Buckets:
| <meta charset="utf-8" /><meta name="hf:doc:metadata" content="{"title":"Kandinsky 2.2","local":"kandinsky-22","sections":[{"title":"Script parameters","local":"script-parameters","sections":[{"title":"Min-SNR weighting","local":"min-snr-weighting","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":"Next steps","local":"next-steps","sections":[],"depth":2}],"depth":1}"> | |
| <link href="/docs/diffusers/main/en/_app/immutable/assets/0.e3b0c442.css" rel="modulepreload"> | |
| <link rel="modulepreload" href="/docs/diffusers/main/en/_app/immutable/entry/start.21e27d66.js"> | |
| <link rel="modulepreload" href="/docs/diffusers/main/en/_app/immutable/chunks/scheduler.8c3d61f6.js"> | |
| <link rel="modulepreload" href="/docs/diffusers/main/en/_app/immutable/chunks/singletons.1db06f6d.js"> | |
| <link rel="modulepreload" href="/docs/diffusers/main/en/_app/immutable/chunks/index.0997d446.js"> | |
| <link rel="modulepreload" href="/docs/diffusers/main/en/_app/immutable/chunks/paths.085e8bc8.js"> | |
| <link rel="modulepreload" href="/docs/diffusers/main/en/_app/immutable/entry/app.de4fb612.js"> | |
| <link rel="modulepreload" href="/docs/diffusers/main/en/_app/immutable/chunks/index.da70eac4.js"> | |
| <link rel="modulepreload" href="/docs/diffusers/main/en/_app/immutable/nodes/0.f6117ae5.js"> | |
| <link rel="modulepreload" href="/docs/diffusers/main/en/_app/immutable/chunks/each.e59479a4.js"> | |
| <link rel="modulepreload" href="/docs/diffusers/main/en/_app/immutable/nodes/189.fb742c11.js"> | |
| <link rel="modulepreload" href="/docs/diffusers/main/en/_app/immutable/chunks/Tip.1d9b8c37.js"> | |
| <link rel="modulepreload" href="/docs/diffusers/main/en/_app/immutable/chunks/CodeBlock.00a903b3.js"> | |
| <link rel="modulepreload" href="/docs/diffusers/main/en/_app/immutable/chunks/EditOnGithub.1e64e623.js"> | |
| <link rel="modulepreload" href="/docs/diffusers/main/en/_app/immutable/chunks/HfOption.c1483eb1.js"> | |
| <link rel="modulepreload" href="/docs/diffusers/main/en/_app/immutable/chunks/stores.d6eecc38.js"><!-- HEAD_svelte-u9bgzb_START --><meta name="hf:doc:metadata" content="{"title":"Kandinsky 2.2","local":"kandinsky-22","sections":[{"title":"Script parameters","local":"script-parameters","sections":[{"title":"Min-SNR weighting","local":"min-snr-weighting","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":"Next steps","local":"next-steps","sections":[],"depth":2}],"depth":1}"><!-- HEAD_svelte-u9bgzb_END --> <p></p> <h1 class="relative group"><a id="kandinsky-22" 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="#kandinsky-22"><span><svg class="" 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>Kandinsky 2.2</span></h1> <div class="course-tip course-tip-orange bg-gradient-to-br dark:bg-gradient-to-r before:border-orange-500 dark:before:border-orange-800 from-orange-50 dark:from-gray-900 to-white dark:to-gray-950 border border-orange-50 text-orange-700 dark:text-gray-400"><p data-svelte-h="svelte-8jxpji">This script is experimental, and it’s easy to overfit and run into issues like catastrophic forgetting. Try exploring different hyperparameters to get the best results on your dataset.</p></div> <p data-svelte-h="svelte-6gxptx">Kandinsky 2.2 is a multilingual text-to-image model capable of producing more photorealistic images. The model includes an image prior model for creating image embeddings from text prompts, and a decoder model that generates images based on the prior model’s embeddings. That’s why you’ll find two separate scripts in Diffusers for Kandinsky 2.2, one for training the prior model and one for training the decoder model. You can train both models separately, but to get the best results, you should train both the prior and decoder models.</p> <p data-svelte-h="svelte-1qig94u">Depending on your GPU, you may need to enable <code>gradient_checkpointing</code> (⚠️ not supported for the prior model!), <code>mixed_precision</code>, and <code>gradient_accumulation_steps</code> to help fit the model into memory and to speedup training. You can reduce your memory-usage even more by enabling memory-efficient attention with <a href="../optimization/xformers">xFormers</a> (version <a href="https://github.com/huggingface/diffusers/issues/2234#issuecomment-1416931212" rel="nofollow">v0.0.16</a> fails for training on some GPUs so you may need to install a development version instead).</p> <p data-svelte-h="svelte-1l8tmhd">This guide explores the <a href="https://github.com/huggingface/diffusers/blob/main/examples/kandinsky2_2/text_to_image/train_text_to_image_prior.py" rel="nofollow">train_text_to_image_prior.py</a> and the <a href="https://github.com/huggingface/diffusers/blob/main/examples/kandinsky2_2/text_to_image/train_text_to_image_decoder.py" rel="nofollow">train_text_to_image_decoder.py</a> scripts to help you become more familiar with it, and how you can adapt it for your own use-case.</p> <p data-svelte-h="svelte-if4ul">Before running the scripts, make sure you install the library from source:</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 class="" 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> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START -->git <span class="hljs-built_in">clone</span> https://github.com/huggingface/diffusers | |
| <span class="hljs-built_in">cd</span> diffusers | |
| pip install .<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-z4z0b9">Then navigate to the example folder containing the training script and install the required dependencies for the script you’re using:</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 class="" 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> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START --><span class="hljs-built_in">cd</span> examples/kandinsky2_2/text_to_image | |
| pip install -r requirements.txt<!-- HTML_TAG_END --></pre></div> <div class="course-tip bg-gradient-to-br dark:bg-gradient-to-r before:border-green-500 dark:before:border-green-800 from-green-50 dark:from-gray-900 to-white dark:to-gray-950 border border-green-50 text-green-700 dark:text-gray-400"><p data-svelte-h="svelte-1qbiqsn">🤗 Accelerate is a library for helping you train on multiple GPUs/TPUs or with mixed-precision. It’ll automatically configure your training setup based on your hardware and environment. Take a look at the 🤗 Accelerate <a href="https://huggingface.co/docs/accelerate/quicktour" rel="nofollow">Quick tour</a> to learn more.</p></div> <p data-svelte-h="svelte-60q53m">Initialize an 🤗 Accelerate environment:</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 class="" 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> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START -->accelerate config<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-di6juu">To setup a default 🤗 Accelerate environment without choosing any configurations:</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 class="" 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> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START -->accelerate config default<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-tsz4qp">Or if your environment doesn’t support an interactive shell, like a notebook, you can use:</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 class="" 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> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START --><span class="hljs-keyword">from</span> accelerate.utils <span class="hljs-keyword">import</span> write_basic_config | |
| write_basic_config()<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1fkfdql">Lastly, if you want to train a model on your own dataset, take a look at the <a href="create_dataset">Create a dataset for training</a> guide to learn how to create a dataset that works with the training script.</p> <div class="course-tip bg-gradient-to-br dark:bg-gradient-to-r before:border-green-500 dark:before:border-green-800 from-green-50 dark:from-gray-900 to-white dark:to-gray-950 border border-green-50 text-green-700 dark:text-gray-400"><p data-svelte-h="svelte-18fk2c2">The following sections highlight parts of the training scripts that are important for understanding how to modify it, but it doesn’t cover every aspect of the scripts in detail. If you’re interested in learning more, feel free to read through the scripts and let us know if you have any questions or concerns.</p></div> <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 class="" 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 parameters</span></h2> <p data-svelte-h="svelte-r2qipu">The training scripts provides many parameters to help you customize your training run. All of the parameters and their descriptions are found in the <a href="https://github.com/huggingface/diffusers/blob/6e68c71503682c8693cb5b06a4da4911dfd655ee/examples/kandinsky2_2/text_to_image/train_text_to_image_prior.py#L190" rel="nofollow"><code>parse_args()</code></a> function. The training scripts provides default values for each parameter, such as the training batch size and learning rate, but you can also set your own values in the training command if you’d like.</p> <p data-svelte-h="svelte-1r0bv1x">For example, to speedup training with mixed precision using the fp16 format, add the <code>--mixed_precision</code> parameter to the training command:</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 class="" 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> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START -->accelerate launch train_text_to_image_prior.py \ | |
| --mixed_precision=<span class="hljs-string">"fp16"</span><!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-fx0sux">Most of the parameters are identical to the parameters in the <a href="text2image#script-parameters">Text-to-image</a> training guide, so let’s get straight to a walkthrough of the Kandinsky training scripts!</p> <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 class="" 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>Min-SNR weighting</span></h3> <p data-svelte-h="svelte-isitbv">The <a href="https://huggingface.co/papers/2303.09556" rel="nofollow">Min-SNR</a> weighting strategy can help with training by rebalancing the loss to achieve faster convergence. The training script supports predicting <code>epsilon</code> (noise) or <code>v_prediction</code>, but Min-SNR is compatible with both prediction types. This weighting strategy is only supported by PyTorch and is unavailable in the Flax training script.</p> <p data-svelte-h="svelte-tp3kp">Add the <code>--snr_gamma</code> parameter and set it to the recommended value of 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 class="" 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> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START -->accelerate launch train_text_to_image_prior.py \ | |
| --snr_gamma=5.0<!-- HTML_TAG_END --></pre></div> <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 class="" 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>Training script</span></h2> <p data-svelte-h="svelte-1c9upjb">The training script is also similar to the <a href="text2image#training-script">Text-to-image</a> training guide, but it’s been modified to support training the prior and decoder models. This guide focuses on the code that is unique to the Kandinsky 2.2 training scripts.</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">prior model </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">decoder model </div></div> <div class="language-select"><p data-svelte-h="svelte-1vaqpft">The <a href="https://github.com/huggingface/diffusers/blob/6e68c71503682c8693cb5b06a4da4911dfd655ee/examples/kandinsky2_2/text_to_image/train_text_to_image_prior.py#L441" rel="nofollow"><code>main()</code></a> function contains the code for preparing the dataset and training the model.</p> <p data-svelte-h="svelte-1q9rvn2">One of the main differences you’ll notice right away is that the training script also loads a <a href="https://huggingface.co/docs/transformers/main/en/model_doc/clip#transformers.CLIPImageProcessor" rel="nofollow">CLIPImageProcessor</a> - in addition to a scheduler and tokenizer - for preprocessing images and a <a href="https://huggingface.co/docs/transformers/main/en/model_doc/clip#transformers.CLIPVisionModelWithProjection" rel="nofollow">CLIPVisionModelWithProjection</a> model for encoding the images:</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 class="" 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> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START -->noise_scheduler = DDPMScheduler(beta_schedule=<span class="hljs-string">"squaredcos_cap_v2"</span>, prediction_type=<span class="hljs-string">"sample"</span>) | |
| image_processor = CLIPImageProcessor.from_pretrained( | |
| args.pretrained_prior_model_name_or_path, subfolder=<span class="hljs-string">"image_processor"</span> | |
| ) | |
| tokenizer = CLIPTokenizer.from_pretrained(args.pretrained_prior_model_name_or_path, subfolder=<span class="hljs-string">"tokenizer"</span>) | |
| <span class="hljs-keyword">with</span> ContextManagers(deepspeed_zero_init_disabled_context_manager()): | |
| image_encoder = CLIPVisionModelWithProjection.from_pretrained( | |
| args.pretrained_prior_model_name_or_path, subfolder=<span class="hljs-string">"image_encoder"</span>, torch_dtype=weight_dtype | |
| ).<span class="hljs-built_in">eval</span>() | |
| text_encoder = CLIPTextModelWithProjection.from_pretrained( | |
| args.pretrained_prior_model_name_or_path, subfolder=<span class="hljs-string">"text_encoder"</span>, torch_dtype=weight_dtype | |
| ).<span class="hljs-built_in">eval</span>()<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-f8cqiu">Kandinsky uses a <a href="/docs/diffusers/main/en/api/models/prior_transformer#diffusers.PriorTransformer">PriorTransformer</a> to generate the image embeddings, so you’ll want to setup the optimizer to learn the prior mode’s parameters.</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 class="" 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> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START -->prior = PriorTransformer.from_pretrained(args.pretrained_prior_model_name_or_path, subfolder=<span class="hljs-string">"prior"</span>) | |
| prior.train() | |
| optimizer = optimizer_cls( | |
| prior.parameters(), | |
| lr=args.learning_rate, | |
| betas=(args.adam_beta1, args.adam_beta2), | |
| weight_decay=args.adam_weight_decay, | |
| eps=args.adam_epsilon, | |
| )<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1vfjyn1">Next, the input captions are tokenized, and images are <a href="https://github.com/huggingface/diffusers/blob/6e68c71503682c8693cb5b06a4da4911dfd655ee/examples/kandinsky2_2/text_to_image/train_text_to_image_prior.py#L632" rel="nofollow">preprocessed</a> by the <a href="https://huggingface.co/docs/transformers/main/en/model_doc/clip#transformers.CLIPImageProcessor" rel="nofollow">CLIPImageProcessor</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 class="" 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> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START --><span class="hljs-keyword">def</span> <span class="hljs-title function_">preprocess_train</span>(<span class="hljs-params">examples</span>): | |
| images = [image.convert(<span class="hljs-string">"RGB"</span>) <span class="hljs-keyword">for</span> image <span class="hljs-keyword">in</span> examples[image_column]] | |
| examples[<span class="hljs-string">"clip_pixel_values"</span>] = image_processor(images, return_tensors=<span class="hljs-string">"pt"</span>).pixel_values | |
| examples[<span class="hljs-string">"text_input_ids"</span>], examples[<span class="hljs-string">"text_mask"</span>] = tokenize_captions(examples) | |
| <span class="hljs-keyword">return</span> examples<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1t2oswx">Finally, the <a href="https://github.com/huggingface/diffusers/blob/6e68c71503682c8693cb5b06a4da4911dfd655ee/examples/kandinsky2_2/text_to_image/train_text_to_image_prior.py#L718" rel="nofollow">training loop</a> converts the input images into latents, adds noise to the image embeddings, and makes a prediction:</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 class="" 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> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START -->model_pred = prior( | |
| noisy_latents, | |
| timestep=timesteps, | |
| proj_embedding=prompt_embeds, | |
| encoder_hidden_states=text_encoder_hidden_states, | |
| attention_mask=text_mask, | |
| ).predicted_image_embedding<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-6gmbd2">If you want to learn more about how the training loop works, check out the <a href="../using-diffusers/write_own_pipeline">Understanding pipelines, models and schedulers</a> tutorial which breaks down the basic pattern of the denoising process.</p> </div> <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 class="" 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>Launch the script</span></h2> <p data-svelte-h="svelte-9dei1q">Once you’ve made all your changes or you’re okay with the default configuration, you’re ready to launch the training script! 🚀</p> <p data-svelte-h="svelte-1iz4o7j">You’ll train on the <a href="https://huggingface.co/datasets/lambdalabs/naruto-blip-captions" rel="nofollow">Naruto BLIP captions</a> dataset to generate your own Naruto characters, but you can also create and train on your own dataset by following the <a href="create_dataset">Create a dataset for training</a> guide. Set the environment variable <code>DATASET_NAME</code> to the name of the dataset on the Hub or if you’re training on your own files, set the environment variable <code>TRAIN_DIR</code> to a path to your dataset.</p> <p data-svelte-h="svelte-mj0rx7">If you’re training on more than one GPU, add the <code>--multi_gpu</code> parameter to the <code>accelerate launch</code> command.</p> <div class="course-tip bg-gradient-to-br dark:bg-gradient-to-r before:border-green-500 dark:before:border-green-800 from-green-50 dark:from-gray-900 to-white dark:to-gray-950 border border-green-50 text-green-700 dark:text-gray-400"><p data-svelte-h="svelte-1sfnrue">To monitor training progress with Weights & Biases, add the <code>--report_to=wandb</code> parameter to the training command. You’ll also need to add the <code>--validation_prompt</code> to the training command to keep track of results. This can be really useful for debugging the model and viewing intermediate results.</p></div> <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">prior model </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">decoder model </div></div> <div class="language-select"><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 class="" 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> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START --><span class="hljs-built_in">export</span> DATASET_NAME=<span class="hljs-string">"lambdalabs/naruto-blip-captions"</span> | |
| accelerate launch --mixed_precision=<span class="hljs-string">"fp16"</span> train_text_to_image_prior.py \ | |
| --dataset_name=<span class="hljs-variable">$DATASET_NAME</span> \ | |
| --resolution=768 \ | |
| --train_batch_size=1 \ | |
| --gradient_accumulation_steps=4 \ | |
| --max_train_steps=15000 \ | |
| --learning_rate=1e-05 \ | |
| --max_grad_norm=1 \ | |
| --checkpoints_total_limit=3 \ | |
| --lr_scheduler=<span class="hljs-string">"constant"</span> \ | |
| --lr_warmup_steps=0 \ | |
| --validation_prompts=<span class="hljs-string">"A robot naruto, 4k photo"</span> \ | |
| --report_to=<span class="hljs-string">"wandb"</span> \ | |
| --push_to_hub \ | |
| --output_dir=<span class="hljs-string">"kandi2-prior-naruto-model"</span><!-- HTML_TAG_END --></pre></div> </div> <p data-svelte-h="svelte-5prdlk">Once training is finished, you can use your newly trained model for inference!</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">prior model </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">decoder model </div></div> <div class="language-select"><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 class="" 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> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START --><span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> AutoPipelineForText2Image, DiffusionPipeline | |
| <span class="hljs-keyword">import</span> torch | |
| prior_pipeline = DiffusionPipeline.from_pretrained(output_dir, torch_dtype=torch.float16) | |
| prior_components = {<span class="hljs-string">"prior_"</span> + k: v <span class="hljs-keyword">for</span> k,v <span class="hljs-keyword">in</span> prior_pipeline.components.items()} | |
| pipeline = AutoPipelineForText2Image.from_pretrained(<span class="hljs-string">"kandinsky-community/kandinsky-2-2-decoder"</span>, **prior_components, torch_dtype=torch.float16) | |
| pipe.enable_model_cpu_offload() | |
| prompt=<span class="hljs-string">"A robot naruto, 4k photo"</span> | |
| image = pipeline(prompt=prompt, negative_prompt=negative_prompt).images[<span class="hljs-number">0</span>]<!-- HTML_TAG_END --></pre></div> <div class="course-tip bg-gradient-to-br dark:bg-gradient-to-r before:border-green-500 dark:before:border-green-800 from-green-50 dark:from-gray-900 to-white dark:to-gray-950 border border-green-50 text-green-700 dark:text-gray-400"><p data-svelte-h="svelte-kjbjgh">Feel free to replace <code>kandinsky-community/kandinsky-2-2-decoder</code> with your own trained decoder checkpoint!</p></div> </div> <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 class="" 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>Next steps</span></h2> <p data-svelte-h="svelte-135a8e3">Congratulations on training a Kandinsky 2.2 model! To learn more about how to use your new model, the following guides may be helpful:</p> <ul data-svelte-h="svelte-1mj75la"><li>Read the <a href="../using-diffusers/kandinsky">Kandinsky</a> guide to learn how to use it for a variety of different tasks (text-to-image, image-to-image, inpainting, interpolation), and how it can be combined with a ControlNet.</li> <li>Check out the <a href="dreambooth">DreamBooth</a> and <a href="lora">LoRA</a> training guides to learn how to train a personalized Kandinsky model with just a few example images. These two training techniques can even be combined!</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/kandinsky.md" target="_blank"><span data-svelte-h="svelte-1kd6by1"><</span> <span data-svelte-h="svelte-x0xyl0">></span> <span data-svelte-h="svelte-1dajgef"><span class="underline ml-1.5">Update</span> on GitHub</span></a> <p></p> | |
| <script> | |
| { | |
| __sveltekit_1eu7tzz = { | |
| assets: "/docs/diffusers/main/en", | |
| base: "/docs/diffusers/main/en", | |
| env: {} | |
| }; | |
| const element = document.currentScript.parentElement; | |
| const data = [null,null]; | |
| Promise.all([ | |
| import("/docs/diffusers/main/en/_app/immutable/entry/start.21e27d66.js"), | |
| import("/docs/diffusers/main/en/_app/immutable/entry/app.de4fb612.js") | |
| ]).then(([kit, app]) => { | |
| kit.start(app, element, { | |
| node_ids: [0, 189], | |
| data, | |
| form: null, | |
| error: null | |
| }); | |
| }); | |
| } | |
| </script> | |
Xet Storage Details
- Size:
- 44.4 kB
- Xet hash:
- 65f75a5f1056bdac3ec355546efea9e12f3381b39d4eb9a14fd02da895ccda97
·
Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.