diffusers/v0.24.0/en/using-diffusers/callback.html

<meta charset="utf-8" /><meta name="hf:doc:metadata" content="{&quot;title&quot;:&quot;Pipeline callbacks&quot;,&quot;local&quot;:&quot;pipeline-callbacks&quot;,&quot;sections&quot;:[],&quot;depth&quot;:1}">
		<link href="/docs/diffusers/v0.24.0/en/_app/immutable/assets/0.e3b0c442.css" rel="modulepreload">
		<link rel="modulepreload" href="/docs/diffusers/v0.24.0/en/_app/immutable/entry/start.b2aeff8b.js">
		<link rel="modulepreload" href="/docs/diffusers/v0.24.0/en/_app/immutable/chunks/scheduler.182ea377.js">
		<link rel="modulepreload" href="/docs/diffusers/v0.24.0/en/_app/immutable/chunks/singletons.247b669c.js">
		<link rel="modulepreload" href="/docs/diffusers/v0.24.0/en/_app/immutable/chunks/index.1f6d62f6.js">
		<link rel="modulepreload" href="/docs/diffusers/v0.24.0/en/_app/immutable/chunks/paths.80a11d15.js">
		<link rel="modulepreload" href="/docs/diffusers/v0.24.0/en/_app/immutable/entry/app.cd130e3d.js">
		<link rel="modulepreload" href="/docs/diffusers/v0.24.0/en/_app/immutable/chunks/index.abf12888.js">
		<link rel="modulepreload" href="/docs/diffusers/v0.24.0/en/_app/immutable/nodes/0.57156b1d.js">
		<link rel="modulepreload" href="/docs/diffusers/v0.24.0/en/_app/immutable/chunks/each.e59479a4.js">
		<link rel="modulepreload" href="/docs/diffusers/v0.24.0/en/_app/immutable/nodes/162.1b50559a.js">
		<link rel="modulepreload" href="/docs/diffusers/v0.24.0/en/_app/immutable/chunks/Tip.230e2334.js">
		<link rel="modulepreload" href="/docs/diffusers/v0.24.0/en/_app/immutable/chunks/CodeBlock.57fe6e13.js">
		<link rel="modulepreload" href="/docs/diffusers/v0.24.0/en/_app/immutable/chunks/Heading.16916d63.js"><!-- HEAD_svelte-u9bgzb_START --><meta name="hf:doc:metadata" content="{&quot;title&quot;:&quot;Pipeline callbacks&quot;,&quot;local&quot;:&quot;pipeline-callbacks&quot;,&quot;sections&quot;:[],&quot;depth&quot;:1}"><!-- HEAD_svelte-u9bgzb_END -->      <p></p>   <h1 class="relative group"><a id="pipeline-callbacks" 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="#pipeline-callbacks"><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>Pipeline callbacks</span></h1> <p data-svelte-h="svelte-z3ufgq">The denoising loop of a pipeline can be modified with custom defined functions using the <code>callback_on_step_end</code> parameter. This can be really useful for <em>dynamically</em> adjusting certain pipeline attributes, or modifying tensor variables. The flexibility of callbacks opens up some interesting use-cases such as changing the prompt embeddings at each timestep, assigning different weights to the prompt embeddings, and editing the guidance scale.</p> <p data-svelte-h="svelte-1e9w3c8">This guide will show you how to use the <code>callback_on_step_end</code> parameter to disable classifier-free guidance (CFG) after 40% of the inference steps to save compute with minimal cost to performance.</p> <p data-svelte-h="svelte-gxr3fz">The callback function should have the following arguments:</p> <ul data-svelte-h="svelte-gz6h7k"><li><code>pipe</code> (or the pipeline instance) provides access to useful properties such as <code>num_timestep</code> and <code>guidance_scale</code>. You can modify these properties by updating the underlying attributes. For this example, you’ll disable CFG by setting <code>pipe._guidance_scale=0.0</code>.</li> <li><code>step_index</code> and <code>timestep</code> tell you where you are in the denoising loop. Use <code>step_index</code> to turn off CFG after reaching 40% of <code>num_timestep</code>.</li> <li><code>callback_kwargs</code> is a dict that contains tensor variables you can modify during the denoising loop. It only includes variables specified in the <code>callback_on_step_end_tensor_inputs</code> argument, which is passed to the pipeline’s <code>__call__</code> method. Different pipelines may use different sets of variables, so please check a pipeline’s <code>_callback_tensor_inputs</code> attribute for the list of variables you can modify. Some common variables include <code>latents</code> and <code>prompt_embeds</code>. For this function, change the batch size of <code>prompt_embeds</code> after setting <code>guidance_scale=0.0</code> in order for it to work properly.</li></ul> <p data-svelte-h="svelte-1s2b9st">Your callback function should look something like this:</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_">callback_dynamic_cfg</span>(<span class="hljs-params">pipe, step_index, timestep, callback_kwargs</span>):
        <span class="hljs-comment"># adjust the batch_size of prompt_embeds according to guidance_scale</span>
        <span class="hljs-keyword">if</span> step_index == <span class="hljs-built_in">int</span>(pipe.num_timestep * <span class="hljs-number">0.4</span>):
                prompt_embeds = callback_kwargs[<span class="hljs-string">&quot;prompt_embeds&quot;</span>]
                prompt_embeds = prompt_embeds.chunk(<span class="hljs-number">2</span>)[-<span class="hljs-number">1</span>]

        <span class="hljs-comment"># update guidance_scale and prompt_embeds</span>
        pipe._guidance_scale = <span class="hljs-number">0.0</span>
        callback_kwargs[<span class="hljs-string">&quot;prompt_embeds&quot;</span>] = prompt_embeds
        <span class="hljs-keyword">return</span> callback_kwargs<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-y3yxaw">Now, you can pass the callback function to the <code>callback_on_step_end</code> parameter and the <code>prompt_embeds</code> to <code>callback_on_step_end_tensor_inputs</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 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">import</span> torch
<span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> StableDiffusionPipeline

pipe = StableDiffusionPipeline.from_pretrained(<span class="hljs-string">&quot;runwayml/stable-diffusion-v1-5&quot;</span>, torch_dtype=torch.float16)
pipe = pipe.to(<span class="hljs-string">&quot;cuda&quot;</span>)

prompt = <span class="hljs-string">&quot;a photo of an astronaut riding a horse on mars&quot;</span>

generator = torch.Generator(device=<span class="hljs-string">&quot;cuda&quot;</span>).manual_seed(<span class="hljs-number">1</span>)
out = pipe(prompt, generator=generator, callback_on_step_end=callback_custom_cfg, callback_on_step_end_tensor_inputs=[<span class="hljs-string">&#x27;prompt_embeds&#x27;</span>])

out.images[<span class="hljs-number">0</span>].save(<span class="hljs-string">&quot;out_custom_cfg.png&quot;</span>)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1531nu7">The callback function is executed at the end of each denoising step, and modifies the pipeline attributes and tensor variables for the next denoising step.</p> <p data-svelte-h="svelte-14y2as5">With callbacks, you can implement features such as dynamic CFG without having to modify the underlying code at all!</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-1wsblrf">🤗 Diffusers currently only supports <code>callback_on_step_end</code>, but feel free to open a <a href="https://github.com/huggingface/diffusers/issues/new/choose" rel="nofollow">feature request</a> if you have a cool use-case and require a callback function with a different execution point!</p></div>  <p></p> 
			
			<script>
				{
					__sveltekit_98q9so = {
						assets: "/docs/diffusers/v0.24.0/en",
						base: "/docs/diffusers/v0.24.0/en",
						env: {}
					};

					const element = document.currentScript.parentElement;

					const data = [null,null];

					Promise.all([
						import("/docs/diffusers/v0.24.0/en/_app/immutable/entry/start.b2aeff8b.js"),
						import("/docs/diffusers/v0.24.0/en/_app/immutable/entry/app.cd130e3d.js")
					]).then(([kit, app]) => {
						kit.start(app, element, {
							node_ids: [0, 162],
							data,
							form: null,
							error: null
						});
					});
				}
			</script>