Buckets:
| <meta charset="utf-8" /><meta name="hf:doc:metadata" content="{"title":"DreamBooth","local":"dreambooth","sections":[{"title":"Parámetros del script","local":"script-parameters","sections":[{"title":"Ponderación Min-SNR","local":"min-snr-weighting","sections":[],"depth":3},{"title":"Pérdida de preservación previa","local":"prior-preservation-loss","sections":[],"depth":3},{"title":"Entrenar el codificador de texto","local":"train-text-encoder","sections":[],"depth":3}],"depth":2},{"title":"Script de entrenamiento","local":"training-script","sections":[],"depth":2},{"title":"Ejecutar el script","local":"launch-the-script","sections":[],"depth":2},{"title":"LoRA","local":"lora","sections":[],"depth":2},{"title":"Stable Diffusion XL","local":"stable-diffusion-xl","sections":[],"depth":2},{"title":"DeepFloyd IF","local":"deepfloyd-if","sections":[{"title":"Consejos de entrenamiento","local":"training-tips","sections":[],"depth":3}],"depth":2},{"title":"Próximos pasos","local":"next-steps","sections":[],"depth":2}],"depth":1}"/> | |
| <link href="/docs/diffusers/main/en/_app/immutable/entry/start.D3LnKZ4R.js" rel="modulepreload"> | |
| <link href="/docs/diffusers/main/en/_app/immutable/chunks/BF9UNEM9.js" rel="modulepreload"> | |
| <link href="/docs/diffusers/main/en/_app/immutable/chunks/DPnyVmXs.js" rel="modulepreload"> | |
| <link href="/docs/diffusers/main/en/_app/immutable/entry/app.C0rBoMxR.js" rel="modulepreload"> | |
| <link href="/docs/diffusers/main/en/_app/immutable/chunks/DSGUT_OW.js" rel="modulepreload"> | |
| <link href="/docs/diffusers/main/en/_app/immutable/chunks/C045MWWx.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.DTZsP_3B.js" rel="modulepreload"> | |
| <link href="/docs/diffusers/main/en/_app/immutable/chunks/IaDvMxLQ.js" rel="modulepreload"> | |
| <link href="/docs/diffusers/main/en/_app/immutable/nodes/316.CuftASja.js" rel="modulepreload"> | |
| <link href="/docs/diffusers/main/en/_app/immutable/chunks/bKdwVYpU.js" rel="modulepreload"> | |
| <link href="/docs/diffusers/main/en/_app/immutable/chunks/DTd9IGSp.js" rel="modulepreload"> | |
| <!--w5pb01--><meta name="hf:doc:metadata" content="{"title":"DreamBooth","local":"dreambooth","sections":[{"title":"Script parameters","local":"script-parameters","sections":[{"title":"Min-SNR weighting","local":"min-snr-weighting","sections":[],"depth":3},{"title":"Prior preservation loss","local":"prior-preservation-loss","sections":[],"depth":3},{"title":"Train text encoder","local":"train-text-encoder","sections":[],"depth":3}],"depth":2},{"title":"Training script","local":"training-script","sections":[],"depth":2},{"title":"Launch the script","local":"launch-the-script","sections":[],"depth":2},{"title":"LoRA","local":"lora","sections":[],"depth":2},{"title":"Stable Diffusion XL","local":"stable-diffusion-xl","sections":[],"depth":2},{"title":"DeepFloyd IF","local":"deepfloyd-if","sections":[{"title":"Training tips","local":"training-tips","sections":[],"depth":3}],"depth":2},{"title":"Next steps","local":"next-steps","sections":[],"depth":2}],"depth":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="dreambooth" 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="#dreambooth"><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><!---->DreamBooth<!----></span></h1><!--]--><!----> <p><a href="https://huggingface.co/papers/2208.12242" rel="nofollow">DreamBooth</a> es una técnica de entrenamiento que actualiza todo el modelo de difusión entrenando con solo unas pocas imágenes de un sujeto o estilo. Funciona asociando una palabra especial en el prompt con las imágenes de ejemplo.</p> <p>Para cargar un punto de control entrenado para inferencia, consulte <a href="../using-diffusers/dreambooth">Cargar un modelo DreamBooth para inferencia</a>.</p> <p>Si estás entrenando en una GPU con vRAM limitada, debes intentar habilitar los parámetros <code>gradient_checkpointing</code> y <code>mixed_precision</code> en el comando de entrenamiento. También puedes reducir tu huella de memoria usando atención eficiente en memoria con <a href="../optimization/attention_backends">xFormers</a>.</p> <p>Esta guía explorará el script <a href="https://github.com/huggingface/diffusers/blob/main/examples/dreambooth/train_dreambooth.py" rel="nofollow">train_dreambooth.py</a> para que se familiarice más con él y para que pueda adaptarlo a su propio caso de uso.</p> <p>Antes de ejecutar el script, asegúrate de instalar la biblioteca desde el código fuente:</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 "><!---->git <span class="hljs-built_in">clone</span> https://github.com/huggingface/diffusers | |
| <span class="hljs-built_in">cd</span> diffusers | |
| pip install .<!----></pre></div><!----> <p>Navegue hasta la carpeta de ejemplo con el script de entrenamiento e instale las dependencias requeridas para el script que esté utilizando:</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">cd</span> examples/dreambooth | |
| pip install -r requirements.txt<!----></pre></div><!----> <blockquote class="tip"><p>🤗 Accelerate es una biblioteca para ayudarte a entrenar en varias GPUs/TPUs o con precisión mixta. Configura automáticamente tu configuración de entrenamiento en función de tu hardware y entorno. Echa un vistazo al <a href="https://huggingface.co/docs/accelerate/quicktour" rel="nofollow">Quick tour</a> de 🤗 Accelerate para saber más.</p></blockquote> <p>Inicializa un entorno de 🤗 Accelerate:</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 "><!---->accelerate config<!----></pre></div><!----> <p>Para configurar un entorno predeterminado de 🤗 Accelerate sin elegir ninguna configuración:</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 "><!---->accelerate config default<!----></pre></div><!----> <p>O si su entorno no admite una shell interactiva, como un notebook, puede usar:</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">from</span> accelerate.utils <span class="hljs-keyword">import</span> write_basic_config | |
| write_basic_config()<!----></pre></div><!----> <p>Por último, si deseas entrenar un modelo con tu propio conjunto de datos, consulta la guía <a href="create_dataset">Crear un conjunto de datos para el entrenamiento</a> para aprender a crear un conjunto de datos que funcione con el script de entrenamiento.</p> <blockquote class="tip"><p>Las siguientes secciones resaltan las partes del script de entrenamiento que son importantes para entender cómo modificarlo, pero no abordan todos los aspectos del script en detalle. Si te interesa aprender más, no dudes en leer el <a href="https://github.com/huggingface/diffusers/blob/main/examples/dreambooth/train_dreambooth.py" rel="nofollow">script</a> e informarnos si tienes alguna pregunta o inquietud.</p></blockquote> <!--[1--><h2 class="relative group"><a id="script-parameters" 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="#script-parameters"><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><!---->Parámetros del script<!----></span></h2><!--]--><!----> <blockquote class="warning"><p>DreamBooth es muy sensible a los hiperparámetros de entrenamiento y es fácil sobreajustar. Lee el artículo de blog <a href="https://huggingface.co/blog/dreambooth" rel="nofollow">Training Stable Diffusion with Dreambooth using 🧨 Diffusers</a> para conocer las configuraciones recomendadas para distintos sujetos y así elegir los hiperparámetros adecuados.</p></blockquote> <p>El script de entrenamiento ofrece muchos parámetros para personalizar tu ejecución de entrenamiento. Todos los parámetros y sus descripciones se encuentran en la función <a href="https://github.com/huggingface/diffusers/blob/072e00897a7cf4302c347a63ec917b4b8add16d4/examples/dreambooth/train_dreambooth.py#L228" rel="nofollow"><code>parse_args()</code></a>. Los parámetros se establecen con valores predeterminados que deberían funcionar bastante bien sin configuración adicional, pero también puedes establecer tus propios valores en el comando de entrenamiento si lo deseas.</p> <p>Por ejemplo, para entrenar en el formato bf16:</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 "><!---->accelerate launch train_dreambooth.py \ | |
| --mixed_precision=<span class="hljs-string">"bf16"</span><!----></pre></div><!----> <p>Algunos parámetros básicos e importantes que conviene conocer y especificar son:</p> <ul><li><code>--pretrained_model_name_or_path</code>: el nombre del modelo en el Hub o una ruta local al modelo preentrenado</li> <li><code>--instance_data_dir</code>: ruta a una carpeta que contiene el conjunto de datos de entrenamiento (imágenes de ejemplo)</li> <li><code>--instance_prompt</code>: el texto de la instrucción que contiene la palabra especial para las imágenes de ejemplo</li> <li><code>--train_text_encoder</code>: si se debe entrenar también el codificador de texto</li> <li><code>--output_dir</code>: dónde guardar el modelo entrenado</li> <li><code>--push_to_hub</code>: si se debe subir el modelo entrenado al Hub</li> <li><code>--checkpointing_steps</code>: frecuencia de guardado de un punto de control durante el entrenamiento del modelo; esto es útil si, por alguna razón, el entrenamiento se interrumpe, ya que puedes continuar el entrenamiento desde ese punto de control añadiendo <code>--resume_from_checkpoint</code> a tu comando de entrenamiento</li></ul> <!--[2--><h3 class="relative group"><a id="min-snr-weighting" 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="#min-snr-weighting"><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><!---->Ponderación Min-SNR<!----></span></h3><!--]--><!----> <p>La estrategia de ponderación <a href="https://huggingface.co/papers/2303.09556" rel="nofollow">Min-SNR</a> puede ayudar en el entrenamiento al reequilibrar la pérdida para lograr una convergencia más rápida. El script de entrenamiento admite la predicción de <code>epsilon</code> (ruido) o <code>v_prediction</code>, pero Min-SNR es compatible con ambos tipos de predicción. Esta estrategia de ponderación solo está soportada por PyTorch.</p> <p>Agrega el parámetro <code>--snr_gamma</code> y establécelo en el valor recomendado de 5.0:</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 "><!---->accelerate launch train_dreambooth.py \ | |
| --snr_gamma=5.0<!----></pre></div><!----> <!--[2--><h3 class="relative group"><a id="prior-preservation-loss" 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="#prior-preservation-loss"><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><!---->Pérdida de preservación previa<!----></span></h3><!--]--><!----> <p>La pérdida de preservación previa es un método que utiliza las propias muestras generadas por un modelo para ayudarlo a aprender a generar imágenes más diversas. Dado que estas imágenes de muestra generadas pertenecen a la misma clase que las imágenes que proporcionaste, ayudan al modelo a retener lo que ha aprendido sobre la clase y a utilizar lo que ya sabe sobre la clase para crear nuevas composiciones.</p> <ul><li><code>--with_prior_preservation</code>: si se debe usar la pérdida de preservación de la distribución previa</li> <li><code>--prior_loss_weight</code>: controla la influencia de la pérdida de preservación del prior en el modelo</li> <li><code>--class_data_dir</code>: ruta a una carpeta que contiene las imágenes de muestra de clase generadas</li> <li><code>--class_prompt</code>: el texto de la instrucción que describe la clase de las imágenes de muestra generadas</li></ul> <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 "><!---->accelerate launch train_dreambooth.py \ | |
| --with_prior_preservation \ | |
| --prior_loss_weight=1.0 \ | |
| --class_data_dir=<span class="hljs-string">"path/to/class/images"</span> \ | |
| --class_prompt=<span class="hljs-string">"text prompt describing class"</span><!----></pre></div><!----> <!--[2--><h3 class="relative group"><a id="train-text-encoder" 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="#train-text-encoder"><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><!---->Entrenar el codificador de texto<!----></span></h3><!--]--><!----> <p>Para mejorar la calidad de las salidas generadas, también puedes entrenar el codificador de texto además de la UNet. Esto requiere memoria adicional y necesitarás una GPU con al menos 24 GB de vRAM. Si tienes el hardware necesario, entrenar el codificador de texto produce mejores resultados, especialmente al generar imágenes de rostros. Habilita esta opción mediante:</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 "><!---->accelerate launch train_dreambooth.py \ | |
| --train_text_encoder<!----></pre></div><!----> <!--[1--><h2 class="relative group"><a id="training-script" 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="#training-script"><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><!---->Script de entrenamiento<!----></span></h2><!--]--><!----> <p>DreamBooth incluye sus propias clases de conjunto de datos:</p> <ul><li><a href="https://github.com/huggingface/diffusers/blob/072e00897a7cf4302c347a63ec917b4b8add16d4/examples/dreambooth/train_dreambooth.py#L604" rel="nofollow"><code>DreamBoothDataset</code></a>: preprocesa las imágenes y las imágenes de clase, y tokeniza los prompts para el entrenamiento</li> <li><a href="https://github.com/huggingface/diffusers/blob/072e00897a7cf4302c347a63ec917b4b8add16d4/examples/dreambooth/train_dreambooth.py#L738" rel="nofollow"><code>PromptDataset</code></a>: genera los embeddings de los prompts para generar las imágenes de clase</li></ul> <p>Si habilitaste la <a href="https://github.com/huggingface/diffusers/blob/072e00897a7cf4302c347a63ec917b4b8add16d4/examples/dreambooth/train_dreambooth.py#L842" rel="nofollow">pérdida de preservación previa</a>, las imágenes de clase se generan aquí:</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 "><!---->sample_dataset = PromptDataset(args.class_prompt, num_new_images) | |
| sample_dataloader = torch.utils.data.DataLoader(sample_dataset, batch_size=args.sample_batch_size) | |
| sample_dataloader = accelerator.prepare(sample_dataloader) | |
| pipeline.to(accelerator.device) | |
| <span class="hljs-keyword">for</span> example <span class="hljs-keyword">in</span> tqdm( | |
| sample_dataloader, desc=<span class="hljs-string">"Generating class images"</span>, disable=<span class="hljs-keyword">not</span> accelerator.is_local_main_process | |
| ): | |
| images = pipeline(example[<span class="hljs-string">"prompt"</span>]).images<!----></pre></div><!----> <p>A continuación se presenta la función <a href="https://github.com/huggingface/diffusers/blob/072e00897a7cf4302c347a63ec917b4b8add16d4/examples/dreambooth/train_dreambooth.py#L799" rel="nofollow"><code>main()</code></a>, que se encarga de configurar el conjunto de datos para el entrenamiento y del propio bucle de entrenamiento. El script carga el <a href="https://github.com/huggingface/diffusers/blob/072e00897a7cf4302c347a63ec917b4b8add16d4/examples/dreambooth/train_dreambooth.py#L898" rel="nofollow">tokenizer</a>, el <a href="https://github.com/huggingface/diffusers/blob/072e00897a7cf4302c347a63ec917b4b8add16d4/examples/dreambooth/train_dreambooth.py#L912C1-L912C1" rel="nofollow">programador y los modelos</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-comment"># Load the tokenizer</span> | |
| <span class="hljs-keyword">if</span> args.tokenizer_name: | |
| tokenizer = AutoTokenizer.from_pretrained(args.tokenizer_name, revision=args.revision, use_fast=<span class="hljs-literal">False</span>) | |
| <span class="hljs-keyword">elif</span> args.pretrained_model_name_or_path: | |
| tokenizer = AutoTokenizer.from_pretrained( | |
| args.pretrained_model_name_or_path, | |
| subfolder=<span class="hljs-string">"tokenizer"</span>, | |
| revision=args.revision, | |
| use_fast=<span class="hljs-literal">False</span>, | |
| ) | |
| <span class="hljs-comment"># Load scheduler and models</span> | |
| noise_scheduler = DDPMScheduler.from_pretrained(args.pretrained_model_name_or_path, subfolder=<span class="hljs-string">"scheduler"</span>) | |
| text_encoder = text_encoder_cls.from_pretrained( | |
| args.pretrained_model_name_or_path, subfolder=<span class="hljs-string">"text_encoder"</span>, revision=args.revision | |
| ) | |
| <span class="hljs-keyword">if</span> model_has_vae(args): | |
| vae = AutoencoderKL.from_pretrained( | |
| args.pretrained_model_name_or_path, subfolder=<span class="hljs-string">"vae"</span>, revision=args.revision | |
| ) | |
| <span class="hljs-keyword">else</span>: | |
| vae = <span class="hljs-literal">None</span> | |
| unet = UNet2DConditionModel.from_pretrained( | |
| args.pretrained_model_name_or_path, subfolder=<span class="hljs-string">"unet"</span>, revision=args.revision | |
| )<!----></pre></div><!----> <p>Luego, es momento de <a href="https://github.com/huggingface/diffusers/blob/072e00897a7cf4302c347a63ec917b4b8add16d4/examples/dreambooth/train_dreambooth.py#L1073" rel="nofollow">crear el conjunto de datos de entrenamiento</a> y el DataLoader a partir de <code>DreamBoothDataset</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 "><!---->train_dataset = DreamBoothDataset( | |
| instance_data_root=args.instance_data_dir, | |
| instance_prompt=args.instance_prompt, | |
| class_data_root=args.class_data_dir <span class="hljs-keyword">if</span> args.with_prior_preservation <span class="hljs-keyword">else</span> <span class="hljs-literal">None</span>, | |
| class_prompt=args.class_prompt, | |
| class_num=args.num_class_images, | |
| tokenizer=tokenizer, | |
| size=args.resolution, | |
| center_crop=args.center_crop, | |
| encoder_hidden_states=pre_computed_encoder_hidden_states, | |
| class_prompt_encoder_hidden_states=pre_computed_class_prompt_encoder_hidden_states, | |
| tokenizer_max_length=args.tokenizer_max_length, | |
| ) | |
| train_dataloader = torch.utils.data.DataLoader( | |
| train_dataset, | |
| batch_size=args.train_batch_size, | |
| shuffle=<span class="hljs-literal">True</span>, | |
| collate_fn=<span class="hljs-keyword">lambda</span> examples: collate_fn(examples, args.with_prior_preservation), | |
| num_workers=args.dataloader_num_workers, | |
| )<!----></pre></div><!----> <p>Por último, el <a href="https://github.com/huggingface/diffusers/blob/072e00897a7cf4302c347a63ec917b4b8add16d4/examples/dreambooth/train_dreambooth.py#L1151" rel="nofollow">bucle de entrenamiento</a> se encarga de los pasos restantes, como convertir las imágenes al espacio latente, añadir ruido a la entrada, predecir el residuo de ruido y calcular la pérdida.</p> <p>Si quieres saber más sobre cómo funciona el bucle de entrenamiento, consulta el tutorial <a href="../using-diffusers/write_own_pipeline">Understanding pipelines, models and schedulers</a>, que desglosa el patrón básico del proceso de desruido.</p> <!--[1--><h2 class="relative group"><a id="launch-the-script" 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="#launch-the-script"><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><!---->Ejecutar el script<!----></span></h2><!--]--><!----> <p>¡Ahora estás listo para lanzar el script de entrenamiento! 🚀</p> <p>Para esta guía, descargará algunas imágenes de un <a href="https://huggingface.co/datasets/diffusers/dog-example" rel="nofollow">perro</a> y las guardará en un directorio. Pero recuerde que puede crear y usar su propio conjunto de datos si lo desea (consulte la guía <a href="create_dataset">Crear un conjunto de datos para el entrenamiento</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">from</span> huggingface_hub <span class="hljs-keyword">import</span> snapshot_download | |
| local_dir = <span class="hljs-string">"./dog"</span> | |
| snapshot_download( | |
| <span class="hljs-string">"diffusers/dog-example"</span>, | |
| local_dir=local_dir, | |
| repo_type=<span class="hljs-string">"dataset"</span>, | |
| ignore_patterns=<span class="hljs-string">".gitattributes"</span>, | |
| )<!----></pre></div><!----> <p>Establezca la variable de entorno <code>MODEL_NAME</code> en un id de modelo del Hub o en una ruta a un modelo local, <code>INSTANCE_DIR</code> en la ruta donde acaba de descargar las imágenes de perros y <code>OUTPUT_DIR</code> en el lugar donde desea guardar el modelo. Usará <code>sks</code> como palabra especial para vincular el entrenamiento.</p> <p>Si te interesa seguir el proceso de entrenamiento, puedes guardar periódicamente las imágenes generadas a medida que avanza el entrenamiento. Añade los siguientes parámetros al comando de entrenamiento:</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 "><!---->--validation_prompt=<span class="hljs-string">"a photo of a sks dog"</span> | |
| --num_validation_images=4 | |
| --validation_steps=100<!----></pre></div><!----> <p>¡Una cosa más antes de lanzar el script! Dependiendo de la GPU que tengas, es posible que necesites habilitar ciertas optimizaciones para entrenar DreamBooth.</p> <div class="flex space-x-2 items-center my-1.5 mr-8 h-7 !pl-0 -mx-3 md:mx-0"><!--[--><div class="flex items-center border rounded-lg px-1.5 py-1 leading-none select-none text-smd border-gray-800 bg-black dark:bg-gray-700 text-white">16GB</div><div class="flex items-center border rounded-lg px-1.5 py-1 leading-none select-none text-smd text-gray-500 cursor-pointer opacity-90 hover:text-gray-700 dark:hover:text-gray-200 hover:shadow-sm">12GB</div><div class="flex items-center border rounded-lg px-1.5 py-1 leading-none select-none text-smd text-gray-500 cursor-pointer opacity-90 hover:text-gray-700 dark:hover:text-gray-200 hover:shadow-sm">8GB</div><!--]--></div> <div class="language-select"><div><p>En una GPU de 16 GB, puedes usar el optimizador de 8 bits de bitsandbytes y gradient checkpointing para ayudarte a entrenar un modelo DreamBooth. Instala bitsandbytes:</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 "><!---->pip install bitsandbytes<!----></pre></div><!----> <p>Luego, añade el siguiente parámetro a tu comando de entrenamiento:</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 "><!---->accelerate launch train_dreambooth.py \ | |
| --gradient_checkpointing \ | |
| --use_8bit_adam \<!----></pre></div><!----><!----></div><!----> <div hidden=""><p>En una GPU de 12 GB, necesitarás el optimizador de 8 bits de bitsandbytes, gradient checkpointing, xFormers y establecer los gradientes en <code>None</code> en lugar de cero para reducir el uso de memoria.</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 "><!---->accelerate launch train_dreambooth.py \ | |
| --use_8bit_adam \ | |
| --gradient_checkpointing \ | |
| --enable_xformers_memory_efficient_attention \ | |
| --set_grads_to_none \<!----></pre></div><!----><!----></div><!----> <div hidden=""><p>En una GPU de 8 GB, necesitarás <a href="https://www.deepspeed.ai/" rel="nofollow">DeepSpeed</a> para descargar algunos de los tensores desde la vRAM hacia la CPU o el NVME, lo que permite entrenar con menos memoria de GPU.</p> <p>Ejecuta el siguiente comando para configurar tu entorno de 🤗 Accelerate:</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 "><!---->accelerate config<!----></pre></div><!----> <p>Durante la configuración, confirme que desea usar DeepSpeed. Ahora debería ser posible entrenar con menos de 8 GB de vRAM combinando DeepSpeed stage 2, precisión mixta fp16 y el descargado de los parámetros del modelo y el estado del optimizador a la CPU. La desventaja es que esto requiere más RAM de sistema (~25 GB). Consulte la <a href="https://huggingface.co/docs/accelerate/usage_guides/deepspeed" rel="nofollow">documentación de DeepSpeed</a> para más opciones de configuración.</p> <p>También debe cambiar el optimizador Adam predeterminado por la versión optimizada de Adam de DeepSpeed, <a href="https://deepspeed.readthedocs.io/en/latest/optimizers.html#adam-cpu" rel="nofollow"><code>deepspeed.ops.adam.DeepSpeedCPUAdam</code></a>, para obtener una mejora sustancial en la velocidad. Para habilitar <code>DeepSpeedCPUAdam</code>, la versión de la cadena de herramientas CUDA de su sistema debe ser la misma que la instalada con PyTorch.</p> <p>Los optimizadores de 8 bits de bitsandbytes no parecen ser compatibles con DeepSpeed en este momento.</p> <p>¡Eso es todo! No necesitas añadir parámetros adicionales a tu comando de entrenamiento.</p><!----></div><!----><!----></div><!----> <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> MODEL_NAME=<span class="hljs-string">"stable-diffusion-v1-5/stable-diffusion-v1-5"</span> | |
| <span class="hljs-built_in">export</span> INSTANCE_DIR=<span class="hljs-string">"./dog"</span> | |
| <span class="hljs-built_in">export</span> OUTPUT_DIR=<span class="hljs-string">"path_to_saved_model"</span> | |
| accelerate launch train_dreambooth.py \ | |
| --pretrained_model_name_or_path=<span class="hljs-variable">$MODEL_NAME</span> \ | |
| --instance_data_dir=<span class="hljs-variable">$INSTANCE_DIR</span> \ | |
| --output_dir=<span class="hljs-variable">$OUTPUT_DIR</span> \ | |
| --instance_prompt=<span class="hljs-string">"a photo of sks dog"</span> \ | |
| --resolution=512 \ | |
| --train_batch_size=1 \ | |
| --gradient_accumulation_steps=1 \ | |
| --learning_rate=5e-6 \ | |
| --lr_scheduler=<span class="hljs-string">"constant"</span> \ | |
| --lr_warmup_steps=0 \ | |
| --max_train_steps=400 \ | |
| --push_to_hub<!----></pre></div><!----> <p>Una vez que el entrenamiento esté completo, ¡puedes usar tu modelo recién entrenado para la inferencia!</p> <blockquote class="tip"><p>¿No puedes esperar a probar tu modelo para inferencia antes de que termine el entrenamiento? 🤭 Asegúrate de tener instalada la última versión de 🤗 Accelerate.</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">from</span> diffusers <span class="hljs-keyword">import</span> DiffusionPipeline, UNet2DConditionModel | |
| <span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> CLIPTextModel | |
| <span class="hljs-keyword">import</span> torch | |
| unet = UNet2DConditionModel.from_pretrained(<span class="hljs-string">"path/to/model/checkpoint-100/unet"</span>) | |
| <span class="hljs-comment"># if you have trained with `--args.train_text_encoder` make sure to also load the text encoder</span> | |
| text_encoder = CLIPTextModel.from_pretrained(<span class="hljs-string">"path/to/model/checkpoint-100/checkpoint-100/text_encoder"</span>) | |
| pipeline = DiffusionPipeline.from_pretrained( | |
| <span class="hljs-string">"stable-diffusion-v1-5/stable-diffusion-v1-5"</span>, unet=unet, text_encoder=text_encoder, dtype=torch.float16, | |
| ).to(<span class="hljs-string">"cuda"</span>) | |
| image = pipeline(<span class="hljs-string">"A photo of sks dog in a bucket"</span>, num_inference_steps=<span class="hljs-number">50</span>, guidance_scale=<span class="hljs-number">7.5</span>).images[<span class="hljs-number">0</span>] | |
| image.save(<span class="hljs-string">"dog-bucket.png"</span>)<!----></pre></div><!----></blockquote> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Copiado</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> DiffusionPipeline | |
| <span class="hljs-keyword">import</span> torch | |
| pipeline = DiffusionPipeline.from_pretrained(<span class="hljs-string">"path_to_saved_model"</span>, dtype=torch.float16, use_safetensors=<span class="hljs-literal">True</span>).to(<span class="hljs-string">"cuda"</span>) <span class="hljs-comment"># or "mps", "xpu", "cpu"</span> | |
| image = pipeline(<span class="hljs-string">"A photo of sks dog in a bucket"</span>, num_inference_steps=<span class="hljs-number">50</span>, guidance_scale=<span class="hljs-number">7.5</span>).images[<span class="hljs-number">0</span>] | |
| image.save(<span class="hljs-string">"dog-bucket.png"</span>)<!----></pre></div><!----> <!--[1--><h2 class="relative group"><a id="lora" 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="#lora"><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><!---->LoRA<!----></span></h2><!--]--><!----> <p>LoRA es una técnica de entrenamiento para reducir significativamente el número de parámetros entrenables. Como resultado, el entrenamiento es más rápido y es más fácil almacenar los pesos resultantes porque son mucho más pequeños (~100MB). Utilice el script <a href="https://github.com/huggingface/diffusers/blob/main/examples/dreambooth/train_dreambooth_lora.py" rel="nofollow">train_dreambooth_lora.py</a> para entrenar con LoRA.</p> <p>El script de entrenamiento LoRA se describe con más detalle en la guía de <a href="lora">entrenamiento LoRA</a>.</p> <!--[1--><h2 class="relative group"><a id="stable-diffusion-xl" 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="#stable-diffusion-xl"><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><!---->Stable Diffusion XL<!----></span></h2><!--]--><!----> <p>Stable Diffusion XL (SDXL) es un potente modelo de texto a imagen que genera imágenes de alta resolución y añade un segundo text-encoder a su arquitectura. Utilice el script <a href="https://github.com/huggingface/diffusers/blob/main/examples/dreambooth/train_dreambooth_lora_sdxl.py" rel="nofollow">train_dreambooth_lora_sdxl.py</a> para entrenar un modelo SDXL con LoRA.</p> <p>El script de entrenamiento de SDXL se describe con más detalle en la guía de <a href="sdxl">entrenamiento de SDXL</a>.</p> <!--[1--><h2 class="relative group"><a id="deepfloyd-if" 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="#deepfloyd-if"><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><!---->DeepFloyd IF<!----></span></h2><!--]--><!----> <p>DeepFloyd IF es un modelo de difusión de píxeles en cascada con tres etapas. La primera etapa genera una imagen base y las segunda y tercera etapas escalan progresivamente la imagen base hasta obtener una imagen de alta resolución de 1024x1024. Utilice los scripts <a href="https://github.com/huggingface/diffusers/blob/main/examples/dreambooth/train_dreambooth_lora.py" rel="nofollow">train_dreambooth_lora.py</a> o <a href="https://github.com/huggingface/diffusers/blob/main/examples/dreambooth/train_dreambooth.py" rel="nofollow">train_dreambooth.py</a> para entrenar un modelo DeepFloyd IF con LoRA o el modelo completo.</p> <p>DeepFloyd IF utiliza varianza predicha, pero los scripts de entrenamiento de Diffusers utilizan error predicho, por lo que los modelos de DeepFloyd IF entrenados se cambian a un programa de varianza fija. Los scripts de entrenamiento actualizarán la configuración del programador del modelo completamente entrenado por ti. Sin embargo, al cargar los pesos de LoRA guardados, también debes actualizar la configuración del programador del pipeline.</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg 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">from</span> diffusers <span class="hljs-keyword">import</span> DiffusionPipeline | |
| pipe = DiffusionPipeline.from_pretrained(<span class="hljs-string">"DeepFloyd/IF-I-XL-v1.0"</span>, use_safetensors=<span class="hljs-literal">True</span>) | |
| pipe.load_lora_weights(<span class="hljs-string">"<lora weights path>"</span>) | |
| <span class="hljs-comment"># Update scheduler config to fixed variance schedule</span> | |
| pipe.scheduler = pipe.scheduler.__class__.from_config(pipe.scheduler.config, variance_type=<span class="hljs-string">"fixed_small"</span>)<!----></pre></div><!----> <p>El modelo de la etapa 2 requiere imágenes de validación adicionales para aumentar la resolución. Puedes descargar y usar una versión reducida de las imágenes de entrenamiento para esto.</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">from</span> huggingface_hub <span class="hljs-keyword">import</span> snapshot_download | |
| local_dir = <span class="hljs-string">"./dog_downsized"</span> | |
| snapshot_download( | |
| <span class="hljs-string">"diffusers/dog-example-downsized"</span>, | |
| local_dir=local_dir, | |
| repo_type=<span class="hljs-string">"dataset"</span>, | |
| ignore_patterns=<span class="hljs-string">".gitattributes"</span>, | |
| )<!----></pre></div><!----> <p>Las muestras de código a continuación ofrecen una breve descripción de cómo entrenar un modelo DeepFloyd IF con una combinación de DreamBooth y LoRA. Algunos parámetros importantes a tener en cuenta son:</p> <ul><li><code>--resolution=64</code>, se requiere una resolución mucho menor porque DeepFloyd IF es un modelo de difusión de píxeles y, para trabajar con píxeles sin comprimir, las imágenes de entrada deben ser más pequeñas</li> <li><code>--pre_compute_text_embeddings</code>, calcula los embeddings de texto de antemano para ahorrar memoria, ya que <a href="https://huggingface.co/docs/transformers/main/en/model_doc/t5#transformers.T5Model" rel="nofollow">T5Model</a> puede consumir mucha memoria</li> <li><code>--tokenizer_max_length=77</code>, puede usar una longitud de texto predeterminada más larga con T5 como codificador de texto, pero el procedimiento de codificación del modelo predeterminado usa una longitud de texto más corta</li> <li><code>--text_encoder_use_attention_mask</code>, para pasar la máscara de atención al codificador de texto</li></ul> <div class="flex space-x-2 items-center my-1.5 mr-8 h-7 !pl-0 -mx-3 md:mx-0"><!--[--><div class="flex items-center border rounded-lg px-1.5 py-1 leading-none select-none text-smd border-gray-800 bg-black dark:bg-gray-700 text-white">Etapa 1 LoRA DreamBooth</div><div class="flex items-center border rounded-lg px-1.5 py-1 leading-none select-none text-smd text-gray-500 cursor-pointer opacity-90 hover:text-gray-700 dark:hover:text-gray-200 hover:shadow-sm">Etapa 2 LoRA DreamBooth</div><div class="flex items-center border rounded-lg px-1.5 py-1 leading-none select-none text-smd text-gray-500 cursor-pointer opacity-90 hover:text-gray-700 dark:hover:text-gray-200 hover:shadow-sm">Etapa 1 DreamBooth</div><div class="flex items-center border rounded-lg px-1.5 py-1 leading-none select-none text-smd text-gray-500 cursor-pointer opacity-90 hover:text-gray-700 dark:hover:text-gray-200 hover:shadow-sm">Etapa 2 DreamBooth</div><!--]--></div> <div class="language-select"><div><p>La etapa 1 de entrenamiento de DeepFloyd IF con LoRA y DreamBooth requiere ~28GB de memoria.</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> MODEL_NAME=<span class="hljs-string">"DeepFloyd/IF-I-XL-v1.0"</span> | |
| <span class="hljs-built_in">export</span> INSTANCE_DIR=<span class="hljs-string">"dog"</span> | |
| <span class="hljs-built_in">export</span> OUTPUT_DIR=<span class="hljs-string">"dreambooth_dog_lora"</span> | |
| accelerate launch train_dreambooth_lora.py \ | |
| --report_to wandb \ | |
| --pretrained_model_name_or_path=<span class="hljs-variable">$MODEL_NAME</span> \ | |
| --instance_data_dir=<span class="hljs-variable">$INSTANCE_DIR</span> \ | |
| --output_dir=<span class="hljs-variable">$OUTPUT_DIR</span> \ | |
| --instance_prompt=<span class="hljs-string">"a sks dog"</span> \ | |
| --resolution=64 \ | |
| --train_batch_size=4 \ | |
| --gradient_accumulation_steps=1 \ | |
| --learning_rate=5e-6 \ | |
| --scale_lr \ | |
| --max_train_steps=1200 \ | |
| --validation_prompt=<span class="hljs-string">"a sks dog"</span> \ | |
| --validation_epochs=25 \ | |
| --checkpointing_steps=100 \ | |
| --pre_compute_text_embeddings \ | |
| --tokenizer_max_length=77 \ | |
| --text_encoder_use_attention_mask<!----></pre></div><!----><!----></div><!----> <div hidden=""><p>Para la etapa 2 de DeepFloyd IF con LoRA y DreamBooth, preste atención a estos parámetros:</p> <ul><li><code>--validation_images</code>, las imágenes para ampliar durante la validación</li> <li><code>--class_labels_conditioning=timesteps</code>, para condicionar adicionalmente el UNet según sea necesario en la etapa 2</li> <li><code>--learning_rate=1e-6</code>, se utiliza una tasa de aprendizaje más baja en comparación con la etapa 1</li> <li><code>--resolution=256</code>, la resolución esperada para el upscaler</li></ul> <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> MODEL_NAME=<span class="hljs-string">"DeepFloyd/IF-II-L-v1.0"</span> | |
| <span class="hljs-built_in">export</span> INSTANCE_DIR=<span class="hljs-string">"dog"</span> | |
| <span class="hljs-built_in">export</span> OUTPUT_DIR=<span class="hljs-string">"dreambooth_dog_upscale"</span> | |
| <span class="hljs-built_in">export</span> VALIDATION_IMAGES=<span class="hljs-string">"dog_downsized/image_1.png dog_downsized/image_2.png dog_downsized/image_3.png dog_downsized/image_4.png"</span> | |
| python train_dreambooth_lora.py \ | |
| --report_to wandb \ | |
| --pretrained_model_name_or_path=<span class="hljs-variable">$MODEL_NAME</span> \ | |
| --instance_data_dir=<span class="hljs-variable">$INSTANCE_DIR</span> \ | |
| --output_dir=<span class="hljs-variable">$OUTPUT_DIR</span> \ | |
| --instance_prompt=<span class="hljs-string">"a sks dog"</span> \ | |
| --resolution=256 \ | |
| --train_batch_size=4 \ | |
| --gradient_accumulation_steps=1 \ | |
| --learning_rate=1e-6 \ | |
| --max_train_steps=2000 \ | |
| --validation_prompt=<span class="hljs-string">"a sks dog"</span> \ | |
| --validation_epochs=100 \ | |
| --checkpointing_steps=500 \ | |
| --pre_compute_text_embeddings \ | |
| --tokenizer_max_length=77 \ | |
| --text_encoder_use_attention_mask \ | |
| --validation_images <span class="hljs-variable">$VALIDATION_IMAGES</span> \ | |
| --class_labels_conditioning=timesteps<!----></pre></div><!----><!----></div><!----> <div hidden=""><p>Para la etapa 1 de DeepFloyd IF con DreamBooth, preste atención a estos parámetros:</p> <ul><li><code>--skip_save_text_encoder</code>, para omitir el guardado del codificador de texto T5 completo junto con el modelo afinado</li> <li><code>--use_8bit_adam</code>, para usar el optimizador Adam de 8 bits y ahorrar memoria debido al tamaño del estado del optimizador al entrenar el modelo completo</li> <li><code>--learning_rate=1e-7</code>, se debe usar una tasa de aprendizaje realmente baja para el entrenamiento completo del modelo, de lo contrario la calidad del modelo se degrada (puede usar una tasa de aprendizaje más alta con un tamaño de lote mayor)</li></ul> <p>Entrenando con 8-bit Adam y un tamaño de lote de 4, el modelo completo se puede entrenar con ~48GB de memoria.</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> MODEL_NAME=<span class="hljs-string">"DeepFloyd/IF-I-XL-v1.0"</span> | |
| <span class="hljs-built_in">export</span> INSTANCE_DIR=<span class="hljs-string">"dog"</span> | |
| <span class="hljs-built_in">export</span> OUTPUT_DIR=<span class="hljs-string">"dreambooth_if"</span> | |
| accelerate launch train_dreambooth.py \ | |
| --pretrained_model_name_or_path=<span class="hljs-variable">$MODEL_NAME</span> \ | |
| --instance_data_dir=<span class="hljs-variable">$INSTANCE_DIR</span> \ | |
| --output_dir=<span class="hljs-variable">$OUTPUT_DIR</span> \ | |
| --instance_prompt=<span class="hljs-string">"a photo of sks dog"</span> \ | |
| --resolution=64 \ | |
| --train_batch_size=4 \ | |
| --gradient_accumulation_steps=1 \ | |
| --learning_rate=1e-7 \ | |
| --max_train_steps=150 \ | |
| --validation_prompt <span class="hljs-string">"a photo of sks dog"</span> \ | |
| --validation_steps 25 \ | |
| --text_encoder_use_attention_mask \ | |
| --tokenizer_max_length 77 \ | |
| --pre_compute_text_embeddings \ | |
| --use_8bit_adam \ | |
| --set_grads_to_none \ | |
| --skip_save_text_encoder \ | |
| --push_to_hub<!----></pre></div><!----><!----></div><!----> <div hidden=""><p>Para la etapa 2 de DeepFloyd IF con DreamBooth, presta atención a estos parámetros:</p> <ul><li><code>--learning_rate=5e-6</code>, use una tasa de aprendizaje más baja con un tamaño de lote efectivo más pequeño</li> <li><code>--resolution=256</code>, la resolución esperada para el upscaler</li> <li><code>--train_batch_size=2</code> y <code>--gradient_accumulation_steps=6</code>, para entrenar de forma efectiva con imágenes que contienen rostros se requieren tamaños de lote más grandes</li></ul> <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> MODEL_NAME=<span class="hljs-string">"DeepFloyd/IF-II-L-v1.0"</span> | |
| <span class="hljs-built_in">export</span> INSTANCE_DIR=<span class="hljs-string">"dog"</span> | |
| <span class="hljs-built_in">export</span> OUTPUT_DIR=<span class="hljs-string">"dreambooth_dog_upscale"</span> | |
| <span class="hljs-built_in">export</span> VALIDATION_IMAGES=<span class="hljs-string">"dog_downsized/image_1.png dog_downsized/image_2.png dog_downsized/image_3.png dog_downsized/image_4.png"</span> | |
| accelerate launch train_dreambooth.py \ | |
| --report_to wandb \ | |
| --pretrained_model_name_or_path=<span class="hljs-variable">$MODEL_NAME</span> \ | |
| --instance_data_dir=<span class="hljs-variable">$INSTANCE_DIR</span> \ | |
| --output_dir=<span class="hljs-variable">$OUTPUT_DIR</span> \ | |
| --instance_prompt=<span class="hljs-string">"a sks dog"</span> \ | |
| --resolution=256 \ | |
| --train_batch_size=2 \ | |
| --gradient_accumulation_steps=6 \ | |
| --learning_rate=5e-6 \ | |
| --max_train_steps=2000 \ | |
| --validation_prompt=<span class="hljs-string">"a sks dog"</span> \ | |
| --validation_steps=150 \ | |
| --checkpointing_steps=500 \ | |
| --pre_compute_text_embeddings \ | |
| --tokenizer_max_length=77 \ | |
| --text_encoder_use_attention_mask \ | |
| --validation_images <span class="hljs-variable">$VALIDATION_IMAGES</span> \ | |
| --class_labels_conditioning timesteps \ | |
| --push_to_hub<!----></pre></div><!----><!----></div><!----><!----></div><!----> <!--[2--><h3 class="relative group"><a id="training-tips" 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="#training-tips"><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><!---->Consejos de entrenamiento<!----></span></h3><!--]--><!----> <p>Entrenar el modelo DeepFloyd IF puede ser un desafío, pero aquí hay algunos consejos que nos han resultado útiles:</p> <ul><li>LoRA es suficiente para entrenar el modelo de la etapa 1, ya que la baja resolución del modelo dificulta la representación de detalles más finos de todos modos.</li> <li>Para objetos comunes o simples, no es necesario ajustar el upscaler. Asegúrate de que el prompt pasado al upscaler esté ajustado para eliminar el nuevo token del prompt de la instancia. Por ejemplo, si tu prompt de la etapa 1 es “a sks dog”, entonces tu prompt de la etapa 2 debe ser “a dog”.</li> <li>Para detalles más finos, como los rostros, es mejor entrenar por completo el upscaler de la etapa 2 que entrenar el modelo de la etapa 2 con LoRA. También ayuda utilizar tasas de aprendizaje más bajas con tamaños de lote más grandes.</li> <li>Se deben usar tasas de aprendizaje más bajas para entrenar el modelo de la etapa 2.</li> <li>El <a href="/docs/diffusers/main/es/api/schedulers/ddpm#diffusers.DDPMScheduler">DDPMScheduler</a> funciona mejor que el DPMSolver utilizado en los scripts de entrenamiento.</li></ul> <!--[1--><h2 class="relative group"><a id="next-steps" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#next-steps"><span><svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg><!----></span></a> <span><!---->Próximos pasos<!----></span></h2><!--]--><!----> <p>¡Felicidades por entrenar tu modelo de DreamBooth! Para aprender más sobre cómo usar tu nuevo modelo, la siguiente guía puede ser útil:</p> <ul><li>Aprende a <a href="../using-diffusers/dreambooth">cargar un checkpoint de DreamBooth</a> para la inferencia.</li> <li>Si entrenaste con LoRA, consulta <a href="../tutorials/using_peft_for_inference">LoRA para inferencia</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/training/dreambooth.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_g37pnb = { | |
| 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.D3LnKZ4R.js"), | |
| import("/docs/diffusers/main/en/_app/immutable/entry/app.C0rBoMxR.js") | |
| ]).then(([kit, app]) => { | |
| kit.start(app, element, { | |
| node_ids: [0, 316], | |
| data: [null,null], | |
| form: null, | |
| error: null | |
| }); | |
| }); | |
| } | |
| </script> | |
Xet Storage Details
- Size:
- 90.5 kB
- Xet hash:
- 0fd54ee92f4f2cfa78488de040dc2298bc416a4169701dc414e7358dc62f8428
·
Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.