AI/diffusers
1
0
Fork 0
mirror of https://github.com/huggingface/diffusers.git synced 2026-01-27 17:22:53 +03:00
Code Activity

[docs] Merge LoRAs (#7213)

Browse Source 
* merge loras

* feedback

* torch.compile

* feedback
This commit is contained in:
Steven Liu
2024-03-07 11:28:50 -08:00
committed by GitHub
parent f539497ab4
commit 3ce905c9d0
4 changed files with 301 additions and 204 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
docs/source/en/_toctree.yml
View File

@@ -18,7 +18,7 @@
- local: tutorials/basic_training
title: Train a diffusion model
- local: tutorials/using_peft_for_inference
title: Inference with PEFT
title: Load LoRAs for inference
- local: tutorials/fast_diffusion
title: Accelerate inference of text-to-image diffusion models
title: Tutorials
@@ -62,6 +62,8 @@
title: Textual inversion
- local: using-diffusers/ip_adapter
title: IP-Adapter
- local: using-diffusers/merge_loras
title: Merge LoRAs
- local: training/distributed_inference
title: Distributed inference with multiple GPUs
- local: using-diffusers/reusing_seeds

132
docs/source/en/tutorials/using_peft_for_inference.md
View File

@@ -14,19 +14,17 @@ specific language governing permissions and limitations under the License.
# Load LoRAs for inference
There are many adapters (with LoRAs being the most common type) trained in different styles to achieve different effects. You can even combine multiple adapters to create new and unique images. With the 🤗 [PEFT](https://huggingface.co/docs/peft/index) integration in 🤗 Diffusers, it is really easy to load and manage adapters for inference. In this guide, you'll learn how to use different adapters with [Stable Diffusion XL (SDXL)](../api/pipelines/stable_diffusion/stable_diffusion_xl) for inference.
There are many adapter types (with [LoRAs](https://huggingface.co/docs/peft/conceptual_guides/adapter#low-rank-adaptation-lora) being the most popular) trained in different styles to achieve different effects. You can even combine multiple adapters to create new and unique images.
Throughout this guide, you'll use LoRA as the main adapter technique, so we'll use the terms LoRA and adapter interchangeably. You should have some familiarity with LoRA, and if you don't, we welcome you to check out the [LoRA guide](https://huggingface.co/docs/peft/conceptual_guides/lora).
In this tutorial, you'll learn how to easily load and manage adapters for inference with the 🤗 [PEFT](https://huggingface.co/docs/peft/index) integration in 🤗 Diffusers. You'll use LoRA as the main adapter technique, so you'll see the terms LoRA and adapter used interchangeably.
Let's first install all the required libraries.
```bash
!pip install -q transformers accelerate
!pip install peft
!pip install diffusers
!pip install -q transformers accelerate peft diffusers
```
Now, let's load a pipeline with a SDXL checkpoint:
Now, load a pipeline with a [Stable Diffusion XL (SDXL)](../api/pipelines/stable_diffusion/stable_diffusion_xl) checkpoint:
```python
from diffusers import DiffusionPipeline
@@ -36,16 +34,13 @@ pipe_id = "stabilityai/stable-diffusion-xl-base-1.0"
pipe = DiffusionPipeline.from_pretrained(pipe_id, torch_dtype=torch.float16).to("cuda")
```
Next, load a LoRA checkpoint with the [`~diffusers.loaders.StableDiffusionXLLoraLoaderMixin.load_lora_weights`] method.
With the 🤗 PEFT integration, you can assign a specific `adapter_name` to the checkpoint, which let's you easily switch between different LoRA checkpoints. Let's call this adapter `"toy"`.
Next, load a [CiroN2022/toy-face](https://huggingface.co/CiroN2022/toy-face) adapter with the [`~diffusers.loaders.StableDiffusionXLLoraLoaderMixin.load_lora_weights`] method. With the 🤗 PEFT integration, you can assign a specific `adapter_name` to the checkpoint, which let's you easily switch between different LoRA checkpoints. Let's call this adapter `"toy"`.
```python
pipe.load_lora_weights("CiroN2022/toy-face", weight_name="toy_face_sdxl.safetensors", adapter_name="toy")
```
And then perform inference:
Make sure to include the token `toy_face` in the prompt and then you can perform inference:
```python
prompt = "toy_face of a hacker with a hoodie"
@@ -59,17 +54,16 @@ image
![toy-face](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/peft_integration/diffusers_peft_lora_inference_8_1.png)
With the `adapter_name` parameter, it is really easy to use another adapter for inference! Load the [nerijs/pixel-art-xl](https://huggingface.co/nerijs/pixel-art-xl) adapter that has been fine-tuned to generate pixel art images and call it `"pixel"`.
With the `adapter_name` parameter, it is really easy to use another adapter for inference! Load the [nerijs/pixel-art-xl](https://huggingface.co/nerijs/pixel-art-xl) adapter that has been fine-tuned to generate pixel art images, and let's call it `"pixel"`.
The pipeline automatically sets the first loaded adapter (`"toy"`) as the active adapter. But you can activate the `"pixel"` adapter with the [`~diffusers.loaders.UNet2DConditionLoadersMixin.set_adapters`] method as shown below:
The pipeline automatically sets the first loaded adapter (`"toy"`) as the active adapter, but you can activate the `"pixel"` adapter with the [`~diffusers.loaders.UNet2DConditionLoadersMixin.set_adapters`] method:
```python
pipe.load_lora_weights("nerijs/pixel-art-xl", weight_name="pixel-art-xl.safetensors", adapter_name="pixel")
pipe.set_adapters("pixel")
```
Let's now generate an image with the second adapter and check the result:
Make sure you include the token `pixel art` in your prompt to generate a pixel art image:
```python
prompt = "a hacker with a hoodie, pixel art"
@@ -81,29 +75,25 @@ image
![pixel-art](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/peft_integration/diffusers_peft_lora_inference_12_1.png)
## Combine multiple adapters
## Merge adapters
You can also perform multi-adapter inference where you combine different adapter checkpoints for inference.
You can also merge different adapter checkpoints for inference to blend their styles together.
Once again, use the [`~diffusers.loaders.UNet2DConditionLoadersMixin.set_adapters`] method to activate two LoRA checkpoints and specify the weight for how the checkpoints should be combined.
Once again, use the [`~diffusers.loaders.UNet2DConditionLoadersMixin.set_adapters`] method to activate the `pixel` and `toy` adapters and specify the weights for how they should be merged.
```python
pipe.set_adapters(["pixel", "toy"], adapter_weights=[0.5, 1.0])
```
Now that we have set these two adapters, let's generate an image from the combined adapters!
<Tip>
LoRA checkpoints in the diffusion community are almost always obtained with [DreamBooth](https://huggingface.co/docs/diffusers/main/en/training/dreambooth). DreamBooth training often relies on "trigger" words in the input text prompts in order for the generation results to look as expected. When you combine multiple LoRA checkpoints, it's important to ensure the trigger words for the corresponding LoRA checkpoints are present in the input text prompts.
</Tip>
The trigger words for [CiroN2022/toy-face](https://hf.co/CiroN2022/toy-face) and [nerijs/pixel-art-xl](https://hf.co/nerijs/pixel-art-xl) are found in their repositories.
Remember to use the trigger words for [CiroN2022/toy-face](https://hf.co/CiroN2022/toy-face) and [nerijs/pixel-art-xl](https://hf.co/nerijs/pixel-art-xl) (these are found in their repositories) in the prompt to generate an image.
```python
# Notice how the prompt is constructed.
prompt = "toy_face of a hacker with a hoodie, pixel art"
image = pipe(
prompt, num_inference_steps=30, cross_attention_kwargs={"scale": 1.0}, generator=torch.manual_seed(0)
@@ -113,15 +103,16 @@ image
![toy-face-pixel-art](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/peft_integration/diffusers_peft_lora_inference_16_1.png)
Impressive! As you can see, the model was able to generate an image that mixes the characteristics of both adapters.
Impressive! As you can see, the model generated an image that mixed the characteristics of both adapters.
If you want to go back to using only one adapter, use the [`~diffusers.loaders.UNet2DConditionLoadersMixin.set_adapters`] method to activate the `"toy"` adapter:
> [!TIP]
> Through its PEFT integration, Diffusers also offers more efficient merging methods which you can learn about in the [Merge LoRAs](../using-diffusers/merge_loras) guide!
To return to only using one adapter, use the [`~diffusers.loaders.UNet2DConditionLoadersMixin.set_adapters`] method to activate the `"toy"` adapter:
```python
# First, set the adapter.
pipe.set_adapters("toy")
# Then, run inference.
prompt = "toy_face of a hacker with a hoodie"
lora_scale= 0.9
image = pipe(
@@ -130,11 +121,7 @@ image = pipe(
image
```
![toy-face-again](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/peft_integration/diffusers_peft_lora_inference_18_1.png)
If you want to switch to only the base model, disable all LoRAs with the [`~diffusers.loaders.UNet2DConditionLoadersMixin.disable_lora`] method.
Or to disable all adapters entirely, use the [`~diffusers.loaders.UNet2DConditionLoadersMixin.disable_lora`] method to return the base model.
```python
pipe.disable_lora()
@@ -145,11 +132,9 @@ image = pipe(prompt, num_inference_steps=30, generator=torch.manual_seed(0)).ima
image
```
![no-lora](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/peft_integration/diffusers_peft_lora_inference_20_1.png)
## Manage active adapters
## Monitoring active adapters
You have attached multiple adapters in this tutorial, and if you're feeling a bit lost on what adapters have been attached to the pipeline's components, you can easily check the list of active adapters using the [`~diffusers.loaders.LoraLoaderMixin.get_active_adapters`] method:
You have attached multiple adapters in this tutorial, and if you're feeling a bit lost on what adapters have been attached to the pipeline's components, use the [`~diffusers.loaders.LoraLoaderMixin.get_active_adapters`] method to check the list of active adapters:
```py
active_adapters = pipe.get_active_adapters()
@@ -164,78 +149,3 @@ list_adapters_component_wise = pipe.get_list_adapters()
list_adapters_component_wise
{"text_encoder": ["toy", "pixel"], "unet": ["toy", "pixel"], "text_encoder_2": ["toy", "pixel"]}
```
## Compatibility with `torch.compile`
If you want to compile your model with `torch.compile` make sure to first fuse the LoRA weights into the base model and unload them.
```diff
pipe.load_lora_weights("nerijs/pixel-art-xl", weight_name="pixel-art-xl.safetensors", adapter_name="pixel")
pipe.load_lora_weights("CiroN2022/toy-face", weight_name="toy_face_sdxl.safetensors", adapter_name="toy")
pipe.set_adapters(["pixel", "toy"], adapter_weights=[0.5, 1.0])
# Fuses the LoRAs into the Unet
pipe.fuse_lora()
pipe.unload_lora_weights()
+ pipe.unet.to(memory_format=torch.channels_last)
+ pipe.unet = torch.compile(pipe.unet, mode="reduce-overhead", fullgraph=True)
prompt = "toy_face of a hacker with a hoodie, pixel art"
image = pipe(prompt, num_inference_steps=30, generator=torch.manual_seed(0)).images[0]
```
> [!TIP]
> You can refer to the `torch.compile()` section [here](https://huggingface.co/docs/diffusers/main/en/optimization/torch2.0#torchcompile) and [here](https://huggingface.co/docs/diffusers/main/en/tutorials/fast_diffusion#torchcompile) for more elaborate examples.
## Fusing adapters into the model
You can use PEFT to easily fuse/unfuse multiple adapters directly into the model weights (both UNet and text encoder) using the [`~diffusers.loaders.LoraLoaderMixin.fuse_lora`] method, which can lead to a speed-up in inference and lower VRAM usage.
```py
pipe.load_lora_weights("nerijs/pixel-art-xl", weight_name="pixel-art-xl.safetensors", adapter_name="pixel")
pipe.load_lora_weights("CiroN2022/toy-face", weight_name="toy_face_sdxl.safetensors", adapter_name="toy")
pipe.set_adapters(["pixel", "toy"], adapter_weights=[0.5, 1.0])
# Fuses the LoRAs into the Unet
pipe.fuse_lora()
prompt = "toy_face of a hacker with a hoodie, pixel art"
image = pipe(prompt, num_inference_steps=30, generator=torch.manual_seed(0)).images[0]
# Gets the Unet back to the original state
pipe.unfuse_lora()
```
You can also fuse some adapters using `adapter_names` for faster generation:
```py
pipe.load_lora_weights("nerijs/pixel-art-xl", weight_name="pixel-art-xl.safetensors", adapter_name="pixel")
pipe.load_lora_weights("CiroN2022/toy-face", weight_name="toy_face_sdxl.safetensors", adapter_name="toy")
pipe.set_adapters(["pixel"], adapter_weights=[0.5, 1.0])
# Fuses the LoRAs into the Unet
pipe.fuse_lora(adapter_names=["pixel"])
prompt = "a hacker with a hoodie, pixel art"
image = pipe(prompt, num_inference_steps=30, generator=torch.manual_seed(0)).images[0]
# Gets the Unet back to the original state
pipe.unfuse_lora()
# Fuse all adapters
pipe.fuse_lora(adapter_names=["pixel", "toy"])
prompt = "toy_face of a hacker with a hoodie, pixel art"
image = pipe(prompt, num_inference_steps=30, generator=torch.manual_seed(0)).images[0]
```
## Saving a pipeline after fusing the adapters
To properly save a pipeline after it's been loaded with the adapters, it should be serialized like so:
```python
pipe.fuse_lora(lora_scale=1.0)
pipe.unload_lora_weights()
pipe.save_pretrained("path-to-pipeline")
```

103
docs/source/en/using-diffusers/loading_adapters.md
View File

@@ -103,7 +103,7 @@ image
<Tip>
LoRA is a very general training technique that can be used with other training methods. For example, it is common to train a model with DreamBooth and LoRA.
LoRA is a very general training technique that can be used with other training methods. For example, it is common to train a model with DreamBooth and LoRA. It is also increasingly common to load and merge multiple LoRAs to create new and unique images. You can learn more about it in the in-depth [Merge LoRAs](merge_loras) guide since merging is outside the scope of this loading guide.
</Tip>
@@ -165,101 +165,14 @@ To unload the LoRA weights, use the [`~loaders.LoraLoaderMixin.unload_lora_weigh
pipeline.unload_lora_weights()
```
### Load multiple LoRAs
It can be fun to use multiple LoRAs together to create something entirely new and unique. The [`~loaders.LoraLoaderMixin.fuse_lora`] method allows you to fuse the LoRA weights with the original weights of the underlying model.
<Tip>
Fusing the weights can lead to a speedup in inference latency because you don't need to separately load the base model and LoRA! You can save your fused pipeline with [`~DiffusionPipeline.save_pretrained`] to avoid loading and fusing the weights every time you want to use the model.
</Tip>
Load an initial model:
```py
from diffusers import StableDiffusionXLPipeline, AutoencoderKL
import torch
vae = AutoencoderKL.from_pretrained("madebyollin/sdxl-vae-fp16-fix", torch_dtype=torch.float16)
pipeline = StableDiffusionXLPipeline.from_pretrained(
"stabilityai/stable-diffusion-xl-base-1.0",
vae=vae,
torch_dtype=torch.float16,
).to("cuda")
```
Next, load the LoRA checkpoint and fuse it with the original weights. The `lora_scale` parameter controls how much to scale the output by with the LoRA weights. It is important to make the `lora_scale` adjustments in the [`~loaders.LoraLoaderMixin.fuse_lora`] method because it won't work if you try to pass `scale` to the `cross_attention_kwargs` in the pipeline.
If you need to reset the original model weights for any reason (use a different `lora_scale`), you should use the [`~loaders.LoraLoaderMixin.unfuse_lora`] method.
```py
pipeline.load_lora_weights("ostris/ikea-instructions-lora-sdxl")
pipeline.fuse_lora(lora_scale=0.7)
# to unfuse the LoRA weights
pipeline.unfuse_lora()
```
Then fuse this pipeline with the next set of LoRA weights:
```py
pipeline.load_lora_weights("ostris/super-cereal-sdxl-lora")
pipeline.fuse_lora(lora_scale=0.7)
```
<Tip warning={true}>
You can't unfuse multiple LoRA checkpoints, so if you need to reset the model to its original weights, you'll need to reload it.
</Tip>
Now you can generate an image that uses the weights from both LoRAs:
```py
prompt = "A cute brown bear eating a slice of pizza, stunning color scheme, masterpiece, illustration"
image = pipeline(prompt).images[0]
image
```
### 🤗 PEFT
<Tip>
Read the [Inference with 🤗 PEFT](../tutorials/using_peft_for_inference) tutorial to learn more about its integration with 🤗 Diffusers and how you can easily work with and juggle multiple adapters. You'll need to install 🤗 Diffusers and PEFT from source to run the example in this section.
</Tip>
Another way you can load and use multiple LoRAs is to specify the `adapter_name` parameter in [`~loaders.LoraLoaderMixin.load_lora_weights`]. This method takes advantage of the 🤗 PEFT integration. For example, load and name both LoRA weights:
```py
from diffusers import DiffusionPipeline
import torch
pipeline = DiffusionPipeline.from_pretrained("stabilityai/stable-diffusion-xl-base-1.0", torch_dtype=torch.float16).to("cuda")
pipeline.load_lora_weights("ostris/ikea-instructions-lora-sdxl", weight_name="ikea_instructions_xl_v1_5.safetensors", adapter_name="ikea")
pipeline.load_lora_weights("ostris/super-cereal-sdxl-lora", weight_name="cereal_box_sdxl_v1.safetensors", adapter_name="cereal")
```
Now use the [`~loaders.UNet2DConditionLoadersMixin.set_adapters`] to activate both LoRAs, and you can configure how much weight each LoRA should have on the output:
```py
pipeline.set_adapters(["ikea", "cereal"], adapter_weights=[0.7, 0.5])
```
Then, generate an image:
```py
prompt = "A cute brown bear eating a slice of pizza, stunning color scheme, masterpiece, illustration"
image = pipeline(prompt, num_inference_steps=30, cross_attention_kwargs={"scale": 1.0}).images[0]
image
```
### Kohya and TheLastBen
Other popular LoRA trainers from the community include those by [Kohya](https://github.com/kohya-ss/sd-scripts/) and [TheLastBen](https://github.com/TheLastBen/fast-stable-diffusion). These trainers create different LoRA checkpoints than those trained by 🤗 Diffusers, but they can still be loaded in the same way.
Let's download the [Blueprintify SD XL 1.0](https://civitai.com/models/150986/blueprintify-sd-xl-10) checkpoint from [Civitai](https://civitai.com/):
<hfoptions id="other-trainers">
<hfoption id="Kohya">
To load a Kohya LoRA, let's download the [Blueprintify SD XL 1.0](https://civitai.com/models/150986/blueprintify-sd-xl-10) checkpoint from [Civitai](https://civitai.com/) as an example:
```sh
!wget https://civitai.com/api/download/models/168776 -O blueprintify-sd-xl-10.safetensors
@@ -293,6 +206,9 @@ Some limitations of using Kohya LoRAs with 🤗 Diffusers include:
</Tip>
</hfoption>
<hfoption id="TheLastBen">
Loading a checkpoint from TheLastBen is very similar. For example, to load the [TheLastBen/William_Eggleston_Style_SDXL](https://huggingface.co/TheLastBen/William_Eggleston_Style_SDXL) checkpoint:
```py
@@ -308,6 +224,9 @@ image = pipeline(prompt=prompt).images[0]
image
```
</hfoption>
</hfoptions>
## IP-Adapter
[IP-Adapter](https://ip-adapter.github.io/) is a lightweight adapter that enables image prompting for any diffusion model. This adapter works by decoupling the cross-attention layers of the image and text features. All the other model components are frozen and only the embedded image features in the UNet are trained. As a result, IP-Adapter files are typically only ~100MBs.

266
docs/source/en/using-diffusers/merge_loras.md Normal file
View File

@@ -0,0 +1,266 @@
<!--Copyright 2024 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.
-->
# Merge LoRAs
It can be fun and creative to use multiple [LoRAs]((https://huggingface.co/docs/peft/conceptual_guides/adapter#low-rank-adaptation-lora)) together to generate something entirely new and unique. This works by merging multiple LoRA weights together to produce images that are a blend of different styles. Diffusers provides a few methods to merge LoRAs depending on *how* you want to merge their weights, which can affect image quality.
This guide will show you how to merge LoRAs using the [`~loaders.UNet2DConditionLoadersMixin.set_adapters`] and [`~peft.LoraModel.add_weighted_adapter`] methods. To improve inference speed and reduce memory-usage of merged LoRAs, you'll also see how to use the [`~loaders.LoraLoaderMixin.fuse_lora`] method to fuse the LoRA weights with the original weights of the underlying model.
For this guide, load a Stable Diffusion XL (SDXL) checkpoint and the [KappaNeuro/studio-ghibli-style]() and [Norod78/sdxl-chalkboarddrawing-lora]() LoRAs with the [`~loaders.LoraLoaderMixin.load_lora_weights`] method. You'll need to assign each LoRA an `adapter_name` to combine them later.
```py
from diffusers import DiffusionPipeline
import torch
pipeline = DiffusionPipeline.from_pretrained("stabilityai/stable-diffusion-xl-base-1.0", torch_dtype=torch.float16).to("cuda")
pipeline.load_lora_weights("ostris/ikea-instructions-lora-sdxl", weight_name="ikea_instructions_xl_v1_5.safetensors", adapter_name="ikea")
pipeline.load_lora_weights("lordjia/by-feng-zikai", weight_name="fengzikai_v1.0_XL.safetensors", adapter_name="feng")
```
## set_adapters
The [`~loaders.UNet2DConditionLoadersMixin.set_adapters`] method merges LoRA adapters by concatenating their weighted matrices. Use the adapter name to specify which LoRAs to merge, and the `adapter_weights` parameter to control the scaling for each LoRA. For example, if `adapter_weights=[0.5, 0.5]`, then the merged LoRA output is an average of both LoRAs. Try adjusting the adapter weights to see how it affects the generated image!
```py
pipeline.set_adapters(["ikea", "feng"], adapter_weights=[0.7, 0.8])
generator = torch.manual_seed(0)
prompt = "A bowl of ramen shaped like a cute kawaii bear, by Feng Zikai"
image = pipeline(prompt, generator=generator, cross_attention_kwargs={"scale": 1.0}).images[0]
image
```
<div class="flex justify-center">
<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/lora_merge_set_adapters.png"/>
</div>
## add_weighted_adapter
> [!WARNING]
> This is an experimental method that adds PEFTs [`~peft.LoraModel.add_weighted_adapter`] method to Diffusers to enable more efficient merging methods. Check out this [issue](https://github.com/huggingface/diffusers/issues/6892) if you're interested in learning more about the motivation and design behind this integration.
The [`~peft.LoraModel.add_weighted_adapter`] method provides access to more efficient merging method such as [TIES and DARE](https://huggingface.co/docs/peft/developer_guides/model_merging). To use these merging methods, make sure you have the latest stable version of Diffusers and PEFT installed.
```bash
pip install -U diffusers peft
```
There are three steps to merge LoRAs with the [`~peft.LoraModel.add_weighted_adapter`] method:
1. Create a [`~peft.PeftModel`] from the underlying model and LoRA checkpoint.
2. Load a base UNet model and the LoRA adapters.
3. Merge the adapters using the [`~peft.LoraModel.add_weighted_adapter`] method and the merging method of your choice.
Let's dive deeper into what these steps entail.
1. Load a UNet that corresponds to the UNet in the LoRA checkpoint. In this case, both LoRAs use the SDXL UNet as their base model.
```python
from diffusers import UNet2DConditionModel
import torch
unet = UNet2DConditionModel.from_pretrained(
"stabilityai/stable-diffusion-xl-base-1.0",
torch_dtype=torch.float16,
use_safetensors=True,
variant="fp16",
subfolder="unet",
).to("cuda")
```
Load the SDXL pipeline and the LoRA checkpoints, starting with the [ostris/ikea-instructions-lora-sdxl](https://huggingface.co/ostris/ikea-instructions-lora-sdxl) LoRA.
```python
from diffusers import DiffusionPipeline
pipeline = DiffusionPipeline.from_pretrained(
"stabilityai/stable-diffusion-xl-base-1.0",
variant="fp16",
torch_dtype=torch.float16,
unet=unet
).to("cuda")
pipeline.load_lora_weights("ostris/ikea-instructions-lora-sdxl", weight_name="ikea_instructions_xl_v1_5.safetensors", adapter_name="ikea")
```
Now you'll create a [`~peft.PeftModel`] from the loaded LoRA checkpoint by combining the SDXL UNet and the LoRA UNet from the pipeline.
```python
from peft import get_peft_model, LoraConfig
import copy
sdxl_unet = copy.deepcopy(unet)
ikea_peft_model = get_peft_model(
sdxl_unet,
pipeline.unet.peft_config["ikea"],
adapter_name="ikea"
)
original_state_dict = {f"base_model.model.{k}": v for k, v in pipeline.unet.state_dict().items()}
ikea_peft_model.load_state_dict(original_state_dict, strict=True)
```
> [!TIP]
> You can optionally push the ikea_peft_model to the Hub by calling `ikea_peft_model.push_to_hub("ikea_peft_model", token=TOKEN)`.
Repeat this process to create a [`~peft.PeftModel`] from the [lordjia/by-feng-zikai](https://huggingface.co/lordjia/by-feng-zikai) LoRA.
```python
pipeline.delete_adapters("ikea")
sdxl_unet.delete_adapters("ikea")
pipeline.load_lora_weights("lordjia/by-feng-zikai", weight_name="fengzikai_v1.0_XL.safetensors", adapter_name="feng")
pipeline.set_adapters(adapter_names="feng")
feng_peft_model = get_peft_model(
sdxl_unet,
pipeline.unet.peft_config["feng"],
adapter_name="feng"
)
original_state_dict = {f"base_model.model.{k}": v for k, v in pipe.unet.state_dict().items()}
feng_peft_model.load_state_dict(original_state_dict, strict=True)
```
2. Load a base UNet model and then load the adapters onto it.
```python
from peft import PeftModel
base_unet = UNet2DConditionModel.from_pretrained(
"stabilityai/stable-diffusion-xl-base-1.0",
torch_dtype=torch.float16,
use_safetensors=True,
variant="fp16",
subfolder="unet",
).to("cuda")
model = PeftModel.from_pretrained(base_unet, "stevhliu/ikea_peft_model", use_safetensors=True, subfolder="ikea", adapter_name="ikea")
model.load_adapter("stevhliu/feng_peft_model", use_safetensors=True, subfolder="feng", adapter_name="feng")
```
3. Merge the adapters using the [`~peft.LoraModel.add_weighted_adapter`] method and the merging method of your choice (learn more about other merging methods in this [blog post](https://huggingface.co/blog/peft_merging)). For this example, let's use the `"dare_linear"` method to merge the LoRAs.
> [!WARNING]
> Keep in mind the LoRAs need to have the same rank to be merged!
```python
model.add_weighted_adapter(
adapters=["ikea", "feng"],
weights=[1.0, 1.0],
combination_type="dare_linear",
adapter_name="ikea-feng"
)
model.set_adapters("ikea-feng")
```
Now you can generate an image with the merged LoRA.
```python
model = model.to(dtype=torch.float16, device="cuda")
pipeline = DiffusionPipeline.from_pretrained(
"stabilityai/stable-diffusion-xl-base-1.0", unet=model, variant="fp16", torch_dtype=torch.float16,
).to("cuda")
image = pipeline("A bowl of ramen shaped like a cute kawaii bear, by Feng Zikai", generator=torch.manual_seed(0)).images[0]
image
```
<div class="flex justify-center">
<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/ikea-feng-dare-linear.png"/>
</div>
## fuse_lora
Both the [`~loaders.UNet2DConditionLoadersMixin.set_adapters`] and [`~peft.LoraModel.add_weighted_adapter`] methods require loading the base model and the LoRA adapters separately which incurs some overhead. The [`~loaders.LoraLoaderMixin.fuse_lora`] method allows you to fuse the LoRA weights directly with the original weights of the underlying model. This way, you're only loading the model once which can increase inference and lower memory-usage.
You can use PEFT to easily fuse/unfuse multiple adapters directly into the model weights (both UNet and text encoder) using the [`~loaders.LoraLoaderMixin.fuse_lora`] method, which can lead to a speed-up in inference and lower VRAM usage.
For example, if you have a base model and adapters loaded and set as active with the following adapter weights:
```py
from diffusers import DiffusionPipeline
import torch
pipeline = DiffusionPipeline.from_pretrained("stabilityai/stable-diffusion-xl-base-1.0", torch_dtype=torch.float16).to("cuda")
pipeline.load_lora_weights("ostris/ikea-instructions-lora-sdxl", weight_name="ikea_instructions_xl_v1_5.safetensors", adapter_name="ikea")
pipeline.load_lora_weights("lordjia/by-feng-zikai", weight_name="fengzikai_v1.0_XL.safetensors", adapter_name="feng")
pipeline.set_adapters(["ikea", "feng"], adapter_weights=[0.7, 0.8])
```
Fuse these LoRAs into the UNet with the [`~loaders.LoraLoaderMixin.fuse_lora`] method. The `lora_scale` parameter controls how much to scale the output by with the LoRA weights. It is important to make the `lora_scale` adjustments in the [`~loaders.LoraLoaderMixin.fuse_lora`] method because it won’t work if you try to pass `scale` to the `cross_attention_kwargs` in the pipeline.
```py
pipeline.fuse_lora(adapter_names=["ikea", "feng"], lora_scale=1.0)
```
Then you should use [`~loaders.LoraLoaderMixin.unload_lora_weights`] to unload the LoRA weights since they've already been fused with the underlying base model. Finally, call [`~DiffusionPipeline.save_pretrained`] to save the fused pipeline locally or you could call [`~DiffusionPipeline.push_to_hub`] to push the fused pipeline to the Hub.
```py
pipeline.unload_lora_weights()
# save locally
pipeline.save_pretrained("path/to/fused-pipeline")
# save to the Hub
pipeline.push_to_hub("fused-ikea-feng")
```
Now you can quickly load the fused pipeline and use it for inference without needing to separately load the LoRA adapters.
```py
pipeline = DiffusionPipeline.from_pretrained(
"username/fused-ikea-feng", torch_dtype=torch.float16,
).to("cuda")
image = pipeline("A bowl of ramen shaped like a cute kawaii bear, by Feng Zikai", generator=torch.manual_seed(0)).images[0]
image
```
You can call [`~loaders.LoraLoaderMixin.unfuse_lora`] to restore the original model's weights (for example, if you want to use a different `lora_scale` value). However, this only works if you've only fused one LoRA adapter to the original model. If you've fused multiple LoRAs, you'll need to reload the model.
```py
pipeline.unfuse_lora()
```
### torch.compile
[torch.compile](../optimization/torch2.0#torchcompile) can speed up your pipeline even more, but the LoRA weights must be fused first and then unloaded. Typically, the UNet is compiled because it is such a computationally intensive component of the pipeline.
```py
from diffusers import DiffusionPipeline
import torch
# load base model and LoRAs
pipeline = DiffusionPipeline.from_pretrained("stabilityai/stable-diffusion-xl-base-1.0", torch_dtype=torch.float16).to("cuda")
pipeline.load_lora_weights("ostris/ikea-instructions-lora-sdxl", weight_name="ikea_instructions_xl_v1_5.safetensors", adapter_name="ikea")
pipeline.load_lora_weights("lordjia/by-feng-zikai", weight_name="fengzikai_v1.0_XL.safetensors", adapter_name="feng")
# activate both LoRAs and set adapter weights
pipeline.set_adapters(["ikea", "feng"], adapter_weights=[0.7, 0.8])
# fuse LoRAs and unload weights
pipeline.fuse_lora(adapter_names=["ikea", "feng"], lora_scale=1.0)
pipeline.unload_lora_weights()
# torch.compile
pipeline.unet.to(memory_format=torch.channels_last)
pipeline.unet = torch.compile(pipeline.unet, mode="reduce-overhead", fullgraph=True)
image = pipeline("A bowl of ramen shaped like a cute kawaii bear, by Feng Zikai", generator=torch.manual_seed(0)).images[0]
```
Learn more about torch.compile in the [Accelerate inference of text-to-image diffusion models](../tutorials/fast_diffusion#torchcompile) guide.
## Next steps
For more conceptual details about how each merging method works, take a look at the [🤗 PEFT welcomes new merging methods](https://huggingface.co/blog/peft_merging#concatenation-cat) blog post!