Image`로 변환하면 생성된 이미지를 확인할 수 있습니다!
+
+```py
+>>> image = (image / 2 + 0.5).clamp(0, 1)
+>>> image = image.detach().cpu().permute(0, 2, 3, 1).numpy()
+>>> images = (image * 255).round().astype("uint8")
+>>> pil_images = [Image.fromarray(image) for image in images]
+>>> pil_images[0]
+```
+
+<div class="flex justify-center">
+    <img src="https://huggingface.co/blog/assets/98_stable_diffusion/stable_diffusion_k_lms.png"/>
+</div>
+
+## 다음 단계
+
+기본 파이프라인부터 복잡한 파이프라인까지, 자신만의 diffusion 시스템을 작성하는 데 필요한 것은 노이즈 제거 루프뿐이라는 것을 알 수 있었습니다. 이 루프는 스케줄러의 timesteps를 설정하고, 이를 반복하며, UNet 모델을 호출하여 noise residual을 예측하고 스케줄러에 전달하여 이전 노이즈 샘플을 계산하는 과정을 번갈아 가며 수행해야 합니다.
+
+이것이 바로 🧨 Diffusers가 설계된 목적입니다: 모델과 스케줄러를 사용해 자신만의 diffusion 시스템을 직관적이고 쉽게 작성할 수 있도록 하기 위해서입니다.
+
+다음 단계를 자유롭게 진행하세요:
+
+* 🧨 Diffusers에 [파이프라인 구축 및 기여](using-diffusers/#contribute_pipeline)하는 방법을 알아보세요. 여러분이 어떤 아이디어를 내놓을지 기대됩니다!
+* 라이브러리에서 [기본 파이프라인](./api/pipelines/overview)을 살펴보고, 모델과 스케줄러를 별도로 사용하여 파이프라인을 처음부터 해체하고 빌드할 수 있는지 확인해 보세요.

examples/community/README.md

@@ -38,6 +38,7 @@ If a community doesn't work as expected, please open an issue and ping the autho
 | Stable Diffusion IPEX Pipeline | Accelerate Stable Diffusion inference pipeline with BF16/FP32 precision on Intel Xeon CPUs with [IPEX](https://github.com/intel/intel-extension-for-pytorch) | [Stable Diffusion on IPEX](#stable-diffusion-on-ipex) | - | [Yingjie Han](https://github.com/yingjie-han/) | 
 | CLIP Guided Images Mixing Stable Diffusion Pipeline | Сombine images using usual diffusion models. | [CLIP Guided Images Mixing Using Stable Diffusion](#clip-guided-images-mixing-with-stable-diffusion) | - | [Karachev Denis](https://github.com/TheDenk) |  
 | TensorRT Stable Diffusion Inpainting Pipeline                                                                                                    | Accelerates the Stable Diffusion Inpainting Pipeline using TensorRT                                                                                                                                                                                                                                                                                                                                                                                                                                      | [TensorRT Stable Diffusion Inpainting Pipeline](#tensorrt-inpainting-stable-diffusion-pipeline)      | - |              [Asfiya Baig](https://github.com/asfiyab-nvidia) |
+|   IADB Pipeline                                                                                                    | Implementation of [Iterative α-(de)Blending: a Minimalist Deterministic Diffusion Model](https://arxiv.org/abs/2305.03486)                                                                                                                                                                                                                                                                                                                                                                                                                                      | [IADB Pipeline](#iadb-pipeline)      | - |              [Thomas Chambon](https://github.com/tchambon) 
 
 To load a custom pipeline you just need to pass the `custom_pipeline` argument to `DiffusionPipeline`, as one of the files in `diffusers/examples/community`. Feel free to send a PR with your own pipelines, we will merge them quickly.
 ```py
@@ -1707,3 +1708,62 @@ output = pipeline(
 ```
 ![Input_Image](https://huggingface.co/datasets/kadirnar/diffusers_readme_images/resolve/main/input_image.png)
 ![mixture_canvas_results](https://huggingface.co/datasets/kadirnar/diffusers_readme_images/resolve/main/canvas.png)
+
+
+### IADB pipeline
+
+This pipeline is the implementation of the [α-(de)Blending: a Minimalist Deterministic Diffusion Model](https://arxiv.org/abs/2305.03486) paper.
+It is a simple and minimalist diffusion model.
+
+The following code shows how to use the IADB pipeline to generate images using a pretrained celebahq-256 model.
+
+```python
+
+pipeline_iadb = DiffusionPipeline.from_pretrained("thomasc4/iadb-celebahq-256", custom_pipeline='iadb')
+
+pipeline_iadb = pipeline_iadb.to('cuda')
+
+output = pipeline_iadb(batch_size=4,num_inference_steps=128)
+for i in range(len(output[0])):
+    plt.imshow(output[0][i])
+    plt.show()
+
+```
+
+Sampling with the IADB formulation is easy, and can be done in a few lines (the pipeline already implements it):
+
+```python
+
+def sample_iadb(model, x0, nb_step):
+    x_alpha = x0
+    for t in range(nb_step):
+        alpha = (t/nb_step)
+        alpha_next =((t+1)/nb_step)
+
+        d = model(x_alpha, torch.tensor(alpha, device=x_alpha.device))['sample']
+        x_alpha = x_alpha + (alpha_next-alpha)*d
+
+    return x_alpha
+
+```
+
+The training loop is also straightforward:
+
+```python
+
+# Training loop
+while True:
+    x0 = sample_noise()
+    x1 = sample_dataset()
+
+    alpha = torch.rand(batch_size)
+
+    # Blend
+    x_alpha = (1-alpha) * x0 + alpha * x1
+
+    # Loss
+    loss = torch.sum((D(x_alpha, alpha)- (x1-x0))**2)
+    optimizer.zero_grad()
+    loss.backward()
+    optimizer.step()
+```

examples/community/iadb.py

@@ -0,0 +1,149 @@
+from typing import List, Optional, Tuple, Union
+
+import torch
+
+from diffusers import DiffusionPipeline
+from diffusers.configuration_utils import ConfigMixin
+from diffusers.pipeline_utils import ImagePipelineOutput
+from diffusers.schedulers.scheduling_utils import SchedulerMixin
+
+
+class IADBScheduler(SchedulerMixin, ConfigMixin):
+    """
+    IADBScheduler is a scheduler for the Iterative α-(de)Blending denoising method. It is simple and minimalist.
+
+    For more details, see the original paper: https://arxiv.org/abs/2305.03486 and the blog post: https://ggx-research.github.io/publication/2023/05/10/publication-iadb.html
+    """
+
+    def step(
+        self,
+        model_output: torch.FloatTensor,
+        timestep: int,
+        x_alpha: torch.FloatTensor,
+    ) -> torch.FloatTensor:
+        """
+        Predict the sample at the previous timestep by reversing the ODE. Core function to propagate the diffusion
+        process from the learned model outputs (most often the predicted noise).
+
+        Args:
+            model_output (`torch.FloatTensor`): direct output from learned diffusion model. It is the direction from x0 to x1.
+            timestep (`float`): current timestep in the diffusion chain.
+            x_alpha (`torch.FloatTensor`): x_alpha sample for the current timestep
+
+        Returns:
+            `torch.FloatTensor`: the sample at the previous timestep
+
+        """
+        if self.num_inference_steps is None:
+            raise ValueError(
+                "Number of inference steps is 'None', you need to run 'set_timesteps' after creating the scheduler"
+            )
+
+        alpha = timestep / self.num_inference_steps
+        alpha_next = (timestep + 1) / self.num_inference_steps
+
+        d = model_output
+
+        x_alpha = x_alpha + (alpha_next - alpha) * d
+
+        return x_alpha
+
+    def set_timesteps(self, num_inference_steps: int):
+        self.num_inference_steps = num_inference_steps
+
+    def add_noise(
+        self,
+        original_samples: torch.FloatTensor,
+        noise: torch.FloatTensor,
+        alpha: torch.FloatTensor,
+    ) -> torch.FloatTensor:
+        return original_samples * alpha + noise * (1 - alpha)
+
+    def __len__(self):
+        return self.config.num_train_timesteps
+
+
+class IADBPipeline(DiffusionPipeline):
+    r"""
+    This model inherits from [`DiffusionPipeline`]. Check the superclass documentation for the generic methods the
+    library implements for all the pipelines (such as downloading or saving, running on a particular device, etc.)
+
+    Parameters:
+        unet ([`UNet2DModel`]): U-Net architecture to denoise the encoded image.
+        scheduler ([`SchedulerMixin`]):
+            A scheduler to be used in combination with `unet` to denoise the encoded image. Can be one of
+            [`DDPMScheduler`], or [`DDIMScheduler`].
+    """
+
+    def __init__(self, unet, scheduler):
+        super().__init__()
+
+        self.register_modules(unet=unet, scheduler=scheduler)
+
+    @torch.no_grad()
+    def __call__(
+        self,
+        batch_size: int = 1,
+        generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
+        num_inference_steps: int = 50,
+        output_type: Optional[str] = "pil",
+        return_dict: bool = True,
+    ) -> Union[ImagePipelineOutput, Tuple]:
+        r"""
+        Args:
+            batch_size (`int`, *optional*, defaults to 1):
+                The number of images to generate.
+            num_inference_steps (`int`, *optional*, defaults to 50):
+                The number of denoising steps. More denoising steps usually lead to a higher quality image at the
+                expense of slower inference.
+            output_type (`str`, *optional*, defaults to `"pil"`):
+                The output format of the generate image. Choose between
+                [PIL](https://pillow.readthedocs.io/en/stable/): `PIL.Image.Image` or `np.array`.
+            return_dict (`bool`, *optional*, defaults to `True`):
+                Whether or not to return a [`~pipelines.ImagePipelineOutput`] instead of a plain tuple.
+
+        Returns:
+            [`~pipelines.ImagePipelineOutput`] or `tuple`: [`~pipelines.utils.ImagePipelineOutput`] if `return_dict` is
+            True, otherwise a `tuple. When returning a tuple, the first element is a list with the generated images.
+        """
+
+        # Sample gaussian noise to begin loop
+        if isinstance(self.unet.config.sample_size, int):
+            image_shape = (
+                batch_size,
+                self.unet.config.in_channels,
+                self.unet.config.sample_size,
+                self.unet.config.sample_size,
+            )
+        else:
+            image_shape = (batch_size, self.unet.config.in_channels, *self.unet.config.sample_size)
+
+        if isinstance(generator, list) and len(generator) != batch_size:
+            raise ValueError(
+                f"You have passed a list of generators of length {len(generator)}, but requested an effective batch"
+                f" size of {batch_size}. Make sure the batch size matches the length of the generators."
+            )
+
+        image = torch.randn(image_shape, generator=generator, device=self.device, dtype=self.unet.dtype)
+
+        # set step values
+        self.scheduler.set_timesteps(num_inference_steps)
+        x_alpha = image.clone()
+        for t in self.progress_bar(range(num_inference_steps)):
+            alpha = t / num_inference_steps
+
+            # 1. predict noise model_output
+            model_output = self.unet(x_alpha, torch.tensor(alpha, device=x_alpha.device)).sample
+
+            # 2. step
+            x_alpha = self.scheduler.step(model_output, t, x_alpha)
+
+        image = (x_alpha * 0.5 + 0.5).clamp(0, 1)
+        image = image.cpu().permute(0, 2, 3, 1).numpy()
+        if output_type == "pil":
+            image = self.numpy_to_pil(image)
+
+        if not return_dict:
+            return (image,)
+
+        return ImagePipelineOutput(images=image)

examples/controlnet/README.md

@@ -274,9 +274,9 @@ pipe = StableDiffusionControlNetPipeline.from_pretrained(
 
 # speed up diffusion process with faster scheduler and memory optimization
 pipe.scheduler = UniPCMultistepScheduler.from_config(pipe.scheduler.config)
-# remove following line if xformers is not installed
+# remove following line if xformers is not installed or when using Torch 2.0.
 pipe.enable_xformers_memory_efficient_attention()
-
+# memory optimization.
 pipe.enable_model_cpu_offload()
 
 control_image = load_image("./conditioning_image_1.png")
@@ -285,9 +285,8 @@ prompt = "pale golden rod circle with old lace background"
 # generate image
 generator = torch.manual_seed(0)
 image = pipe(
-     prompt, num_inference_steps=20, generator=generator, image=control_image
+    prompt, num_inference_steps=20, generator=generator, image=control_image
 ).images[0]
-
 image.save("./output.png")
 ```
 
@@ -460,3 +459,7 @@ The profile can then be inspected at http://localhost:6006/#profile
 Sometimes you'll get version conflicts (error messages like `Duplicate plugins for name projector`), which means that you have to uninstall and reinstall all versions of Tensorflow/Tensorboard (e.g. with `pip uninstall tensorflow tf-nightly tensorboard tb-nightly tensorboard-plugin-profile && pip install tf-nightly tbp-nightly tensorboard-plugin-profile`).
 
 Note that the debugging functionality of the Tensorboard `profile` plugin is still under active development. Not all views are fully functional, and for example the `trace_viewer` cuts off events after 1M (which can result in all your device traces getting lost if you for example profile the compilation step by accident).
+
+## Support for Stable Diffusion XL
+
+We provide a training script for training a ControlNet with [Stable Diffusion XL](https://huggingface.co/papers/2307.01952). Please refer to [README_sdxl.md](./README_sdxl.md) for more details.

examples/controlnet/README_sdxl.md

@@ -0,0 +1,131 @@
+# DreamBooth training example for Stable Diffusion XL (SDXL)
+
+The `train_controlnet_sdxl.py` script shows how to implement the training procedure and adapt it for [Stable Diffusion XL](https://huggingface.co/papers/2307.01952).
+
+## Running locally with PyTorch
+
+### Installing the dependencies
+
+Before running the scripts, make sure to install the library's training dependencies:
+
+**Important**
+
+To make sure you can successfully run the latest versions of the example scripts, we highly recommend **installing from source** and keeping the install up to date as we update the example scripts frequently and install some example-specific requirements. To do this, execute the following steps in a new virtual environment:
+
+```bash
+git clone https://github.com/huggingface/diffusers
+cd diffusers
+pip install -e .
+```
+
+Then cd in the `examples/controlnet` folder and run
+```bash
+pip install -r requirements_sdxl.txt
+```
+
+And initialize an [🤗Accelerate](https://github.com/huggingface/accelerate/) environment with:
+
+```bash
+accelerate config
+```
+
+Or for a default accelerate configuration without answering questions about your environment
+
+```bash
+accelerate config default
+```
+
+Or if your environment doesn't support an interactive shell (e.g., a notebook)
+
+```python
+from accelerate.utils import write_basic_config
+write_basic_config()
+```
+
+When running `accelerate config`, if we specify torch compile mode to True there can be dramatic speedups. 
+
+## Circle filling dataset
+
+The original dataset is hosted in the [ControlNet repo](https://huggingface.co/lllyasviel/ControlNet/blob/main/training/fill50k.zip). We re-uploaded it to be compatible with `datasets` [here](https://huggingface.co/datasets/fusing/fill50k). Note that `datasets` handles dataloading within the training script.
+
+## Training
+
+Our training examples use two test conditioning images. They can be downloaded by running
+
+```sh
+wget https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/controlnet_training/conditioning_image_1.png
+
+wget https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/controlnet_training/conditioning_image_2.png
+```
+
+Then run `huggingface-cli login` to log into your Hugging Face account. This is needed to be able to push the trained ControlNet parameters to Hugging Face Hub.
+
+```bash
+export MODEL_DIR="stabilityai/stable-diffusion-xl-base-0.9"
+export OUTPUT_DIR="path to save model"
+
+accelerate launch train_controlnet_sdxl.py \
+ --pretrained_model_name_or_path=$MODEL_DIR \
+ --output_dir=$OUTPUT_DIR \
+ --dataset_name=fusing/fill50k \
+ --mixed_precision="fp16" \
+ --resolution=1024 \
+ --learning_rate=1e-5 \
+ --max_train_steps=15000 \
+ --validation_image "./conditioning_image_1.png" "./conditioning_image_2.png" \
+ --validation_prompt "red circle with blue background" "cyan circle with brown floral background" \
+ --validation_steps=100 \
+ --train_batch_size=1 \
+ --gradient_accumulation_steps=4 \
+ --report_to="wandb" \
+ --seed=42 \
+ --push_to_hub
+```
+
+To better track our training experiments, we're using the following flags in the command above:
+
+* `report_to="wandb` will ensure the training runs are tracked on Weights and Biases. To use it, be sure to install `wandb` with `pip install wandb`.
+* `validation_image`, `validation_prompt`, and `validation_steps` to allow the script to do a few validation inference runs. This allows us to qualitatively check if the training is progressing as expected. 
+
+Our experiments were conducted on a single 40GB A100 GPU.
+
+### Inference
+
+Once training is done, we can perform inference like so:
+
+```python
+from diffusers import StableDiffusionXLControlNetPipeline, ControlNetModel, UniPCMultistepScheduler
+from diffusers.utils import load_image
+import torch
+
+base_model_path = "stabilityai/stable-diffusion-xl-base-0.9"
+controlnet_path = "path to controlnet"
+
+controlnet = ControlNetModel.from_pretrained(controlnet_path, torch_dtype=torch.float16)
+pipe = StableDiffusionXLControlNetPipeline.from_pretrained(
+    base_model_path, controlnet=controlnet, torch_dtype=torch.float16
+)
+
+# speed up diffusion process with faster scheduler and memory optimization
+pipe.scheduler = UniPCMultistepScheduler.from_config(pipe.scheduler.config)
+# remove following line if xformers is not installed or when using Torch 2.0.
+pipe.enable_xformers_memory_efficient_attention()
+# memory optimization.
+pipe.enable_model_cpu_offload()
+
+control_image = load_image("./conditioning_image_1.png")
+prompt = "pale golden rod circle with old lace background"
+
+# generate image
+generator = torch.manual_seed(0)
+image = pipe(
+    prompt, num_inference_steps=20, generator=generator, image=control_image
+).images[0]
+image.save("./output.png")
+```
+
+## Notes
+
+### Specifying a better VAE
+
+SDXL's VAE is known to suffer from numerical instability issues. This is why we also expose a CLI argument namely `--pretrained_vae_model_name_or_path` that lets you specify the location of a better VAE (such as [this one](https://huggingface.co/madebyollin/sdxl-vae-fp16-fix)).

examples/controlnet/data.py

@@ -0,0 +1,235 @@
+# coding=utf-8
+# Copyright 2023 The HuggingFace Inc. team.
+#
+# 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.
+
+# This file is heavily inspired by https://github.com/mlfoundations/open_clip/blob/main/src/training/data.py
+
+import itertools
+import json
+import math
+import random
+import re
+from typing import List, Optional, Union
+
+import webdataset as wds
+from braceexpand import braceexpand
+from torch.utils.data import default_collate
+from torchvision import transforms
+from transformers import PreTrainedTokenizer
+from webdataset.tariterators import (
+    base_plus_ext,
+    tar_file_expander,
+    url_opener,
+    valid_sample,
+)
+import numpy as np
+import cv2
+from PIL import Image
+
+
+def filter_keys(key_set):
+    def _f(dictionary):
+        return {k: v for k, v in dictionary.items() if k in key_set}
+
+    return _f
+
+
+def group_by_keys_nothrow(data, keys=base_plus_ext, lcase=True, suffixes=None, handler=None):
+    """Return function over iterator that groups key, value pairs into samples.
+
+    :param keys: function that splits the key into key and extension (base_plus_ext)
+    :param lcase: convert suffixes to lower case (Default value = True)
+    """
+    current_sample = None
+    for filesample in data:
+        assert isinstance(filesample, dict)
+        fname, value = filesample["fname"], filesample["data"]
+        prefix, suffix = keys(fname)
+        if prefix is None:
+            continue
+        if lcase:
+            suffix = suffix.lower()
+        # FIXME webdataset version throws if suffix in current_sample, but we have a potential for
+        #  this happening in the current LAION400m dataset if a tar ends with same prefix as the next
+        #  begins, rare, but can happen since prefix aren't unique across tar files in that dataset
+        if current_sample is None or prefix != current_sample["__key__"] or suffix in current_sample:
+            if valid_sample(current_sample):
+                yield current_sample
+            current_sample = dict(__key__=prefix, __url__=filesample["__url__"])
+        if suffixes is None or suffix in suffixes:
+            current_sample[suffix] = value
+    if valid_sample(current_sample):
+        yield current_sample
+
+
+def tarfile_to_samples_nothrow(src, handler=wds.warn_and_continue):
+    # NOTE this is a re-impl of the webdataset impl with group_by_keys that doesn't throw
+    streams = url_opener(src, handler=handler)
+    files = tar_file_expander(streams, handler=handler)
+    samples = group_by_keys_nothrow(files, handler=handler)
+    return samples
+
+
+def control_transform(image):
+    image = np.array(image)
+
+    low_threshold = 100
+    high_threshold = 200
+
+    image = cv2.Canny(image, low_threshold, high_threshold)
+    image = image[:, :, None]
+    image = np.concatenate([image, image, image], axis=2)
+    control_image = Image.fromarray(image)
+    return control_image
+
+
+class ImageNetTransform:
+    def __init__(self, resolution, center_crop=True, random_flip=False):
+        self.train_transform = transforms.Compose(
+            [
+                transforms.Resize(resolution, interpolation=transforms.InterpolationMode.BILINEAR),
+                transforms.CenterCrop(resolution),
+                transforms.ToTensor(),
+                transforms.Normalize([0.5], [0.5]),
+            ]
+        )
+        self.train_control_transform = transforms.Compose(
+            [
+                control_transform,
+                transforms.Resize(resolution, interpolation=transforms.InterpolationMode.BILINEAR),
+                transforms.CenterCrop(resolution),
+                transforms.ToTensor(),
+            ]
+        )
+        self.eval_transform = transforms.Compose(
+            [
+                transforms.Resize(resolution, interpolation=transforms.InterpolationMode.BILINEAR),
+                transforms.CenterCrop(resolution),
+                transforms.ToTensor(),
+            ]
+        )
+
+
+class Text2ImageDataset:
+    def __init__(
+        self,
+        train_shards_path_or_url: Union[str, List[str]],
+        eval_shards_path_or_url: Union[str, List[str]],
+        tokenizer: PreTrainedTokenizer,
+        max_seq_length: int,
+        num_train_examples: int,
+        per_gpu_batch_size: int,
+        global_batch_size: int,
+        num_workers: int,
+        tokenizer_two: Optional[PreTrainedTokenizer] = None,
+        resolution: int = 256,
+        center_crop: bool = True,
+        random_flip: bool = False,
+        shuffle_buffer_size: int = 1000,
+        pin_memory: bool = False,
+        persistent_workers: bool = False,
+    ):
+        transform = ImageNetTransform(resolution, center_crop, random_flip)
+
+        def tokenize(text):
+            input_ids = tokenizer(
+                text, max_length=max_seq_length, padding="max_length", truncation=True, return_tensors="pt"
+            ).input_ids
+            return input_ids[0]
+
+        def tokenize_2(text):
+            input_ids = tokenizer_two(
+                text, max_length=max_seq_length, padding="max_length", truncation=True, return_tensors="pt"
+            ).input_ids
+            return input_ids[0]
+
+        if not isinstance(train_shards_path_or_url, str):
+            train_shards_path_or_url = [list(braceexpand(urls)) for urls in train_shards_path_or_url]
+            # flatten list using itertools
+            train_shards_path_or_url = list(itertools.chain.from_iterable(train_shards_path_or_url))
+
+        if not isinstance(eval_shards_path_or_url, str):
+            eval_shards_path_or_url = [list(braceexpand(urls)) for urls in eval_shards_path_or_url]
+            # flatten list using itertools
+            eval_shards_path_or_url = list(itertools.chain.from_iterable(eval_shards_path_or_url))
+
+        processing_pipeline = [
+            wds.decode("pil", handler=wds.ignore_and_continue),
+            wds.rename(image="jpg;png;jpeg;webp", control_image="jpg;png;jpeg;webp", input_ids="text;txt;caption", input_ids_2="text;txt;caption", handler=wds.warn_and_continue),
+            wds.map(filter_keys(set(["image", "control_image", "input_ids", "input_ids_2"]))),
+            wds.map_dict(image=transform.train_transform, control_image=transform.train_control_transform, input_ids=tokenize, input_ids_2=tokenize_2),
+            wds.to_tuple("image", "control_image", "input_ids", "input_ids_2"),
+        ]
+
+        # Create train dataset and loader
+        pipeline = [
+            wds.ResampledShards(train_shards_path_or_url),
+            tarfile_to_samples_nothrow,
+            wds.shuffle(shuffle_buffer_size),
+            *processing_pipeline,
+            wds.batched(per_gpu_batch_size, partial=False, collation_fn=default_collate),
+        ]
+
+        num_batches = math.ceil(num_train_examples / global_batch_size)
+        num_worker_batches = math.ceil(num_train_examples / (global_batch_size * num_workers))  # per dataloader worker
+        num_batches = num_worker_batches * num_workers
+        num_samples = num_batches * global_batch_size
+
+        # each worker is iterating over this
+        self._train_dataset = wds.DataPipeline(*pipeline).with_epoch(num_worker_batches)
+        self._train_dataloader = wds.WebLoader(
+            self._train_dataset,
+            batch_size=None,
+            shuffle=False,
+            num_workers=num_workers,
+            pin_memory=pin_memory,
+            persistent_workers=persistent_workers,
+        )
+        # add meta-data to dataloader instance for convenience
+        self._train_dataloader.num_batches = num_batches
+        self._train_dataloader.num_samples = num_samples
+
+        # Create eval dataset and loader
+        pipeline = [
+            wds.SimpleShardList(eval_shards_path_or_url),
+            wds.split_by_worker,
+            wds.tarfile_to_samples(handler=wds.ignore_and_continue),
+            *processing_pipeline,
+            wds.batched(per_gpu_batch_size, partial=False, collation_fn=default_collate),
+        ]
+        self._eval_dataset = wds.DataPipeline(*pipeline)
+        self._eval_dataloader = wds.WebLoader(
+            self._eval_dataset,
+            batch_size=None,
+            shuffle=False,
+            num_workers=num_workers,
+            pin_memory=pin_memory,
+            persistent_workers=persistent_workers,
+        )
+
+    @property
+    def train_dataset(self):
+        return self._train_dataset
+
+    @property
+    def train_dataloader(self):
+        return self._train_dataloader
+
+    @property
+    def eval_dataset(self):
+        return self._eval_dataset
+
+    @property
+    def eval_dataloader(self):
+        return self._eval_dataloader

examples/controlnet/requirements_sdxl.txt

@@ -0,0 +1,9 @@
+accelerate>=0.16.0
+torchvision
+transformers>=4.25.1
+ftfy
+tensorboard
+Jinja2
+invisible-watermark>=0.2.0
+datasets
+wandb

examples/controlnet/run.sh

@@ -0,0 +1,28 @@
+export MODEL_DIR="stabilityai/stable-diffusion-xl-base-0.9"
+export OUTPUT_DIR="controlnet-0-9-canny"
+
+# --max_train_steps=15000 \
+accelerate launch train_controlnet_webdatasets.py \
+ --pretrained_model_name_or_path=$MODEL_DIR \
+ --pretrained_vae_model_name_or_path="madebyollin/sdxl-vae-fp16-fix" \
+ --output_dir=$OUTPUT_DIR \
+ --mixed_precision="fp16" \
+ --resolution=1024 \
+ --learning_rate=1e-5 \
+ --max_train_steps=30000 \
+ --max_train_samples=12000000 \
+ --dataloader_num_workers=4 \
+ --validation_image "./c_image_0.png" "./c_image_1.png" "./c_image_2.png" "./c_image_3.png" "./c_image_4.png" "./c_image_5.png" "./c_image_6.png" "./c_image_7.png" \
+ --validation_prompt "beautiful room" "two paradise birds" "a snowy house behind a forest" "a couple watching a romantic sunset" "boats in the Amazonas" "a beautiful face of a woman" "a skater in Brooklyn" "a tornado in Iowa" \
+ --train_shards_path_or_url "pipe:aws s3 cp s3://muse-datasets/laion-aesthetic6plus-data/{00000..01208}.tar -" \
+ --eval_shards_path_or_url "pipe:aws s3 cp s3://muse-datasets/laion-aesthetic6plus-data/{01209..01210}.tar -" \
+ --proportion_empty_prompts 0.5 \
+ --validation_steps=1000 \
+ --train_batch_size=12 \
+ --gradient_checkpointing \
+ --use_8bit_adam \
+ --enable_xformers_memory_efficient_attention \
+ --gradient_accumulation_steps=1 \
+ --seed=42 \
+ --report_to="wandb" \
+ --push_to_hub

examples/controlnet/train_controlnet.py

@@ -56,7 +56,7 @@ if is_wandb_available():
     import wandb
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.18.0.dev0")
+check_min_version("0.19.0.dev0")
 
 logger = get_logger(__name__)
 
@@ -897,8 +897,8 @@ def main(args):
     lr_scheduler = get_scheduler(
         args.lr_scheduler,
         optimizer=optimizer,
-        num_warmup_steps=args.lr_warmup_steps * args.gradient_accumulation_steps,
-        num_training_steps=args.max_train_steps * args.gradient_accumulation_steps,
+        num_warmup_steps=args.lr_warmup_steps * accelerator.num_processes,
+        num_training_steps=args.max_train_steps * accelerator.num_processes,
         num_cycles=args.lr_num_cycles,
         power=args.lr_power,
     )

examples/controlnet/train_controlnet_flax.py

@@ -59,7 +59,7 @@ if is_wandb_available():
     import wandb
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.18.0.dev0")
+check_min_version("0.19.0.dev0")
 
 logger = logging.getLogger(__name__)
 

examples/custom_diffusion/train_custom_diffusion.py

@@ -57,7 +57,7 @@ from diffusers.utils.import_utils import is_xformers_available
 
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.18.0.dev0")
+check_min_version("0.19.0.dev0")
 
 logger = get_logger(__name__)
 
@@ -1007,8 +1007,8 @@ def main(args):
     lr_scheduler = get_scheduler(
         args.lr_scheduler,
         optimizer=optimizer,
-        num_warmup_steps=args.lr_warmup_steps * args.gradient_accumulation_steps,
-        num_training_steps=args.max_train_steps * args.gradient_accumulation_steps,
+        num_warmup_steps=args.lr_warmup_steps * accelerator.num_processes,
+        num_training_steps=args.max_train_steps * accelerator.num_processes,
     )
 
     # Prepare everything with our `accelerator`.

examples/dreambooth/README.md

@@ -737,3 +737,7 @@ accelerate launch train_dreambooth.py \
   --class_labels_conditioning timesteps \
   --push_to_hub
 ```
+
+## Stable Diffusion XL
+
+We support fine-tuning of the UNet shipped in [Stable Diffusion XL](https://huggingface.co/papers/2307.01952) with DreamBooth and LoRA via the `train_dreambooth_lora_sdxl.py` script. Please refer to the docs [here](./README_sdxl.md). 

examples/dreambooth/README_sdxl.md

@@ -0,0 +1,178 @@
+# DreamBooth training example for Stable Diffusion XL (SDXL)
+
+[DreamBooth](https://arxiv.org/abs/2208.12242) is a method to personalize text2image models like stable diffusion given just a few (3~5) images of a subject.
+
+The `train_dreambooth_lora_sdxl.py` script shows how to implement the training procedure and adapt it for [Stable Diffusion XL](https://huggingface.co/papers/2307.01952).
+
+> 💡 **Note**: For now, we only allow DreamBooth fine-tuning of the SDXL UNet via LoRA. LoRA is a parameter-efficient fine-tuning technique introduced in [LoRA: Low-Rank Adaptation of Large Language Models](https://arxiv.org/abs/2106.09685) by *Edward J. Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, Weizhu Chen*. 
+
+## Running locally with PyTorch
+
+### Installing the dependencies
+
+Before running the scripts, make sure to install the library's training dependencies:
+
+**Important**
+
+To make sure you can successfully run the latest versions of the example scripts, we highly recommend **installing from source** and keeping the install up to date as we update the example scripts frequently and install some example-specific requirements. To do this, execute the following steps in a new virtual environment:
+
+```bash
+git clone https://github.com/huggingface/diffusers
+cd diffusers
+pip install -e .
+```
+
+Then cd in the `examples/dreambooth` folder and run
+```bash
+pip install -r requirements_sdxl.txt
+```
+
+And initialize an [🤗Accelerate](https://github.com/huggingface/accelerate/) environment with:
+
+```bash
+accelerate config
+```
+
+Or for a default accelerate configuration without answering questions about your environment
+
+```bash
+accelerate config default
+```
+
+Or if your environment doesn't support an interactive shell (e.g., a notebook)
+
+```python
+from accelerate.utils import write_basic_config
+write_basic_config()
+```
+
+When running `accelerate config`, if we specify torch compile mode to True there can be dramatic speedups. 
+
+### Dog toy example
+
+Now let's get our dataset. For this example we will use some dog images: https://huggingface.co/datasets/diffusers/dog-example.
+
+Let's first download it locally:
+
+```python
+from huggingface_hub import snapshot_download
+
+local_dir = "./dog"
+snapshot_download(
+    "diffusers/dog-example",
+    local_dir=local_dir, repo_type="dataset",
+    ignore_patterns=".gitattributes",
+)
+```
+
+Since SDXL 0.9 weights are gated, we need to be authenticated to be able to use them. So, let's run:
+
+```bash
+huggingface-cli login
+```
+
+This will also allow us to push the trained LoRA parameters to the Hugging Face Hub platform. 
+
+Now, we can launch training using:
+
+```bash
+export MODEL_NAME="diffusers/stable-diffusion-xl-base-0.9"
+export INSTANCE_DIR="dog"
+export OUTPUT_DIR="lora-trained-xl"
+
+accelerate launch train_dreambooth_lora_sdxl.py \
+  --pretrained_model_name_or_path=$MODEL_NAME  \
+  --instance_data_dir=$INSTANCE_DIR \
+  --output_dir=$OUTPUT_DIR \
+  --mixed_precision="fp16" \
+  --instance_prompt="a photo of sks dog" \
+  --resolution=1024 \
+  --train_batch_size=1 \
+  --gradient_accumulation_steps=4 \
+  --learning_rate=1e-4 \
+  --report_to="wandb" \
+  --lr_scheduler="constant" \
+  --lr_warmup_steps=0 \
+  --max_train_steps=500 \
+  --validation_prompt="A photo of sks dog in a bucket" \
+  --validation_epochs=25 \
+  --seed="0" \
+  --push_to_hub
+```
+
+To better track our training experiments, we're using the following flags in the command above:
+
+* `report_to="wandb` will ensure the training runs are tracked on Weights and Biases. To use it, be sure to install `wandb` with `pip install wandb`.
+* `validation_prompt` and `validation_epochs` to allow the script to do a few validation inference runs. This allows us to qualitatively check if the training is progressing as expected. 
+
+Our experiments were conducted on a single 40GB A100 GPU.
+
+### Inference
+
+Once training is done, we can perform inference like so:
+
+```python
+from huggingface_hub.repocard import RepoCard
+from diffusers import DiffusionPipeline
+import torch
+
+lora_model_id = <"lora-sdxl-dreambooth-id">
+card = RepoCard.load(lora_model_id)
+base_model_id = card.data.to_dict()["base_model"]
+
+pipe = DiffusionPipeline.from_pretrained(base_model_id, torch_dtype=torch.float16)
+pipe = pipe.to("cuda")
+pipe.load_lora_weights(lora_model_id)
+image = pipe("A picture of a sks dog in a bucket", num_inference_steps=25).images[0]
+image.save("sks_dog.png")
+```
+
+We can further refine the outputs with the [Refiner](https://huggingface.co/stabilityai/stable-diffusion-xl-refiner-0.9):
+
+```python
+from huggingface_hub.repocard import RepoCard
+from diffusers import DiffusionPipeline, StableDiffusionXLImg2ImgPipeline
+import torch
+
+lora_model_id = <"lora-sdxl-dreambooth-id">
+card = RepoCard.load(lora_model_id)
+base_model_id = card.data.to_dict()["base_model"]
+
+# Load the base pipeline and load the LoRA parameters into it. 
+pipe = DiffusionPipeline.from_pretrained(base_model_id, torch_dtype=torch.float16)
+pipe = pipe.to("cuda")
+pipe.load_lora_weights(lora_model_id)
+
+# Load the refiner.
+refiner = StableDiffusionXLImg2ImgPipeline.from_pretrained(
+    "stabilityai/stable-diffusion-xl-refiner-0.9", torch_dtype=torch.float16, use_safetensors=True, variant="fp16"
+)
+refiner.to("cuda")
+
+prompt = "A picture of a sks dog in a bucket"
+generator = torch.Generator("cuda").manual_seed(0)
+
+# Run inference.
+image = pipe(prompt=prompt, output_type="latent", generator=generator).images[0]
+image = refiner(prompt=prompt, image=image[None, :], generator=generator).images[0]
+image.save("refined_sks_dog.png")
+```
+
+Here's a side-by-side comparison of the with and without Refiner pipeline outputs:
+
+| Without Refiner | With Refiner |
+|---|---|
+| ![](https://huggingface.co/datasets/diffusers/docs-images/resolve/main/sd_xl/sks_dog.png) | ![](https://huggingface.co/datasets/diffusers/docs-images/resolve/main/sd_xl/refined_sks_dog.png) |
+
+## Notes
+
+In our experiments we found that SDXL yields very good initial results using the default settings of the script. We didn't explore further hyper-parameter tuning experiments, but we do encourage the community to explore this avenue further and share their results with us 🤗
+
+## Results
+
+You can explore the results from a couple of our internal experiments by checking out this link: [https://wandb.ai/sayakpaul/dreambooth-lora-sd-xl](https://wandb.ai/sayakpaul/dreambooth-lora-sd-xl). Specifically, we used the same script with the exact same hyperparameters on the following datasets:
+
+* [Dogs](https://huggingface.co/datasets/diffusers/dog-example)
+* [Starbucks logo](https://huggingface.co/datasets/diffusers/starbucks-example)
+* [Mr. Potato Head](https://huggingface.co/datasets/diffusers/potato-head-example)
+* [Keramer face](https://huggingface.co/datasets/diffusers/keramer-face-example)

examples/dreambooth/requirements_sdxl.txt

@@ -0,0 +1,7 @@
+accelerate>=0.16.0
+torchvision
+transformers>=4.25.1
+ftfy
+tensorboard
+Jinja2
+invisible-watermark>=0.2.0

examples/dreambooth/train_dreambooth.py

@@ -59,7 +59,7 @@ if is_wandb_available():
     import wandb
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.18.0.dev0")
+check_min_version("0.19.0.dev0")
 
 logger = get_logger(__name__)
 
@@ -1075,8 +1075,8 @@ def main(args):
     lr_scheduler = get_scheduler(
         args.lr_scheduler,
         optimizer=optimizer,
-        num_warmup_steps=args.lr_warmup_steps * args.gradient_accumulation_steps,
-        num_training_steps=args.max_train_steps * args.gradient_accumulation_steps,
+        num_warmup_steps=args.lr_warmup_steps * accelerator.num_processes,
+        num_training_steps=args.max_train_steps * accelerator.num_processes,
         num_cycles=args.lr_num_cycles,
         power=args.lr_power,
     )

examples/dreambooth/train_dreambooth_flax.py

@@ -36,7 +36,7 @@ from diffusers.utils import check_min_version
 
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.18.0.dev0")
+check_min_version("0.19.0.dev0")
 
 # Cache compiled models across invocations of this script.
 cc.initialize_cache(os.path.expanduser("~/.cache/jax/compilation_cache"))

examples/dreambooth/train_dreambooth_lora.py

@@ -23,6 +23,7 @@ import os
 import shutil
 import warnings
 from pathlib import Path
+from typing import Dict
 
 import numpy as np
 import torch
@@ -50,7 +51,10 @@ from diffusers import (
     StableDiffusionPipeline,
     UNet2DConditionModel,
 )
-from diffusers.loaders import AttnProcsLayers, LoraLoaderMixin
+from diffusers.loaders import (
+    LoraLoaderMixin,
+    text_encoder_lora_state_dict,
+)
 from diffusers.models.attention_processor import (
     AttnAddedKVProcessor,
     AttnAddedKVProcessor2_0,
@@ -60,12 +64,12 @@ from diffusers.models.attention_processor import (
     SlicedAttnAddedKVProcessor,
 )
 from diffusers.optimization import get_scheduler
-from diffusers.utils import TEXT_ENCODER_ATTN_MODULE, check_min_version, is_wandb_available
+from diffusers.utils import check_min_version, is_wandb_available
 from diffusers.utils.import_utils import is_xformers_available
 
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.18.0.dev0")
+check_min_version("0.19.0.dev0")
 
 logger = get_logger(__name__)
 
@@ -653,6 +657,22 @@ def encode_prompt(text_encoder, input_ids, attention_mask, text_encoder_use_atte
     return prompt_embeds
 
 
+def unet_attn_processors_state_dict(unet) -> Dict[str, torch.tensor]:
+    r"""
+    Returns:
+        a state dict containing just the attention processor parameters.
+    """
+    attn_processors = unet.attn_processors
+
+    attn_processors_state_dict = {}
+
+    for attn_processor_key, attn_processor in attn_processors.items():
+        for parameter_key, parameter in attn_processor.state_dict().items():
+            attn_processors_state_dict[f"{attn_processor_key}.{parameter_key}"] = parameter
+
+    return attn_processors_state_dict
+
+
 def main(args):
     logging_dir = Path(args.output_dir, args.logging_dir)
 
@@ -833,6 +853,7 @@ def main(args):
 
     # Set correct lora layers
     unet_lora_attn_procs = {}
+    unet_lora_parameters = []
     for name, attn_processor in unet.attn_processors.items():
         cross_attention_dim = None if name.endswith("attn1.processor") else unet.config.cross_attention_dim
         if name.startswith("mid_block"):
@@ -850,35 +871,20 @@ def main(args):
             lora_attn_processor_class = (
                 LoRAAttnProcessor2_0 if hasattr(F, "scaled_dot_product_attention") else LoRAAttnProcessor
             )
-        unet_lora_attn_procs[name] = lora_attn_processor_class(
-            hidden_size=hidden_size,
-            cross_attention_dim=cross_attention_dim,
-            rank=args.rank,
+
+        module = lora_attn_processor_class(
+            hidden_size=hidden_size, cross_attention_dim=cross_attention_dim, rank=args.rank
         )
+        unet_lora_attn_procs[name] = module
+        unet_lora_parameters.extend(module.parameters())
 
     unet.set_attn_processor(unet_lora_attn_procs)
-    unet_lora_layers = AttnProcsLayers(unet.attn_processors)
 
     # The text encoder comes from 🤗 transformers, so we cannot directly modify it.
-    # So, instead, we monkey-patch the forward calls of its attention-blocks. For this,
-    # we first load a dummy pipeline with the text encoder and then do the monkey-patching.
-    text_encoder_lora_layers = None
+    # So, instead, we monkey-patch the forward calls of its attention-blocks.
     if args.train_text_encoder:
-        text_lora_attn_procs = {}
-        for name, module in text_encoder.named_modules():
-            if name.endswith(TEXT_ENCODER_ATTN_MODULE):
-                text_lora_attn_procs[name] = LoRAAttnProcessor(
-                    hidden_size=module.out_proj.out_features,
-                    cross_attention_dim=None,
-                    rank=args.rank,
-                )
-        text_encoder_lora_layers = AttnProcsLayers(text_lora_attn_procs)
-        temp_pipeline = DiffusionPipeline.from_pretrained(
-            args.pretrained_model_name_or_path, text_encoder=text_encoder
-        )
-        temp_pipeline._modify_text_encoder(text_lora_attn_procs)
-        text_encoder = temp_pipeline.text_encoder
-        del temp_pipeline
+        # ensure that dtype is float32, even if rest of the model that isn't trained is loaded in fp16
+        text_lora_parameters = LoraLoaderMixin._modify_text_encoder(text_encoder, dtype=torch.float32, rank=args.rank)
 
     # create custom saving & loading hooks so that `accelerator.save_state(...)` serializes in a nice format
     def save_model_hook(models, weights, output_dir):
@@ -887,23 +893,13 @@ def main(args):
         unet_lora_layers_to_save = None
         text_encoder_lora_layers_to_save = None
 
-        if args.train_text_encoder:
-            text_encoder_keys = accelerator.unwrap_model(text_encoder_lora_layers).state_dict().keys()
-        unet_keys = accelerator.unwrap_model(unet_lora_layers).state_dict().keys()
-
         for model in models:
-            state_dict = model.state_dict()
-
-            if (
-                text_encoder_lora_layers is not None
-                and text_encoder_keys is not None
-                and state_dict.keys() == text_encoder_keys
-            ):
-                # text encoder
-                text_encoder_lora_layers_to_save = state_dict
-            elif state_dict.keys() == unet_keys:
-                # unet
-                unet_lora_layers_to_save = state_dict
+            if isinstance(model, type(accelerator.unwrap_model(unet))):
+                unet_lora_layers_to_save = unet_attn_processors_state_dict(model)
+            elif isinstance(model, type(accelerator.unwrap_model(text_encoder))):
+                text_encoder_lora_layers_to_save = text_encoder_lora_state_dict(model)
+            else:
+                raise ValueError(f"unexpected save model: {model.__class__}")
 
             # make sure to pop weight so that corresponding model is not saved again
             weights.pop()
@@ -915,27 +911,24 @@ def main(args):
         )
 
     def load_model_hook(models, input_dir):
-        # Note we DON'T pass the unet and text encoder here an purpose
-        # so that the we don't accidentally override the LoRA layers of
-        # unet_lora_layers and text_encoder_lora_layers which are stored in `models`
-        # with new torch.nn.Modules / weights. We simply use the pipeline class as
-        # an easy way to load the lora checkpoints
-        temp_pipeline = DiffusionPipeline.from_pretrained(
-            args.pretrained_model_name_or_path,
-            revision=args.revision,
-            torch_dtype=weight_dtype,
+        unet_ = None
+        text_encoder_ = None
+
+        while len(models) > 0:
+            model = models.pop()
+
+            if isinstance(model, type(accelerator.unwrap_model(unet))):
+                unet_ = model
+            elif isinstance(model, type(accelerator.unwrap_model(text_encoder))):
+                text_encoder_ = model
+            else:
+                raise ValueError(f"unexpected save model: {model.__class__}")
+
+        lora_state_dict, network_alpha = LoraLoaderMixin.lora_state_dict(input_dir)
+        LoraLoaderMixin.load_lora_into_unet(lora_state_dict, network_alpha=network_alpha, unet=unet_)
+        LoraLoaderMixin.load_lora_into_text_encoder(
+            lora_state_dict, network_alpha=network_alpha, text_encoder=text_encoder_
         )
-        temp_pipeline.load_lora_weights(input_dir)
-
-        # load lora weights into models
-        models[0].load_state_dict(AttnProcsLayers(temp_pipeline.unet.attn_processors).state_dict())
-        if len(models) > 1:
-            models[1].load_state_dict(AttnProcsLayers(temp_pipeline.text_encoder_lora_attn_procs).state_dict())
-
-        # delete temporary pipeline and pop models
-        del temp_pipeline
-        for _ in range(len(models)):
-            models.pop()
 
     accelerator.register_save_state_pre_hook(save_model_hook)
     accelerator.register_load_state_pre_hook(load_model_hook)
@@ -965,9 +958,9 @@ def main(args):
 
     # Optimizer creation
     params_to_optimize = (
-        itertools.chain(unet_lora_layers.parameters(), text_encoder_lora_layers.parameters())
+        itertools.chain(unet_lora_parameters, text_lora_parameters)
         if args.train_text_encoder
-        else unet_lora_layers.parameters()
+        else unet_lora_parameters
     )
     optimizer = optimizer_class(
         params_to_optimize,
@@ -1048,20 +1041,20 @@ def main(args):
     lr_scheduler = get_scheduler(
         args.lr_scheduler,
         optimizer=optimizer,
-        num_warmup_steps=args.lr_warmup_steps * args.gradient_accumulation_steps,
-        num_training_steps=args.max_train_steps * args.gradient_accumulation_steps,
+        num_warmup_steps=args.lr_warmup_steps * accelerator.num_processes,
+        num_training_steps=args.max_train_steps * accelerator.num_processes,
         num_cycles=args.lr_num_cycles,
         power=args.lr_power,
     )
 
     # Prepare everything with our `accelerator`.
     if args.train_text_encoder:
-        unet_lora_layers, text_encoder_lora_layers, optimizer, train_dataloader, lr_scheduler = accelerator.prepare(
-            unet_lora_layers, text_encoder_lora_layers, optimizer, train_dataloader, lr_scheduler
+        unet, text_encoder, optimizer, train_dataloader, lr_scheduler = accelerator.prepare(
+            unet, text_encoder, optimizer, train_dataloader, lr_scheduler
         )
     else:
-        unet_lora_layers, optimizer, train_dataloader, lr_scheduler = accelerator.prepare(
-            unet_lora_layers, optimizer, train_dataloader, lr_scheduler
+        unet, optimizer, train_dataloader, lr_scheduler = accelerator.prepare(
+            unet, optimizer, train_dataloader, lr_scheduler
         )
 
     # We need to recalculate our total training steps as the size of the training dataloader may have changed.
@@ -1210,9 +1203,9 @@ def main(args):
                 accelerator.backward(loss)
                 if accelerator.sync_gradients:
                     params_to_clip = (
-                        itertools.chain(unet_lora_layers.parameters(), text_encoder_lora_layers.parameters())
+                        itertools.chain(unet_lora_parameters, text_lora_parameters)
                         if args.train_text_encoder
-                        else unet_lora_layers.parameters()
+                        else unet_lora_parameters
                     )
                     accelerator.clip_grad_norm_(params_to_clip, args.max_grad_norm)
                 optimizer.step()
@@ -1301,15 +1294,17 @@ def main(args):
                     pipeline_args = {"prompt": args.validation_prompt}
 
                 if args.validation_images is None:
-                    images = [
-                        pipeline(**pipeline_args, generator=generator).images[0]
-                        for _ in range(args.num_validation_images)
-                    ]
+                    images = []
+                    for _ in range(args.num_validation_images):
+                        with torch.cuda.amp.autocast():
+                            image = pipeline(**pipeline_args, generator=generator).images[0]
+                            images.append(image)
                 else:
                     images = []
                     for image in args.validation_images:
                         image = Image.open(image)
-                        image = pipeline(**pipeline_args, image=image, generator=generator).images[0]
+                        with torch.cuda.amp.autocast():
+                            image = pipeline(**pipeline_args, image=image, generator=generator).images[0]
                         images.append(image)
 
                 for tracker in accelerator.trackers:
@@ -1332,12 +1327,16 @@ def main(args):
     # Save the lora layers
     accelerator.wait_for_everyone()
     if accelerator.is_main_process:
+        unet = accelerator.unwrap_model(unet)
         unet = unet.to(torch.float32)
-        unet_lora_layers = accelerator.unwrap_model(unet_lora_layers)
+        unet_lora_layers = unet_attn_processors_state_dict(unet)
 
-        if text_encoder is not None:
+        if text_encoder is not None and args.train_text_encoder:
+            text_encoder = accelerator.unwrap_model(text_encoder)
             text_encoder = text_encoder.to(torch.float32)
-            text_encoder_lora_layers = accelerator.unwrap_model(text_encoder_lora_layers)
+            text_encoder_lora_layers = text_encoder_lora_state_dict(text_encoder)
+        else:
+            text_encoder_lora_layers = None
 
         LoraLoaderMixin.save_lora_weights(
             save_directory=args.output_dir,
@@ -1367,7 +1366,7 @@ def main(args):
         pipeline = pipeline.to(accelerator.device)
 
         # load attention processors
-        pipeline.load_lora_weights(args.output_dir)
+        pipeline.load_lora_weights(args.output_dir, weight_name="pytorch_lora_weights.bin")
 
         # run inference
         images = []

examples/instruct_pix2pix/train_instruct_pix2pix.py

@@ -52,7 +52,7 @@ from diffusers.utils.import_utils import is_xformers_available
 
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.18.0.dev0")
+check_min_version("0.19.0.dev0")
 
 logger = get_logger(__name__, log_level="INFO")
 
@@ -690,8 +690,8 @@ def main():
     lr_scheduler = get_scheduler(
         args.lr_scheduler,
         optimizer=optimizer,
-        num_warmup_steps=args.lr_warmup_steps * args.gradient_accumulation_steps,
-        num_training_steps=args.max_train_steps * args.gradient_accumulation_steps,
+        num_warmup_steps=args.lr_warmup_steps * accelerator.num_processes,
+        num_training_steps=args.max_train_steps * accelerator.num_processes,
     )
 
     # Prepare everything with our `accelerator`.

examples/research_projects/dreambooth_inpaint/train_dreambooth_inpaint.py

@@ -600,8 +600,8 @@ def main():
     lr_scheduler = get_scheduler(
         args.lr_scheduler,
         optimizer=optimizer,
-        num_warmup_steps=args.lr_warmup_steps * args.gradient_accumulation_steps,
-        num_training_steps=args.max_train_steps * args.gradient_accumulation_steps,
+        num_warmup_steps=args.lr_warmup_steps * accelerator.num_processes,
+        num_training_steps=args.max_train_steps * accelerator.num_processes,
     )
 
     if args.train_text_encoder:

examples/research_projects/dreambooth_inpaint/train_dreambooth_inpaint_lora.py

@@ -644,8 +644,8 @@ def main():
     lr_scheduler = get_scheduler(
         args.lr_scheduler,
         optimizer=optimizer,
-        num_warmup_steps=args.lr_warmup_steps * args.gradient_accumulation_steps,
-        num_training_steps=args.max_train_steps * args.gradient_accumulation_steps,
+        num_warmup_steps=args.lr_warmup_steps * accelerator.num_processes,
+        num_training_steps=args.max_train_steps * accelerator.num_processes,
     )
 
     # Prepare everything with our `accelerator`.

examples/research_projects/intel_opts/textual_inversion/textual_inversion_bf16.py

@@ -481,8 +481,8 @@ def main():
     lr_scheduler = get_scheduler(
         args.lr_scheduler,
         optimizer=optimizer,
-        num_warmup_steps=args.lr_warmup_steps * args.gradient_accumulation_steps,
-        num_training_steps=args.max_train_steps * args.gradient_accumulation_steps,
+        num_warmup_steps=args.lr_warmup_steps * accelerator.num_processes,
+        num_training_steps=args.max_train_steps * accelerator.num_processes,
     )
 
     text_encoder, optimizer, train_dataloader, lr_scheduler = accelerator.prepare(

examples/research_projects/intel_opts/textual_inversion_dfq/textual_inversion.py

@@ -588,8 +588,8 @@ def main():
     lr_scheduler = get_scheduler(
         args.lr_scheduler,
         optimizer=optimizer,
-        num_warmup_steps=args.lr_warmup_steps * args.gradient_accumulation_steps,
-        num_training_steps=args.max_train_steps * args.gradient_accumulation_steps,
+        num_warmup_steps=args.lr_warmup_steps * accelerator.num_processes,
+        num_training_steps=args.max_train_steps * accelerator.num_processes,
     )
 
     if not train_unet:

examples/research_projects/lora/train_text_to_image_lora.py

@@ -701,8 +701,8 @@ def main():
     lr_scheduler = get_scheduler(
         args.lr_scheduler,
         optimizer=optimizer,
-        num_warmup_steps=args.lr_warmup_steps * args.gradient_accumulation_steps,
-        num_training_steps=args.max_train_steps * args.gradient_accumulation_steps,
+        num_warmup_steps=args.lr_warmup_steps * accelerator.num_processes,
+        num_training_steps=args.max_train_steps * accelerator.num_processes,
     )
 
     # Prepare everything with our `accelerator`.

examples/research_projects/mulit_token_textual_inversion/textual_inversion.py

@@ -690,8 +690,8 @@ def main():
     lr_scheduler = get_scheduler(
         args.lr_scheduler,
         optimizer=optimizer,
-        num_warmup_steps=args.lr_warmup_steps * args.gradient_accumulation_steps,
-        num_training_steps=args.max_train_steps * args.gradient_accumulation_steps,
+        num_warmup_steps=args.lr_warmup_steps * accelerator.num_processes,
+        num_training_steps=args.max_train_steps * accelerator.num_processes,
     )
 
     # Prepare everything with our `accelerator`.

examples/research_projects/multi_subject_dreambooth/train_multi_subject_dreambooth.py

@@ -970,8 +970,8 @@ def main(args):
     lr_scheduler = get_scheduler(
         args.lr_scheduler,
         optimizer=optimizer,
-        num_warmup_steps=args.lr_warmup_steps * args.gradient_accumulation_steps,
-        num_training_steps=args.max_train_steps * args.gradient_accumulation_steps,
+        num_warmup_steps=args.lr_warmup_steps * accelerator.num_processes,
+        num_training_steps=args.max_train_steps * accelerator.num_processes,
         num_cycles=args.lr_num_cycles,
         power=args.lr_power,
     )

examples/research_projects/onnxruntime/text_to_image/train_text_to_image.py

@@ -732,8 +732,8 @@ def main():
     lr_scheduler = get_scheduler(
         args.lr_scheduler,
         optimizer=optimizer,
-        num_warmup_steps=args.lr_warmup_steps * args.gradient_accumulation_steps,
-        num_training_steps=args.max_train_steps * args.gradient_accumulation_steps,
+        num_warmup_steps=args.lr_warmup_steps * accelerator.num_processes,
+        num_training_steps=args.max_train_steps * accelerator.num_processes,
     )
 
     # Prepare everything with our `accelerator`.

examples/research_projects/onnxruntime/textual_inversion/textual_inversion.py

@@ -741,8 +741,8 @@ def main():
     lr_scheduler = get_scheduler(
         args.lr_scheduler,
         optimizer=optimizer,
-        num_warmup_steps=args.lr_warmup_steps * args.gradient_accumulation_steps,
-        num_training_steps=args.max_train_steps * args.gradient_accumulation_steps,
+        num_warmup_steps=args.lr_warmup_steps * accelerator.num_processes,
+        num_training_steps=args.max_train_steps * accelerator.num_processes,
     )
 
     # Prepare everything with our `accelerator`.

examples/test_examples.py

@@ -353,6 +353,38 @@ class ExamplesTestsAccelerate(unittest.TestCase):
             starts_with_unet = all(key.startswith("unet") for key in lora_state_dict.keys())
             self.assertTrue(starts_with_unet)
 
+    def test_dreambooth_lora_sdxl(self):
+        with tempfile.TemporaryDirectory() as tmpdir:
+            test_args = f"""
+                examples/dreambooth/train_dreambooth_lora_sdxl.py
+                --pretrained_model_name_or_path hf-internal-testing/tiny-stable-diffusion-xl-pipe
+                --instance_data_dir docs/source/en/imgs
+                --instance_prompt photo
+                --resolution 64
+                --train_batch_size 1
+                --gradient_accumulation_steps 1
+                --max_train_steps 2
+                --learning_rate 5.0e-04
+                --scale_lr
+                --lr_scheduler constant
+                --lr_warmup_steps 0
+                --output_dir {tmpdir}
+                """.split()
+
+            run_command(self._launch_args + test_args)
+            # save_pretrained smoke test
+            self.assertTrue(os.path.isfile(os.path.join(tmpdir, "pytorch_lora_weights.bin")))
+
+            # make sure the state_dict has the correct naming in the parameters.
+            lora_state_dict = torch.load(os.path.join(tmpdir, "pytorch_lora_weights.bin"))
+            is_lora = all("lora" in k for k in lora_state_dict.keys())
+            self.assertTrue(is_lora)
+
+            # when not training the text encoder, all the parameters in the state dict should start
+            # with `"unet"` in their names.
+            starts_with_unet = all(key.startswith("unet") for key in lora_state_dict.keys())
+            self.assertTrue(starts_with_unet)
+
     def test_custom_diffusion(self):
         with tempfile.TemporaryDirectory() as tmpdir:
             test_args = f"""

examples/text_to_image/train_text_to_image.py

@@ -54,7 +54,7 @@ if is_wandb_available():
 
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.18.0.dev0")
+check_min_version("0.19.0.dev0")
 
 logger = get_logger(__name__, log_level="INFO")
 
@@ -819,8 +819,8 @@ def main():
     lr_scheduler = get_scheduler(
         args.lr_scheduler,
         optimizer=optimizer,
-        num_warmup_steps=args.lr_warmup_steps * args.gradient_accumulation_steps,
-        num_training_steps=args.max_train_steps * args.gradient_accumulation_steps,
+        num_warmup_steps=args.lr_warmup_steps * accelerator.num_processes,
+        num_training_steps=args.max_train_steps * accelerator.num_processes,
     )
 
     # Prepare everything with our `accelerator`.

examples/text_to_image/train_text_to_image_flax.py

@@ -33,7 +33,7 @@ from diffusers.utils import check_min_version
 
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.18.0.dev0")
+check_min_version("0.19.0.dev0")
 
 logger = logging.getLogger(__name__)
 

examples/text_to_image/train_text_to_image_lora.py

@@ -48,7 +48,7 @@ from diffusers.utils.import_utils import is_xformers_available
 
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.18.0.dev0")
+check_min_version("0.19.0.dev0")
 
 logger = get_logger(__name__, log_level="INFO")
 
@@ -662,8 +662,8 @@ def main():
     lr_scheduler = get_scheduler(
         args.lr_scheduler,
         optimizer=optimizer,
-        num_warmup_steps=args.lr_warmup_steps * args.gradient_accumulation_steps,
-        num_training_steps=args.max_train_steps * args.gradient_accumulation_steps,
+        num_warmup_steps=args.lr_warmup_steps * accelerator.num_processes,
+        num_training_steps=args.max_train_steps * accelerator.num_processes,
     )
 
     # Prepare everything with our `accelerator`.

examples/textual_inversion/textual_inversion.py

@@ -78,7 +78,7 @@ else:
 
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.18.0.dev0")
+check_min_version("0.19.0.dev0")
 
 logger = get_logger(__name__)
 
@@ -737,9 +737,9 @@ def main():
     lr_scheduler = get_scheduler(
         args.lr_scheduler,
         optimizer=optimizer,
-        num_warmup_steps=args.lr_warmup_steps * args.gradient_accumulation_steps,
-        num_training_steps=args.max_train_steps * args.gradient_accumulation_steps,
-        num_cycles=args.lr_num_cycles * args.gradient_accumulation_steps,
+        num_warmup_steps=args.lr_warmup_steps * accelerator.num_processes,
+        num_training_steps=args.max_train_steps * accelerator.num_processes,
+        num_cycles=args.lr_num_cycles,
     )
 
     # Prepare everything with our `accelerator`.

examples/textual_inversion/textual_inversion_flax.py

@@ -56,7 +56,7 @@ else:
 # ------------------------------------------------------------------------------
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.18.0.dev0")
+check_min_version("0.19.0.dev0")
 
 logger = logging.getLogger(__name__)
 

examples/unconditional_image_generation/train_unconditional.py

@@ -4,6 +4,7 @@ import logging
 import math
 import os
 import shutil
+from datetime import timedelta
 from pathlib import Path
 from typing import Optional
 
@@ -11,7 +12,7 @@ import accelerate
 import datasets
 import torch
 import torch.nn.functional as F
-from accelerate import Accelerator
+from accelerate import Accelerator, InitProcessGroupKwargs
 from accelerate.logging import get_logger
 from accelerate.utils import ProjectConfiguration
 from datasets import load_dataset
@@ -29,7 +30,7 @@ from diffusers.utils.import_utils import is_xformers_available
 
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.18.0.dev0")
+check_min_version("0.19.0.dev0")
 
 logger = get_logger(__name__, log_level="INFO")
 
@@ -286,11 +287,13 @@ def main(args):
     logging_dir = os.path.join(args.output_dir, args.logging_dir)
     accelerator_project_config = ProjectConfiguration(project_dir=args.output_dir, logging_dir=logging_dir)
 
+    kwargs = InitProcessGroupKwargs(timeout=timedelta(seconds=7200))  # a big number for high resolution or big dataset
     accelerator = Accelerator(
         gradient_accumulation_steps=args.gradient_accumulation_steps,
         mixed_precision=args.mixed_precision,
         log_with=args.logger,
         project_config=accelerator_project_config,
+        kwargs_handlers=[kwargs],
     )
 
     if args.logger == "tensorboard":

scripts/convert_asymmetric_vqgan_to_diffusers.py

@@ -0,0 +1,184 @@
+import argparse
+import time
+from pathlib import Path
+from typing import Any, Dict, Literal
+
+import torch
+
+from diffusers import AsymmetricAutoencoderKL
+
+
+ASYMMETRIC_AUTOENCODER_KL_x_1_5_CONFIG = {
+    "in_channels": 3,
+    "out_channels": 3,
+    "down_block_types": [
+        "DownEncoderBlock2D",
+        "DownEncoderBlock2D",
+        "DownEncoderBlock2D",
+        "DownEncoderBlock2D",
+    ],
+    "down_block_out_channels": [128, 256, 512, 512],
+    "layers_per_down_block": 2,
+    "up_block_types": [
+        "UpDecoderBlock2D",
+        "UpDecoderBlock2D",
+        "UpDecoderBlock2D",
+        "UpDecoderBlock2D",
+    ],
+    "up_block_out_channels": [192, 384, 768, 768],
+    "layers_per_up_block": 3,
+    "act_fn": "silu",
+    "latent_channels": 4,
+    "norm_num_groups": 32,
+    "sample_size": 256,
+    "scaling_factor": 0.18215,
+}
+
+ASYMMETRIC_AUTOENCODER_KL_x_2_CONFIG = {
+    "in_channels": 3,
+    "out_channels": 3,
+    "down_block_types": [
+        "DownEncoderBlock2D",
+        "DownEncoderBlock2D",
+        "DownEncoderBlock2D",
+        "DownEncoderBlock2D",
+    ],
+    "down_block_out_channels": [128, 256, 512, 512],
+    "layers_per_down_block": 2,
+    "up_block_types": [
+        "UpDecoderBlock2D",
+        "UpDecoderBlock2D",
+        "UpDecoderBlock2D",
+        "UpDecoderBlock2D",
+    ],
+    "up_block_out_channels": [256, 512, 1024, 1024],
+    "layers_per_up_block": 5,
+    "act_fn": "silu",
+    "latent_channels": 4,
+    "norm_num_groups": 32,
+    "sample_size": 256,
+    "scaling_factor": 0.18215,
+}
+
+
+def convert_asymmetric_autoencoder_kl_state_dict(original_state_dict: Dict[str, Any]) -> Dict[str, Any]:
+    converted_state_dict = {}
+    for k, v in original_state_dict.items():
+        if k.startswith("encoder."):
+            converted_state_dict[
+                k.replace("encoder.down.", "encoder.down_blocks.")
+                .replace("encoder.mid.", "encoder.mid_block.")
+                .replace("encoder.norm_out.", "encoder.conv_norm_out.")
+                .replace(".downsample.", ".downsamplers.0.")
+                .replace(".nin_shortcut.", ".conv_shortcut.")
+                .replace(".block.", ".resnets.")
+                .replace(".block_1.", ".resnets.0.")
+                .replace(".block_2.", ".resnets.1.")
+                .replace(".attn_1.k.", ".attentions.0.to_k.")
+                .replace(".attn_1.q.", ".attentions.0.to_q.")
+                .replace(".attn_1.v.", ".attentions.0.to_v.")
+                .replace(".attn_1.proj_out.", ".attentions.0.to_out.0.")
+                .replace(".attn_1.norm.", ".attentions.0.group_norm.")
+            ] = v
+        elif k.startswith("decoder.") and "up_layers" not in k:
+            converted_state_dict[
+                k.replace("decoder.encoder.", "decoder.condition_encoder.")
+                .replace(".norm_out.", ".conv_norm_out.")
+                .replace(".up.0.", ".up_blocks.3.")
+                .replace(".up.1.", ".up_blocks.2.")
+                .replace(".up.2.", ".up_blocks.1.")
+                .replace(".up.3.", ".up_blocks.0.")
+                .replace(".block.", ".resnets.")
+                .replace("mid", "mid_block")
+                .replace(".0.upsample.", ".0.upsamplers.0.")
+                .replace(".1.upsample.", ".1.upsamplers.0.")
+                .replace(".2.upsample.", ".2.upsamplers.0.")
+                .replace(".nin_shortcut.", ".conv_shortcut.")
+                .replace(".block_1.", ".resnets.0.")
+                .replace(".block_2.", ".resnets.1.")
+                .replace(".attn_1.k.", ".attentions.0.to_k.")
+                .replace(".attn_1.q.", ".attentions.0.to_q.")
+                .replace(".attn_1.v.", ".attentions.0.to_v.")
+                .replace(".attn_1.proj_out.", ".attentions.0.to_out.0.")
+                .replace(".attn_1.norm.", ".attentions.0.group_norm.")
+            ] = v
+        elif k.startswith("quant_conv."):
+            converted_state_dict[k] = v
+        elif k.startswith("post_quant_conv."):
+            converted_state_dict[k] = v
+        else:
+            print(f"  skipping key `{k}`")
+    # fix weights shape
+    for k, v in converted_state_dict.items():
+        if (
+            (k.startswith("encoder.mid_block.attentions.0") or k.startswith("decoder.mid_block.attentions.0"))
+            and k.endswith("weight")
+            and ("to_q" in k or "to_k" in k or "to_v" in k or "to_out" in k)
+        ):
+            converted_state_dict[k] = converted_state_dict[k][:, :, 0, 0]
+
+    return converted_state_dict
+
+
+def get_asymmetric_autoencoder_kl_from_original_checkpoint(
+    scale: Literal["1.5", "2"], original_checkpoint_path: str, map_location: torch.device
+) -> AsymmetricAutoencoderKL:
+    print("Loading original state_dict")
+    original_state_dict = torch.load(original_checkpoint_path, map_location=map_location)
+    original_state_dict = original_state_dict["state_dict"]
+    print("Converting state_dict")
+    converted_state_dict = convert_asymmetric_autoencoder_kl_state_dict(original_state_dict)
+    kwargs = ASYMMETRIC_AUTOENCODER_KL_x_1_5_CONFIG if scale == "1.5" else ASYMMETRIC_AUTOENCODER_KL_x_2_CONFIG
+    print("Initializing AsymmetricAutoencoderKL model")
+    asymmetric_autoencoder_kl = AsymmetricAutoencoderKL(**kwargs)
+    print("Loading weight from converted state_dict")
+    asymmetric_autoencoder_kl.load_state_dict(converted_state_dict)
+    asymmetric_autoencoder_kl.eval()
+    print("AsymmetricAutoencoderKL successfully initialized")
+    return asymmetric_autoencoder_kl
+
+
+if __name__ == "__main__":
+    start = time.time()
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "--scale",
+        default=None,
+        type=str,
+        required=True,
+        help="Asymmetric VQGAN scale: `1.5` or `2`",
+    )
+    parser.add_argument(
+        "--original_checkpoint_path",
+        default=None,
+        type=str,
+        required=True,
+        help="Path to the original Asymmetric VQGAN checkpoint",
+    )
+    parser.add_argument(
+        "--output_path",
+        default=None,
+        type=str,
+        required=True,
+        help="Path to save pretrained AsymmetricAutoencoderKL model",
+    )
+    parser.add_argument(
+        "--map_location",
+        default="cpu",
+        type=str,
+        required=False,
+        help="The device passed to `map_location` when loading the checkpoint",
+    )
+    args = parser.parse_args()
+
+    assert args.scale in ["1.5", "2"], f"{args.scale} should be `1.5` of `2`"
+    assert Path(args.original_checkpoint_path).is_file()
+
+    asymmetric_autoencoder_kl = get_asymmetric_autoencoder_kl_from_original_checkpoint(
+        scale=args.scale,
+        original_checkpoint_path=args.original_checkpoint_path,
+        map_location=torch.device(args.map_location),
+    )
+    print("Saving pretrained AsymmetricAutoencoderKL")
+    asymmetric_autoencoder_kl.save_pretrained(args.output_path)
+    print(f"Done in {time.time() - start:.2f} seconds")

scripts/convert_original_t2i_adapter.py

@@ -0,0 +1,250 @@
+# coding=utf-8
+# Copyright 2023 The HuggingFace Inc. team.
+#
+# 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.
+"""
+Conversion script for the T2I-Adapter checkpoints.
+"""
+
+import argparse
+
+import torch
+
+from diffusers import T2IAdapter
+
+
+def convert_adapter(src_state, in_channels):
+    original_body_length = max([int(x.split(".")[1]) for x in src_state.keys() if "body." in x]) + 1
+
+    assert original_body_length == 8
+
+    # (0, 1) -> channels 1
+    assert src_state["body.0.block1.weight"].shape == (320, 320, 3, 3)
+
+    # (2, 3) -> channels 2
+    assert src_state["body.2.in_conv.weight"].shape == (640, 320, 1, 1)
+
+    # (4, 5) -> channels 3
+    assert src_state["body.4.in_conv.weight"].shape == (1280, 640, 1, 1)
+
+    # (6, 7) -> channels 4
+    assert src_state["body.6.block1.weight"].shape == (1280, 1280, 3, 3)
+
+    res_state = {
+        "adapter.conv_in.weight": src_state.pop("conv_in.weight"),
+        "adapter.conv_in.bias": src_state.pop("conv_in.bias"),
+        # 0.resnets.0
+        "adapter.body.0.resnets.0.block1.weight": src_state.pop("body.0.block1.weight"),
+        "adapter.body.0.resnets.0.block1.bias": src_state.pop("body.0.block1.bias"),
+        "adapter.body.0.resnets.0.block2.weight": src_state.pop("body.0.block2.weight"),
+        "adapter.body.0.resnets.0.block2.bias": src_state.pop("body.0.block2.bias"),
+        # 0.resnets.1
+        "adapter.body.0.resnets.1.block1.weight": src_state.pop("body.1.block1.weight"),
+        "adapter.body.0.resnets.1.block1.bias": src_state.pop("body.1.block1.bias"),
+        "adapter.body.0.resnets.1.block2.weight": src_state.pop("body.1.block2.weight"),
+        "adapter.body.0.resnets.1.block2.bias": src_state.pop("body.1.block2.bias"),
+        # 1
+        "adapter.body.1.in_conv.weight": src_state.pop("body.2.in_conv.weight"),
+        "adapter.body.1.in_conv.bias": src_state.pop("body.2.in_conv.bias"),
+        # 1.resnets.0
+        "adapter.body.1.resnets.0.block1.weight": src_state.pop("body.2.block1.weight"),
+        "adapter.body.1.resnets.0.block1.bias": src_state.pop("body.2.block1.bias"),
+        "adapter.body.1.resnets.0.block2.weight": src_state.pop("body.2.block2.weight"),
+        "adapter.body.1.resnets.0.block2.bias": src_state.pop("body.2.block2.bias"),
+        # 1.resnets.1
+        "adapter.body.1.resnets.1.block1.weight": src_state.pop("body.3.block1.weight"),
+        "adapter.body.1.resnets.1.block1.bias": src_state.pop("body.3.block1.bias"),
+        "adapter.body.1.resnets.1.block2.weight": src_state.pop("body.3.block2.weight"),
+        "adapter.body.1.resnets.1.block2.bias": src_state.pop("body.3.block2.bias"),
+        # 2
+        "adapter.body.2.in_conv.weight": src_state.pop("body.4.in_conv.weight"),
+        "adapter.body.2.in_conv.bias": src_state.pop("body.4.in_conv.bias"),
+        # 2.resnets.0
+        "adapter.body.2.resnets.0.block1.weight": src_state.pop("body.4.block1.weight"),
+        "adapter.body.2.resnets.0.block1.bias": src_state.pop("body.4.block1.bias"),
+        "adapter.body.2.resnets.0.block2.weight": src_state.pop("body.4.block2.weight"),
+        "adapter.body.2.resnets.0.block2.bias": src_state.pop("body.4.block2.bias"),
+        # 2.resnets.1
+        "adapter.body.2.resnets.1.block1.weight": src_state.pop("body.5.block1.weight"),
+        "adapter.body.2.resnets.1.block1.bias": src_state.pop("body.5.block1.bias"),
+        "adapter.body.2.resnets.1.block2.weight": src_state.pop("body.5.block2.weight"),
+        "adapter.body.2.resnets.1.block2.bias": src_state.pop("body.5.block2.bias"),
+        # 3.resnets.0
+        "adapter.body.3.resnets.0.block1.weight": src_state.pop("body.6.block1.weight"),
+        "adapter.body.3.resnets.0.block1.bias": src_state.pop("body.6.block1.bias"),
+        "adapter.body.3.resnets.0.block2.weight": src_state.pop("body.6.block2.weight"),
+        "adapter.body.3.resnets.0.block2.bias": src_state.pop("body.6.block2.bias"),
+        # 3.resnets.1
+        "adapter.body.3.resnets.1.block1.weight": src_state.pop("body.7.block1.weight"),
+        "adapter.body.3.resnets.1.block1.bias": src_state.pop("body.7.block1.bias"),
+        "adapter.body.3.resnets.1.block2.weight": src_state.pop("body.7.block2.weight"),
+        "adapter.body.3.resnets.1.block2.bias": src_state.pop("body.7.block2.bias"),
+    }
+
+    assert len(src_state) == 0
+
+    adapter = T2IAdapter(in_channels=in_channels, adapter_type="full_adapter")
+
+    adapter.load_state_dict(res_state)
+
+    return adapter
+
+
+def convert_light_adapter(src_state):
+    original_body_length = max([int(x.split(".")[1]) for x in src_state.keys() if "body." in x]) + 1
+
+    assert original_body_length == 4
+
+    res_state = {
+        # body.0.in_conv
+        "adapter.body.0.in_conv.weight": src_state.pop("body.0.in_conv.weight"),
+        "adapter.body.0.in_conv.bias": src_state.pop("body.0.in_conv.bias"),
+        # body.0.resnets.0
+        "adapter.body.0.resnets.0.block1.weight": src_state.pop("body.0.body.0.block1.weight"),
+        "adapter.body.0.resnets.0.block1.bias": src_state.pop("body.0.body.0.block1.bias"),
+        "adapter.body.0.resnets.0.block2.weight": src_state.pop("body.0.body.0.block2.weight"),
+        "adapter.body.0.resnets.0.block2.bias": src_state.pop("body.0.body.0.block2.bias"),
+        # body.0.resnets.1
+        "adapter.body.0.resnets.1.block1.weight": src_state.pop("body.0.body.1.block1.weight"),
+        "adapter.body.0.resnets.1.block1.bias": src_state.pop("body.0.body.1.block1.bias"),
+        "adapter.body.0.resnets.1.block2.weight": src_state.pop("body.0.body.1.block2.weight"),
+        "adapter.body.0.resnets.1.block2.bias": src_state.pop("body.0.body.1.block2.bias"),
+        # body.0.resnets.2
+        "adapter.body.0.resnets.2.block1.weight": src_state.pop("body.0.body.2.block1.weight"),
+        "adapter.body.0.resnets.2.block1.bias": src_state.pop("body.0.body.2.block1.bias"),
+        "adapter.body.0.resnets.2.block2.weight": src_state.pop("body.0.body.2.block2.weight"),
+        "adapter.body.0.resnets.2.block2.bias": src_state.pop("body.0.body.2.block2.bias"),
+        # body.0.resnets.3
+        "adapter.body.0.resnets.3.block1.weight": src_state.pop("body.0.body.3.block1.weight"),
+        "adapter.body.0.resnets.3.block1.bias": src_state.pop("body.0.body.3.block1.bias"),
+        "adapter.body.0.resnets.3.block2.weight": src_state.pop("body.0.body.3.block2.weight"),
+        "adapter.body.0.resnets.3.block2.bias": src_state.pop("body.0.body.3.block2.bias"),
+        # body.0.out_conv
+        "adapter.body.0.out_conv.weight": src_state.pop("body.0.out_conv.weight"),
+        "adapter.body.0.out_conv.bias": src_state.pop("body.0.out_conv.bias"),
+        # body.1.in_conv
+        "adapter.body.1.in_conv.weight": src_state.pop("body.1.in_conv.weight"),
+        "adapter.body.1.in_conv.bias": src_state.pop("body.1.in_conv.bias"),
+        # body.1.resnets.0
+        "adapter.body.1.resnets.0.block1.weight": src_state.pop("body.1.body.0.block1.weight"),
+        "adapter.body.1.resnets.0.block1.bias": src_state.pop("body.1.body.0.block1.bias"),
+        "adapter.body.1.resnets.0.block2.weight": src_state.pop("body.1.body.0.block2.weight"),
+        "adapter.body.1.resnets.0.block2.bias": src_state.pop("body.1.body.0.block2.bias"),
+        # body.1.resnets.1
+        "adapter.body.1.resnets.1.block1.weight": src_state.pop("body.1.body.1.block1.weight"),
+        "adapter.body.1.resnets.1.block1.bias": src_state.pop("body.1.body.1.block1.bias"),
+        "adapter.body.1.resnets.1.block2.weight": src_state.pop("body.1.body.1.block2.weight"),
+        "adapter.body.1.resnets.1.block2.bias": src_state.pop("body.1.body.1.block2.bias"),
+        # body.1.body.2
+        "adapter.body.1.resnets.2.block1.weight": src_state.pop("body.1.body.2.block1.weight"),
+        "adapter.body.1.resnets.2.block1.bias": src_state.pop("body.1.body.2.block1.bias"),
+        "adapter.body.1.resnets.2.block2.weight": src_state.pop("body.1.body.2.block2.weight"),
+        "adapter.body.1.resnets.2.block2.bias": src_state.pop("body.1.body.2.block2.bias"),
+        # body.1.body.3
+        "adapter.body.1.resnets.3.block1.weight": src_state.pop("body.1.body.3.block1.weight"),
+        "adapter.body.1.resnets.3.block1.bias": src_state.pop("body.1.body.3.block1.bias"),
+        "adapter.body.1.resnets.3.block2.weight": src_state.pop("body.1.body.3.block2.weight"),
+        "adapter.body.1.resnets.3.block2.bias": src_state.pop("body.1.body.3.block2.bias"),
+        # body.1.out_conv
+        "adapter.body.1.out_conv.weight": src_state.pop("body.1.out_conv.weight"),
+        "adapter.body.1.out_conv.bias": src_state.pop("body.1.out_conv.bias"),
+        # body.2.in_conv
+        "adapter.body.2.in_conv.weight": src_state.pop("body.2.in_conv.weight"),
+        "adapter.body.2.in_conv.bias": src_state.pop("body.2.in_conv.bias"),
+        # body.2.body.0
+        "adapter.body.2.resnets.0.block1.weight": src_state.pop("body.2.body.0.block1.weight"),
+        "adapter.body.2.resnets.0.block1.bias": src_state.pop("body.2.body.0.block1.bias"),
+        "adapter.body.2.resnets.0.block2.weight": src_state.pop("body.2.body.0.block2.weight"),
+        "adapter.body.2.resnets.0.block2.bias": src_state.pop("body.2.body.0.block2.bias"),
+        # body.2.body.1
+        "adapter.body.2.resnets.1.block1.weight": src_state.pop("body.2.body.1.block1.weight"),
+        "adapter.body.2.resnets.1.block1.bias": src_state.pop("body.2.body.1.block1.bias"),
+        "adapter.body.2.resnets.1.block2.weight": src_state.pop("body.2.body.1.block2.weight"),
+        "adapter.body.2.resnets.1.block2.bias": src_state.pop("body.2.body.1.block2.bias"),
+        # body.2.body.2
+        "adapter.body.2.resnets.2.block1.weight": src_state.pop("body.2.body.2.block1.weight"),
+        "adapter.body.2.resnets.2.block1.bias": src_state.pop("body.2.body.2.block1.bias"),
+        "adapter.body.2.resnets.2.block2.weight": src_state.pop("body.2.body.2.block2.weight"),
+        "adapter.body.2.resnets.2.block2.bias": src_state.pop("body.2.body.2.block2.bias"),
+        # body.2.body.3
+        "adapter.body.2.resnets.3.block1.weight": src_state.pop("body.2.body.3.block1.weight"),
+        "adapter.body.2.resnets.3.block1.bias": src_state.pop("body.2.body.3.block1.bias"),
+        "adapter.body.2.resnets.3.block2.weight": src_state.pop("body.2.body.3.block2.weight"),
+        "adapter.body.2.resnets.3.block2.bias": src_state.pop("body.2.body.3.block2.bias"),
+        # body.2.out_conv
+        "adapter.body.2.out_conv.weight": src_state.pop("body.2.out_conv.weight"),
+        "adapter.body.2.out_conv.bias": src_state.pop("body.2.out_conv.bias"),
+        # body.3.in_conv
+        "adapter.body.3.in_conv.weight": src_state.pop("body.3.in_conv.weight"),
+        "adapter.body.3.in_conv.bias": src_state.pop("body.3.in_conv.bias"),
+        # body.3.body.0
+        "adapter.body.3.resnets.0.block1.weight": src_state.pop("body.3.body.0.block1.weight"),
+        "adapter.body.3.resnets.0.block1.bias": src_state.pop("body.3.body.0.block1.bias"),
+        "adapter.body.3.resnets.0.block2.weight": src_state.pop("body.3.body.0.block2.weight"),
+        "adapter.body.3.resnets.0.block2.bias": src_state.pop("body.3.body.0.block2.bias"),
+        # body.3.body.1
+        "adapter.body.3.resnets.1.block1.weight": src_state.pop("body.3.body.1.block1.weight"),
+        "adapter.body.3.resnets.1.block1.bias": src_state.pop("body.3.body.1.block1.bias"),
+        "adapter.body.3.resnets.1.block2.weight": src_state.pop("body.3.body.1.block2.weight"),
+        "adapter.body.3.resnets.1.block2.bias": src_state.pop("body.3.body.1.block2.bias"),
+        # body.3.body.2
+        "adapter.body.3.resnets.2.block1.weight": src_state.pop("body.3.body.2.block1.weight"),
+        "adapter.body.3.resnets.2.block1.bias": src_state.pop("body.3.body.2.block1.bias"),
+        "adapter.body.3.resnets.2.block2.weight": src_state.pop("body.3.body.2.block2.weight"),
+        "adapter.body.3.resnets.2.block2.bias": src_state.pop("body.3.body.2.block2.bias"),
+        # body.3.body.3
+        "adapter.body.3.resnets.3.block1.weight": src_state.pop("body.3.body.3.block1.weight"),
+        "adapter.body.3.resnets.3.block1.bias": src_state.pop("body.3.body.3.block1.bias"),
+        "adapter.body.3.resnets.3.block2.weight": src_state.pop("body.3.body.3.block2.weight"),
+        "adapter.body.3.resnets.3.block2.bias": src_state.pop("body.3.body.3.block2.bias"),
+        # body.3.out_conv
+        "adapter.body.3.out_conv.weight": src_state.pop("body.3.out_conv.weight"),
+        "adapter.body.3.out_conv.bias": src_state.pop("body.3.out_conv.bias"),
+    }
+
+    assert len(src_state) == 0
+
+    adapter = T2IAdapter(in_channels=3, channels=[320, 640, 1280], num_res_blocks=4, adapter_type="light_adapter")
+
+    adapter.load_state_dict(res_state)
+
+    return adapter
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+
+    parser.add_argument(
+        "--checkpoint_path", default=None, type=str, required=True, help="Path to the checkpoint to convert."
+    )
+    parser.add_argument(
+        "--output_path", default=None, type=str, required=True, help="Path to the store the result checkpoint."
+    )
+    parser.add_argument(
+        "--is_adapter_light",
+        action="store_true",
+        help="Is checkpoint come from Adapter-Light architecture. ex: color-adapter",
+    )
+    parser.add_argument("--in_channels", required=False, type=int, help="Input channels for non-light adapter")
+
+    args = parser.parse_args()
+    src_state = torch.load(args.checkpoint_path)
+
+    if args.is_adapter_light:
+        adapter = convert_light_adapter(src_state)
+    else:
+        if args.in_channels is None:
+            raise ValueError("set `--in_channels=<n>`")
+        adapter = convert_adapter(src_state, args.in_channels)
+
+    adapter.save_pretrained(args.output_path)

scripts/convert_shap_e_to_diffusers.py

@@ -22,7 +22,7 @@ $ python scripts/convert_shap_e_to_diffusers.py \
       --prior_checkpoint_path  /home/yiyi_huggingface_co/shap-e/shap_e_model_cache/text_cond.pt \
       --prior_image_checkpoint_path /home/yiyi_huggingface_co/shap-e/shap_e_model_cache/image_cond.pt \
       --transmitter_checkpoint_path /home/yiyi_huggingface_co/shap-e/shap_e_model_cache/transmitter.pt\
-      --dump_path /home/yiyi_huggingface_co/model_repo/shap-e/renderer\
+      --dump_path /home/yiyi_huggingface_co/model_repo/shap-e-img2img/shap_e_renderer\
       --debug renderer
 ```
 """
@@ -373,6 +373,487 @@ def prior_image_original_checkpoint_to_diffusers_checkpoint(model, checkpoint):
 
 # renderer
 
+## create the lookup table for marching cubes method used in MeshDecoder
+
+MC_TABLE = [
+    [],
+    [[0, 1, 0, 2, 0, 4]],
+    [[1, 0, 1, 5, 1, 3]],
+    [[0, 4, 1, 5, 0, 2], [1, 5, 1, 3, 0, 2]],
+    [[2, 0, 2, 3, 2, 6]],
+    [[0, 1, 2, 3, 0, 4], [2, 3, 2, 6, 0, 4]],
+    [[1, 0, 1, 5, 1, 3], [2, 6, 0, 2, 3, 2]],
+    [[3, 2, 2, 6, 3, 1], [3, 1, 2, 6, 1, 5], [1, 5, 2, 6, 0, 4]],
+    [[3, 1, 3, 7, 3, 2]],
+    [[0, 2, 0, 4, 0, 1], [3, 7, 2, 3, 1, 3]],
+    [[1, 5, 3, 7, 1, 0], [3, 7, 3, 2, 1, 0]],
+    [[2, 0, 0, 4, 2, 3], [2, 3, 0, 4, 3, 7], [3, 7, 0, 4, 1, 5]],
+    [[2, 0, 3, 1, 2, 6], [3, 1, 3, 7, 2, 6]],
+    [[1, 3, 3, 7, 1, 0], [1, 0, 3, 7, 0, 4], [0, 4, 3, 7, 2, 6]],
+    [[0, 1, 1, 5, 0, 2], [0, 2, 1, 5, 2, 6], [2, 6, 1, 5, 3, 7]],
+    [[0, 4, 1, 5, 3, 7], [0, 4, 3, 7, 2, 6]],
+    [[4, 0, 4, 6, 4, 5]],
+    [[0, 2, 4, 6, 0, 1], [4, 6, 4, 5, 0, 1]],
+    [[1, 5, 1, 3, 1, 0], [4, 6, 5, 4, 0, 4]],
+    [[5, 1, 1, 3, 5, 4], [5, 4, 1, 3, 4, 6], [4, 6, 1, 3, 0, 2]],
+    [[2, 0, 2, 3, 2, 6], [4, 5, 0, 4, 6, 4]],
+    [[6, 4, 4, 5, 6, 2], [6, 2, 4, 5, 2, 3], [2, 3, 4, 5, 0, 1]],
+    [[2, 6, 2, 0, 3, 2], [1, 0, 1, 5, 3, 1], [6, 4, 5, 4, 0, 4]],
+    [[1, 3, 5, 4, 1, 5], [1, 3, 4, 6, 5, 4], [1, 3, 3, 2, 4, 6], [3, 2, 2, 6, 4, 6]],
+    [[3, 1, 3, 7, 3, 2], [6, 4, 5, 4, 0, 4]],
+    [[4, 5, 0, 1, 4, 6], [0, 1, 0, 2, 4, 6], [7, 3, 2, 3, 1, 3]],
+    [[3, 2, 1, 0, 3, 7], [1, 0, 1, 5, 3, 7], [6, 4, 5, 4, 0, 4]],
+    [[3, 7, 3, 2, 1, 5], [3, 2, 6, 4, 1, 5], [1, 5, 6, 4, 5, 4], [3, 2, 2, 0, 6, 4]],
+    [[3, 7, 2, 6, 3, 1], [2, 6, 2, 0, 3, 1], [5, 4, 0, 4, 6, 4]],
+    [[1, 0, 1, 3, 5, 4], [1, 3, 2, 6, 5, 4], [1, 3, 3, 7, 2, 6], [5, 4, 2, 6, 4, 6]],
+    [[0, 1, 1, 5, 0, 2], [0, 2, 1, 5, 2, 6], [2, 6, 1, 5, 3, 7], [4, 5, 0, 4, 4, 6]],
+    [[6, 2, 4, 6, 4, 5], [4, 5, 5, 1, 6, 2], [6, 2, 5, 1, 7, 3]],
+    [[5, 1, 5, 4, 5, 7]],
+    [[0, 1, 0, 2, 0, 4], [5, 7, 1, 5, 4, 5]],
+    [[1, 0, 5, 4, 1, 3], [5, 4, 5, 7, 1, 3]],
+    [[4, 5, 5, 7, 4, 0], [4, 0, 5, 7, 0, 2], [0, 2, 5, 7, 1, 3]],
+    [[2, 0, 2, 3, 2, 6], [7, 5, 1, 5, 4, 5]],
+    [[2, 6, 0, 4, 2, 3], [0, 4, 0, 1, 2, 3], [7, 5, 1, 5, 4, 5]],
+    [[5, 7, 1, 3, 5, 4], [1, 3, 1, 0, 5, 4], [6, 2, 0, 2, 3, 2]],
+    [[3, 1, 3, 2, 7, 5], [3, 2, 0, 4, 7, 5], [3, 2, 2, 6, 0, 4], [7, 5, 0, 4, 5, 4]],
+    [[3, 7, 3, 2, 3, 1], [5, 4, 7, 5, 1, 5]],
+    [[0, 4, 0, 1, 2, 0], [3, 1, 3, 7, 2, 3], [4, 5, 7, 5, 1, 5]],
+    [[7, 3, 3, 2, 7, 5], [7, 5, 3, 2, 5, 4], [5, 4, 3, 2, 1, 0]],
+    [[0, 4, 2, 3, 0, 2], [0, 4, 3, 7, 2, 3], [0, 4, 4, 5, 3, 7], [4, 5, 5, 7, 3, 7]],
+    [[2, 0, 3, 1, 2, 6], [3, 1, 3, 7, 2, 6], [4, 5, 7, 5, 1, 5]],
+    [[1, 3, 3, 7, 1, 0], [1, 0, 3, 7, 0, 4], [0, 4, 3, 7, 2, 6], [5, 7, 1, 5, 5, 4]],
+    [[2, 6, 2, 0, 3, 7], [2, 0, 4, 5, 3, 7], [3, 7, 4, 5, 7, 5], [2, 0, 0, 1, 4, 5]],
+    [[4, 0, 5, 4, 5, 7], [5, 7, 7, 3, 4, 0], [4, 0, 7, 3, 6, 2]],
+    [[4, 6, 5, 7, 4, 0], [5, 7, 5, 1, 4, 0]],
+    [[1, 0, 0, 2, 1, 5], [1, 5, 0, 2, 5, 7], [5, 7, 0, 2, 4, 6]],
+    [[0, 4, 4, 6, 0, 1], [0, 1, 4, 6, 1, 3], [1, 3, 4, 6, 5, 7]],
+    [[0, 2, 4, 6, 5, 7], [0, 2, 5, 7, 1, 3]],
+    [[5, 1, 4, 0, 5, 7], [4, 0, 4, 6, 5, 7], [3, 2, 6, 2, 0, 2]],
+    [[2, 3, 2, 6, 0, 1], [2, 6, 7, 5, 0, 1], [0, 1, 7, 5, 1, 5], [2, 6, 6, 4, 7, 5]],
+    [[0, 4, 4, 6, 0, 1], [0, 1, 4, 6, 1, 3], [1, 3, 4, 6, 5, 7], [2, 6, 0, 2, 2, 3]],
+    [[3, 1, 2, 3, 2, 6], [2, 6, 6, 4, 3, 1], [3, 1, 6, 4, 7, 5]],
+    [[4, 6, 5, 7, 4, 0], [5, 7, 5, 1, 4, 0], [2, 3, 1, 3, 7, 3]],
+    [[1, 0, 0, 2, 1, 5], [1, 5, 0, 2, 5, 7], [5, 7, 0, 2, 4, 6], [3, 2, 1, 3, 3, 7]],
+    [[0, 1, 0, 4, 2, 3], [0, 4, 5, 7, 2, 3], [0, 4, 4, 6, 5, 7], [2, 3, 5, 7, 3, 7]],
+    [[7, 5, 3, 7, 3, 2], [3, 2, 2, 0, 7, 5], [7, 5, 2, 0, 6, 4]],
+    [[0, 4, 4, 6, 5, 7], [0, 4, 5, 7, 1, 5], [0, 2, 1, 3, 3, 7], [3, 7, 2, 6, 0, 2]],
+    [
+        [3, 1, 7, 3, 6, 2],
+        [6, 2, 0, 1, 3, 1],
+        [6, 4, 0, 1, 6, 2],
+        [6, 4, 5, 1, 0, 1],
+        [6, 4, 7, 5, 5, 1],
+    ],
+    [
+        [4, 0, 6, 4, 7, 5],
+        [7, 5, 1, 0, 4, 0],
+        [7, 3, 1, 0, 7, 5],
+        [7, 3, 2, 0, 1, 0],
+        [7, 3, 6, 2, 2, 0],
+    ],
+    [[7, 3, 6, 2, 6, 4], [7, 5, 7, 3, 6, 4]],
+    [[6, 2, 6, 7, 6, 4]],
+    [[0, 4, 0, 1, 0, 2], [6, 7, 4, 6, 2, 6]],
+    [[1, 0, 1, 5, 1, 3], [7, 6, 4, 6, 2, 6]],
+    [[1, 3, 0, 2, 1, 5], [0, 2, 0, 4, 1, 5], [7, 6, 4, 6, 2, 6]],
+    [[2, 3, 6, 7, 2, 0], [6, 7, 6, 4, 2, 0]],
+    [[4, 0, 0, 1, 4, 6], [4, 6, 0, 1, 6, 7], [6, 7, 0, 1, 2, 3]],
+    [[6, 4, 2, 0, 6, 7], [2, 0, 2, 3, 6, 7], [5, 1, 3, 1, 0, 1]],
+    [[1, 5, 1, 3, 0, 4], [1, 3, 7, 6, 0, 4], [0, 4, 7, 6, 4, 6], [1, 3, 3, 2, 7, 6]],
+    [[3, 2, 3, 1, 3, 7], [6, 4, 2, 6, 7, 6]],
+    [[3, 7, 3, 2, 1, 3], [0, 2, 0, 4, 1, 0], [7, 6, 4, 6, 2, 6]],
+    [[1, 5, 3, 7, 1, 0], [3, 7, 3, 2, 1, 0], [4, 6, 2, 6, 7, 6]],
+    [[2, 0, 0, 4, 2, 3], [2, 3, 0, 4, 3, 7], [3, 7, 0, 4, 1, 5], [6, 4, 2, 6, 6, 7]],
+    [[7, 6, 6, 4, 7, 3], [7, 3, 6, 4, 3, 1], [3, 1, 6, 4, 2, 0]],
+    [[0, 1, 4, 6, 0, 4], [0, 1, 6, 7, 4, 6], [0, 1, 1, 3, 6, 7], [1, 3, 3, 7, 6, 7]],
+    [[0, 2, 0, 1, 4, 6], [0, 1, 3, 7, 4, 6], [0, 1, 1, 5, 3, 7], [4, 6, 3, 7, 6, 7]],
+    [[7, 3, 6, 7, 6, 4], [6, 4, 4, 0, 7, 3], [7, 3, 4, 0, 5, 1]],
+    [[4, 0, 6, 2, 4, 5], [6, 2, 6, 7, 4, 5]],
+    [[2, 6, 6, 7, 2, 0], [2, 0, 6, 7, 0, 1], [0, 1, 6, 7, 4, 5]],
+    [[6, 7, 4, 5, 6, 2], [4, 5, 4, 0, 6, 2], [3, 1, 0, 1, 5, 1]],
+    [[2, 0, 2, 6, 3, 1], [2, 6, 4, 5, 3, 1], [2, 6, 6, 7, 4, 5], [3, 1, 4, 5, 1, 5]],
+    [[0, 2, 2, 3, 0, 4], [0, 4, 2, 3, 4, 5], [4, 5, 2, 3, 6, 7]],
+    [[0, 1, 2, 3, 6, 7], [0, 1, 6, 7, 4, 5]],
+    [[0, 2, 2, 3, 0, 4], [0, 4, 2, 3, 4, 5], [4, 5, 2, 3, 6, 7], [1, 3, 0, 1, 1, 5]],
+    [[5, 4, 1, 5, 1, 3], [1, 3, 3, 2, 5, 4], [5, 4, 3, 2, 7, 6]],
+    [[4, 0, 6, 2, 4, 5], [6, 2, 6, 7, 4, 5], [1, 3, 7, 3, 2, 3]],
+    [[2, 6, 6, 7, 2, 0], [2, 0, 6, 7, 0, 1], [0, 1, 6, 7, 4, 5], [3, 7, 2, 3, 3, 1]],
+    [[0, 1, 1, 5, 3, 7], [0, 1, 3, 7, 2, 3], [0, 4, 2, 6, 6, 7], [6, 7, 4, 5, 0, 4]],
+    [
+        [6, 2, 7, 6, 5, 4],
+        [5, 4, 0, 2, 6, 2],
+        [5, 1, 0, 2, 5, 4],
+        [5, 1, 3, 2, 0, 2],
+        [5, 1, 7, 3, 3, 2],
+    ],
+    [[3, 1, 3, 7, 2, 0], [3, 7, 5, 4, 2, 0], [2, 0, 5, 4, 0, 4], [3, 7, 7, 6, 5, 4]],
+    [[1, 0, 3, 1, 3, 7], [3, 7, 7, 6, 1, 0], [1, 0, 7, 6, 5, 4]],
+    [
+        [1, 0, 5, 1, 7, 3],
+        [7, 3, 2, 0, 1, 0],
+        [7, 6, 2, 0, 7, 3],
+        [7, 6, 4, 0, 2, 0],
+        [7, 6, 5, 4, 4, 0],
+    ],
+    [[7, 6, 5, 4, 5, 1], [7, 3, 7, 6, 5, 1]],
+    [[5, 7, 5, 1, 5, 4], [6, 2, 7, 6, 4, 6]],
+    [[0, 2, 0, 4, 1, 0], [5, 4, 5, 7, 1, 5], [2, 6, 7, 6, 4, 6]],
+    [[1, 0, 5, 4, 1, 3], [5, 4, 5, 7, 1, 3], [2, 6, 7, 6, 4, 6]],
+    [[4, 5, 5, 7, 4, 0], [4, 0, 5, 7, 0, 2], [0, 2, 5, 7, 1, 3], [6, 7, 4, 6, 6, 2]],
+    [[2, 3, 6, 7, 2, 0], [6, 7, 6, 4, 2, 0], [1, 5, 4, 5, 7, 5]],
+    [[4, 0, 0, 1, 4, 6], [4, 6, 0, 1, 6, 7], [6, 7, 0, 1, 2, 3], [5, 1, 4, 5, 5, 7]],
+    [[0, 2, 2, 3, 6, 7], [0, 2, 6, 7, 4, 6], [0, 1, 4, 5, 5, 7], [5, 7, 1, 3, 0, 1]],
+    [
+        [5, 4, 7, 5, 3, 1],
+        [3, 1, 0, 4, 5, 4],
+        [3, 2, 0, 4, 3, 1],
+        [3, 2, 6, 4, 0, 4],
+        [3, 2, 7, 6, 6, 4],
+    ],
+    [[5, 4, 5, 7, 1, 5], [3, 7, 3, 2, 1, 3], [4, 6, 2, 6, 7, 6]],
+    [[1, 0, 0, 2, 0, 4], [1, 5, 5, 4, 5, 7], [3, 2, 1, 3, 3, 7], [2, 6, 7, 6, 4, 6]],
+    [[7, 3, 3, 2, 7, 5], [7, 5, 3, 2, 5, 4], [5, 4, 3, 2, 1, 0], [6, 2, 7, 6, 6, 4]],
+    [
+        [0, 4, 2, 3, 0, 2],
+        [0, 4, 3, 7, 2, 3],
+        [0, 4, 4, 5, 3, 7],
+        [4, 5, 5, 7, 3, 7],
+        [6, 7, 4, 6, 2, 6],
+    ],
+    [[7, 6, 6, 4, 7, 3], [7, 3, 6, 4, 3, 1], [3, 1, 6, 4, 2, 0], [5, 4, 7, 5, 5, 1]],
+    [
+        [0, 1, 4, 6, 0, 4],
+        [0, 1, 6, 7, 4, 6],
+        [0, 1, 1, 3, 6, 7],
+        [1, 3, 3, 7, 6, 7],
+        [5, 7, 1, 5, 4, 5],
+    ],
+    [
+        [6, 7, 4, 6, 0, 2],
+        [0, 2, 3, 7, 6, 7],
+        [0, 1, 3, 7, 0, 2],
+        [0, 1, 5, 7, 3, 7],
+        [0, 1, 4, 5, 5, 7],
+    ],
+    [[4, 0, 6, 7, 4, 6], [4, 0, 7, 3, 6, 7], [4, 0, 5, 7, 7, 3], [4, 5, 5, 7, 4, 0]],
+    [[7, 5, 5, 1, 7, 6], [7, 6, 5, 1, 6, 2], [6, 2, 5, 1, 4, 0]],
+    [[0, 2, 1, 5, 0, 1], [0, 2, 5, 7, 1, 5], [0, 2, 2, 6, 5, 7], [2, 6, 6, 7, 5, 7]],
+    [[1, 3, 1, 0, 5, 7], [1, 0, 2, 6, 5, 7], [5, 7, 2, 6, 7, 6], [1, 0, 0, 4, 2, 6]],
+    [[2, 0, 6, 2, 6, 7], [6, 7, 7, 5, 2, 0], [2, 0, 7, 5, 3, 1]],
+    [[0, 4, 0, 2, 1, 5], [0, 2, 6, 7, 1, 5], [0, 2, 2, 3, 6, 7], [1, 5, 6, 7, 5, 7]],
+    [[7, 6, 5, 7, 5, 1], [5, 1, 1, 0, 7, 6], [7, 6, 1, 0, 3, 2]],
+    [
+        [2, 0, 3, 2, 7, 6],
+        [7, 6, 4, 0, 2, 0],
+        [7, 5, 4, 0, 7, 6],
+        [7, 5, 1, 0, 4, 0],
+        [7, 5, 3, 1, 1, 0],
+    ],
+    [[7, 5, 3, 1, 3, 2], [7, 6, 7, 5, 3, 2]],
+    [[7, 5, 5, 1, 7, 6], [7, 6, 5, 1, 6, 2], [6, 2, 5, 1, 4, 0], [3, 1, 7, 3, 3, 2]],
+    [
+        [0, 2, 1, 5, 0, 1],
+        [0, 2, 5, 7, 1, 5],
+        [0, 2, 2, 6, 5, 7],
+        [2, 6, 6, 7, 5, 7],
+        [3, 7, 2, 3, 1, 3],
+    ],
+    [
+        [3, 7, 2, 3, 0, 1],
+        [0, 1, 5, 7, 3, 7],
+        [0, 4, 5, 7, 0, 1],
+        [0, 4, 6, 7, 5, 7],
+        [0, 4, 2, 6, 6, 7],
+    ],
+    [[2, 0, 3, 7, 2, 3], [2, 0, 7, 5, 3, 7], [2, 0, 6, 7, 7, 5], [2, 6, 6, 7, 2, 0]],
+    [
+        [5, 7, 1, 5, 0, 4],
+        [0, 4, 6, 7, 5, 7],
+        [0, 2, 6, 7, 0, 4],
+        [0, 2, 3, 7, 6, 7],
+        [0, 2, 1, 3, 3, 7],
+    ],
+    [[1, 0, 5, 7, 1, 5], [1, 0, 7, 6, 5, 7], [1, 0, 3, 7, 7, 6], [1, 3, 3, 7, 1, 0]],
+    [[0, 2, 0, 1, 0, 4], [3, 7, 6, 7, 5, 7]],
+    [[7, 5, 7, 3, 7, 6]],
+    [[7, 3, 7, 5, 7, 6]],
+    [[0, 1, 0, 2, 0, 4], [6, 7, 3, 7, 5, 7]],
+    [[1, 3, 1, 0, 1, 5], [7, 6, 3, 7, 5, 7]],
+    [[0, 4, 1, 5, 0, 2], [1, 5, 1, 3, 0, 2], [6, 7, 3, 7, 5, 7]],
+    [[2, 6, 2, 0, 2, 3], [7, 5, 6, 7, 3, 7]],
+    [[0, 1, 2, 3, 0, 4], [2, 3, 2, 6, 0, 4], [5, 7, 6, 7, 3, 7]],
+    [[1, 5, 1, 3, 0, 1], [2, 3, 2, 6, 0, 2], [5, 7, 6, 7, 3, 7]],
+    [[3, 2, 2, 6, 3, 1], [3, 1, 2, 6, 1, 5], [1, 5, 2, 6, 0, 4], [7, 6, 3, 7, 7, 5]],
+    [[3, 1, 7, 5, 3, 2], [7, 5, 7, 6, 3, 2]],
+    [[7, 6, 3, 2, 7, 5], [3, 2, 3, 1, 7, 5], [4, 0, 1, 0, 2, 0]],
+    [[5, 7, 7, 6, 5, 1], [5, 1, 7, 6, 1, 0], [1, 0, 7, 6, 3, 2]],
+    [[2, 3, 2, 0, 6, 7], [2, 0, 1, 5, 6, 7], [2, 0, 0, 4, 1, 5], [6, 7, 1, 5, 7, 5]],
+    [[6, 2, 2, 0, 6, 7], [6, 7, 2, 0, 7, 5], [7, 5, 2, 0, 3, 1]],
+    [[0, 4, 0, 1, 2, 6], [0, 1, 5, 7, 2, 6], [2, 6, 5, 7, 6, 7], [0, 1, 1, 3, 5, 7]],
+    [[1, 5, 0, 2, 1, 0], [1, 5, 2, 6, 0, 2], [1, 5, 5, 7, 2, 6], [5, 7, 7, 6, 2, 6]],
+    [[5, 1, 7, 5, 7, 6], [7, 6, 6, 2, 5, 1], [5, 1, 6, 2, 4, 0]],
+    [[4, 5, 4, 0, 4, 6], [7, 3, 5, 7, 6, 7]],
+    [[0, 2, 4, 6, 0, 1], [4, 6, 4, 5, 0, 1], [3, 7, 5, 7, 6, 7]],
+    [[4, 6, 4, 5, 0, 4], [1, 5, 1, 3, 0, 1], [6, 7, 3, 7, 5, 7]],
+    [[5, 1, 1, 3, 5, 4], [5, 4, 1, 3, 4, 6], [4, 6, 1, 3, 0, 2], [7, 3, 5, 7, 7, 6]],
+    [[2, 3, 2, 6, 0, 2], [4, 6, 4, 5, 0, 4], [3, 7, 5, 7, 6, 7]],
+    [[6, 4, 4, 5, 6, 2], [6, 2, 4, 5, 2, 3], [2, 3, 4, 5, 0, 1], [7, 5, 6, 7, 7, 3]],
+    [[0, 1, 1, 5, 1, 3], [0, 2, 2, 3, 2, 6], [4, 5, 0, 4, 4, 6], [5, 7, 6, 7, 3, 7]],
+    [
+        [1, 3, 5, 4, 1, 5],
+        [1, 3, 4, 6, 5, 4],
+        [1, 3, 3, 2, 4, 6],
+        [3, 2, 2, 6, 4, 6],
+        [7, 6, 3, 7, 5, 7],
+    ],
+    [[3, 1, 7, 5, 3, 2], [7, 5, 7, 6, 3, 2], [0, 4, 6, 4, 5, 4]],
+    [[1, 0, 0, 2, 4, 6], [1, 0, 4, 6, 5, 4], [1, 3, 5, 7, 7, 6], [7, 6, 3, 2, 1, 3]],
+    [[5, 7, 7, 6, 5, 1], [5, 1, 7, 6, 1, 0], [1, 0, 7, 6, 3, 2], [4, 6, 5, 4, 4, 0]],
+    [
+        [7, 5, 6, 7, 2, 3],
+        [2, 3, 1, 5, 7, 5],
+        [2, 0, 1, 5, 2, 3],
+        [2, 0, 4, 5, 1, 5],
+        [2, 0, 6, 4, 4, 5],
+    ],
+    [[6, 2, 2, 0, 6, 7], [6, 7, 2, 0, 7, 5], [7, 5, 2, 0, 3, 1], [4, 0, 6, 4, 4, 5]],
+    [
+        [4, 6, 5, 4, 1, 0],
+        [1, 0, 2, 6, 4, 6],
+        [1, 3, 2, 6, 1, 0],
+        [1, 3, 7, 6, 2, 6],
+        [1, 3, 5, 7, 7, 6],
+    ],
+    [
+        [1, 5, 0, 2, 1, 0],
+        [1, 5, 2, 6, 0, 2],
+        [1, 5, 5, 7, 2, 6],
+        [5, 7, 7, 6, 2, 6],
+        [4, 6, 5, 4, 0, 4],
+    ],
+    [[5, 1, 4, 6, 5, 4], [5, 1, 6, 2, 4, 6], [5, 1, 7, 6, 6, 2], [5, 7, 7, 6, 5, 1]],
+    [[5, 4, 7, 6, 5, 1], [7, 6, 7, 3, 5, 1]],
+    [[7, 3, 5, 1, 7, 6], [5, 1, 5, 4, 7, 6], [2, 0, 4, 0, 1, 0]],
+    [[3, 1, 1, 0, 3, 7], [3, 7, 1, 0, 7, 6], [7, 6, 1, 0, 5, 4]],
+    [[0, 2, 0, 4, 1, 3], [0, 4, 6, 7, 1, 3], [1, 3, 6, 7, 3, 7], [0, 4, 4, 5, 6, 7]],
+    [[5, 4, 7, 6, 5, 1], [7, 6, 7, 3, 5, 1], [0, 2, 3, 2, 6, 2]],
+    [[1, 5, 5, 4, 7, 6], [1, 5, 7, 6, 3, 7], [1, 0, 3, 2, 2, 6], [2, 6, 0, 4, 1, 0]],
+    [[3, 1, 1, 0, 3, 7], [3, 7, 1, 0, 7, 6], [7, 6, 1, 0, 5, 4], [2, 0, 3, 2, 2, 6]],
+    [
+        [2, 3, 6, 2, 4, 0],
+        [4, 0, 1, 3, 2, 3],
+        [4, 5, 1, 3, 4, 0],
+        [4, 5, 7, 3, 1, 3],
+        [4, 5, 6, 7, 7, 3],
+    ],
+    [[1, 5, 5, 4, 1, 3], [1, 3, 5, 4, 3, 2], [3, 2, 5, 4, 7, 6]],
+    [[1, 5, 5, 4, 1, 3], [1, 3, 5, 4, 3, 2], [3, 2, 5, 4, 7, 6], [0, 4, 1, 0, 0, 2]],
+    [[1, 0, 5, 4, 7, 6], [1, 0, 7, 6, 3, 2]],
+    [[2, 3, 0, 2, 0, 4], [0, 4, 4, 5, 2, 3], [2, 3, 4, 5, 6, 7]],
+    [[1, 3, 1, 5, 0, 2], [1, 5, 7, 6, 0, 2], [1, 5, 5, 4, 7, 6], [0, 2, 7, 6, 2, 6]],
+    [
+        [5, 1, 4, 5, 6, 7],
+        [6, 7, 3, 1, 5, 1],
+        [6, 2, 3, 1, 6, 7],
+        [6, 2, 0, 1, 3, 1],
+        [6, 2, 4, 0, 0, 1],
+    ],
+    [[6, 7, 2, 6, 2, 0], [2, 0, 0, 1, 6, 7], [6, 7, 0, 1, 4, 5]],
+    [[6, 2, 4, 0, 4, 5], [6, 7, 6, 2, 4, 5]],
+    [[6, 7, 7, 3, 6, 4], [6, 4, 7, 3, 4, 0], [4, 0, 7, 3, 5, 1]],
+    [[1, 5, 1, 0, 3, 7], [1, 0, 4, 6, 3, 7], [1, 0, 0, 2, 4, 6], [3, 7, 4, 6, 7, 6]],
+    [[1, 0, 3, 7, 1, 3], [1, 0, 7, 6, 3, 7], [1, 0, 0, 4, 7, 6], [0, 4, 4, 6, 7, 6]],
+    [[6, 4, 7, 6, 7, 3], [7, 3, 3, 1, 6, 4], [6, 4, 3, 1, 2, 0]],
+    [[6, 7, 7, 3, 6, 4], [6, 4, 7, 3, 4, 0], [4, 0, 7, 3, 5, 1], [2, 3, 6, 2, 2, 0]],
+    [
+        [7, 6, 3, 7, 1, 5],
+        [1, 5, 4, 6, 7, 6],
+        [1, 0, 4, 6, 1, 5],
+        [1, 0, 2, 6, 4, 6],
+        [1, 0, 3, 2, 2, 6],
+    ],
+    [
+        [1, 0, 3, 7, 1, 3],
+        [1, 0, 7, 6, 3, 7],
+        [1, 0, 0, 4, 7, 6],
+        [0, 4, 4, 6, 7, 6],
+        [2, 6, 0, 2, 3, 2],
+    ],
+    [[3, 1, 7, 6, 3, 7], [3, 1, 6, 4, 7, 6], [3, 1, 2, 6, 6, 4], [3, 2, 2, 6, 3, 1]],
+    [[3, 2, 3, 1, 7, 6], [3, 1, 0, 4, 7, 6], [7, 6, 0, 4, 6, 4], [3, 1, 1, 5, 0, 4]],
+    [
+        [0, 1, 2, 0, 6, 4],
+        [6, 4, 5, 1, 0, 1],
+        [6, 7, 5, 1, 6, 4],
+        [6, 7, 3, 1, 5, 1],
+        [6, 7, 2, 3, 3, 1],
+    ],
+    [[0, 1, 4, 0, 4, 6], [4, 6, 6, 7, 0, 1], [0, 1, 6, 7, 2, 3]],
+    [[6, 7, 2, 3, 2, 0], [6, 4, 6, 7, 2, 0]],
+    [
+        [2, 6, 0, 2, 1, 3],
+        [1, 3, 7, 6, 2, 6],
+        [1, 5, 7, 6, 1, 3],
+        [1, 5, 4, 6, 7, 6],
+        [1, 5, 0, 4, 4, 6],
+    ],
+    [[1, 5, 1, 0, 1, 3], [4, 6, 7, 6, 2, 6]],
+    [[0, 1, 2, 6, 0, 2], [0, 1, 6, 7, 2, 6], [0, 1, 4, 6, 6, 7], [0, 4, 4, 6, 0, 1]],
+    [[6, 7, 6, 2, 6, 4]],
+    [[6, 2, 7, 3, 6, 4], [7, 3, 7, 5, 6, 4]],
+    [[7, 5, 6, 4, 7, 3], [6, 4, 6, 2, 7, 3], [1, 0, 2, 0, 4, 0]],
+    [[6, 2, 7, 3, 6, 4], [7, 3, 7, 5, 6, 4], [0, 1, 5, 1, 3, 1]],
+    [[2, 0, 0, 4, 1, 5], [2, 0, 1, 5, 3, 1], [2, 6, 3, 7, 7, 5], [7, 5, 6, 4, 2, 6]],
+    [[3, 7, 7, 5, 3, 2], [3, 2, 7, 5, 2, 0], [2, 0, 7, 5, 6, 4]],
+    [[3, 2, 3, 7, 1, 0], [3, 7, 6, 4, 1, 0], [3, 7, 7, 5, 6, 4], [1, 0, 6, 4, 0, 4]],
+    [[3, 7, 7, 5, 3, 2], [3, 2, 7, 5, 2, 0], [2, 0, 7, 5, 6, 4], [1, 5, 3, 1, 1, 0]],
+    [
+        [7, 3, 5, 7, 4, 6],
+        [4, 6, 2, 3, 7, 3],
+        [4, 0, 2, 3, 4, 6],
+        [4, 0, 1, 3, 2, 3],
+        [4, 0, 5, 1, 1, 3],
+    ],
+    [[2, 3, 3, 1, 2, 6], [2, 6, 3, 1, 6, 4], [6, 4, 3, 1, 7, 5]],
+    [[2, 3, 3, 1, 2, 6], [2, 6, 3, 1, 6, 4], [6, 4, 3, 1, 7, 5], [0, 1, 2, 0, 0, 4]],
+    [[1, 0, 1, 5, 3, 2], [1, 5, 4, 6, 3, 2], [3, 2, 4, 6, 2, 6], [1, 5, 5, 7, 4, 6]],
+    [
+        [0, 2, 4, 0, 5, 1],
+        [5, 1, 3, 2, 0, 2],
+        [5, 7, 3, 2, 5, 1],
+        [5, 7, 6, 2, 3, 2],
+        [5, 7, 4, 6, 6, 2],
+    ],
+    [[2, 0, 3, 1, 7, 5], [2, 0, 7, 5, 6, 4]],
+    [[4, 6, 0, 4, 0, 1], [0, 1, 1, 3, 4, 6], [4, 6, 1, 3, 5, 7]],
+    [[0, 2, 1, 0, 1, 5], [1, 5, 5, 7, 0, 2], [0, 2, 5, 7, 4, 6]],
+    [[5, 7, 4, 6, 4, 0], [5, 1, 5, 7, 4, 0]],
+    [[5, 4, 4, 0, 5, 7], [5, 7, 4, 0, 7, 3], [7, 3, 4, 0, 6, 2]],
+    [[0, 1, 0, 2, 4, 5], [0, 2, 3, 7, 4, 5], [4, 5, 3, 7, 5, 7], [0, 2, 2, 6, 3, 7]],
+    [[5, 4, 4, 0, 5, 7], [5, 7, 4, 0, 7, 3], [7, 3, 4, 0, 6, 2], [1, 0, 5, 1, 1, 3]],
+    [
+        [1, 5, 3, 1, 2, 0],
+        [2, 0, 4, 5, 1, 5],
+        [2, 6, 4, 5, 2, 0],
+        [2, 6, 7, 5, 4, 5],
+        [2, 6, 3, 7, 7, 5],
+    ],
+    [[2, 3, 0, 4, 2, 0], [2, 3, 4, 5, 0, 4], [2, 3, 3, 7, 4, 5], [3, 7, 7, 5, 4, 5]],
+    [[3, 2, 7, 3, 7, 5], [7, 5, 5, 4, 3, 2], [3, 2, 5, 4, 1, 0]],
+    [
+        [2, 3, 0, 4, 2, 0],
+        [2, 3, 4, 5, 0, 4],
+        [2, 3, 3, 7, 4, 5],
+        [3, 7, 7, 5, 4, 5],
+        [1, 5, 3, 1, 0, 1],
+    ],
+    [[3, 2, 1, 5, 3, 1], [3, 2, 5, 4, 1, 5], [3, 2, 7, 5, 5, 4], [3, 7, 7, 5, 3, 2]],
+    [[2, 6, 2, 3, 0, 4], [2, 3, 7, 5, 0, 4], [2, 3, 3, 1, 7, 5], [0, 4, 7, 5, 4, 5]],
+    [
+        [3, 2, 1, 3, 5, 7],
+        [5, 7, 6, 2, 3, 2],
+        [5, 4, 6, 2, 5, 7],
+        [5, 4, 0, 2, 6, 2],
+        [5, 4, 1, 0, 0, 2],
+    ],
+    [
+        [4, 5, 0, 4, 2, 6],
+        [2, 6, 7, 5, 4, 5],
+        [2, 3, 7, 5, 2, 6],
+        [2, 3, 1, 5, 7, 5],
+        [2, 3, 0, 1, 1, 5],
+    ],
+    [[2, 3, 2, 0, 2, 6], [1, 5, 7, 5, 4, 5]],
+    [[5, 7, 4, 5, 4, 0], [4, 0, 0, 2, 5, 7], [5, 7, 0, 2, 1, 3]],
+    [[5, 4, 1, 0, 1, 3], [5, 7, 5, 4, 1, 3]],
+    [[0, 2, 4, 5, 0, 4], [0, 2, 5, 7, 4, 5], [0, 2, 1, 5, 5, 7], [0, 1, 1, 5, 0, 2]],
+    [[5, 4, 5, 1, 5, 7]],
+    [[4, 6, 6, 2, 4, 5], [4, 5, 6, 2, 5, 1], [5, 1, 6, 2, 7, 3]],
+    [[4, 6, 6, 2, 4, 5], [4, 5, 6, 2, 5, 1], [5, 1, 6, 2, 7, 3], [0, 2, 4, 0, 0, 1]],
+    [[3, 7, 3, 1, 2, 6], [3, 1, 5, 4, 2, 6], [3, 1, 1, 0, 5, 4], [2, 6, 5, 4, 6, 4]],
+    [
+        [6, 4, 2, 6, 3, 7],
+        [3, 7, 5, 4, 6, 4],
+        [3, 1, 5, 4, 3, 7],
+        [3, 1, 0, 4, 5, 4],
+        [3, 1, 2, 0, 0, 4],
+    ],
+    [[2, 0, 2, 3, 6, 4], [2, 3, 1, 5, 6, 4], [6, 4, 1, 5, 4, 5], [2, 3, 3, 7, 1, 5]],
+    [
+        [0, 4, 1, 0, 3, 2],
+        [3, 2, 6, 4, 0, 4],
+        [3, 7, 6, 4, 3, 2],
+        [3, 7, 5, 4, 6, 4],
+        [3, 7, 1, 5, 5, 4],
+    ],
+    [
+        [1, 3, 0, 1, 4, 5],
+        [4, 5, 7, 3, 1, 3],
+        [4, 6, 7, 3, 4, 5],
+        [4, 6, 2, 3, 7, 3],
+        [4, 6, 0, 2, 2, 3],
+    ],
+    [[3, 7, 3, 1, 3, 2], [5, 4, 6, 4, 0, 4]],
+    [[3, 1, 2, 6, 3, 2], [3, 1, 6, 4, 2, 6], [3, 1, 1, 5, 6, 4], [1, 5, 5, 4, 6, 4]],
+    [
+        [3, 1, 2, 6, 3, 2],
+        [3, 1, 6, 4, 2, 6],
+        [3, 1, 1, 5, 6, 4],
+        [1, 5, 5, 4, 6, 4],
+        [0, 4, 1, 0, 2, 0],
+    ],
+    [[4, 5, 6, 4, 6, 2], [6, 2, 2, 3, 4, 5], [4, 5, 2, 3, 0, 1]],
+    [[2, 3, 6, 4, 2, 6], [2, 3, 4, 5, 6, 4], [2, 3, 0, 4, 4, 5], [2, 0, 0, 4, 2, 3]],
+    [[1, 3, 5, 1, 5, 4], [5, 4, 4, 6, 1, 3], [1, 3, 4, 6, 0, 2]],
+    [[1, 3, 0, 4, 1, 0], [1, 3, 4, 6, 0, 4], [1, 3, 5, 4, 4, 6], [1, 5, 5, 4, 1, 3]],
+    [[4, 6, 0, 2, 0, 1], [4, 5, 4, 6, 0, 1]],
+    [[4, 6, 4, 0, 4, 5]],
+    [[4, 0, 6, 2, 7, 3], [4, 0, 7, 3, 5, 1]],
+    [[1, 5, 0, 1, 0, 2], [0, 2, 2, 6, 1, 5], [1, 5, 2, 6, 3, 7]],
+    [[3, 7, 1, 3, 1, 0], [1, 0, 0, 4, 3, 7], [3, 7, 0, 4, 2, 6]],
+    [[3, 1, 2, 0, 2, 6], [3, 7, 3, 1, 2, 6]],
+    [[0, 4, 2, 0, 2, 3], [2, 3, 3, 7, 0, 4], [0, 4, 3, 7, 1, 5]],
+    [[3, 7, 1, 5, 1, 0], [3, 2, 3, 7, 1, 0]],
+    [[0, 4, 1, 3, 0, 1], [0, 4, 3, 7, 1, 3], [0, 4, 2, 3, 3, 7], [0, 2, 2, 3, 0, 4]],
+    [[3, 7, 3, 1, 3, 2]],
+    [[2, 6, 3, 2, 3, 1], [3, 1, 1, 5, 2, 6], [2, 6, 1, 5, 0, 4]],
+    [[1, 5, 3, 2, 1, 3], [1, 5, 2, 6, 3, 2], [1, 5, 0, 2, 2, 6], [1, 0, 0, 2, 1, 5]],
+    [[2, 3, 0, 1, 0, 4], [2, 6, 2, 3, 0, 4]],
+    [[2, 3, 2, 0, 2, 6]],
+    [[1, 5, 0, 4, 0, 2], [1, 3, 1, 5, 0, 2]],
+    [[1, 5, 1, 0, 1, 3]],
+    [[0, 2, 0, 1, 0, 4]],
+    [],
+]
+
+
+def create_mc_lookup_table():
+    cases = torch.zeros(256, 5, 3, dtype=torch.long)
+    masks = torch.zeros(256, 5, dtype=torch.bool)
+
+    edge_to_index = {
+        (0, 1): 0,
+        (2, 3): 1,
+        (4, 5): 2,
+        (6, 7): 3,
+        (0, 2): 4,
+        (1, 3): 5,
+        (4, 6): 6,
+        (5, 7): 7,
+        (0, 4): 8,
+        (1, 5): 9,
+        (2, 6): 10,
+        (3, 7): 11,
+    }
+
+    for i, case in enumerate(MC_TABLE):
+        for j, tri in enumerate(case):
+            for k, (c1, c2) in enumerate(zip(tri[::2], tri[1::2])):
+                cases[i, j, k] = edge_to_index[(c1, c2) if c1 < c2 else (c2, c1)]
+            masks[i, j] = True
+    return cases, masks
+
+
 RENDERER_CONFIG = {}
 
 
@@ -400,7 +881,12 @@ def renderer_model_original_checkpoint_to_diffusers_checkpoint(model, checkpoint
         }
     )
 
-    diffusers_checkpoint.update({"void.background": torch.tensor([0.0, 0.0, 0.0], dtype=torch.float32)})
+    diffusers_checkpoint.update({"void.background": model.state_dict()["void.background"]})
+
+    cases, masks = create_mc_lookup_table()
+
+    diffusers_checkpoint.update({"mesh_decoder.cases": cases})
+    diffusers_checkpoint.update({"mesh_decoder.masks": masks})
 
     return diffusers_checkpoint
 

setup.py

@@ -89,7 +89,7 @@ _deps = [
     "huggingface-hub>=0.13.2",
     "requests-mock==1.10.0",
     "importlib_metadata",
-    "invisible-watermark",
+    "invisible-watermark>=0.2.0",
     "isort>=5.5.4",
     "jax>=0.2.8,!=0.3.2",
     "jaxlib>=0.1.65",
@@ -232,7 +232,7 @@ install_requires = [
 
 setup(
     name="diffusers",
-    version="0.18.0.dev0",  # expected format is one of x.y.z.dev0, or x.y.z.rc1 or x.y.z (no to dashes, yes to dots)
+    version="0.19.0.dev0",  # expected format is one of x.y.z.dev0, or x.y.z.rc1 or x.y.z (no to dashes, yes to dots)
     description="Diffusers",
     long_description=open("README.md", "r", encoding="utf-8").read(),
     long_description_content_type="text/markdown",
@@ -245,7 +245,7 @@ setup(
     packages=find_packages("src"),
     include_package_data=True,
     python_requires=">=3.7.0",
-    install_requires=install_requires,
+    install_requires=list(install_requires),
     extras_require=extras,
     entry_points={"console_scripts": ["diffusers-cli=diffusers.commands.diffusers_cli:main"]},
     classifiers=[

src/diffusers/__init__.py

@@ -1,4 +1,4 @@
-__version__ = "0.18.0.dev0"
+__version__ = "0.19.0.dev0"
 
 from .configuration_utils import ConfigMixin
 from .utils import (
@@ -36,10 +36,13 @@ except OptionalDependencyNotAvailable:
     from .utils.dummy_pt_objects import *  # noqa F403
 else:
     from .models import (
+        AsymmetricAutoencoderKL,
         AutoencoderKL,
         ControlNetModel,
         ModelMixin,
+        MultiAdapter,
         PriorTransformer,
+        T2IAdapter,
         T5FilmDecoder,
         Transformer2DModel,
         UNet1DModel,
@@ -151,6 +154,7 @@ else:
         SemanticStableDiffusionPipeline,
         ShapEImg2ImgPipeline,
         ShapEPipeline,
+        StableDiffusionAdapterPipeline,
         StableDiffusionAttendAndExcitePipeline,
         StableDiffusionControlNetImg2ImgPipeline,
         StableDiffusionControlNetInpaintPipeline,
@@ -195,7 +199,12 @@ try:
 except OptionalDependencyNotAvailable:
     from .utils.dummy_torch_and_transformers_and_invisible_watermark_objects import *  # noqa F403
 else:
-    from .pipelines import StableDiffusionXLImg2ImgPipeline, StableDiffusionXLPipeline
+    from .pipelines import (
+        StableDiffusionXLControlNetPipeline,
+        StableDiffusionXLImg2ImgPipeline,
+        StableDiffusionXLInpaintPipeline,
+        StableDiffusionXLPipeline,
+    )
 
 try:
     if not (is_torch_available() and is_transformers_available() and is_k_diffusion_available()):

src/diffusers/commands/diffusers_cli.py

@@ -16,6 +16,7 @@
 from argparse import ArgumentParser
 
 from .env import EnvironmentCommand
+from .fp16_safetensors import FP16SafetensorsCommand
 
 
 def main():
@@ -24,6 +25,7 @@ def main():
 
     # Register commands
     EnvironmentCommand.register_subcommand(commands_parser)
+    FP16SafetensorsCommand.register_subcommand(commands_parser)
 
     # Let's go
     args = parser.parse_args()

src/diffusers/commands/fp16_safetensors.py

@@ -0,0 +1,138 @@
+# Copyright 2023 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.
+
+"""
+Usage example:
+    diffusers-cli fp16_safetensors --ckpt_id=openai/shap-e --fp16 --use_safetensors
+"""
+
+import glob
+import json
+from argparse import ArgumentParser, Namespace
+from importlib import import_module
+
+import huggingface_hub
+import torch
+from huggingface_hub import hf_hub_download
+from packaging import version
+
+from ..utils import is_safetensors_available, logging
+from . import BaseDiffusersCLICommand
+
+
+def conversion_command_factory(args: Namespace):
+    return FP16SafetensorsCommand(
+        args.ckpt_id,
+        args.fp16,
+        args.use_safetensors,
+        args.use_auth_token,
+    )
+
+
+class FP16SafetensorsCommand(BaseDiffusersCLICommand):
+    @staticmethod
+    def register_subcommand(parser: ArgumentParser):
+        conversion_parser = parser.add_parser("fp16_safetensors")
+        conversion_parser.add_argument(
+            "--ckpt_id",
+            type=str,
+            help="Repo id of the checkpoints on which to run the conversion. Example: 'openai/shap-e'.",
+        )
+        conversion_parser.add_argument(
+            "--fp16", action="store_true", help="If serializing the variables in FP16 precision."
+        )
+        conversion_parser.add_argument(
+            "--use_safetensors", action="store_true", help="If serializing in the safetensors format."
+        )
+        conversion_parser.add_argument(
+            "--use_auth_token",
+            action="store_true",
+            help="When working with checkpoints having private visibility. When used `huggingface-cli login` needs to be run beforehand.",
+        )
+        conversion_parser.set_defaults(func=conversion_command_factory)
+
+    def __init__(self, ckpt_id: str, fp16: bool, use_safetensors: bool, use_auth_token: bool):
+        self.logger = logging.get_logger("diffusers-cli/fp16_safetensors")
+        self.ckpt_id = ckpt_id
+        self.local_ckpt_dir = f"/tmp/{ckpt_id}"
+        self.fp16 = fp16
+
+        if is_safetensors_available():
+            self.use_safetensors = use_safetensors
+        else:
+            raise ImportError(
+                "When `use_safetensors` is set to True, the `safetensors` library needs to be installed. Install it via `pip install safetensors`."
+            )
+
+        if not self.use_safetensors and not self.fp16:
+            raise NotImplementedError(
+                "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(
+                "The huggingface_hub version must be >= 0.9.0 to use this command. Please update your huggingface_hub"
+                " installation."
+            )
+        else:
+            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)
+        with open(model_index, "r") as f:
+            pipeline_class_name = json.load(f)["_class_name"]
+        pipeline_class = getattr(import_module("diffusers"), pipeline_class_name)
+        self.logger.info(f"Pipeline class imported: {pipeline_class_name}.")
+
+        # 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
+        )
+        pipeline.save_pretrained(
+            self.local_ckpt_dir,
+            safe_serialization=True if self.use_safetensors else False,
+            variant="fp16" if self.fp16 else None,
+        )
+        self.logger.info(f"Pipeline locally saved to {self.local_ckpt_dir}.")
+
+        # Fetch all the paths.
+        if self.fp16:
+            modified_paths = glob.glob(f"{self.local_ckpt_dir}/*/*.fp16.*")
+        elif self.use_safetensors:
+            modified_paths = glob.glob(f"{self.local_ckpt_dir}/*/*.safetensors")
+
+        # Prepare for the PR.
+        commit_message = f"Serialize variables with FP16: {self.fp16} and safetensors: {self.use_safetensors}."
+        operations = []
+        for path in modified_paths:
+            operations.append(CommitOperationAdd(path_in_repo="/".join(path.split("/")[4:]), path_or_fileobj=path))
+
+        # Open the PR.
+        commit_description = (
+            "Variables converted by the [`diffusers`' `fp16_safetensors`"
+            " CLI](https://github.com/huggingface/diffusers/blob/main/src/diffusers/commands/fp16_safetensors.py)."
+        )
+        hub_pr_url = create_commit(
+            repo_id=self.ckpt_id,
+            operations=operations,
+            commit_message=commit_message,
+            commit_description=commit_description,
+            repo_type="model",
+            create_pr=True,
+        ).pr_url
+        self.logger.info(f"PR created here: {hub_pr_url}.")

src/diffusers/configuration_utils.py

@@ -607,7 +607,7 @@ def register_to_config(init):
 
         # Take note of the parameters that were not present in the loaded config
         if len(set(new_kwargs.keys()) - set(init_kwargs)) > 0:
-            new_kwargs["_use_default_values"] = set(new_kwargs.keys()) - set(init_kwargs)
+            new_kwargs["_use_default_values"] = list(set(new_kwargs.keys()) - set(init_kwargs))
 
         new_kwargs = {**config_init_kwargs, **new_kwargs}
         getattr(self, "register_to_config")(**new_kwargs)
@@ -655,7 +655,7 @@ def flax_register_to_config(cls):
 
         # Take note of the parameters that were not present in the loaded config
         if len(set(new_kwargs.keys()) - set(init_kwargs)) > 0:
-            new_kwargs["_use_default_values"] = set(new_kwargs.keys()) - set(init_kwargs)
+            new_kwargs["_use_default_values"] = list(set(new_kwargs.keys()) - set(init_kwargs))
 
         getattr(self, "register_to_config")(**new_kwargs)
         original_init(self, *args, **kwargs)

src/diffusers/dependency_versions_table.py

@@ -13,7 +13,7 @@ deps = {
     "huggingface-hub": "huggingface-hub>=0.13.2",
     "requests-mock": "requests-mock==1.10.0",
     "importlib_metadata": "importlib_metadata",
-    "invisible-watermark": "invisible-watermark",
+    "invisible-watermark": "invisible-watermark>=0.2.0",
     "isort": "isort>=5.5.4",
     "jax": "jax>=0.2.8,!=0.3.2",
     "jaxlib": "jaxlib>=0.1.65",

src/diffusers/models/README.md

@@ -1,3 +1,3 @@
 # Models
 
-For more detail on the models, please refer to the [docs](https://huggingface.co/docs/diffusers/api/models).
+For more detail on the models, please refer to the [docs](https://huggingface.co/docs/diffusers/api/models/overview).

src/diffusers/models/__init__.py

@@ -16,6 +16,8 @@ from ..utils import is_flax_available, is_torch_available
 
 
 if is_torch_available():
+    from .adapter import MultiAdapter, T2IAdapter
+    from .autoencoder_asym_kl import AsymmetricAutoencoderKL
     from .autoencoder_kl import AutoencoderKL
     from .controlnet import ControlNetModel
     from .dual_transformer_2d import DualTransformer2DModel

src/diffusers/models/adapter.py

@@ -0,0 +1,291 @@
+# 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.
+
+from typing import List, Optional
+
+import torch
+import torch.nn as nn
+
+from ..configuration_utils import ConfigMixin, register_to_config
+from .modeling_utils import ModelMixin
+from .resnet import Downsample2D
+
+
+class MultiAdapter(ModelMixin):
+    r"""
+    MultiAdapter is a wrapper model that contains multiple adapter models and merges their outputs according to
+    user-assigned weighting.
+
+    This model inherits from [`ModelMixin`]. Check the superclass documentation for the generic methods the library
+    implements for all the model (such as downloading or saving, etc.)
+
+    Parameters:
+        adapters (`List[T2IAdapter]`, *optional*, defaults to None):
+            A list of `T2IAdapter` model instances.
+    """
+
+    def __init__(self, adapters: List["T2IAdapter"]):
+        super(MultiAdapter, self).__init__()
+
+        self.num_adapter = len(adapters)
+        self.adapters = nn.ModuleList(adapters)
+
+    def forward(self, xs: torch.Tensor, adapter_weights: Optional[List[float]] = None) -> List[torch.Tensor]:
+        r"""
+        Args:
+            xs (`torch.Tensor`):
+                (batch, channel, height, width) input images for multiple adapter models concated along dimension 1,
+                `channel` should equal to `num_adapter` * "number of channel of image".
+            adapter_weights (`List[float]`, *optional*, defaults to None):
+                List of floats representing the weight which will be multiply to each adapter's output before adding
+                them together.
+        """
+        if adapter_weights is None:
+            adapter_weights = torch.tensor([1 / self.num_adapter] * self.num_adapter)
+        else:
+            adapter_weights = torch.tensor(adapter_weights)
+
+        if xs.shape[1] % self.num_adapter != 0:
+            raise ValueError(
+                f"Expecting multi-adapter's input have number of channel that cab be evenly divisible "
+                f"by num_adapter: {xs.shape[1]} % {self.num_adapter} != 0"
+            )
+        x_list = torch.chunk(xs, self.num_adapter, dim=1)
+        accume_state = None
+        for x, w, adapter in zip(x_list, adapter_weights, self.adapters):
+            features = adapter(x)
+            if accume_state is None:
+                accume_state = features
+            else:
+                for i in range(len(features)):
+                    accume_state[i] += w * features[i]
+        return accume_state
+
+
+class T2IAdapter(ModelMixin, ConfigMixin):
+    r"""
+    A simple ResNet-like model that accepts images containing control signals such as keyposes and depth. The model
+    generates multiple feature maps that are used as additional conditioning in [`UNet2DConditionModel`]. The model's
+    architecture follows the original implementation of
+    [Adapter](https://github.com/TencentARC/T2I-Adapter/blob/686de4681515662c0ac2ffa07bf5dda83af1038a/ldm/modules/encoders/adapter.py#L97)
+     and
+     [AdapterLight](https://github.com/TencentARC/T2I-Adapter/blob/686de4681515662c0ac2ffa07bf5dda83af1038a/ldm/modules/encoders/adapter.py#L235).
+
+    This model inherits from [`ModelMixin`]. Check the superclass documentation for the generic methods the library
+    implements for all the model (such as downloading or saving, etc.)
+
+    Parameters:
+        in_channels (`int`, *optional*, defaults to 3):
+            Number of channels of Aapter's input(*control image*). Set this parameter to 1 if you're using gray scale
+            image as *control image*.
+        channels (`List[int]`, *optional*, defaults to `(320, 640, 1280, 1280)`):
+            The number of channel of each downsample block's output hidden state. The `len(block_out_channels)` will
+            also determine the number of downsample blocks in the Adapter.
+        num_res_blocks (`int`, *optional*, defaults to 2):
+            Number of ResNet blocks in each downsample block
+    """
+
+    @register_to_config
+    def __init__(
+        self,
+        in_channels: int = 3,
+        channels: List[int] = [320, 640, 1280, 1280],
+        num_res_blocks: int = 2,
+        downscale_factor: int = 8,
+        adapter_type: str = "full_adapter",
+    ):
+        super().__init__()
+
+        if adapter_type == "full_adapter":
+            self.adapter = FullAdapter(in_channels, channels, num_res_blocks, downscale_factor)
+        elif adapter_type == "light_adapter":
+            self.adapter = LightAdapter(in_channels, channels, num_res_blocks, downscale_factor)
+        else:
+            raise ValueError(f"unknown adapter_type: {type}. Choose either 'full_adapter' or 'simple_adapter'")
+
+    def forward(self, x: torch.Tensor) -> List[torch.Tensor]:
+        return self.adapter(x)
+
+    @property
+    def total_downscale_factor(self):
+        return self.adapter.total_downscale_factor
+
+
+# full adapter
+
+
+class FullAdapter(nn.Module):
+    def __init__(
+        self,
+        in_channels: int = 3,
+        channels: List[int] = [320, 640, 1280, 1280],
+        num_res_blocks: int = 2,
+        downscale_factor: int = 8,
+    ):
+        super().__init__()
+
+        in_channels = in_channels * downscale_factor**2
+
+        self.unshuffle = nn.PixelUnshuffle(downscale_factor)
+        self.conv_in = nn.Conv2d(in_channels, channels[0], kernel_size=3, padding=1)
+
+        self.body = nn.ModuleList(
+            [
+                AdapterBlock(channels[0], channels[0], num_res_blocks),
+                *[
+                    AdapterBlock(channels[i - 1], channels[i], num_res_blocks, down=True)
+                    for i in range(1, len(channels))
+                ],
+            ]
+        )
+
+        self.total_downscale_factor = downscale_factor * 2 ** (len(channels) - 1)
+
+    def forward(self, x: torch.Tensor) -> List[torch.Tensor]:
+        x = self.unshuffle(x)
+        x = self.conv_in(x)
+
+        features = []
+
+        for block in self.body:
+            x = block(x)
+            features.append(x)
+
+        return features
+
+
+class AdapterBlock(nn.Module):
+    def __init__(self, in_channels, out_channels, num_res_blocks, down=False):
+        super().__init__()
+
+        self.downsample = None
+        if down:
+            self.downsample = Downsample2D(in_channels)
+
+        self.in_conv = None
+        if in_channels != out_channels:
+            self.in_conv = nn.Conv2d(in_channels, out_channels, kernel_size=1)
+
+        self.resnets = nn.Sequential(
+            *[AdapterResnetBlock(out_channels) for _ in range(num_res_blocks)],
+        )
+
+    def forward(self, x):
+        if self.downsample is not None:
+            x = self.downsample(x)
+
+        if self.in_conv is not None:
+            x = self.in_conv(x)
+
+        x = self.resnets(x)
+
+        return x
+
+
+class AdapterResnetBlock(nn.Module):
+    def __init__(self, channels):
+        super().__init__()
+        self.block1 = nn.Conv2d(channels, channels, kernel_size=3, padding=1)
+        self.act = nn.ReLU()
+        self.block2 = nn.Conv2d(channels, channels, kernel_size=1)
+
+    def forward(self, x):
+        h = x
+        h = self.block1(h)
+        h = self.act(h)
+        h = self.block2(h)
+
+        return h + x
+
+
+# light adapter
+
+
+class LightAdapter(nn.Module):
+    def __init__(
+        self,
+        in_channels: int = 3,
+        channels: List[int] = [320, 640, 1280],
+        num_res_blocks: int = 4,
+        downscale_factor: int = 8,
+    ):
+        super().__init__()
+
+        in_channels = in_channels * downscale_factor**2
+
+        self.unshuffle = nn.PixelUnshuffle(downscale_factor)
+
+        self.body = nn.ModuleList(
+            [
+                LightAdapterBlock(in_channels, channels[0], num_res_blocks),
+                *[
+                    LightAdapterBlock(channels[i], channels[i + 1], num_res_blocks, down=True)
+                    for i in range(len(channels) - 1)
+                ],
+                LightAdapterBlock(channels[-1], channels[-1], num_res_blocks, down=True),
+            ]
+        )
+
+        self.total_downscale_factor = downscale_factor * (2 ** len(channels))
+
+    def forward(self, x):
+        x = self.unshuffle(x)
+
+        features = []
+
+        for block in self.body:
+            x = block(x)
+            features.append(x)
+
+        return features
+
+
+class LightAdapterBlock(nn.Module):
+    def __init__(self, in_channels, out_channels, num_res_blocks, down=False):
+        super().__init__()
+        mid_channels = out_channels // 4
+
+        self.downsample = None
+        if down:
+            self.downsample = Downsample2D(in_channels)
+
+        self.in_conv = nn.Conv2d(in_channels, mid_channels, kernel_size=1)
+        self.resnets = nn.Sequential(*[LightAdapterResnetBlock(mid_channels) for _ in range(num_res_blocks)])
+        self.out_conv = nn.Conv2d(mid_channels, out_channels, kernel_size=1)
+
+    def forward(self, x):
+        if self.downsample is not None:
+            x = self.downsample(x)
+
+        x = self.in_conv(x)
+        x = self.resnets(x)
+        x = self.out_conv(x)
+
+        return x
+
+
+class LightAdapterResnetBlock(nn.Module):
+    def __init__(self, channels):
+        super().__init__()
+        self.block1 = nn.Conv2d(channels, channels, kernel_size=3, padding=1)
+        self.act = nn.ReLU()
+        self.block2 = nn.Conv2d(channels, channels, kernel_size=3, padding=1)
+
+    def forward(self, x):
+        h = x
+        h = self.block1(h)
+        h = self.act(h)
+        h = self.block2(h)
+
+        return h + x

src/diffusers/models/attention_processor.py

@@ -506,14 +506,14 @@ class AttnProcessor:
 
 
 class LoRALinearLayer(nn.Module):
-    def __init__(self, in_features, out_features, rank=4, network_alpha=None):
+    def __init__(self, in_features, out_features, rank=4, network_alpha=None, device=None, dtype=None):
         super().__init__()
 
         if rank > min(in_features, out_features):
             raise ValueError(f"LoRA rank {rank} must be less or equal than {min(in_features, out_features)}")
 
-        self.down = nn.Linear(in_features, rank, bias=False)
-        self.up = nn.Linear(rank, out_features, bias=False)
+        self.down = nn.Linear(in_features, rank, bias=False, device=device, dtype=dtype)
+        self.up = nn.Linear(rank, out_features, bias=False, device=device, dtype=dtype)
         # This value has the same meaning as the `--network_alpha` option in the kohya-ss trainer script.
         # See https://github.com/darkstorm2150/sd-scripts/blob/main/docs/train_network_README-en.md#execute-learning
         self.network_alpha = network_alpha
@@ -1096,7 +1096,6 @@ class AttnProcessor2_0:
         batch_size, sequence_length, _ = (
             hidden_states.shape if encoder_hidden_states is None else encoder_hidden_states.shape
         )
-        inner_dim = hidden_states.shape[-1]
 
         if attention_mask is not None:
             attention_mask = attn.prepare_attention_mask(attention_mask, sequence_length, batch_size)
@@ -1117,6 +1116,7 @@ class AttnProcessor2_0:
         key = attn.to_k(encoder_hidden_states)
         value = attn.to_v(encoder_hidden_states)
 
+        inner_dim = key.shape[-1]
         head_dim = inner_dim // attn.heads
 
         query = query.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)

src/diffusers/models/autoencoder_asym_kl.py

@@ -0,0 +1,180 @@
+# Copyright 2023 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.
+from typing import Optional, Tuple, Union
+
+import torch
+import torch.nn as nn
+
+from ..configuration_utils import ConfigMixin, register_to_config
+from ..utils import apply_forward_hook
+from .autoencoder_kl import AutoencoderKLOutput
+from .modeling_utils import ModelMixin
+from .vae import DecoderOutput, DiagonalGaussianDistribution, Encoder, MaskConditionDecoder
+
+
+class AsymmetricAutoencoderKL(ModelMixin, ConfigMixin):
+    r"""
+    Designing a Better Asymmetric VQGAN for StableDiffusion https://arxiv.org/abs/2306.04632 . A VAE model with KL loss
+    for encoding images into latents and decoding latent representations into images.
+
+    This model inherits from [`ModelMixin`]. Check the superclass documentation for it's generic methods implemented
+    for all models (such as downloading or saving).
+
+    Parameters:
+        in_channels (int, *optional*, defaults to 3): Number of channels in the input image.
+        out_channels (int,  *optional*, defaults to 3): Number of channels in the output.
+        down_block_types (`Tuple[str]`, *optional*, defaults to `("DownEncoderBlock2D",)`):
+            Tuple of downsample block types.
+        down_block_out_channels (`Tuple[int]`, *optional*, defaults to `(64,)`):
+            Tuple of down block output channels.
+        layers_per_down_block (`int`, *optional*, defaults to `1`):
+            Number layers for down block.
+        up_block_types (`Tuple[str]`, *optional*, defaults to `("UpDecoderBlock2D",)`):
+            Tuple of upsample block types.
+        up_block_out_channels (`Tuple[int]`, *optional*, defaults to `(64,)`):
+            Tuple of up block output channels.
+        layers_per_up_block (`int`, *optional*, defaults to `1`):
+            Number layers for up block.
+        act_fn (`str`, *optional*, defaults to `"silu"`): The activation function to use.
+        latent_channels (`int`, *optional*, defaults to 4): Number of channels in the latent space.
+        sample_size (`int`, *optional*, defaults to `32`): Sample input size.
+        norm_num_groups (`int`, *optional*, defaults to `32`):
+            Number of groups to use for the first normalization layer in ResNet blocks.
+        scaling_factor (`float`, *optional*, defaults to 0.18215):
+            The component-wise standard deviation of the trained latent space computed using the first batch of the
+            training set. This is used to scale the latent space to have unit variance when training the diffusion
+            model. The latents are scaled with the formula `z = z * scaling_factor` before being passed to the
+            diffusion model. When decoding, the latents are scaled back to the original scale with the formula: `z = 1
+            / scaling_factor * z`. For more details, refer to sections 4.3.2 and D.1 of the [High-Resolution Image
+            Synthesis with Latent Diffusion Models](https://arxiv.org/abs/2112.10752) paper.
+    """
+
+    @register_to_config
+    def __init__(
+        self,
+        in_channels: int = 3,
+        out_channels: int = 3,
+        down_block_types: Tuple[str] = ("DownEncoderBlock2D",),
+        down_block_out_channels: Tuple[int] = (64,),
+        layers_per_down_block: int = 1,
+        up_block_types: Tuple[str] = ("UpDecoderBlock2D",),
+        up_block_out_channels: Tuple[int] = (64,),
+        layers_per_up_block: int = 1,
+        act_fn: str = "silu",
+        latent_channels: int = 4,
+        norm_num_groups: int = 32,
+        sample_size: int = 32,
+        scaling_factor: float = 0.18215,
+    ) -> None:
+        super().__init__()
+
+        # pass init params to Encoder
+        self.encoder = Encoder(
+            in_channels=in_channels,
+            out_channels=latent_channels,
+            down_block_types=down_block_types,
+            block_out_channels=down_block_out_channels,
+            layers_per_block=layers_per_down_block,
+            act_fn=act_fn,
+            norm_num_groups=norm_num_groups,
+            double_z=True,
+        )
+
+        # pass init params to Decoder
+        self.decoder = MaskConditionDecoder(
+            in_channels=latent_channels,
+            out_channels=out_channels,
+            up_block_types=up_block_types,
+            block_out_channels=up_block_out_channels,
+            layers_per_block=layers_per_up_block,
+            act_fn=act_fn,
+            norm_num_groups=norm_num_groups,
+        )
+
+        self.quant_conv = nn.Conv2d(2 * latent_channels, 2 * latent_channels, 1)
+        self.post_quant_conv = nn.Conv2d(latent_channels, latent_channels, 1)
+
+        self.use_slicing = False
+        self.use_tiling = False
+
+    @apply_forward_hook
+    def encode(self, x: torch.FloatTensor, return_dict: bool = True) -> AutoencoderKLOutput:
+        h = self.encoder(x)
+        moments = self.quant_conv(h)
+        posterior = DiagonalGaussianDistribution(moments)
+
+        if not return_dict:
+            return (posterior,)
+
+        return AutoencoderKLOutput(latent_dist=posterior)
+
+    def _decode(
+        self,
+        z: torch.FloatTensor,
+        image: Optional[torch.FloatTensor] = None,
+        mask: Optional[torch.FloatTensor] = None,
+        return_dict: bool = True,
+    ) -> Union[DecoderOutput, torch.FloatTensor]:
+        z = self.post_quant_conv(z)
+        dec = self.decoder(z, image, mask)
+
+        if not return_dict:
+            return (dec,)
+
+        return DecoderOutput(sample=dec)
+
+    @apply_forward_hook
+    def decode(
+        self,
+        z: torch.FloatTensor,
+        image: Optional[torch.FloatTensor] = None,
+        mask: Optional[torch.FloatTensor] = None,
+        return_dict: bool = True,
+    ) -> Union[DecoderOutput, torch.FloatTensor]:
+        decoded = self._decode(z, image, mask).sample
+
+        if not return_dict:
+            return (decoded,)
+
+        return DecoderOutput(sample=decoded)
+
+    def forward(
+        self,
+        sample: torch.FloatTensor,
+        mask: Optional[torch.FloatTensor] = None,
+        sample_posterior: bool = False,
+        return_dict: bool = True,
+        generator: Optional[torch.Generator] = None,
+    ) -> Union[DecoderOutput, torch.FloatTensor]:
+        r"""
+        Args:
+            sample (`torch.FloatTensor`): Input sample.
+            mask (`torch.FloatTensor`, *optional*, defaults to `None`): Optional inpainting mask.
+            sample_posterior (`bool`, *optional*, defaults to `False`):
+                Whether to sample from the posterior.
+            return_dict (`bool`, *optional*, defaults to `True`):
+                Whether or not to return a [`DecoderOutput`] instead of a plain tuple.
+        """
+        x = sample
+        posterior = self.encode(x).latent_dist
+        if sample_posterior:
+            z = posterior.sample(generator=generator)
+        else:
+            z = posterior.mode()
+        dec = self.decode(z, sample, mask).sample
+
+        if not return_dict:
+            return (dec,)
+
+        return DecoderOutput(sample=dec)

src/diffusers/models/autoencoder_kl.py

@@ -18,6 +18,7 @@ import torch
 import torch.nn as nn
 
 from ..configuration_utils import ConfigMixin, register_to_config
+from ..loaders import FromOriginalVAEMixin
 from ..utils import BaseOutput, apply_forward_hook
 from .attention_processor import AttentionProcessor, AttnProcessor
 from .modeling_utils import ModelMixin
@@ -38,7 +39,7 @@ class AutoencoderKLOutput(BaseOutput):
     latent_dist: "DiagonalGaussianDistribution"
 
 
-class AutoencoderKL(ModelMixin, ConfigMixin):
+class AutoencoderKL(ModelMixin, ConfigMixin, FromOriginalVAEMixin):
     r"""
     A VAE model with KL loss for encoding images into latents and decoding latent representations into images.
 
@@ -64,6 +65,10 @@ class AutoencoderKL(ModelMixin, ConfigMixin):
             diffusion model. When decoding, the latents are scaled back to the original scale with the formula: `z = 1
             / scaling_factor * z`. For more details, refer to sections 4.3.2 and D.1 of the [High-Resolution Image
             Synthesis with Latent Diffusion Models](https://arxiv.org/abs/2112.10752) paper.
+        force_upcast (`bool`, *optional*, default to `True`):
+            If enabled it will force the VAE to run in float32 for high image resolution pipelines, such as SD-XL. VAE
+            can be fine-tuned / trained to a lower range without loosing too much precision in which case
+            `force_upcast` can be set to `False` - see: https://huggingface.co/madebyollin/sdxl-vae-fp16-fix
     """
 
     _supports_gradient_checkpointing = True
@@ -82,6 +87,7 @@ class AutoencoderKL(ModelMixin, ConfigMixin):
         norm_num_groups: int = 32,
         sample_size: int = 32,
         scaling_factor: float = 0.18215,
+        force_upcast: float = True,
     ):
         super().__init__()
 

src/diffusers/models/controlnet.py

@@ -19,9 +19,10 @@ from torch import nn
 from torch.nn import functional as F
 
 from ..configuration_utils import ConfigMixin, register_to_config
+from ..loaders import FromOriginalControlnetMixin
 from ..utils import BaseOutput, logging
 from .attention_processor import AttentionProcessor, AttnProcessor
-from .embeddings import TimestepEmbedding, Timesteps
+from .embeddings import TextImageProjection, TextImageTimeEmbedding, TextTimeEmbedding, TimestepEmbedding, Timesteps
 from .modeling_utils import ModelMixin
 from .unet_2d_blocks import (
     CrossAttnDownBlock2D,
@@ -100,7 +101,7 @@ class ControlNetConditioningEmbedding(nn.Module):
         return embedding
 
 
-class ControlNetModel(ModelMixin, ConfigMixin):
+class ControlNetModel(ModelMixin, ConfigMixin, FromOriginalControlnetMixin):
     """
     A ControlNet model.
 
@@ -131,12 +132,25 @@ class ControlNetModel(ModelMixin, ConfigMixin):
             The epsilon to use for the normalization.
         cross_attention_dim (`int`, defaults to 1280):
             The dimension of the cross attention features.
+        transformer_layers_per_block (`int` or `Tuple[int]`, *optional*, defaults to 1):
+            The number of transformer blocks of type [`~models.attention.BasicTransformerBlock`]. Only relevant for
+            [`~models.unet_2d_blocks.CrossAttnDownBlock2D`], [`~models.unet_2d_blocks.CrossAttnUpBlock2D`],
+            [`~models.unet_2d_blocks.UNetMidBlock2DCrossAttn`].
+        encoder_hid_dim (`int`, *optional*, defaults to None):
+            If `encoder_hid_dim_type` is defined, `encoder_hidden_states` will be projected from `encoder_hid_dim`
+            dimension to `cross_attention_dim`.
+        encoder_hid_dim_type (`str`, *optional*, defaults to `None`):
+            If given, the `encoder_hidden_states` and potentially other embeddings are down-projected to text
+            embeddings of dimension `cross_attention` according to `encoder_hid_dim_type`.
         attention_head_dim (`Union[int, Tuple[int]]`, defaults to 8):
             The dimension of the attention heads.
         use_linear_projection (`bool`, defaults to `False`):
         class_embed_type (`str`, *optional*, defaults to `None`):
             The type of class embedding to use which is ultimately summed with the time embeddings. Choose from None,
             `"timestep"`, `"identity"`, `"projection"`, or `"simple_projection"`.
+        addition_embed_type (`str`, *optional*, defaults to `None`):
+            Configures an optional embedding which will be summed with the time embeddings. Choose from `None` or
+            "text". "text" will use the `TextTimeEmbedding` layer.
         num_class_embeds (`int`, *optional*, defaults to 0):
             Input dimension of the learnable embedding matrix to be projected to `time_embed_dim`, when performing
             class conditioning with `class_embed_type` equal to `None`.
@@ -177,10 +191,15 @@ class ControlNetModel(ModelMixin, ConfigMixin):
         norm_num_groups: Optional[int] = 32,
         norm_eps: float = 1e-5,
         cross_attention_dim: int = 1280,
+        transformer_layers_per_block: Union[int, Tuple[int]] = 1,
+        encoder_hid_dim: Optional[int] = None,
+        encoder_hid_dim_type: Optional[str] = None,
         attention_head_dim: Union[int, Tuple[int]] = 8,
         num_attention_heads: Optional[Union[int, Tuple[int]]] = None,
         use_linear_projection: bool = False,
         class_embed_type: Optional[str] = None,
+        addition_embed_type: Optional[str] = None,
+        addition_time_embed_dim: Optional[int] = None,
         num_class_embeds: Optional[int] = None,
         upcast_attention: bool = False,
         resnet_time_scale_shift: str = "default",
@@ -188,6 +207,7 @@ class ControlNetModel(ModelMixin, ConfigMixin):
         controlnet_conditioning_channel_order: str = "rgb",
         conditioning_embedding_out_channels: Optional[Tuple[int]] = (16, 32, 96, 256),
         global_pool_conditions: bool = False,
+        addition_embed_type_num_heads=64,
     ):
         super().__init__()
 
@@ -215,6 +235,9 @@ class ControlNetModel(ModelMixin, ConfigMixin):
                 f"Must provide the same number of `num_attention_heads` as `down_block_types`. `num_attention_heads`: {num_attention_heads}. `down_block_types`: {down_block_types}."
             )
 
+        if isinstance(transformer_layers_per_block, int):
+            transformer_layers_per_block = [transformer_layers_per_block] * len(down_block_types)
+
         # input
         conv_in_kernel = 3
         conv_in_padding = (conv_in_kernel - 1) // 2
@@ -224,16 +247,43 @@ class ControlNetModel(ModelMixin, ConfigMixin):
 
         # time
         time_embed_dim = block_out_channels[0] * 4
-
         self.time_proj = Timesteps(block_out_channels[0], flip_sin_to_cos, freq_shift)
         timestep_input_dim = block_out_channels[0]
-
         self.time_embedding = TimestepEmbedding(
             timestep_input_dim,
             time_embed_dim,
             act_fn=act_fn,
         )
 
+        if encoder_hid_dim_type is None and encoder_hid_dim is not None:
+            encoder_hid_dim_type = "text_proj"
+            self.register_to_config(encoder_hid_dim_type=encoder_hid_dim_type)
+            logger.info("encoder_hid_dim_type defaults to 'text_proj' as `encoder_hid_dim` is defined.")
+
+        if encoder_hid_dim is None and encoder_hid_dim_type is not None:
+            raise ValueError(
+                f"`encoder_hid_dim` has to be defined when `encoder_hid_dim_type` is set to {encoder_hid_dim_type}."
+            )
+
+        if encoder_hid_dim_type == "text_proj":
+            self.encoder_hid_proj = nn.Linear(encoder_hid_dim, cross_attention_dim)
+        elif encoder_hid_dim_type == "text_image_proj":
+            # image_embed_dim DOESN'T have to be `cross_attention_dim`. To not clutter the __init__ too much
+            # they are set to `cross_attention_dim` here as this is exactly the required dimension for the currently only use
+            # case when `addition_embed_type == "text_image_proj"` (Kadinsky 2.1)`
+            self.encoder_hid_proj = TextImageProjection(
+                text_embed_dim=encoder_hid_dim,
+                image_embed_dim=cross_attention_dim,
+                cross_attention_dim=cross_attention_dim,
+            )
+
+        elif encoder_hid_dim_type is not None:
+            raise ValueError(
+                f"encoder_hid_dim_type: {encoder_hid_dim_type} must be None, 'text_proj' or 'text_image_proj'."
+            )
+        else:
+            self.encoder_hid_proj = None
+
         # class embedding
         if class_embed_type is None and num_class_embeds is not None:
             self.class_embedding = nn.Embedding(num_class_embeds, time_embed_dim)
@@ -257,6 +307,29 @@ class ControlNetModel(ModelMixin, ConfigMixin):
         else:
             self.class_embedding = None
 
+        if addition_embed_type == "text":
+            if encoder_hid_dim is not None:
+                text_time_embedding_from_dim = encoder_hid_dim
+            else:
+                text_time_embedding_from_dim = cross_attention_dim
+
+            self.add_embedding = TextTimeEmbedding(
+                text_time_embedding_from_dim, time_embed_dim, num_heads=addition_embed_type_num_heads
+            )
+        elif addition_embed_type == "text_image":
+            # text_embed_dim and image_embed_dim DON'T have to be `cross_attention_dim`. To not clutter the __init__ too much
+            # they are set to `cross_attention_dim` here as this is exactly the required dimension for the currently only use
+            # case when `addition_embed_type == "text_image"` (Kadinsky 2.1)`
+            self.add_embedding = TextImageTimeEmbedding(
+                text_embed_dim=cross_attention_dim, image_embed_dim=cross_attention_dim, time_embed_dim=time_embed_dim
+            )
+        elif addition_embed_type == "text_time":
+            self.add_time_proj = Timesteps(addition_time_embed_dim, flip_sin_to_cos, freq_shift)
+            self.add_embedding = TimestepEmbedding(projection_class_embeddings_input_dim, time_embed_dim)
+
+        elif addition_embed_type is not None:
+            raise ValueError(f"addition_embed_type: {addition_embed_type} must be None, 'text' or 'text_image'.")
+
         # control net conditioning embedding
         self.controlnet_cond_embedding = ControlNetConditioningEmbedding(
             conditioning_embedding_channels=block_out_channels[0],
@@ -291,6 +364,7 @@ class ControlNetModel(ModelMixin, ConfigMixin):
             down_block = get_down_block(
                 down_block_type,
                 num_layers=layers_per_block,
+                transformer_layers_per_block=transformer_layers_per_block[i],
                 in_channels=input_channel,
                 out_channels=output_channel,
                 temb_channels=time_embed_dim,
@@ -327,6 +401,7 @@ class ControlNetModel(ModelMixin, ConfigMixin):
         self.controlnet_mid_block = controlnet_block
 
         self.mid_block = UNetMidBlock2DCrossAttn(
+            transformer_layers_per_block=transformer_layers_per_block[-1],
             in_channels=mid_block_channel,
             temb_channels=time_embed_dim,
             resnet_eps=norm_eps,
@@ -356,7 +431,22 @@ class ControlNetModel(ModelMixin, ConfigMixin):
                 The UNet model weights to copy to the [`ControlNetModel`]. All configuration options are also copied
                 where applicable.
         """
+        transformer_layers_per_block = (
+            unet.config.transformer_layers_per_block if "transformer_layers_per_block" in unet.config else 1
+        )
+        encoder_hid_dim = unet.config.encoder_hid_dim if "encoder_hid_dim" in unet.config else None
+        encoder_hid_dim_type = unet.config.encoder_hid_dim_type if "encoder_hid_dim_type" in unet.config else None
+        addition_embed_type = unet.config.addition_embed_type if "addition_embed_type" in unet.config else None
+        addition_time_embed_dim = (
+            unet.config.addition_time_embed_dim if "addition_time_embed_dim" in unet.config else None
+        )
+
         controlnet = cls(
+            encoder_hid_dim=encoder_hid_dim,
+            encoder_hid_dim_type=encoder_hid_dim_type,
+            addition_embed_type=addition_embed_type,
+            addition_time_embed_dim=addition_time_embed_dim,
+            transformer_layers_per_block=transformer_layers_per_block,
             in_channels=unet.config.in_channels,
             flip_sin_to_cos=unet.config.flip_sin_to_cos,
             freq_shift=unet.config.freq_shift,
@@ -542,6 +632,7 @@ class ControlNetModel(ModelMixin, ConfigMixin):
         class_labels: Optional[torch.Tensor] = None,
         timestep_cond: Optional[torch.Tensor] = None,
         attention_mask: Optional[torch.Tensor] = None,
+        added_cond_kwargs: Optional[Dict[str, torch.Tensor]] = None,
         cross_attention_kwargs: Optional[Dict[str, Any]] = None,
         guess_mode: bool = False,
         return_dict: bool = True,
@@ -564,7 +655,9 @@ class ControlNetModel(ModelMixin, ConfigMixin):
                 Optional class labels for conditioning. Their embeddings will be summed with the timestep embeddings.
             timestep_cond (`torch.Tensor`, *optional*, defaults to `None`):
             attention_mask (`torch.Tensor`, *optional*, defaults to `None`):
-            cross_attention_kwargs(`dict[str]`, *optional*, defaults to `None`):
+            added_cond_kwargs (`dict`):
+                Additional conditions for the Stable Diffusion XL UNet.
+            cross_attention_kwargs (`dict[str]`, *optional*, defaults to `None`):
                 A kwargs dictionary that if specified is passed along to the `AttnProcessor`.
             guess_mode (`bool`, defaults to `False`):
                 In this mode, the ControlNet encoder tries its best to recognize the input content of the input even if
@@ -618,6 +711,7 @@ class ControlNetModel(ModelMixin, ConfigMixin):
         t_emb = t_emb.to(dtype=sample.dtype)
 
         emb = self.time_embedding(t_emb, timestep_cond)
+        aug_emb = None
 
         if self.class_embedding is not None:
             if class_labels is None:
@@ -629,6 +723,30 @@ class ControlNetModel(ModelMixin, ConfigMixin):
             class_emb = self.class_embedding(class_labels).to(dtype=self.dtype)
             emb = emb + class_emb
 
+        if "addition_embed_type" in self.config:
+            if self.config.addition_embed_type == "text":
+                aug_emb = self.add_embedding(encoder_hidden_states)
+
+            elif self.config.addition_embed_type == "text_time":
+                if "text_embeds" not in added_cond_kwargs:
+                    raise ValueError(
+                        f"{self.__class__} has the config param `addition_embed_type` set to 'text_time' which requires the keyword argument `text_embeds` to be passed in `added_cond_kwargs`"
+                    )
+                text_embeds = added_cond_kwargs.get("text_embeds")
+                if "time_ids" not in added_cond_kwargs:
+                    raise ValueError(
+                        f"{self.__class__} has the config param `addition_embed_type` set to 'text_time' which requires the keyword argument `time_ids` to be passed in `added_cond_kwargs`"
+                    )
+                time_ids = added_cond_kwargs.get("time_ids")
+                time_embeds = self.add_time_proj(time_ids.flatten())
+                time_embeds = time_embeds.reshape((text_embeds.shape[0], -1))
+
+                add_embeds = torch.concat([text_embeds, time_embeds], dim=-1)
+                add_embeds = add_embeds.to(emb.dtype)
+                aug_emb = self.add_embedding(add_embeds)
+
+        emb = emb + aug_emb if aug_emb is not None else emb
+
         # 2. pre-process
         sample = self.conv_in(sample)
 

src/diffusers/models/unet_1d_blocks.py

@@ -235,12 +235,12 @@ class OutConv1DBlock(nn.Module):
 
 
 class OutValueFunctionBlock(nn.Module):
-    def __init__(self, fc_dim, embed_dim):
+    def __init__(self, fc_dim, embed_dim, act_fn="mish"):
         super().__init__()
         self.final_block = nn.ModuleList(
             [
                 nn.Linear(fc_dim + embed_dim, fc_dim // 2),
-                nn.Mish(),
+                get_activation(act_fn),
                 nn.Linear(fc_dim // 2, 1),
             ]
         )
@@ -652,5 +652,5 @@ def get_out_block(*, out_block_type, num_groups_out, embed_dim, out_channels, ac
     if out_block_type == "OutConv1DBlock":
         return OutConv1DBlock(num_groups_out, out_channels, embed_dim, act_fn)
     elif out_block_type == "ValueFunction":
-        return OutValueFunctionBlock(fc_dim, embed_dim)
+        return OutValueFunctionBlock(fc_dim, embed_dim, act_fn)
     return None

src/diffusers/models/unet_2d_blocks.py

@@ -955,10 +955,13 @@ class CrossAttnDownBlock2D(nn.Module):
         attention_mask: Optional[torch.FloatTensor] = None,
         cross_attention_kwargs: Optional[Dict[str, Any]] = None,
         encoder_attention_mask: Optional[torch.FloatTensor] = None,
+        additional_residuals=None,
     ):
         output_states = ()
 
-        for resnet, attn in zip(self.resnets, self.attentions):
+        blocks = list(zip(self.resnets, self.attentions))
+
+        for i, (resnet, attn) in enumerate(blocks):
             if self.training and self.gradient_checkpointing:
 
                 def create_custom_forward(module, return_dict=None):
@@ -999,6 +1002,10 @@ class CrossAttnDownBlock2D(nn.Module):
                     return_dict=False,
                 )[0]
 
+            # apply additional residuals to the output of the last pair of resnet and attention blocks
+            if i == len(blocks) - 1 and additional_residuals is not None:
+                hidden_states = hidden_states + additional_residuals
+
             output_states = output_states + (hidden_states,)
 
         if self.downsamplers is not None: