Buckets:

hf-doc-build
/

doc

Files
xet
hf-doc-build/doc / diffusers /main /pt /optimization /attention_backends.html
mishig
1 day ago
download
raw
37.3 kB
<meta charset="utf-8" /><meta name="hf:doc:metadata" content="{&quot;title&quot;:&quot;Backends de atenção&quot;,&quot;local&quot;:&quot;attention-backends&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;Defina um backend no modelo&quot;,&quot;local&quot;:&quot;set-a-backend-on-the-model&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Tente um backend temporariamente&quot;,&quot;local&quot;:&quot;try-a-backend-temporarily&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Confiança em kernels remotos&quot;,&quot;local&quot;:&quot;trusting-remote-kernels&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Verificações&quot;,&quot;local&quot;:&quot;checks&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Backends disponíveis&quot;,&quot;local&quot;:&quot;available-backends&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2}],&quot;depth&quot;:1}"/>
		<link href="/docs/diffusers/main/en/_app/immutable/entry/start.B6roO9Zr.js" rel="modulepreload">
		<link href="/docs/diffusers/main/en/_app/immutable/chunks/CjYFLaCl.js" rel="modulepreload">
		<link href="/docs/diffusers/main/en/_app/immutable/chunks/Cc4T7sti.js" rel="modulepreload">
		<link href="/docs/diffusers/main/en/_app/immutable/entry/app.CxxEVG8A.js" rel="modulepreload">
		<link href="/docs/diffusers/main/en/_app/immutable/chunks/C8F28shN.js" rel="modulepreload">
		<link href="/docs/diffusers/main/en/_app/immutable/chunks/BTnLXsn-.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.CmHT0Edy.js" rel="modulepreload">
		<link href="/docs/diffusers/main/en/_app/immutable/chunks/C7Ky3Ped.js" rel="modulepreload">
		<link href="/docs/diffusers/main/en/_app/immutable/nodes/7.CnCvnb8O.js" rel="modulepreload">
		<link href="/docs/diffusers/main/en/_app/immutable/chunks/D2YP2_0L.js" rel="modulepreload">
		<!--gxcprv--><meta name="hf:doc:metadata" content="{&quot;title&quot;:&quot;Attention backends&quot;,&quot;local&quot;:&quot;attention-backends&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;Set a backend on the model&quot;,&quot;local&quot;:&quot;set-a-backend-on-the-model&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Try a backend temporarily&quot;,&quot;local&quot;:&quot;try-a-backend-temporarily&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Trusting remote kernels&quot;,&quot;local&quot;:&quot;trusting-remote-kernels&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Checks&quot;,&quot;local&quot;:&quot;checks&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Available backends&quot;,&quot;local&quot;:&quot;available-backends&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><!----> <!--[0--><h1 class="relative group"><a id="attention-backends" 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="#attention-backends"><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><!---->Backends de atenção<!----></span></h1><!--]--><!----> <blockquote class="note"><p>O despachante de atenção é um recurso experimental. Abra uma issue se tiver algum feedback ou encontrar algum problema.</p></blockquote> <p>A maioria dos modelos transformer do Diffusers roteia a atenção por meio de um <em>dispatcher de atenção</em>, permitindo alternar entre backends otimizados por trás de uma única API. O dispatcher gerencia as implementações registradas e expõe um caminho de chamada unificado para elas. Alguns modelos, como muitos autoencoders, não utilizam o dispatcher devido à sua estrutura interna, e a troca do backend não afeta suas camadas de atenção.</p> <p>Consulte a tabela abaixo para uma visão geral das famílias de atenção disponíveis e a seção <a href="#available-backends">Backends disponíveis</a> para uma lista mais completa. O backend mais rápido depende do modelo, da GPU, do dtype e da forma da entrada.</p> <table><thead><tr><th>família de atenção</th><th>principal recurso</th></tr></thead><tbody><tr><td>FlashAttention</td><td>minimiza leituras/escritas de memória por meio de tiling e recomputação</td></tr><tr><td>AI Tensor Engine para ROCm</td><td>Implementação do FlashAttention otimizada para aceleradores AMD ROCm</td></tr><tr><td>SageAttention</td><td>quantiza a atenção para int8</td></tr><tr><td>FlexAttention</td><td>PyTorch FlexAttention</td></tr><tr><td>PyTorch nativo</td><td>implementação embutida do PyTorch usando <a href="./fp16#scaled-dot-product-attention">scaled_dot_product_attention</a></td></tr><tr><td>xFormers</td><td>atenção eficiente em memória com suporte para vários kernels de atenção</td></tr></tbody></table> <p>Os backends do Hub (os nomes <code>*_hub</code>) precisam da biblioteca <a href="https://github.com/huggingface/kernels" rel="nofollow">Kernels</a> e baixam o kernel no primeiro uso. Os backends do Hub do SageAttention também precisam de <code>DIFFUSERS_TRUST_REMOTE_KERNELS=true</code> (veja <a href="#trusting-remote-kernels">Confiança em kernels remotos</a>). Os outros backends precisam do próprio pacote, como <code>flash-attn</code> ou <code>sageattention</code>. A tabela <a href="#available-backends">Backends disponíveis</a> lista os requisitos que o Diffusers verifica ao você habilitar um backend.</p> <!--[1--><h2 class="relative group"><a id="set-a-backend-on-the-model" 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="#set-a-backend-on-the-model"><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><!---->Defina um backend no modelo<!----></span></h2><!--]--><!----> <p>O método <a href="/docs/diffusers/main/pt/api/models/overview#diffusers.ModelMixin.set_attention_backend">set_attention_backend()</a> percorre as camadas de atenção do modelo e aplica o backend escolhido em cada uma delas. Ele também define o backend ativo do dispatcher em todo o processo para o mesmo valor.</p> <p><a href="/docs/diffusers/main/pt/api/models/overview#diffusers.ModelMixin.reset_attention_backend">reset_attention_backend()</a> limpa o backend apenas nas camadas de atenção. Ele não limpa o backend ativo em todo o processo. Para uma troca temporária que restaura o backend ativo anterior ao sair, use o gerenciador de contexto <a href="#try-a-backend-temporarily">attention_backend</a>.</p> <p>O exemplo abaixo habilita <code>_flash_3_hub</code> (FlashAttention-3 do Hub) com <code>device_map="cuda"</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-keyword">import</span> torch
<span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> QwenImagePipeline

pipeline = QwenImagePipeline.from_pretrained(
    <span class="hljs-string">&quot;Qwen/Qwen-Image&quot;</span>, dtype=torch.bfloat16, device_map=<span class="hljs-string">&quot;cuda&quot;</span>
)
pipeline.transformer.set_attention_backend(<span class="hljs-string">&quot;_flash_3_hub&quot;</span>)

prompt = <span class="hljs-string">&quot;&quot;&quot;
cinematic film still of a cat sipping a margarita in a pool in Palm Springs, California
highly detailed, high budget hollywood movie, cinemascope, moody, epic, gorgeous, film grain
&quot;&quot;&quot;</span>
pipeline(prompt).images[<span class="hljs-number">0</span>]<!----></pre></div><!----> <blockquote class="note"><p>Os backends de FlashAttention-3 não-Hub (<code>_flash_3</code>, <code>_flash_varlen_3</code>) exigem a compilação do FlashAttention-3 a partir do código-fonte e serão descontinuados em breve. Use <code>_flash_3_hub</code> ou <code>_flash_3_varlen_hub</code> no lugar deles.</p></blockquote> <!--[1--><h2 class="relative group"><a id="try-a-backend-temporarily" 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="#try-a-backend-temporarily"><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><!---->Tente um backend temporariamente<!----></span></h2><!--]--><!----> <p>O gerenciador de contexto <code>attention_backend()</code> define o backend ativo em todo o processo durante a duração do bloco e restaura o backend anterior quando o bloco é encerrado. Use-o para testar um backend em uma única chamada sem deixar um backend permanente aplicado a partir de <a href="/docs/diffusers/main/pt/api/models/overview#diffusers.ModelMixin.set_attention_backend">set_attention_backend()</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-keyword">import</span> torch
<span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> QwenImagePipeline, attention_backend

pipeline = QwenImagePipeline.from_pretrained(
    <span class="hljs-string">&quot;Qwen/Qwen-Image&quot;</span>, dtype=torch.bfloat16, device_map=<span class="hljs-string">&quot;cuda&quot;</span>
)
prompt = <span class="hljs-string">&quot;&quot;&quot;
cinematic film still of a cat sipping a margarita in a pool in Palm Springs, California
highly detailed, high budget hollywood movie, cinemascope, moody, epic, gorgeous, film grain
&quot;&quot;&quot;</span>

<span class="hljs-keyword">with</span> attention_backend(<span class="hljs-string">&quot;_flash_3_hub&quot;</span>):
    image = pipeline(prompt).images[<span class="hljs-number">0</span>]<!----></pre></div><!----> <blockquote class="tip"><p>A maioria dos backends de atenção funciona com <code>torch.compile</code>. Se isso acelera seu pipeline depende do modelo e do backend. Consulte <a href="./fp16#torchcompile">Precisão e compilação</a>.</p></blockquote> <!--[1--><h2 class="relative group"><a id="trusting-remote-kernels" 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="#trusting-remote-kernels"><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><!---->Confiança em kernels remotos<!----></span></h2><!--]--><!----> <p>Os backends do Hub precisam primeiro da biblioteca <a href="https://github.com/huggingface/kernels" rel="nofollow">Kernels</a>, e em seguida o Diffusers busca o kernel do Hub no primeiro uso.</p> <p>Os backends de atenção do Hub baixam kernels de computação com a biblioteca <a href="https://github.com/huggingface/kernels" rel="nofollow">Kernels</a> e os executam localmente. A maioria dos nomes de atenção do Hub (<code>_flash_3_hub</code>, <code>flash_hub</code> e os outros backends <code>*_hub</code>) resolve para a organização <a href="https://huggingface.co/kernels-community" rel="nofollow">kernels-community</a>. Essa organização é um publicador confiável no Kernels, portanto o Diffusers carrega esses kernels de atenção sem definir <code>DIFFUSERS_TRUST_REMOTE_KERNELS</code>.</p> <blockquote class="note"><p>Os backends do SageAttention Hub (<code>sage_hub</code> e <code>sage_blackwell_hub</code>) são carregados a partir da organização <a href="https://huggingface.co/SageAttention" rel="nofollow">SageAttention</a>. Defina <code>DIFFUSERS_TRUST_REMOTE_KERNELS=true</code> para usá-los.</p></blockquote> <p>Outros recursos baseados em kernels, como <a href="../quantization/gguf">GGUF</a> e <a href="../quantization/nunchaku">Nunchaku Lite</a>, podem obter kernels de publicadores fora do kernels-community. Esses caminhos permanecem bloqueados, a menos que você opte por habilitá-los com <code>DIFFUSERS_TRUST_REMOTE_KERNELS</code>. Quando definido, o Diffusers encaminha <code>trust_remote_code=True</code> para o Kernels, permitindo que publicadores não confiáveis também sejam carregados.</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-bash "><!----><span class="hljs-built_in">export</span> DIFFUSERS_TRUST_REMOTE_KERNELS=<span class="hljs-literal">true</span><!----></pre></div><!----> <p>Ative isso apenas após inspecionar o repositório do kernel. Sem ele, carregar um kernel de um editor não confiável gera um erro. O Diffusers executa essa verificação por conta própria, portanto ela também se aplica a <code>kernels&lt;0.14.0</code>, que antecede o argumento <code>trust_remote_code</code>. Definir <code>DIFFUSERS_DISABLE_REMOTE_CODE=true</code> desativa o código remoto globalmente e tem precedência sobre <code>DIFFUSERS_TRUST_REMOTE_KERNELS</code>.</p> <!--[1--><h2 class="relative group"><a id="checks" 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="#checks"><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><!---->Verificações<!----></span></h2><!--]--><!----> <p>O despachador de atenção pode executar verificações de depuração antes de cada chamada de atenção despachada. Quais verificações são executadas depende das restrições registradas para o backend ativo.</p> <ol><li>As verificações de dispositivo confirmam que os tensores de consulta, chave e valor estão no mesmo dispositivo.</li> <li>As verificações de tipo de dados, quando registradas, confirmam a correspondência dos dtypes e frequentemente exigem <code>bfloat16</code> ou <code>float16</code>.</li> <li>As verificações de forma validam as dimensões dos tensores e impedem a mistura de máscaras de atenção com bandeiras causais.</li></ol> <p>Ative as verificações com a variável de ambiente <code>DIFFUSERS_ATTN_CHECKS</code>. As verificações adicionam sobrecarga, portanto, estão desativadas por padrão.</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-bash "><!----><span class="hljs-built_in">export</span> DIFFUSERS_ATTN_CHECKS=<span class="hljs-built_in">yes</span><!----></pre></div><!----> <p>Com as verificações ativadas, o Diffusers executa essas restrições antes de cada chamada de atenção despachada. O exemplo de baixo nível abaixo chama <code>dispatch_attention_fn</code> diretamente. A inferência de pipeline não precisa dessa importação. Ela só precisa do backend definido via <a href="/docs/diffusers/main/pt/api/models/overview#diffusers.ModelMixin.set_attention_backend">set_attention_backend()</a> ou <code>attention_backend()</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-keyword">import</span> torch
<span class="hljs-keyword">from</span> diffusers.models.attention_dispatch <span class="hljs-keyword">import</span> attention_backend, dispatch_attention_fn

query = torch.randn(<span class="hljs-number">1</span>, <span class="hljs-number">10</span>, <span class="hljs-number">8</span>, <span class="hljs-number">64</span>, dtype=torch.bfloat16, device=<span class="hljs-string">&quot;cuda&quot;</span>)
key = torch.randn(<span class="hljs-number">1</span>, <span class="hljs-number">10</span>, <span class="hljs-number">8</span>, <span class="hljs-number">64</span>, dtype=torch.bfloat16, device=<span class="hljs-string">&quot;cuda&quot;</span>)
value = torch.randn(<span class="hljs-number">1</span>, <span class="hljs-number">10</span>, <span class="hljs-number">8</span>, <span class="hljs-number">64</span>, dtype=torch.bfloat16, device=<span class="hljs-string">&quot;cuda&quot;</span>)

<span class="hljs-keyword">try</span>:
    <span class="hljs-keyword">with</span> attention_backend(<span class="hljs-string">&quot;flash&quot;</span>):
        output = dispatch_attention_fn(query, key, value)
        <span class="hljs-built_in">print</span>(<span class="hljs-string">&quot;✓ Flash Attention works with checks enabled&quot;</span>)
<span class="hljs-keyword">except</span> Exception <span class="hljs-keyword">as</span> e:
    <span class="hljs-built_in">print</span>(<span class="hljs-string">f&quot;✗ Flash Attention failed: <span class="hljs-subst">{e}</span>&quot;</span>)<!----></pre></div><!----> <!--[1--><h2 class="relative group"><a id="available-backends" 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="#available-backends"><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><!---->Backends disponíveis<!----></span></h2><!--]--><!----> <p>Consulte a tabela abaixo para obter a lista completa dos backends de atenção disponíveis e suas variantes. O Diffusers verifica a disponibilidade do pacote e as versões fixadas quando você habilita um backend.</p> <table><thead><tr><th>Nome do Backend</th><th>Família</th><th>Descrição</th><th>Pré-requisito</th></tr></thead><tbody><tr><td><code>native</code></td><td><a href="https://docs.pytorch.org/docs/stable/generated/torch.nn.attention.SDPBackend.html#torch.nn.attention.SDPBackend" rel="nofollow">PyTorch nativo</a></td><td>Backend padrão usando scaled_dot_product_attention do PyTorch</td><td>None</td></tr><tr><td><code>flex</code></td><td><a href="https://docs.pytorch.org/docs/stable/nn.attention.flex_attention.html#module-torch.nn.attention.flex_attention" rel="nofollow">FlexAttention</a></td><td>PyTorch FlexAttention</td><td><code>torch>=2.5.0</code></td></tr><tr><td><code>_native_cudnn</code></td><td><a href="https://docs.pytorch.org/docs/stable/generated/torch.nn.attention.SDPBackend.html#torch.nn.attention.SDPBackend" rel="nofollow">PyTorch nativo</a></td><td>Atenção otimizada pelo CuDNN</td><td>CUDA + CuDNN</td></tr><tr><td><code>_native_efficient</code></td><td><a href="https://docs.pytorch.org/docs/stable/generated/torch.nn.attention.SDPBackend.html#torch.nn.attention.SDPBackend" rel="nofollow">PyTorch nativo</a></td><td>Atenção eficiente em memória</td><td>Nenhuma além do PyTorch</td></tr><tr><td><code>_native_flash</code></td><td><a href="https://docs.pytorch.org/docs/stable/generated/torch.nn.attention.SDPBackend.html#torch.nn.attention.SDPBackend" rel="nofollow">PyTorch nativo</a></td><td>FlashAttention do PyTorch</td><td>CUDA</td></tr><tr><td><code>_native_math</code></td><td><a href="https://docs.pytorch.org/docs/stable/generated/torch.nn.attention.SDPBackend.html#torch.nn.attention.SDPBackend" rel="nofollow">PyTorch nativo</a></td><td>Atenção baseada em matemática (fallback)</td><td>None</td></tr><tr><td><code>_native_npu</code></td><td><a href="https://docs.pytorch.org/docs/stable/generated/torch.nn.attention.SDPBackend.html#torch.nn.attention.SDPBackend" rel="nofollow">PyTorch nativo</a></td><td>Atenção otimizada para NPU</td><td><code>torch_npu</code></td></tr><tr><td><code>_native_xla</code></td><td><a href="https://docs.pytorch.org/docs/stable/generated/torch.nn.attention.SDPBackend.html#torch.nn.attention.SDPBackend" rel="nofollow">PyTorch nativo</a></td><td>Atenção otimizada com XLA</td><td><code>torch_xla>=2.2</code></td></tr><tr><td><code>flash</code></td><td><a href="https://github.com/Dao-AILab/flash-attention" rel="nofollow">FlashAttention</a></td><td>FlashAttention-2</td><td><code>flash-attn>=2.6.3</code></td></tr><tr><td><code>flash_hub</code></td><td><a href="https://github.com/Dao-AILab/flash-attention" rel="nofollow">FlashAttention</a></td><td>FlashAttention-2 dos kernels do Hub</td><td><code>kernels>=0.12</code></td></tr><tr><td><code>flash_varlen</code></td><td><a href="https://github.com/Dao-AILab/flash-attention" rel="nofollow">FlashAttention</a></td><td>FlashAttention de comprimento variável</td><td><code>flash-attn>=2.6.3</code></td></tr><tr><td><code>flash_varlen_hub</code></td><td><a href="https://github.com/Dao-AILab/flash-attention" rel="nofollow">FlashAttention</a></td><td>FlashAttention de comprimento variável dos kernels do Hub</td><td><code>kernels>=0.12</code></td></tr><tr><td><code>aiter_fa2_hub</code></td><td><a href="https://github.com/ROCm/aiter" rel="nofollow">Motor de Tensor de IA para ROCm</a></td><td>FlashAttention-2 para AMD ROCm a partir dos kernels do Hub (<code>bfloat16</code>)</td><td><code>kernels>=0.12</code>, ROCm</td></tr><tr><td><code>flash_4_hub</code></td><td><a href="https://github.com/Dao-AILab/flash-attention" rel="nofollow">FlashAttention</a></td><td>FlashAttention-4 dos kernels do Hub</td><td><code>kernels>=0.12.3</code></td></tr><tr><td><code>_flash_3</code></td><td><a href="https://github.com/Dao-AILab/flash-attention" rel="nofollow">FlashAttention</a></td><td>FlashAttention-3 (local; em breve será descontinuado)</td><td>Compilar o FA3 a partir do código-fonte</td></tr><tr><td><code>_flash_varlen_3</code></td><td><a href="https://github.com/Dao-AILab/flash-attention" rel="nofollow">FlashAttention</a></td><td>FlashAttention-3 de comprimento variável (local; será descontinuado em breve)</td><td>Compilar o FA3 a partir do código-fonte</td></tr><tr><td><code>_flash_3_hub</code></td><td><a href="https://github.com/Dao-AILab/flash-attention" rel="nofollow">FlashAttention</a></td><td>FlashAttention-3 dos kernels do Hub</td><td><code>kernels>=0.12</code></td></tr><tr><td><code>_flash_3_varlen_hub</code></td><td><a href="https://github.com/Dao-AILab/flash-attention" rel="nofollow">FlashAttention</a></td><td>FlashAttention-3 de comprimento variável dos kernels do Hub</td><td><code>kernels>=0.12</code></td></tr><tr><td><code>sage</code></td><td><a href="https://github.com/thu-ml/SageAttention" rel="nofollow">SageAttention</a></td><td>Atenção quantizada (INT8 QK)</td><td><code>sageattention>=2.1.1</code></td></tr><tr><td><code>sage_hub</code></td><td><a href="https://github.com/thu-ml/SageAttention" rel="nofollow">SageAttention</a></td><td>Atenção quantizada (INT8 QK) dos kernels do Hub</td><td><code>kernels>=0.12</code>, <code>DIFFUSERS_TRUST_REMOTE_KERNELS=true</code></td></tr><tr><td><code>sage_blackwell_hub</code></td><td><a href="https://github.com/thu-ml/SageAttention" rel="nofollow">SageAttention</a></td><td>Atenção SageAttention3 FP4 para GPUs Blackwell SM120 a partir de kernels do Hub</td><td><code>kernels>=0.12</code>, <code>DIFFUSERS_TRUST_REMOTE_KERNELS=true</code></td></tr><tr><td><code>sage_varlen</code></td><td><a href="https://github.com/thu-ml/SageAttention" rel="nofollow">SageAttention</a></td><td>SageAttention de comprimento variável</td><td><code>sageattention>=2.1.1</code></td></tr><tr><td><code>_sage_qk_int8_pv_fp8_cuda</code></td><td><a href="https://github.com/thu-ml/SageAttention" rel="nofollow">SageAttention</a></td><td>INT8 QK + FP8 PV (CUDA)</td><td><code>sageattention>=2.1.1</code></td></tr><tr><td><code>_sage_qk_int8_pv_fp8_cuda_sm90</code></td><td><a href="https://github.com/thu-ml/SageAttention" rel="nofollow">SageAttention</a></td><td>INT8 QK + FP8 PV (SM90)</td><td><code>sageattention>=2.1.1</code>; SM90</td></tr><tr><td><code>_sage_qk_int8_pv_fp16_cuda</code></td><td><a href="https://github.com/thu-ml/SageAttention" rel="nofollow">SageAttention</a></td><td>INT8 QK + FP16 PV (CUDA)</td><td><code>sageattention>=2.1.1</code></td></tr><tr><td><code>_sage_qk_int8_pv_fp16_triton</code></td><td><a href="https://github.com/thu-ml/SageAttention" rel="nofollow">SageAttention</a></td><td>INT8 QK + FP16 PV (Triton)</td><td><code>sageattention>=2.1.1</code></td></tr><tr><td><code>xformers</code></td><td><a href="https://github.com/facebookresearch/xformers" rel="nofollow">xFormers</a></td><td>Atenção eficiente em memória</td><td><code>xformers>=0.0.29</code></td></tr></tbody></table> <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/optimization/attention_backends.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_kft2r9 = {
						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.B6roO9Zr.js"),
						import("/docs/diffusers/main/en/_app/immutable/entry/app.CxxEVG8A.js")
					]).then(([kit, app]) => {
						kit.start(app, element, {
							node_ids: [0, 7],
							data: [null,null],
							form: null,
							error: null
						});
					});
				}
			</script>
		
Xet Storage Details

Size:: 37.3 kB
Xet hash:: dd6b237e87b3b0756fbb392fb78f12f8920a5679f72549846da39caa555c303f
Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.