Buckets:

hf-doc-build
/

doc

Files
xet
hf-doc-build/doc / diffusers /main /pt /using-diffusers /write_own_pipeline.html
mishig
3 days ago
download
raw
65.7 kB
<meta charset="utf-8" /><meta name="hf:doc:metadata" content="{&quot;title&quot;:&quot;Entendendo pipelines, modelos e agendadores&quot;,&quot;local&quot;:&quot;understanding-pipelines-models-and-schedulers&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;Desconstruir um pipeline básico&quot;,&quot;local&quot;:&quot;deconstruct-a-basic-pipeline&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Desconstrua o pipeline do Stable Diffusion&quot;,&quot;local&quot;:&quot;deconstruct-the-stable-diffusion-pipeline&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;Criar embeddings de texto&quot;,&quot;local&quot;:&quot;create-text-embeddings&quot;,&quot;sections&quot;:[],&quot;depth&quot;:3},{&quot;title&quot;:&quot;Criar ruído aleatório&quot;,&quot;local&quot;:&quot;create-random-noise&quot;,&quot;sections&quot;:[],&quot;depth&quot;:3},{&quot;title&quot;:&quot;Remova o ruído da imagem&quot;,&quot;local&quot;:&quot;denoise-the-image&quot;,&quot;sections&quot;:[],&quot;depth&quot;:3},{&quot;title&quot;:&quot;Decodifique a imagem&quot;,&quot;local&quot;:&quot;decode-the-image&quot;,&quot;sections&quot;:[],&quot;depth&quot;:3}],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Próximas etapas&quot;,&quot;local&quot;:&quot;next-steps&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2}],&quot;depth&quot;:1}"/>
		<link href="/docs/diffusers/main/en/_app/immutable/entry/start.0S0_iEoG.js" rel="modulepreload">
		<link href="/docs/diffusers/main/en/_app/immutable/chunks/I9eooN77.js" rel="modulepreload">
		<link href="/docs/diffusers/main/en/_app/immutable/chunks/Bupkjr9f.js" rel="modulepreload">
		<link href="/docs/diffusers/main/en/_app/immutable/entry/app.CPdOxCbw.js" rel="modulepreload">
		<link href="/docs/diffusers/main/en/_app/immutable/chunks/CyvZ3FS5.js" rel="modulepreload">
		<link href="/docs/diffusers/main/en/_app/immutable/chunks/l0QnnOED.js" rel="modulepreload">
		<link href="/docs/diffusers/main/en/_app/immutable/chunks/DnVIJfC5.js" rel="modulepreload">
		<link href="/docs/diffusers/main/en/_app/immutable/chunks/DsnmJJEf.js" rel="modulepreload">
		<link href="/docs/diffusers/main/en/_app/immutable/nodes/0.DteTvmQZ.js" rel="modulepreload">
		<link href="/docs/diffusers/main/en/_app/immutable/chunks/CoSrdQax.js" rel="modulepreload">
		<link href="/docs/diffusers/main/en/_app/immutable/nodes/410.CRip7-vY.js" rel="modulepreload">
		<link href="/docs/diffusers/main/en/_app/immutable/chunks/DSPDl1gz.js" rel="modulepreload">
		<link href="/docs/diffusers/main/en/_app/immutable/chunks/B7CN6qzM.js" rel="modulepreload">
		<!--1bhomej--><meta name="hf:doc:metadata" content="{&quot;title&quot;:&quot;Understanding pipelines, models and schedulers&quot;,&quot;local&quot;:&quot;understanding-pipelines-models-and-schedulers&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;Deconstruct a basic pipeline&quot;,&quot;local&quot;:&quot;deconstruct-a-basic-pipeline&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Deconstruct the Stable Diffusion pipeline&quot;,&quot;local&quot;:&quot;deconstruct-the-stable-diffusion-pipeline&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;Create text embeddings&quot;,&quot;local&quot;:&quot;create-text-embeddings&quot;,&quot;sections&quot;:[],&quot;depth&quot;:3},{&quot;title&quot;:&quot;Create random noise&quot;,&quot;local&quot;:&quot;create-random-noise&quot;,&quot;sections&quot;:[],&quot;depth&quot;:3},{&quot;title&quot;:&quot;Denoise the image&quot;,&quot;local&quot;:&quot;denoise-the-image&quot;,&quot;sections&quot;:[],&quot;depth&quot;:3},{&quot;title&quot;:&quot;Decode the image&quot;,&quot;local&quot;:&quot;decode-the-image&quot;,&quot;sections&quot;:[],&quot;depth&quot;:3}],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Next steps&quot;,&quot;local&quot;:&quot;next-steps&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2}],&quot;depth&quot;:1}"/><!---->
		<link href="/docs/diffusers/main/en/_app/immutable/assets/0.tn0RQdqM.css" rel="modulepreload"> <!--[--><!--[0--><!--[--><!--[0--><!--[--><p></p> <div class="items-center shrink-0 min-w-[100px] max-sm:min-w-[50px] justify-end ml-auto flex" style="float: inline-end; margin-inline-start: 10px; display: inline-flex; position: relative; z-index: 10;"><div dir="ltr" class="inline-flex rounded-md max-sm:rounded-sm"><button class="inline-flex items-center gap-1 h-7 max-sm:h-7 px-2 max-sm:px-1.5 text-sm font-medium text-gray-800 border border-r-0 rounded-l-md max-sm:rounded-l-sm border-gray-200 bg-white hover:shadow-inner dark:border-gray-850 dark:bg-gray-950 dark:text-gray-200 dark:hover:bg-gray-800" aria-live="polite"><span class="inline-flex items-center justify-center rounded-md p-0.5 max-sm:p-0 hover:text-gray-800 dark:hover:text-gray-200"><svg class="sm:size-3.5 size-3" 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><!----></span> <span dir="auto">Copy page</span></button> <button class="inline-flex items-center justify-center w-6 max-sm:w-5 h-7 max-sm:h-7 disabled:pointer-events-none text-sm text-gray-500 hover:text-gray-700 dark:hover:text-white rounded-r-md max-sm:rounded-r-sm border border-l transition border-gray-200 bg-white hover:shadow-inner dark:border-gray-850 dark:bg-gray-950 dark:text-gray-200 dark:hover:bg-gray-800" aria-haspopup="menu" aria-expanded="false" aria-label="Open copy menu"><svg class="transition-transform text-gray-400 overflow-visible sm:size-3.5 size-3 rotate-0" width="1em" height="1em" viewBox="0 0 12 7" fill="none" xmlns="http://www.w3.org/2000/svg"><path d="M1 1L6 6L11 1" stroke="currentColor"></path></svg><!----></button></div> <!--[-1--><!--]--></div><!----> <div class="flex space-x-1 " style="float: inline-end; margin-inline-start: 10px; display: inline-flex; position: relative; z-index: 10;" dir="ltr"><!----> <!--[1--><div class="relative colab-dropdown "><button class="   " type="button"><!--[0--><!--[-1--><img alt="Open In Colab" class="!m-0" src="https://colab.research.google.com/assets/colab-badge.svg"/><!--]--><!----><!--]--></button>  <!--[-1--><!--]--></div><!--]--> <!--[1--><div class="relative colab-dropdown "><button class="   " type="button"><!--[0--><!--[-1--><img alt="Open In Studio Lab" class="!m-0" src="https://studiolab.sagemaker.aws/studiolab.svg"/><!--]--><!----><!--]--></button>  <!--[-1--><!--]--></div><!--]--></div><!----> <!--[0--><h1 class="relative group"><a id="understanding-pipelines-models-and-schedulers" 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="#understanding-pipelines-models-and-schedulers"><span><svg 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><!---->Entendendo pipelines, modelos e agendadores<!----></span></h1><!--]--><!----> <p>🧨 O Diffusers foi projetado para ser uma caixa de ferramentas amigável e flexível para a construção de sistemas de difusão adaptados ao seu caso de uso. No núcleo da caixa de ferramentas estão os modelos e os agendadores. Embora o <a href="/docs/diffusers/main/pt/api/pipelines/overview#diffusers.DiffusionPipeline">DiffusionPipeline</a> agrupe esses componentes para facilitar o uso, você também pode desmembrar o pipeline e usar os modelos e os agendadores separadamente para criar novos sistemas de difusão.</p> <p>Neste tutorial, você aprenderá a usar modelos e agendadores para montar um sistema de difusão para inferência, começando com um pipeline básico e, em seguida, avançando para o pipeline do Stable Diffusion.</p> <!--[1--><h2 class="relative group"><a id="deconstruct-a-basic-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="#deconstruct-a-basic-pipeline"><span><svg 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><!---->Desconstruir um pipeline básico<!----></span></h2><!--]--><!----> <p>Uma pipeline é uma maneira rápida e fácil de executar um modelo para inferência, exigindo no máximo quatro linhas de código para gerar uma imagem:</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 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> Copiado</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> DDPMPipeline

<span class="hljs-meta">&gt;&gt;&gt; </span>ddpm = DDPMPipeline.from_pretrained(<span class="hljs-string">&quot;google/ddpm-cat-256&quot;</span>, use_safetensors=<span class="hljs-literal">True</span>).to(<span class="hljs-string">&quot;cuda&quot;</span>)  <span class="hljs-comment"># or &quot;mps&quot;, &quot;xpu&quot;, &quot;cpu&quot;</span>
<span class="hljs-meta">&gt;&gt;&gt; </span>image = ddpm(num_inference_steps=<span class="hljs-number">25</span>).images[<span class="hljs-number">0</span>]
<span class="hljs-meta">&gt;&gt;&gt; </span>image<!----></pre></div><!----> <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> <p>Foi super fácil, mas como o pipeline fez isso? Vamos decompor o pipeline e dar uma olhada no que está acontecendo por baixo dos panos.</p> <p>No exemplo acima, o pipeline contém um modelo <a href="/docs/diffusers/main/pt/api/models/unet2d#diffusers.UNet2DModel">UNet2DModel</a> e um <a href="/docs/diffusers/main/pt/api/schedulers/ddpm#diffusers.DDPMScheduler">DDPMScheduler</a>. O pipeline remove o ruído de uma imagem ao pegar ruído aleatório do tamanho da saída desejada e passá-lo pelo modelo várias vezes. Em cada timestep, o modelo prevê o <em>resíduo de ruído</em> e o scheduler usa isso para prever uma imagem com menos ruído. O pipeline repete esse processo até atingir o final do número especificado de etapas de inferência.</p> <p>Para recriar o pipeline com o modelo e o agendador separadamente, vamos escrever nosso próprio processo de remoção de ruído.</p> <ol><li>Carregue o modelo e o agendador:</li></ol> <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 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> Copiado</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> DDPMScheduler, UNet2DModel

<span class="hljs-meta">&gt;&gt;&gt; </span>scheduler = DDPMScheduler.from_pretrained(<span class="hljs-string">&quot;google/ddpm-cat-256&quot;</span>)
<span class="hljs-meta">&gt;&gt;&gt; </span>model = UNet2DModel.from_pretrained(<span class="hljs-string">&quot;google/ddpm-cat-256&quot;</span>, use_safetensors=<span class="hljs-literal">True</span>).to(<span class="hljs-string">&quot;cuda&quot;</span>)  <span class="hljs-comment"># or &quot;mps&quot;, &quot;xpu&quot;, &quot;cpu&quot;</span><!----></pre></div><!----> <ol start="2"><li>Defina o número de passos de tempo para executar o processo de remoção de ruído:</li></ol> <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 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> Copiado</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-meta">&gt;&gt;&gt; </span>scheduler.set_timesteps(<span class="hljs-number">50</span>)<!----></pre></div><!----> <ol start="3"><li>Definir os timesteps do scheduler cria um tensor com elementos igualmente espaçados, 50 neste exemplo. Cada elemento corresponde a um timestep no qual o modelo remove o ruído de uma imagem. Quando você criar o loop de remoção de ruído mais tarde, iterará sobre este tensor para remover o ruído de uma imagem:</li></ol> <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 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> Copiado</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-meta">&gt;&gt;&gt; </span>scheduler.timesteps
tensor([<span class="hljs-number">980</span>, <span class="hljs-number">960</span>, <span class="hljs-number">940</span>, <span class="hljs-number">920</span>, <span class="hljs-number">900</span>, <span class="hljs-number">880</span>, <span class="hljs-number">860</span>, <span class="hljs-number">840</span>, <span class="hljs-number">820</span>, <span class="hljs-number">800</span>, <span class="hljs-number">780</span>, <span class="hljs-number">760</span>, <span class="hljs-number">740</span>, <span class="hljs-number">720</span>,
    <span class="hljs-number">700</span>, <span class="hljs-number">680</span>, <span class="hljs-number">660</span>, <span class="hljs-number">640</span>, <span class="hljs-number">620</span>, <span class="hljs-number">600</span>, <span class="hljs-number">580</span>, <span class="hljs-number">560</span>, <span class="hljs-number">540</span>, <span class="hljs-number">520</span>, <span class="hljs-number">500</span>, <span class="hljs-number">480</span>, <span class="hljs-number">460</span>, <span class="hljs-number">440</span>,
    <span class="hljs-number">420</span>, <span class="hljs-number">400</span>, <span class="hljs-number">380</span>, <span class="hljs-number">360</span>, <span class="hljs-number">340</span>, <span class="hljs-number">320</span>, <span class="hljs-number">300</span>, <span class="hljs-number">280</span>, <span class="hljs-number">260</span>, <span class="hljs-number">240</span>, <span class="hljs-number">220</span>, <span class="hljs-number">200</span>, <span class="hljs-number">180</span>, <span class="hljs-number">160</span>,
    <span class="hljs-number">140</span>, <span class="hljs-number">120</span>, <span class="hljs-number">100</span>,  <span class="hljs-number">80</span>,  <span class="hljs-number">60</span>,  <span class="hljs-number">40</span>,  <span class="hljs-number">20</span>,   <span class="hljs-number">0</span>])<!----></pre></div><!----> <ol start="4"><li>Crie um ruído aleatório com a mesma forma que a saída desejada:</li></ol> <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 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> Copiado</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">import</span> torch

<span class="hljs-meta">&gt;&gt;&gt; </span>sample_size = model.config.sample_size
<span class="hljs-meta">&gt;&gt;&gt; </span>noise = torch.randn((<span class="hljs-number">1</span>, <span class="hljs-number">3</span>, sample_size, sample_size), device=<span class="hljs-string">&quot;cuda&quot;</span>)  <span class="hljs-comment"># or &quot;mps&quot;, &quot;xpu&quot;, &quot;cpu&quot;</span><!----></pre></div><!----> <ol start="5"><li>Agora escreva um loop para iterar pelos timesteps. Em cada timestep, o modelo executa uma passagem <a href="/docs/diffusers/main/pt/api/models/unet2d#diffusers.UNet2DModel.forward">UNet2DModel.forward()</a> e retorna o residual ruidoso. O método <a href="/docs/diffusers/main/pt/api/schedulers/ddpm#diffusers.DDPMScheduler.step">step()</a> do scheduler recebe o residual ruidoso, o timestep e a entrada, e prevê a imagem no timestep anterior. Essa saída se torna a próxima entrada do modelo no loop de remoção de ruído, e o processo se repete até atingir o final do array <code>timesteps</code>.</li></ol> <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 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> Copiado</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-built_in">input</span> = noise

<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">for</span> t <span class="hljs-keyword">in</span> scheduler.timesteps:
<span class="hljs-meta">... </span>    <span class="hljs-keyword">with</span> torch.no_grad():
<span class="hljs-meta">... </span>        noisy_residual = model(<span class="hljs-built_in">input</span>, t).sample
<span class="hljs-meta">... </span>    previous_noisy_sample = scheduler.step(noisy_residual, t, <span class="hljs-built_in">input</span>).prev_sample
<span class="hljs-meta">... </span>    <span class="hljs-built_in">input</span> = previous_noisy_sample<!----></pre></div><!----> <p>Este é todo o processo de remoção de ruído, e você pode usar esse mesmo padrão para escrever qualquer sistema de difusão.</p> <ol start="6"><li>A última etapa é converter a saída desruída em uma imagem:</li></ol> <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 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> Copiado</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">from</span> PIL <span class="hljs-keyword">import</span> Image
<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">import</span> numpy <span class="hljs-keyword">as</span> np

<span class="hljs-meta">&gt;&gt;&gt; </span>image = (<span class="hljs-built_in">input</span> / <span class="hljs-number">2</span> + <span class="hljs-number">0.5</span>).clamp(<span class="hljs-number">0</span>, <span class="hljs-number">1</span>).squeeze()
<span class="hljs-meta">&gt;&gt;&gt; </span>image = (image.permute(<span class="hljs-number">1</span>, <span class="hljs-number">2</span>, <span class="hljs-number">0</span>) * <span class="hljs-number">255</span>).<span class="hljs-built_in">round</span>().to(torch.uint8).cpu().numpy()
<span class="hljs-meta">&gt;&gt;&gt; </span>image = Image.fromarray(image)
<span class="hljs-meta">&gt;&gt;&gt; </span>image<!----></pre></div><!----> <p>Na próxima seção, você vai colocar suas habilidades à prova e analisar o pipeline mais complexo do Stable Diffusion. Os passos são mais ou menos os mesmos. Você vai inicializar os componentes necessários e definir o número de timesteps para criar um array <code>timestep</code>. O array <code>timestep</code> é usado no loop de remoção de ruído, e para cada elemento deste array, o modelo prevê uma imagem com menos ruído. O loop de remoção de ruído itera pelos <code>timestep</code>s e, em cada timestep, ele produz um residual com ruído e o scheduler o usa para prever uma imagem com menos ruído no timestep anterior. Esse processo é repetido até você chegar ao final do array <code>timestep</code>.</p> <p>Vamos testar!</p> <!--[1--><h2 class="relative group"><a id="deconstruct-the-stable-diffusion-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="#deconstruct-the-stable-diffusion-pipeline"><span><svg 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><!---->Desconstrua o pipeline do Stable Diffusion<!----></span></h2><!--]--><!----> <p>Stable Diffusion é um modelo de <em>latent diffusion</em> de texto para imagem. É chamado de modelo de latent diffusion porque trabalha com uma representação de menor dimensão da imagem, em vez do espaço de pixels real, o que o torna mais eficiente em termos de memória. O encoder comprime a imagem em uma representação menor, e um decoder converte a representação comprimida de volta em uma imagem. Para modelos de texto para imagem, você precisará de um tokenizer e de um encoder para gerar embeddings de texto. A partir do exemplo anterior, você já sabe que precisa de um modelo UNet e de um scheduler.</p> <p>Como você pode ver, isso já é mais complexo que o pipeline DDPM, que contém apenas um modelo UNet. O modelo Stable Diffusion possui três modelos pré-treinados separados.</p> <blockquote class="tip"><p>💡 Leia o blog <a href="https://huggingface.co/blog/stable_diffusion#how-does-stable-diffusion-work" rel="nofollow">Como o Stable Diffusion funciona?</a> para mais detalhes sobre como funcionam os modelos VAE, UNet e text encoder.</p></blockquote> <p>Agora que você sabe o que precisa para o pipeline do Stable Diffusion, carregue todos esses componentes com o método <a href="/docs/diffusers/main/pt/api/models/overview#diffusers.ModelMixin.from_pretrained">from_pretrained()</a>. Você pode encontrá-los no checkpoint pré-treinado <a href="https://huggingface.co/stable-diffusion-v1-5/stable-diffusion-v1-5" rel="nofollow"><code>stable-diffusion-v1-5/stable-diffusion-v1-5</code></a>, e cada componente é armazenado em uma subpasta separada:</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 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> Copiado</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">from</span> PIL <span class="hljs-keyword">import</span> Image
<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">import</span> torch
<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> CLIPTextModel, CLIPTokenizer
<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> AutoencoderKL, UNet2DConditionModel, PNDMScheduler

<span class="hljs-meta">&gt;&gt;&gt; </span>vae = AutoencoderKL.from_pretrained(<span class="hljs-string">&quot;CompVis/stable-diffusion-v1-4&quot;</span>, subfolder=<span class="hljs-string">&quot;vae&quot;</span>, use_safetensors=<span class="hljs-literal">True</span>)
<span class="hljs-meta">&gt;&gt;&gt; </span>tokenizer = CLIPTokenizer.from_pretrained(<span class="hljs-string">&quot;CompVis/stable-diffusion-v1-4&quot;</span>, subfolder=<span class="hljs-string">&quot;tokenizer&quot;</span>)
<span class="hljs-meta">&gt;&gt;&gt; </span>text_encoder = CLIPTextModel.from_pretrained(
<span class="hljs-meta">... </span>    <span class="hljs-string">&quot;CompVis/stable-diffusion-v1-4&quot;</span>, subfolder=<span class="hljs-string">&quot;text_encoder&quot;</span>, use_safetensors=<span class="hljs-literal">True</span>
<span class="hljs-meta">... </span>)
<span class="hljs-meta">&gt;&gt;&gt; </span>unet = UNet2DConditionModel.from_pretrained(
<span class="hljs-meta">... </span>    <span class="hljs-string">&quot;CompVis/stable-diffusion-v1-4&quot;</span>, subfolder=<span class="hljs-string">&quot;unet&quot;</span>, use_safetensors=<span class="hljs-literal">True</span>
<span class="hljs-meta">... </span>)<!----></pre></div><!----> <p>Em vez do <a href="/docs/diffusers/main/pt/api/schedulers/pndm#diffusers.PNDMScheduler">PNDMScheduler</a> padrão, substitua-o pelo <a href="/docs/diffusers/main/pt/api/schedulers/unipc#diffusers.UniPCMultistepScheduler">UniPCMultistepScheduler</a> para ver como é fácil plugar um scheduler diferente:</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 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> Copiado</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> UniPCMultistepScheduler

<span class="hljs-meta">&gt;&gt;&gt; </span>scheduler = UniPCMultistepScheduler.from_pretrained(<span class="hljs-string">&quot;CompVis/stable-diffusion-v1-4&quot;</span>, subfolder=<span class="hljs-string">&quot;scheduler&quot;</span>)<!----></pre></div><!----> <p>Para acelerar a inferência, mova os modelos para uma GPU, pois, ao contrário do agendador, eles possuem pesos treináveis:</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 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> Copiado</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-meta">&gt;&gt;&gt; </span>torch_device = <span class="hljs-string">&quot;cuda&quot;</span>  <span class="hljs-comment"># or &quot;mps&quot;, &quot;xpu&quot;, &quot;cpu&quot;</span>
<span class="hljs-meta">&gt;&gt;&gt; </span>vae.to(torch_device)
<span class="hljs-meta">&gt;&gt;&gt; </span>text_encoder.to(torch_device)
<span class="hljs-meta">&gt;&gt;&gt; </span>unet.to(torch_device)<!----></pre></div><!----> <!--[2--><h3 class="relative group"><a id="create-text-embeddings" 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="#create-text-embeddings"><span><svg 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><!---->Criar embeddings de texto<!----></span></h3><!--]--><!----> <p>O próximo passo é tokenizar o texto para gerar embeddings. O texto é usado para condicionar o modelo UNet e direcionar o processo de difusão para algo que se assemelhe ao prompt de entrada.</p> <blockquote class="tip"><p>💡 O parâmetro <code>guidance_scale</code> determina o quanto de peso deve ser dado ao prompt ao gerar uma imagem.</p></blockquote> <p>Sinta-se à vontade para escolher qualquer prompt que quiser, caso queira gerar algo diferente!</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 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> Copiado</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-meta">&gt;&gt;&gt; </span>prompt = [<span class="hljs-string">&quot;a photograph of an astronaut riding a horse&quot;</span>]
<span class="hljs-meta">&gt;&gt;&gt; </span>height = <span class="hljs-number">512</span>  <span class="hljs-comment"># default height of Stable Diffusion</span>
<span class="hljs-meta">&gt;&gt;&gt; </span>width = <span class="hljs-number">512</span>  <span class="hljs-comment"># default width of Stable Diffusion</span>
<span class="hljs-meta">&gt;&gt;&gt; </span>num_inference_steps = <span class="hljs-number">25</span>  <span class="hljs-comment"># Number of denoising steps</span>
<span class="hljs-meta">&gt;&gt;&gt; </span>guidance_scale = <span class="hljs-number">7.5</span>  <span class="hljs-comment"># Scale for classifier-free guidance</span>
<span class="hljs-meta">&gt;&gt;&gt; </span>generator = torch.manual_seed(<span class="hljs-number">0</span>)  <span class="hljs-comment"># Seed generator to create the initial latent noise</span>
<span class="hljs-meta">&gt;&gt;&gt; </span>batch_size = <span class="hljs-built_in">len</span>(prompt)<!----></pre></div><!----> <p>Tokenize o texto e gere os embeddings a partir do prompt:</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 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> Copiado</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-meta">&gt;&gt;&gt; </span>text_input = tokenizer(
<span class="hljs-meta">... </span>    prompt, padding=<span class="hljs-string">&quot;max_length&quot;</span>, max_length=tokenizer.model_max_length, truncation=<span class="hljs-literal">True</span>, return_tensors=<span class="hljs-string">&quot;pt&quot;</span>
<span class="hljs-meta">... </span>)

<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">with</span> torch.no_grad():
<span class="hljs-meta">... </span>    text_embeddings = text_encoder(text_input.input_ids.to(torch_device))[<span class="hljs-number">0</span>]<!----></pre></div><!----> <p>Você também precisará gerar os <em>embeddings de texto incondicionais</em>, que são os embeddings para o token de padding. Eles precisam ter a mesma forma (<code>batch_size</code> e <code>seq_length</code>) que os <code>text_embeddings</code> condicionais:</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 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> Copiado</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-meta">&gt;&gt;&gt; </span>max_length = text_input.input_ids.shape[-<span class="hljs-number">1</span>]
<span class="hljs-meta">&gt;&gt;&gt; </span>uncond_input = tokenizer([<span class="hljs-string">&quot;&quot;</span>] * batch_size, padding=<span class="hljs-string">&quot;max_length&quot;</span>, max_length=max_length, return_tensors=<span class="hljs-string">&quot;pt&quot;</span>)
<span class="hljs-meta">&gt;&gt;&gt; </span>uncond_embeddings = text_encoder(uncond_input.input_ids.to(torch_device))[<span class="hljs-number">0</span>]<!----></pre></div><!----> <p>Vamos concatenar os embeddings condicionais e incondicionais em um lote para evitar duas passadas de frente:</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 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> Copiado</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-meta">&gt;&gt;&gt; </span>text_embeddings = torch.cat([uncond_embeddings, text_embeddings])<!----></pre></div><!----> <!--[2--><h3 class="relative group"><a id="create-random-noise" 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="#create-random-noise"><span><svg 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><!---->Criar ruído aleatório<!----></span></h3><!--]--><!----> <p>Em seguida, gere um ruído aleatório inicial como ponto de partida para o processo de difusão. Essa é a representação latente da imagem, que será gradualmente desruidada. Neste ponto, a imagem <code>latent</code> é menor que o tamanho final da imagem, mas isso não é um problema, pois o modelo a transformará nas dimensões finais de 512x512 mais tarde.</p> <blockquote class="tip"><p>💡 A altura e a largura são divididas por 8 porque o modelo <code>vae</code> possui 3 camadas de redução de amostragem. Você pode verificar executando o seguinte:</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 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> Copiado</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-number">2</span> ** (<span class="hljs-built_in">len</span>(vae.config.block_out_channels) - <span class="hljs-number">1</span>) == <span class="hljs-number">8</span><!----></pre></div><!----></blockquote> <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 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> Copiado</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-meta">&gt;&gt;&gt; </span>latents = torch.randn(
<span class="hljs-meta">... </span>    (batch_size, unet.config.in_channels, height // <span class="hljs-number">8</span>, width // <span class="hljs-number">8</span>),
<span class="hljs-meta">... </span>    generator=generator,
<span class="hljs-meta">... </span>    device=torch_device,
<span class="hljs-meta">... </span>)<!----></pre></div><!----> <!--[2--><h3 class="relative group"><a id="denoise-the-image" 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="#denoise-the-image"><span><svg 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><!---->Remova o ruído da imagem<!----></span></h3><!--]--><!----> <p>Comece escalando a entrada com a distribuição inicial de ruído, <em>sigma</em>, o valor de escala do ruído, que é necessário para agendadores aprimorados como <a href="/docs/diffusers/main/pt/api/schedulers/unipc#diffusers.UniPCMultistepScheduler">UniPCMultistepScheduler</a>:</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 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> Copiado</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-meta">&gt;&gt;&gt; </span>latents = latents * scheduler.init_noise_sigma<!----></pre></div><!----> <p>A última etapa é criar o loop de remoção de ruído que transformará progressivamente o ruído puro em <code>latents</code> em uma imagem descrita pelo seu prompt. Lembre-se, o loop de remoção de ruído precisa fazer três coisas:</p> <ol><li>Defina os timesteps do scheduler a serem usados durante o desruidamento.</li> <li>Itere sobre os timesteps.</li> <li>Em cada passo de tempo, chame o modelo UNet para prever o resíduo de ruído e passe-o ao agendador para calcular a amostra ruidosa anterior.</li></ol> <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 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> Copiado</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">from</span> tqdm.auto <span class="hljs-keyword">import</span> tqdm

<span class="hljs-meta">&gt;&gt;&gt; </span>scheduler.set_timesteps(num_inference_steps)

<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">for</span> t <span class="hljs-keyword">in</span> tqdm(scheduler.timesteps):
<span class="hljs-meta">... </span>    <span class="hljs-comment"># expand the latents if we are doing classifier-free guidance to avoid doing two forward passes.</span>
<span class="hljs-meta">... </span>    latent_model_input = torch.cat([latents] * <span class="hljs-number">2</span>)

<span class="hljs-meta">... </span>    latent_model_input = scheduler.scale_model_input(latent_model_input, timestep=t)

<span class="hljs-meta">... </span>    <span class="hljs-comment"># predict the noise residual</span>
<span class="hljs-meta">... </span>    <span class="hljs-keyword">with</span> torch.no_grad():
<span class="hljs-meta">... </span>        noise_pred = unet(latent_model_input, t, encoder_hidden_states=text_embeddings).sample

<span class="hljs-meta">... </span>    <span class="hljs-comment"># perform guidance</span>
<span class="hljs-meta">... </span>    noise_pred_uncond, noise_pred_text = noise_pred.chunk(<span class="hljs-number">2</span>)
<span class="hljs-meta">... </span>    noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond)

<span class="hljs-meta">... </span>    <span class="hljs-comment"># compute the previous noisy sample x_t -&gt; x_t-1</span>
<span class="hljs-meta">... </span>    latents = scheduler.step(noise_pred, t, latents).prev_sample<!----></pre></div><!----> <!--[2--><h3 class="relative group"><a id="decode-the-image" 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="#decode-the-image"><span><svg 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><!---->Decodifique a imagem<!----></span></h3><!--]--><!----> <p>A etapa final é usar o <code>vae</code> para decodificar a representação latente em uma imagem e obter a saída decodificada com <code>sample</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 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> Copiado</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-comment"># scale and decode the image latents with vae</span>
latents = <span class="hljs-number">1</span> / <span class="hljs-number">0.18215</span> * latents
<span class="hljs-keyword">with</span> torch.no_grad():
    image = vae.decode(latents).sample<!----></pre></div><!----> <p>Por fim, converta a imagem para <code>PIL.Image</code> para ver a imagem gerada!</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 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> Copiado</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-meta">&gt;&gt;&gt; </span>image = (image / <span class="hljs-number">2</span> + <span class="hljs-number">0.5</span>).clamp(<span class="hljs-number">0</span>, <span class="hljs-number">1</span>).squeeze()
<span class="hljs-meta">&gt;&gt;&gt; </span>image = (image.permute(<span class="hljs-number">1</span>, <span class="hljs-number">2</span>, <span class="hljs-number">0</span>) * <span class="hljs-number">255</span>).to(torch.uint8).cpu().numpy()
<span class="hljs-meta">&gt;&gt;&gt; </span>image = Image.fromarray(image)
<span class="hljs-meta">&gt;&gt;&gt; </span>image<!----></pre></div><!----> <div class="flex justify-center"><img src="https://huggingface.co/blog/assets/98_stable_diffusion/stable_diffusion_k_lms.png"/></div> <!--[1--><h2 class="relative group"><a id="next-steps" 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="#next-steps"><span><svg 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><!---->Próximas etapas<!----></span></h2><!--]--><!----> <p>De pipelines básicos a complexos, você viu que tudo o que realmente precisa para escrever seu próprio sistema de difusão é um loop de desruidamento. O loop deve definir os timesteps do scheduler, iterar sobre eles e alternar entre chamar o modelo UNet para prever o resíduo de ruído e passá-lo ao scheduler para calcular a amostra ruidosa anterior.</p> <p>É exatamente para isso que o 🧨 Diffusers foi projetado: para tornar intuitivo e fácil escrever o seu próprio sistema de difusão usando modelos e agendadores.</p> <p>Para os próximos passos, sinta-se à vontade para:</p> <ul><li>Aprenda como <a href="../conceptual/contribution">construir e contribuir com um pipeline</a> no 🧨 Diffusers. Mal podemos esperar para ver o que você vai criar!</li> <li>Explore os <a href="../api/pipelines/overview">pipelines existentes</a> na biblioteca e veja se você consegue deconstruí-los e construir um pipeline do zero usando os modelos e os agendadores separadamente.</li></ul> <a class="!text-gray-400 !no-underline text-sm flex items-center not-prose mt-4" href="https://github.com/huggingface/diffusers/blob/main/docs/source/en/using-diffusers/write_own_pipeline.md" target="_blank"><svg class="mr-1" 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="M31,16l-7,7l-1.41-1.41L28.17,16l-5.58-5.59L24,9l7,7z"></path><path d="M1,16l7-7l1.41,1.41L3.83,16l5.58,5.59L8,23l-7-7z"></path><path d="M12.419,25.484L17.639,6.552l1.932,0.518L14.351,26.002z"></path></svg><!----> <span><span class="underline">Update</span> on GitHub</span></a><!----> <p></p><!--]--><!----><!--]--><!--]--><!--]--> <!--[-1--><!--]--><!--]-->
			
			<script>
				{
					__sveltekit_1qsdbwf = {
						base: "/docs/diffusers/main/en",
						assets: "/docs/diffusers/main/en"
					};
					const element = document.currentScript.parentElement;
					Promise.all([
						import("/docs/diffusers/main/en/_app/immutable/entry/start.0S0_iEoG.js"),
						import("/docs/diffusers/main/en/_app/immutable/entry/app.CPdOxCbw.js")
					]).then(([kit, app]) => {
						kit.start(app, element, {
							node_ids: [0, 410],
							data: [null,null],
							form: null,
							error: null
						});
					});
				}
			</script>
		
Xet Storage Details

Size:: 65.7 kB
Xet hash:: dd42612d6a827ae38d91e73d1852e670add734d9ab22a960586d56c20600ea6b
Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.