Buckets:

hf-doc-build
/

doc

Files
xet
hf-doc-build/doc / diffusers /main /es /optimization /attention_backends.html
mishig
1 day ago
download
raw
37.4 kB
<meta charset="utf-8" /><meta name="hf:doc:metadata" content="{&quot;title&quot;:&quot;Backends de atención&quot;,&quot;local&quot;:&quot;attention-backends&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;Establecer un backend en el modelo&quot;,&quot;local&quot;:&quot;set-a-backend-on-the-model&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Prueba un backend temporalmente&quot;,&quot;local&quot;:&quot;try-a-backend-temporarily&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Confianza en kernels remotos&quot;,&quot;local&quot;:&quot;trusting-remote-kernels&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Comprobaciones&quot;,&quot;local&quot;:&quot;checks&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Backends disponibles&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 atención<!----></span></h1><!--]--><!----> <blockquote class="note"><p>El despachador de atención es una función experimental. Abra un problema si tiene algún comentario o encuentra algún problema.</p></blockquote> <p>La mayoría de los modelos transformer de Diffusers enrutan la atención a través de un <em>attention dispatcher</em> para que puedas cambiar entre backends optimizados detrás de una única API. El dispatcher gestiona las implementaciones registradas y expone una ruta de llamada unificada para ellas. Algunos modelos, como muchos autoencoders, no utilizan el dispatcher debido a su estructura interna, y cambiar el backend no tiene efecto en sus capas de atención.</p> <p>Consulte la tabla siguiente para obtener un resumen de las familias de atención disponibles y consulte la sección <a href="#available-backends">Backends disponibles</a> para una lista más completa. El backend más rápido depende del modelo, la GPU, el dtype y la forma de la entrada.</p> <table><thead><tr><th>familia de atención</th><th>característica principal</th></tr></thead><tbody><tr><td>FlashAttention</td><td>minimiza las lecturas/escrituras de memoria mediante el uso de teselas y la recomputación</td></tr><tr><td>Motor de Tensor de IA para ROCm</td><td>Implementación de FlashAttention optimizada para aceleradores AMD ROCm</td></tr><tr><td>SageAttention</td><td>cuantiza la atención a int8</td></tr><tr><td>FlexAttention</td><td>PyTorch FlexAttention</td></tr><tr><td>PyTorch nativo</td><td>implementación integrada de PyTorch que utiliza <a href="./fp16#scaled-dot-product-attention">scaled_dot_product_attention</a></td></tr><tr><td>xFormers</td><td>atención eficiente en memoria con soporte para varios núcleos de atención</td></tr></tbody></table> <p>Los backends de Hub (los nombres <code>*_hub</code>) requieren la biblioteca <a href="https://github.com/huggingface/kernels" rel="nofollow">Kernels</a> y descargan el kernel en el primer uso. Los backends de Hub de SageAttention también requieren <code>DIFFUSERS_TRUST_REMOTE_KERNELS=true</code> (ver <a href="#trusting-remote-kernels">Confianza en kernels remotos</a>). Otros backends requieren su propio paquete, como <code>flash-attn</code> o <code>sageattention</code>. La tabla <a href="#available-backends">Backends disponibles</a> enumera los requisitos que Diffusers comprueba al habilitar un 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><!---->Establecer un backend en el modelo<!----></span></h2><!--]--><!----> <p>El método <a href="/docs/diffusers/main/es/api/models/overview#diffusers.ModelMixin.set_attention_backend">set_attention_backend()</a> recorre las capas de atención del modelo y aplica el backend elegido en cada una. También establece el backend activo a nivel de proceso del despachador con el mismo valor.</p> <p><a href="/docs/diffusers/main/es/api/models/overview#diffusers.ModelMixin.reset_attention_backend">reset_attention_backend()</a> solo limpia el backend en las capas de atención. No limpia el backend activo a nivel de proceso. Para un cambio temporal que restaure el backend activo anterior al salir, use el gestor de contexto <a href="#try-a-backend-temporarily">attention_backend</a>.</p> <p>El ejemplo siguiente habilita <code>_flash_3_hub</code> (FlashAttention-3 desde el Hub) con <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>Los backends de FlashAttention-3 no-Hub (<code>_flash_3</code>, <code>_flash_varlen_3</code>) requieren compilar FlashAttention-3 desde el código fuente y se deprecarán pronto. Utilice <code>_flash_3_hub</code> o <code>_flash_3_varlen_hub</code> en su lugar.</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><!---->Prueba un backend temporalmente<!----></span></h2><!--]--><!----> <p>El gestor de contexto <code>attention_backend()</code> establece el backend activo a nivel de proceso durante la duración del bloque y restaura el backend anterior al salir del bloque. Úselo para probar un backend en una sola llamada sin dejar un backend permanente aplicado desde <a href="/docs/diffusers/main/es/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>La mayoría de los backends de atención funcionan con <code>torch.compile</code>. Si eso acelera tu pipeline depende del modelo y del backend. Consulta <a href="./fp16#torchcompile">Precisión y compilación</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><!---->Confianza en kernels remotos<!----></span></h2><!--]--><!----> <p>Los backends de Hub requieren primero la biblioteca <a href="https://github.com/huggingface/kernels" rel="nofollow">Kernels</a> y, a continuación, Diffusers descarga el kernel de Hub en su primer uso.</p> <p>Los backends de atención de Hub descargan kernels de cómputo con la biblioteca <a href="https://github.com/huggingface/kernels" rel="nofollow">Kernels</a> y los ejecutan localmente. La mayoría de los nombres de atención de Hub (<code>_flash_3_hub</code>, <code>flash_hub</code> y los demás backends <code>*_hub</code>) se resuelven a la organización <a href="https://huggingface.co/kernels-community" rel="nofollow">kernels-community</a>. Dicha organización es un editor de confianza en Kernels, por lo que Diffusers carga esos kernels de atención sin establecer <code>DIFFUSERS_TRUST_REMOTE_KERNELS</code>.</p> <blockquote class="note"><p>Los backends de SageAttention Hub (<code>sage_hub</code> y <code>sage_blackwell_hub</code>) se cargan desde la organización <a href="https://huggingface.co/SageAttention" rel="nofollow">SageAttention</a>. Establezca <code>DIFFUSERS_TRUST_REMOTE_KERNELS=true</code> para usarlos.</p></blockquote> <p>Otras funciones respaldadas por kernels, como <a href="../quantization/gguf">GGUF</a> y <a href="../quantization/nunchaku">Nunchaku Lite</a>, pueden obtener kernels de editores externos a kernels-community. Esas rutas permanecen bloqueadas a menos que se activen con <code>DIFFUSERS_TRUST_REMOTE_KERNELS</code>. Cuando se establece, Diffusers reenvía <code>trust_remote_code=True</code> a Kernels para que también puedan cargarse editores no confiables.</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>Solo active esto después de inspeccionar el repositorio del kernel. Sin ello, cargar un kernel de un editor no confiable genera un error. Diffusers realiza esta comprobación por sí mismo, por lo que también se aplica a <code>kernels&lt;0.14.0</code>, que es anterior al argumento <code>trust_remote_code</code>. Establecer <code>DIFFUSERS_DISABLE_REMOTE_CODE=true</code> deshabilita el código remoto de forma global y tiene prioridad 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><!---->Comprobaciones<!----></span></h2><!--]--><!----> <p>El despachador de atención puede ejecutar comprobaciones de depuración antes de cada llamada de atención despachada. Las comprobaciones que se ejecutan dependen de las restricciones registradas para el backend activo.</p> <ol><li>Las comprobaciones de dispositivo verifican que los tensores de consulta, clave y valor estén en el mismo dispositivo.</li> <li>Las comprobaciones de tipo de datos, cuando están registradas, confirman la coincidencia de dtypes y a menudo requieren <code>bfloat16</code> o <code>float16</code>.</li> <li>Las comprobaciones de forma validan las dimensiones de los tensores y evitan mezclar máscaras de atención con banderas causales.</li></ol> <p>Habilite las comprobaciones con la variable de entorno <code>DIFFUSERS_ATTN_CHECKS</code>. Las comprobaciones añaden sobrecarga, por lo que están desactivadas de forma predeterminada.</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>Con las comprobaciones activadas, Diffusers ejecuta esas restricciones antes de cada llamada de atención despachada. El ejemplo de bajo nivel a continuación llama directamente a <code>dispatch_attention_fn</code>. La inferencia de la pipeline no necesita esa importación. Solo necesita que el backend esté configurado mediante <a href="/docs/diffusers/main/es/api/models/overview#diffusers.ModelMixin.set_attention_backend">set_attention_backend()</a> o <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 disponibles<!----></span></h2><!--]--><!----> <p>Consulte la tabla siguiente para obtener una lista completa de los backends de atención disponibles y sus variantes. Diffusers comprueba la disponibilidad del paquete y las versiones fijadas cuando se habilita un backend.</p> <table><thead><tr><th>Nombre del backend</th><th>Familia</th><th>Descripción</th><th>Prerrequisito</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 predeterminado que utiliza scaled_dot_product_attention de PyTorch</td><td>Ninguno</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>Atención optimizada con 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>Atención eficiente en memoria</td><td>Ninguna más allá de 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 de 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>Atención basada en matemáticas (alternativa)</td><td>Ninguno</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>Atención optimizada 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>Atención optimizada con 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 de los núcleos de 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 longitud variable</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 longitud variable desde kernels de 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 tensores de IA para ROCm</a></td><td>FlashAttention-2 para AMD ROCm desde los kernels de 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 desde los kernels de 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; pronto se deprecará)</td><td>Compilar FA3 desde el código fuente</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 longitud variable (local; pronto se deprecará)</td><td>Compilar FA3 desde el código fuente</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 de los núcleos de 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 longitud variable desde kernels de 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>Atención cuantizada (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>Atención cuantizada (INT8 QK) desde kernels de 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>Atención SageAttention3 FP4 para GPUs Blackwell SM120 desde kernels de 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 longitud variable</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>Atención eficiente en memoria</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:: 37.4 kB
Xet hash:: ee71f5e7681b230e8fb025762982deb9d734515cdd20fbaf03b66f6ecaab3d8e
Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.