Hugging Face's logo

Buckets:
hf-doc-build
/
doc

Files
xet
hf-doc-build/doc / diffusers /v0.43.0 /es /quantization /torchao.html
mishig's picture
mishig
download
raw
33.8 kB
<meta charset="utf-8" /><meta name="hf:doc:metadata" content="{&quot;title&quot;:&quot;torchao&quot;,&quot;local&quot;:&quot;torchao&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;torch.compile&quot;,&quot;local&quot;:&quot;torchcompile&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Tipos de cuantización admitidos&quot;,&quot;local&quot;:&quot;supported-quantization-types&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Serialización y deserialización de modelos cuantizados&quot;,&quot;local&quot;:&quot;serializing-and-deserializing-quantized-models&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Recursos&quot;,&quot;local&quot;:&quot;resources&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/359.DjT-ulWG.js" rel="modulepreload">
<link href="/docs/diffusers/main/en/_app/immutable/chunks/DSPDl1gz.js" rel="modulepreload">
<!--1q5ej5h--><meta name="hf:doc:metadata" content="{&quot;title&quot;:&quot;torchao&quot;,&quot;local&quot;:&quot;torchao&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;torch.compile&quot;,&quot;local&quot;:&quot;torchcompile&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Supported quantization types&quot;,&quot;local&quot;:&quot;supported-quantization-types&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Serializing and Deserializing quantized models&quot;,&quot;local&quot;:&quot;serializing-and-deserializing-quantized-models&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Resources&quot;,&quot;local&quot;:&quot;resources&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="torchao" 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="#torchao"><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><!---->torchao<!----></span></h1><!--]--><!----> <p><a href="https://github.com/pytorch/ao" rel="nofollow">torchao</a> ofrece dtypes de alto rendimiento y optimizaciones basadas en cuantización y esparsidad para la inferencia y el entrenamiento de modelos de PyTorch. Es compatible con cualquier modelo de cualquier modalidad, siempre que admita la carga con <a href="https://hf.co/docs/accelerate/index" rel="nofollow">Accelerate</a> y contenga capas <code>torch.nn.Linear</code>.</p> <p>Asegúrate de que Pytorch 2.5+ y torchao estén instalados con el comando siguiente.</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 "><!---->uv pip install -U torch torchao<!----></pre></div><!----> <p>Cada dtype de cuantización está disponible como una instancia separada de la clase <a href="https://docs.pytorch.org/ao/main/api_reference/generated/torchao.core.config.AOBaseConfig.html" rel="nofollow">AOBaseConfig</a>. Esto ofrece opciones de configuración más flexibles al exponer más argumentos disponibles.</p> <p>Pase el <code>AOBaseConfig</code> de un dtype de cuantización, como <a href="https://docs.pytorch.org/ao/main/api_reference/generated/torchao.quantization.Int4WeightOnlyConfig.html" rel="nofollow">Int4WeightOnlyConfig</a>, a <a href="/docs/diffusers/v0.43.0/es/api/quantization#diffusers.TorchAoConfig">TorchAoConfig</a> en <a href="/docs/diffusers/v0.43.0/es/api/models/overview#diffusers.ModelMixin.from_pretrained">from_pretrained()</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> DiffusionPipeline, PipelineQuantizationConfig, TorchAoConfig
<span class="hljs-keyword">from</span> torchao.quantization <span class="hljs-keyword">import</span> Int8WeightOnlyConfig
pipeline_quant_config = PipelineQuantizationConfig(
quant_mapping={<span class="hljs-string">&quot;transformer&quot;</span>: TorchAoConfig(Int8WeightOnlyConfig(group_size=<span class="hljs-number">128</span>, version=<span class="hljs-number">2</span>))}
)
pipeline = DiffusionPipeline.from_pretrained(
<span class="hljs-string">&quot;black-forest-labs/FLUX.1-dev&quot;</span>,
quantization_config=pipeline_quant_config,
dtype=torch.bfloat16,
device_map=<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><!----> <p><code>device_map="cuda"</code> cuantiza cada capa en la GPU mientras se carga. Esto es rápido, pero requiere temporalmente memoria adicional de GPU para los pesos originales y cuantizados. Si el modelo ya utiliza la mayor parte de la memoria de tu GPU, la carga puede fallar con un error de memoria insuficiente. En ese caso, elimina <code>device_map</code> y mueve el pipeline a la GPU después de cargarlo:</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 "><!---->pipeline = DiffusionPipeline.from_pretrained(
<span class="hljs-string">&quot;black-forest-labs/FLUX.1-dev&quot;</span>,
quantization_config=pipeline_quant_config,
torch_dtype=torch.bfloat16,
)
pipeline.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><!----> <p>Sin <code>device_map</code>, Diffusers cuantiza las capas en la CPU. Esto es más lento, pero evita el pico temporal de memoria de GPU durante la cuantización. Para reducir aún más el uso de memoria de GPU, usa <a href="/docs/diffusers/v0.43.0/es/api/pipelines/overview#diffusers.DiffusionPipeline.enable_model_cpu_offload">enable_model_cpu_offload()</a> en su lugar. También puedes cuantizar componentes adicionales, como el codificador de texto.</p> <!--[1--><h2 class="relative group"><a id="torchcompile" 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="#torchcompile"><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><!---->torch.compile<!----></span></h2><!--]--><!----> <p>torchao admite <a href="../optimization/fp16#torchcompile">torch.compile</a>, que puede acelerar la inferencia con una sola línea de código.</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-python "><!----><span class="hljs-keyword">import</span> torch
<span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> DiffusionPipeline, PipelineQuantizationConfig, TorchAoConfig
<span class="hljs-keyword">from</span> torchao.quantization <span class="hljs-keyword">import</span> Int4WeightOnlyConfig
pipeline_quant_config = PipelineQuantizationConfig(
quant_mapping={<span class="hljs-string">&quot;transformer&quot;</span>: TorchAoConfig(Int4WeightOnlyConfig(group_size=<span class="hljs-number">128</span>))}
)
pipeline = DiffusionPipeline.from_pretrained(
<span class="hljs-string">&quot;black-forest-labs/FLUX.1-dev&quot;</span>,
quantization_config=pipeline_quant_config,
dtype=torch.bfloat16,
device_map=<span class="hljs-string">&quot;cuda&quot;</span> <span class="hljs-comment"># or &quot;mps&quot;, &quot;xpu&quot;, &quot;cpu&quot;</span>
)
pipeline.transformer.<span class="hljs-built_in">compile</span>(transformer, mode=<span class="hljs-string">&quot;max-autotune&quot;</span>, fullgraph=<span class="hljs-literal">True</span>)<!----></pre></div><!----> <p>Consulte esta <a href="https://github.com/huggingface/diffusers/pull/10009#issue-2688781450" rel="nofollow">tabla</a> para los benchmarks de velocidad de inferencia y uso de memoria con Flux y CogVideoX. También hay más benchmarks en varios hardware disponibles en el <a href="https://github.com/pytorch/ao/tree/main/torchao/quantization#benchmarks" rel="nofollow">repositorio</a> de torchao.</p> <blockquote class="tip"><p>Los esquemas de cuantización post-entrenamiento en FP8 de torchao son efectivos para GPUs con capacidad de cómputo de al menos 8.9 (RTX-4090, Hopper, etc.). FP8 suele ofrecer el mejor equilibrio entre velocidad, memoria y calidad al generar imágenes y videos. Recomendamos combinar FP8 y torch.compile si tu GPU es compatible.</p></blockquote> <!--[1--><h2 class="relative group"><a id="supported-quantization-types" 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="#supported-quantization-types"><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><!---->Tipos de cuantización admitidos<!----></span></h2><!--]--><!----> <p>torchao admite la cuantización solo de pesos y la cuantización de pesos y activaciones dinámicas para int8, float3-float8 y uint1-uint7.</p> <p>La cuantización solo de pesos almacena los pesos del modelo en un tipo de datos de bajo número de bits, pero realiza los cálculos con un tipo de datos de mayor precisión, como <code>bfloat16</code>. Esto reduce los requisitos de memoria derivados de los pesos del modelo, pero mantiene los picos de memoria para el cálculo de las activaciones.</p> <p>La cuantización dinámica de activaciones almacena los pesos del modelo en un dtype de bajo número de bits, mientras que también cuantiza las activaciones sobre la marcha para ahorrar memoria adicional. Esto reduce los requisitos de memoria de los pesos del modelo, al tiempo que reduce la sobrecarga de memoria de los cálculos de activación. Sin embargo, esto puede implicar un compromiso en la calidad en algunos casos, por lo que se recomienda probar a fondo diferentes modelos.</p> <p>Consulte la <a href="https://docs.pytorch.org/ao/stable/index.html" rel="nofollow">documentación oficial de torchao</a> para comprender mejor los métodos de cuantización disponibles. Una lista exhaustiva de las opciones de configuración está disponible <a href="https://docs.pytorch.org/ao/main/workflows/inference.html#inference-workflows" rel="nofollow">aquí</a>.</p> <p>Algunas configuraciones de cuantización populares y de ejemplo son las siguientes:</p> <table><thead><tr><th><strong>Categoría</strong></th><th><strong>Clases de configuración</strong></th></tr></thead><tbody><tr><td><strong>Cuantización entera</strong></td><td><a href="https://docs.pytorch.org/ao/stable/api_reference/generated/torchao.quantization.Int4WeightOnlyConfig.html" rel="nofollow"><code>Int4WeightOnlyConfig</code></a>, <a href="https://docs.pytorch.org/ao/stable/api_reference/generated/torchao.quantization.Int8WeightOnlyConfig.html" rel="nofollow"><code>Int8WeightOnlyConfig</code></a>, <a href="https://docs.pytorch.org/ao/stable/api_reference/generated/torchao.quantization.Int8DynamicActivationInt8WeightConfig.html" rel="nofollow"><code>Int8DynamicActivationInt8WeightConfig</code></a></td></tr><tr><td><strong>Cuantización de punto flotante de 8 bits</strong></td><td><a href="https://docs.pytorch.org/ao/stable/api_reference/generated/torchao.quantization.Float8WeightOnlyConfig.html" rel="nofollow"><code>Float8WeightOnlyConfig</code></a>, <a href="https://docs.pytorch.org/ao/stable/api_reference/generated/torchao.quantization.Float8DynamicActivationFloat8WeightConfig.html" rel="nofollow"><code>Float8DynamicActivationFloat8WeightConfig</code></a></td></tr><tr><td><strong>Cuantización de enteros sin signo</strong></td><td><a href="https://docs.pytorch.org/ao/stable/api_reference/generated/torchao.quantization.IntxWeightOnlyConfig.html" rel="nofollow"><code>IntxWeightOnlyConfig</code></a></td></tr></tbody></table> <!--[1--><h2 class="relative group"><a id="serializing-and-deserializing-quantized-models" 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="#serializing-and-deserializing-quantized-models"><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><!---->Serialización y deserialización de modelos cuantizados<!----></span></h2><!--]--><!----> <p>Para serializar un modelo cuantizado en un dtype dado, primero cargue el modelo con el dtype de cuantización deseado y luego guárdelo usando el método <a href="/docs/diffusers/v0.43.0/es/api/models/overview#diffusers.ModelMixin.save_pretrained">save_pretrained()</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-python "><!----><span class="hljs-keyword">import</span> torch
<span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> AutoModel, TorchAoConfig
<span class="hljs-keyword">from</span> torchao.quantization <span class="hljs-keyword">import</span> Int8WeightOnlyConfig
quantization_config = TorchAoConfig(Int8WeightOnlyConfig())
transformer = AutoModel.from_pretrained(
<span class="hljs-string">&quot;black-forest-labs/Flux.1-Dev&quot;</span>,
subfolder=<span class="hljs-string">&quot;transformer&quot;</span>,
quantization_config=quantization_config,
dtype=torch.bfloat16,
)
transformer.save_pretrained(<span class="hljs-string">&quot;/path/to/flux_int8wo&quot;</span>, safe_serialization=<span class="hljs-literal">False</span>)<!----></pre></div><!----> <p>Para cargar un modelo cuantizado serializado, use el método <a href="/docs/diffusers/v0.43.0/es/api/models/overview#diffusers.ModelMixin.from_pretrained">from_pretrained()</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-python "><!----><span class="hljs-keyword">import</span> torch
<span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> FluxPipeline, AutoModel
transformer = AutoModel.from_pretrained(<span class="hljs-string">&quot;/path/to/flux_int8wo&quot;</span>, dtype=torch.bfloat16, use_safetensors=<span class="hljs-literal">False</span>)
pipe = FluxPipeline.from_pretrained(<span class="hljs-string">&quot;black-forest-labs/Flux.1-Dev&quot;</span>, transformer=transformer, dtype=torch.bfloat16)
pipe.to(<span class="hljs-string">&quot;cuda&quot;</span>) <span class="hljs-comment"># or &quot;mps&quot;, &quot;xpu&quot;, &quot;cpu&quot;</span>
prompt = <span class="hljs-string">&quot;A cat holding a sign that says hello world&quot;</span>
image = pipe(prompt, num_inference_steps=<span class="hljs-number">30</span>, guidance_scale=<span class="hljs-number">7.0</span>).images[<span class="hljs-number">0</span>]
image.save(<span class="hljs-string">&quot;output.png&quot;</span>)<!----></pre></div><!----> <p>Si estás usando <code>torch&lt;=2.6.0</code>, algunos métodos de cuantización, como <code>uint4</code> solo de pesos, no se pueden cargar directamente y pueden provocar un <code>UnpicklingError</code> al intentar cargar los modelos, aunque funcionan como se espera al guardarlos. Para evitar esto, se puede cargar el state dict manualmente en el modelo. Ten en cuenta, sin embargo, que esto requiere usar <code>weights_only=False</code> en <code>torch.load</code>, por lo que solo debe ejecutarse si los pesos se obtuvieron de una fuente de confianza.</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-python "><!----><span class="hljs-keyword">import</span> torch
<span class="hljs-keyword">from</span> accelerate <span class="hljs-keyword">import</span> init_empty_weights
<span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> FluxPipeline, AutoModel, TorchAoConfig
<span class="hljs-keyword">from</span> torchao.quantization <span class="hljs-keyword">import</span> IntxWeightOnlyConfig
<span class="hljs-comment"># Serialize the model</span>
transformer = AutoModel.from_pretrained(
<span class="hljs-string">&quot;black-forest-labs/Flux.1-Dev&quot;</span>,
subfolder=<span class="hljs-string">&quot;transformer&quot;</span>,
quantization_config=TorchAoConfig(IntxWeightOnlyConfig(dtype=torch.uint4)),
dtype=torch.bfloat16,
)
transformer.save_pretrained(<span class="hljs-string">&quot;/path/to/flux_uint4wo&quot;</span>, safe_serialization=<span class="hljs-literal">False</span>, max_shard_size=<span class="hljs-string">&quot;50GB&quot;</span>)
<span class="hljs-comment"># ...</span>
<span class="hljs-comment"># Load the model</span>
state_dict = torch.load(<span class="hljs-string">&quot;/path/to/flux_uint4wo/diffusion_pytorch_model.bin&quot;</span>, weights_only=<span class="hljs-literal">False</span>, map_location=<span class="hljs-string">&quot;cpu&quot;</span>)
<span class="hljs-keyword">with</span> init_empty_weights():
transformer = AutoModel.from_config(<span class="hljs-string">&quot;/path/to/flux_uint4wo/config.json&quot;</span>)
transformer.load_state_dict(state_dict, strict=<span class="hljs-literal">True</span>, assign=<span class="hljs-literal">True</span>)<!----></pre></div><!----> <blockquote class="tip"><p>La API de <a href="/docs/diffusers/v0.43.0/es/api/models/auto_model#diffusers.AutoModel">AutoModel</a> es compatible con PyTorch &gt;= 2.6, como se muestra en los ejemplos a continuación.</p></blockquote> <!--[1--><h2 class="relative group"><a id="resources" 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="#resources"><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><!---->Recursos<!----></span></h2><!--]--><!----> <ul><li><a href="https://docs.pytorch.org/ao/stable/index.html" rel="nofollow">API de cuantización de TorchAO</a></li> <li><a href="https://github.com/sayakpaul/diffusers-torchao" rel="nofollow">Ejemplos de Diffusers-TorchAO</a></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/quantization/torchao.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, 359],
data: [null,null],
form: null,
error: null
});
});
}
</script>

Xet Storage Details

Size:
33.8 kB
·
Xet hash:
9574b5ab0f869ab9b7e7ae0e70f86abcf04b0bb17ee501aa61ad006e2c701b18

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