Buckets:

hf-doc-build/doc / diffusers /v0.19.3 /en /using-diffusers /contribute_pipeline.html
rtrm's picture
download
raw
29.4 kB
<meta charset="utf-8" /><meta http-equiv="content-security-policy" content=""><meta name="hf:doc:metadata" content="{&quot;local&quot;:&quot;how-to-contribute-a-community-pipeline&quot;,&quot;sections&quot;:[{&quot;local&quot;:&quot;initialize-the-pipeline&quot;,&quot;title&quot;:&quot;Initialize the pipeline&quot;},{&quot;local&quot;:&quot;define-the-forward-pass&quot;,&quot;title&quot;:&quot;Define the forward pass&quot;},{&quot;local&quot;:&quot;share-your-pipeline&quot;,&quot;title&quot;:&quot;Share your pipeline&quot;},{&quot;local&quot;:&quot;how-do-community-pipelines-work&quot;,&quot;title&quot;:&quot;How do community pipelines work?&quot;}],&quot;title&quot;:&quot;How to contribute a community pipeline&quot;}" data-svelte="svelte-1phssyn">
<link rel="modulepreload" href="/docs/diffusers/v0.19.3/en/_app/assets/pages/__layout.svelte-hf-doc-builder.css">
<link rel="modulepreload" href="/docs/diffusers/v0.19.3/en/_app/start-hf-doc-builder.js">
<link rel="modulepreload" href="/docs/diffusers/v0.19.3/en/_app/chunks/vendor-hf-doc-builder.js">
<link rel="modulepreload" href="/docs/diffusers/v0.19.3/en/_app/chunks/paths-hf-doc-builder.js">
<link rel="modulepreload" href="/docs/diffusers/v0.19.3/en/_app/pages/__layout.svelte-hf-doc-builder.js">
<link rel="modulepreload" href="/docs/diffusers/v0.19.3/en/_app/pages/using-diffusers/contribute_pipeline.mdx-hf-doc-builder.js">
<link rel="modulepreload" href="/docs/diffusers/v0.19.3/en/_app/chunks/Tip-hf-doc-builder.js">
<link rel="modulepreload" href="/docs/diffusers/v0.19.3/en/_app/chunks/IconCopyLink-hf-doc-builder.js">
<link rel="modulepreload" href="/docs/diffusers/v0.19.3/en/_app/chunks/CodeBlock-hf-doc-builder.js">
<h1 class="relative group"><a id="how-to-contribute-a-community-pipeline" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#how-to-contribute-a-community-pipeline"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a>
<span>How to contribute a community pipeline
</span></h1>
<div class="course-tip bg-gradient-to-br dark:bg-gradient-to-r before:border-green-500 dark:before:border-green-800 from-green-50 dark:from-gray-900 to-white dark:to-gray-950 border border-green-50 text-green-700 dark:text-gray-400"><p>💡 Take a look at GitHub Issue <a href="https://github.com/huggingface/diffusers/issues/841" rel="nofollow">#841</a> for more context about why we’re adding community pipelines to help everyone easily share their work without being slowed down.</p></div>
<p>Community pipelines allow you to add any additional features you’d like on top of the <a href="/docs/diffusers/v0.19.3/en/api/pipelines/overview#diffusers.DiffusionPipeline">DiffusionPipeline</a>. The main benefit of building on top of the <code>DiffusionPipeline</code> is anyone can load and use your pipeline by only adding one more argument, making it super easy for the community to access.</p>
<p>This guide will show you how to create a community pipeline and explain how they work. To keep things simple, you’ll create a “one-step” pipeline where the <code>UNet</code> does a single forward pass and calls the scheduler once.</p>
<h2 class="relative group"><a id="initialize-the-pipeline" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#initialize-the-pipeline"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a>
<span>Initialize the pipeline
</span></h2>
<p>You should start by creating a <code>one_step_unet.py</code> file for your community pipeline. In this file, create a pipeline class that inherits from the <a href="/docs/diffusers/v0.19.3/en/api/pipelines/overview#diffusers.DiffusionPipeline">DiffusionPipeline</a> to be able to load model weights and the scheduler configuration from the Hub. The one-step pipeline needs a <code>UNet</code> and a scheduler, so you’ll need to add these as arguments to the <code>__init__</code> function:</p>
<div class="code-block relative"><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg>
<div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div>
Copied</div></button></div>
<pre><!-- HTML_TAG_START --><span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> DiffusionPipeline
<span class="hljs-keyword">import</span> torch
<span class="hljs-keyword">class</span> <span class="hljs-title class_">UnetSchedulerOneForwardPipeline</span>(<span class="hljs-title class_ inherited__">DiffusionPipeline</span>):
<span class="hljs-keyword">def</span> <span class="hljs-title function_">__init__</span>(<span class="hljs-params">self, unet, scheduler</span>):
<span class="hljs-built_in">super</span>().__init__()<!-- HTML_TAG_END --></pre></div>
<p>To ensure your pipeline and its components (<code>unet</code> and <code>scheduler</code>) can be saved with <a href="/docs/diffusers/v0.19.3/en/api/pipelines/overview#diffusers.DiffusionPipeline.save_pretrained">save_pretrained()</a>, add them to the <code>register_modules</code> function:</p>
<div class="code-block relative"><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg>
<div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div>
Copied</div></button></div>
<pre><!-- HTML_TAG_START --> from diffusers import DiffusionPipeline
import torch
class UnetSchedulerOneForwardPipeline(DiffusionPipeline):
def __init__(self, unet, scheduler):
super().__init__()
<span class="hljs-addition">+ self.register_modules(unet=unet, scheduler=scheduler)</span><!-- HTML_TAG_END --></pre></div>
<p>Cool, the <code>__init__</code> step is done and you can move to the forward pass now! 🔥 </p>
<h2 class="relative group"><a id="define-the-forward-pass" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#define-the-forward-pass"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a>
<span>Define the forward pass
</span></h2>
<p>In the forward pass, which we recommend defining as <code>__call__</code>, you have complete creative freedom to add whatever feature you’d like. For our amazing one-step pipeline, create a random image and only call the <code>unet</code> and <code>scheduler</code> once by setting <code>timestep=1</code>:</p>
<div class="code-block relative"><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg>
<div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div>
Copied</div></button></div>
<pre><!-- HTML_TAG_START --> from diffusers import DiffusionPipeline
import torch
class UnetSchedulerOneForwardPipeline(DiffusionPipeline):
def __init__(self, unet, scheduler):
super().__init__()
self.register_modules(unet=unet, scheduler=scheduler)
<span class="hljs-addition">+ def __call__(self):</span>
<span class="hljs-addition">+ image = torch.randn(</span>
<span class="hljs-addition">+ (1, self.unet.config.in_channels, self.unet.config.sample_size, self.unet.config.sample_size),</span>
<span class="hljs-addition">+ )</span>
<span class="hljs-addition">+ timestep = 1</span>
<span class="hljs-addition">+ model_output = self.unet(image, timestep).sample</span>
<span class="hljs-addition">+ scheduler_output = self.scheduler.step(model_output, timestep, image).prev_sample</span>
<span class="hljs-addition">+ return scheduler_output</span><!-- HTML_TAG_END --></pre></div>
<p>That’s it! 🚀 You can now run this pipeline by passing a <code>unet</code> and <code>scheduler</code> to it:</p>
<div class="code-block relative"><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg>
<div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div>
Copied</div></button></div>
<pre><!-- HTML_TAG_START --><span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> DDPMScheduler, UNet2DModel
scheduler = DDPMScheduler()
unet = UNet2DModel()
pipeline = UnetSchedulerOneForwardPipeline(unet=unet, scheduler=scheduler)
output = pipeline()<!-- HTML_TAG_END --></pre></div>
<p>But what’s even better is you can load pre-existing weights into the pipeline if the pipeline structure is identical. For example, you can load the <a href="https://huggingface.co/google/ddpm-cifar10-32" rel="nofollow"><code>google/ddpm-cifar10-32</code></a> weights into the one-step pipeline:</p>
<div class="code-block relative"><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg>
<div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div>
Copied</div></button></div>
<pre><!-- HTML_TAG_START -->pipeline = UnetSchedulerOneForwardPipeline.from_pretrained(<span class="hljs-string">&quot;google/ddpm-cifar10-32&quot;</span>)
output = pipeline()<!-- HTML_TAG_END --></pre></div>
<h2 class="relative group"><a id="share-your-pipeline" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#share-your-pipeline"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a>
<span>Share your pipeline
</span></h2>
<p>Open a Pull Request on the 🧨 Diffusers <a href="https://github.com/huggingface/diffusers" rel="nofollow">repository</a> to add your awesome pipeline in <code>one_step_unet.py</code> to the <a href="https://github.com/huggingface/diffusers/tree/main/examples/community" rel="nofollow">examples/community</a> subfolder.</p>
<p>Once it is merged, anyone with <code>diffusers &gt;= 0.4.0</code> installed can use this pipeline magically 🪄 by specifying it in the <code>custom_pipeline</code> argument:</p>
<div class="code-block relative"><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg>
<div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div>
Copied</div></button></div>
<pre><!-- HTML_TAG_START --><span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> DiffusionPipeline
pipe = DiffusionPipeline.from_pretrained(<span class="hljs-string">&quot;google/ddpm-cifar10-32&quot;</span>, custom_pipeline=<span class="hljs-string">&quot;one_step_unet&quot;</span>)
pipe()<!-- HTML_TAG_END --></pre></div>
<p>Another way to share your community pipeline is to upload the <code>one_step_unet.py</code> file directly to your preferred <a href="https://huggingface.co/docs/hub/models-uploading" rel="nofollow">model repository</a> on the Hub. Instead of specifying the <code>one_step_unet.py</code> file, pass the model repository id to the <code>custom_pipeline</code> argument:</p>
<div class="code-block relative"><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg>
<div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div>
Copied</div></button></div>
<pre><!-- HTML_TAG_START --><span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> DiffusionPipeline
pipeline = DiffusionPipeline.from_pretrained(<span class="hljs-string">&quot;google/ddpm-cifar10-32&quot;</span>, custom_pipeline=<span class="hljs-string">&quot;stevhliu/one_step_unet&quot;</span>)<!-- HTML_TAG_END --></pre></div>
<p>Take a look at the following table to compare the two sharing workflows to help you decide the best option for you:</p>
<table><thead><tr><th></th>
<th>GitHub community pipeline</th>
<th>HF Hub community pipeline</th></tr></thead>
<tbody><tr><td>usage</td>
<td>same</td>
<td>same</td></tr>
<tr><td>review process</td>
<td>open a Pull Request on GitHub and undergo a review process from the Diffusers team before merging; may be slower</td>
<td>upload directly to a Hub repository without any review; this is the fastest workflow</td></tr>
<tr><td>visibility</td>
<td>included in the official Diffusers repository and documentation</td>
<td>included on your HF Hub profile and relies on your own usage/promotion to gain visibility</td></tr></tbody></table>
<div class="course-tip bg-gradient-to-br dark:bg-gradient-to-r before:border-green-500 dark:before:border-green-800 from-green-50 dark:from-gray-900 to-white dark:to-gray-950 border border-green-50 text-green-700 dark:text-gray-400"><p>💡 You can use whatever package you want in your community pipeline file - as long as the user has it installed, everything will work fine. Make sure you have one and only one pipeline class that inherits from <code>DiffusionPipeline</code> because this is automatically detected.</p></div>
<h2 class="relative group"><a id="how-do-community-pipelines-work" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#how-do-community-pipelines-work"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a>
<span>How do community pipelines work?
</span></h2>
<p>A community pipeline is a class that inherits from <a href="/docs/diffusers/v0.19.3/en/api/pipelines/overview#diffusers.DiffusionPipeline">DiffusionPipeline</a> which means:</p>
<ul><li>It can be loaded with the <code>custom_pipeline</code> argument.</li>
<li>The model weights and scheduler configuration are loaded from <code>pretrained_model_name_or_path</code>.</li>
<li>The code that implements a feature in the community pipeline is defined in a <code>pipeline.py</code> file.</li></ul>
<p>Sometimes you can’t load all the pipeline components weights from an official repository. In this case, the other components should be passed directly to the pipeline:</p>
<div class="code-block relative"><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg>
<div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div>
Copied</div></button></div>
<pre><!-- HTML_TAG_START --><span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> DiffusionPipeline
<span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> CLIPFeatureExtractor, CLIPModel
model_id = <span class="hljs-string">&quot;CompVis/stable-diffusion-v1-4&quot;</span>
clip_model_id = <span class="hljs-string">&quot;laion/CLIP-ViT-B-32-laion2B-s34B-b79K&quot;</span>
feature_extractor = CLIPFeatureExtractor.from_pretrained(clip_model_id)
clip_model = CLIPModel.from_pretrained(clip_model_id, torch_dtype=torch.float16)
pipeline = DiffusionPipeline.from_pretrained(
model_id,
custom_pipeline=<span class="hljs-string">&quot;clip_guided_stable_diffusion&quot;</span>,
clip_model=clip_model,
feature_extractor=feature_extractor,
scheduler=scheduler,
torch_dtype=torch.float16,
)<!-- HTML_TAG_END --></pre></div>
<p>The magic behind community pipelines is contained in the following code. It allows the community pipeline to be loaded from GitHub or the Hub, and it’ll be available to all 🧨 Diffusers packages.</p>
<div class="code-block relative"><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg>
<div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div>
Copied</div></button></div>
<pre><!-- HTML_TAG_START --><span class="hljs-comment"># 2. Load the pipeline class, if using custom module then load it from the hub</span>
<span class="hljs-comment"># if we load from explicit class, let&#x27;s use it</span>
<span class="hljs-keyword">if</span> custom_pipeline <span class="hljs-keyword">is</span> <span class="hljs-keyword">not</span> <span class="hljs-literal">None</span>:
pipeline_class = get_class_from_dynamic_module(
custom_pipeline, module_file=CUSTOM_PIPELINE_FILE_NAME, cache_dir=custom_pipeline
)
<span class="hljs-keyword">elif</span> cls != DiffusionPipeline:
pipeline_class = cls
<span class="hljs-keyword">else</span>:
diffusers_module = importlib.import_module(cls.__module__.split(<span class="hljs-string">&quot;.&quot;</span>)[<span class="hljs-number">0</span>])
pipeline_class = <span class="hljs-built_in">getattr</span>(diffusers_module, config_dict[<span class="hljs-string">&quot;_class_name&quot;</span>])<!-- HTML_TAG_END --></pre></div>
<script type="module" data-hydrate="mlb908">
import { start } from "/docs/diffusers/v0.19.3/en/_app/start-hf-doc-builder.js";
start({
target: document.querySelector('[data-hydrate="mlb908"]').parentNode,
paths: {"base":"/docs/diffusers/v0.19.3/en","assets":"/docs/diffusers/v0.19.3/en"},
session: {},
route: false,
spa: false,
trailing_slash: "never",
hydrate: {
status: 200,
error: null,
nodes: [
import("/docs/diffusers/v0.19.3/en/_app/pages/__layout.svelte-hf-doc-builder.js"),
import("/docs/diffusers/v0.19.3/en/_app/pages/using-diffusers/contribute_pipeline.mdx-hf-doc-builder.js")
],
params: {}
}
});
</script>

Xet Storage Details

Size:
29.4 kB
·
Xet hash:
419ce680a8552cbee9cab84877cae5c06802aaad43eadf942ab341bbab281537

Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.