Buckets:

hf-doc-build
/

doc

Files
xet
hf-doc-build/doc / diffusers /v0.43.0 /zh /using-diffusers /callback.html
mishig
2 days ago
download
raw
34.1 kB
<meta charset="utf-8" /><meta name="hf:doc:metadata" content="{&quot;title&quot;:&quot;Pipeline 回调&quot;,&quot;local&quot;:&quot;pipeline-callbacks&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;早停&quot;,&quot;local&quot;:&quot;early-stopping&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;显示中间图像&quot;,&quot;local&quot;:&quot;display-intermediate-images&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;模块化流水线&quot;,&quot;local&quot;:&quot;modular-pipelines&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2}],&quot;depth&quot;:1}"/>
		<link href="/docs/diffusers/v0.43.0/en/_app/immutable/entry/start.1gaXe03k.js" rel="modulepreload">
		<link href="/docs/diffusers/v0.43.0/en/_app/immutable/chunks/BObs6oQO.js" rel="modulepreload">
		<link href="/docs/diffusers/v0.43.0/en/_app/immutable/chunks/DPnyVmXs.js" rel="modulepreload">
		<link href="/docs/diffusers/v0.43.0/en/_app/immutable/entry/app.BwdYRbCn.js" rel="modulepreload">
		<link href="/docs/diffusers/v0.43.0/en/_app/immutable/chunks/DSGUT_OW.js" rel="modulepreload">
		<link href="/docs/diffusers/v0.43.0/en/_app/immutable/chunks/C045MWWx.js" rel="modulepreload">
		<link href="/docs/diffusers/v0.43.0/en/_app/immutable/chunks/6tuJ3nyi.js" rel="modulepreload">
		<link href="/docs/diffusers/v0.43.0/en/_app/immutable/chunks/DsnmJJEf.js" rel="modulepreload">
		<link href="/docs/diffusers/v0.43.0/en/_app/immutable/nodes/0.CornNeQR.js" rel="modulepreload">
		<link href="/docs/diffusers/v0.43.0/en/_app/immutable/chunks/IaDvMxLQ.js" rel="modulepreload">
		<link href="/docs/diffusers/v0.43.0/en/_app/immutable/nodes/320.O4hvelgi.js" rel="modulepreload">
		<link href="/docs/diffusers/v0.43.0/en/_app/immutable/chunks/bKdwVYpU.js" rel="modulepreload">
		<!--apt59n--><meta name="hf:doc:metadata" content="{&quot;title&quot;:&quot;Pipeline callbacks&quot;,&quot;local&quot;:&quot;pipeline-callbacks&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;Early stopping&quot;,&quot;local&quot;:&quot;early-stopping&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Display intermediate images&quot;,&quot;local&quot;:&quot;display-intermediate-images&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Modular pipelines&quot;,&quot;local&quot;:&quot;modular-pipelines&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2}],&quot;depth&quot;:1}"/><!---->
		<link href="/docs/diffusers/v0.43.0/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="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 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 回调<!----></span></h1><!--]--><!----> <p>回调在去噪步骤结束时运行，可以修改管道状态或张量，以影响后续步骤。使用它来调整属性或张量变量，从而获得新的行为，而无需重写管道。</p> <p>这些回调函数开箱即可与经典的 <a href="/docs/diffusers/v0.43.0/zh/api/pipelines/overview#diffusers.DiffusionPipeline">DiffusionPipeline</a> 循环配合使用。在去噪循环中插入一个执行它们的块之后，它们也能与 <a href="../modular_diffusers/overview">Modular Diffusers</a> 流水线配合使用；请参阅 <a href="#modular-pipelines">Modular pipelines</a>。</p> <p>Diffusers 在流水线<a href="../api/pipelines/overview#diffusers.callbacks.PipelineCallback">概览</a>中提供了多种回调。</p> <p>要启用回调，请使用以下参数之一，配置在特定数量的去噪步骤后执行该回调的时机。</p> <ul><li><code>cutoff_step_ratio</code> 以总去噪步数的百分比形式指定回调的激活时机。当截断点需要随 <code>num_inference_steps</code> 缩放时使用（例如，在运行 40% 后丢弃 CFG）。</li> <li><code>cutoff_step_index</code> 指定回调被激活的确切步数。当您关注绝对步数时（例如，在固定的 25 步计划中的第 <code>10</code> 步），请使用此参数。</li></ul> <p>以下示例使用了 <code>cutoff_step_ratio=0.4</code>，这意味着当去噪过程达到总推理步数的 40% 时，回调将被激活。<a href="/docs/diffusers/v0.43.0/zh/api/pipelines/overview#diffusers.callbacks.SDXLCFGCutoffCallback">SDXLCFGCutoffCallback</a> 会在执行一定步数后禁用无分类器引导（CFG），这有助于在不显著影响性能的情况下节省计算资源。</p> <p>使用其中一个 <code>cutoff</code> 参数定义一个回调，并将其传递给流水线中的 <code>callback_on_step_end</code> 参数。</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> 已复制</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-keyword">import</span> torch
<span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> DPMSolverMultistepScheduler, StableDiffusionXLPipeline
<span class="hljs-keyword">from</span> diffusers.callbacks <span class="hljs-keyword">import</span> SDXLCFGCutoffCallback

callback = SDXLCFGCutoffCallback(cutoff_step_ratio=<span class="hljs-number">0.4</span>)
<span class="hljs-comment"># if using cutoff_step_index</span>
<span class="hljs-comment"># callback = SDXLCFGCutoffCallback(cutoff_step_ratio=None, cutoff_step_index=10)</span>

pipeline = StableDiffusionXLPipeline.from_pretrained(
    <span class="hljs-string">&quot;stabilityai/stable-diffusion-xl-base-1.0&quot;</span>,
    dtype=torch.float16,
    device_map=<span class="hljs-string">&quot;cuda&quot;</span>  <span class="hljs-comment"># or &quot;mps&quot;, &quot;xpu&quot;, &quot;cpu&quot;</span>
)
pipeline.scheduler = DPMSolverMultistepScheduler.from_config(pipeline.scheduler.config, use_karras_sigmas=<span class="hljs-literal">True</span>)

prompt = <span class="hljs-string">&quot;a sports car on the road, best quality, high quality, high detail, 8k resolution&quot;</span>
output = pipeline(
    prompt=prompt,
    negative_prompt=<span class="hljs-string">&quot;&quot;</span>,
    guidance_scale=<span class="hljs-number">6.5</span>,
    num_inference_steps=<span class="hljs-number">25</span>,
    callback_on_step_end=callback,
)<!----></pre></div><!----> <p>官方回调会自行设置其张量输入。对于自定义函数，请按照<a href="#display-intermediate-images">显示中间图像</a>中的示例，传入<code>callback_on_step_end_tensor_inputs</code>。</p> <p>如果您想添加一个新的官方回调，请随时提交一个<a href="https://github.com/huggingface/diffusers/issues/new/choose" rel="nofollow">功能请求</a>或<a href="https://huggingface.co/docs/diffusers/main/en/conceptual/contribution#how-to-open-a-pr" rel="nofollow">提交 PR</a>。否则，您也可以按照下文所示创建自己的回调。</p> <!--[1--><h2 class="relative group"><a id="early-stopping" 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="#early-stopping"><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><!---->早停<!----></span></h2><!--]--><!----> <p>如果您对生成过程中的中间结果不满意，早停（Early stopping）会很有用。此回调通过将 <code>_interrupt</code> 属性设置为 <code>True</code> 来设置一个硬编码的停止点，从而使去噪循环跳过剩余的步骤。</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> 已复制</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-keyword">import</span> torch
<span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> DiffusionPipeline

<span class="hljs-keyword">def</span> <span class="hljs-title function_">interrupt_callback</span>(<span class="hljs-params">pipeline, i, t, callback_kwargs</span>):
    stop_idx = <span class="hljs-number">10</span>
    <span class="hljs-keyword">if</span> i == stop_idx:
        pipeline._interrupt = <span class="hljs-literal">True</span>
    <span class="hljs-keyword">return</span> callback_kwargs

pipeline = DiffusionPipeline.from_pretrained(
    <span class="hljs-string">&quot;Qwen/Qwen-Image&quot;</span>,
    dtype=torch.bfloat16,
    device_map=<span class="hljs-string">&quot;cuda&quot;</span>,  <span class="hljs-comment"># or &quot;mps&quot;, &quot;xpu&quot;, &quot;cpu&quot;</span>
)
pipeline(
    prompt=<span class="hljs-string">&quot;A photo of a cat&quot;</span>,
    num_inference_steps=<span class="hljs-number">50</span>,
    callback_on_step_end=interrupt_callback,
)<!----></pre></div><!----> <!--[1--><h2 class="relative group"><a id="display-intermediate-images" 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="#display-intermediate-images"><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><!---->显示中间图像<!----></span></h2><!--]--><!----> <p>可视化中间图像有助于监控进度。下方的预览仅适用于 SDXL。它通过线性变换将 SDXL 的潜变量映射为 RGB，以便在去噪过程中快速查看。这些权重无法迁移到其他模型。对于 Qwen-Image 及类似的检查点，请使用模型自身的 VAE 进行解码，而不是使用此辅助函数。</p> <p><a href="https://huggingface.co/blog/TimothyAlexisVass/explaining-the-sdxl-latent-space" rel="nofollow">转换</a> Stable Diffusion XL 潜变量（4 个通道）为 RGB 张量（3 个通道）。</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> 已复制</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-keyword">import</span> torch
<span class="hljs-keyword">from</span> PIL <span class="hljs-keyword">import</span> Image
<span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> AutoPipelineForText2Image

<span class="hljs-keyword">def</span> <span class="hljs-title function_">latents_to_rgb</span>(<span class="hljs-params">latents</span>):
    weights = (
        (<span class="hljs-number">60</span>, -<span class="hljs-number">60</span>, <span class="hljs-number">25</span>, -<span class="hljs-number">70</span>),
        (<span class="hljs-number">60</span>,  -<span class="hljs-number">5</span>, <span class="hljs-number">15</span>, -<span class="hljs-number">50</span>),
        (<span class="hljs-number">60</span>,  <span class="hljs-number">10</span>, -<span class="hljs-number">5</span>, -<span class="hljs-number">35</span>),
    )

    weights_tensor = torch.t(torch.tensor(weights, dtype=latents.dtype).to(latents.device))
    biases_tensor = torch.tensor((<span class="hljs-number">150</span>, <span class="hljs-number">140</span>, <span class="hljs-number">130</span>), dtype=latents.dtype).to(latents.device)
    rgb_tensor = torch.einsum(<span class="hljs-string">&quot;...lxy,lr -&gt; ...rxy&quot;</span>, latents, weights_tensor) + biases_tensor.unsqueeze(-<span class="hljs-number">1</span>).unsqueeze(-<span class="hljs-number">1</span>)
    image_array = rgb_tensor.clamp(<span class="hljs-number">0</span>, <span class="hljs-number">255</span>).byte().cpu().numpy().transpose(<span class="hljs-number">1</span>, <span class="hljs-number">2</span>, <span class="hljs-number">0</span>)

    <span class="hljs-keyword">return</span> Image.fromarray(image_array)<!----></pre></div><!----> <p>提取潜在变量，并将批次中的第一张图像转换为 RGB 格式。将该图像保存为 PNG 文件，文件名中包含步数。</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> 已复制</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-keyword">def</span> <span class="hljs-title function_">decode_tensors</span>(<span class="hljs-params">pipe, step, timestep, callback_kwargs</span>):
    latents = callback_kwargs[<span class="hljs-string">&quot;latents&quot;</span>]

    image = latents_to_rgb(latents[<span class="hljs-number">0</span>])
    image.save(<span class="hljs-string">f&quot;<span class="hljs-subst">{step}</span>.png&quot;</span>)

    <span class="hljs-keyword">return</span> callback_kwargs<!----></pre></div><!----> <p>使用 <code>callback_on_step_end_tensor_inputs</code> 来选择回调函数接收哪些张量，在本例中即为潜变量。</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> 已复制</div><!----></button><!----></div> <pre class="language-py "><!---->pipeline = AutoPipelineForText2Image.from_pretrained(
    <span class="hljs-string">&quot;stabilityai/stable-diffusion-xl-base-1.0&quot;</span>,
    dtype=torch.float16,
    device_map=<span class="hljs-string">&quot;cuda&quot;</span>  <span class="hljs-comment"># or &quot;mps&quot;, &quot;xpu&quot;, &quot;cpu&quot;</span>
)

image = pipeline(
    prompt=<span class="hljs-string">&quot;A croissant shaped like a cute bear.&quot;</span>,
    negative_prompt=<span class="hljs-string">&quot;Deformed, ugly, bad anatomy&quot;</span>,
    callback_on_step_end=decode_tensors,
    callback_on_step_end_tensor_inputs=[<span class="hljs-string">&quot;latents&quot;</span>],
).images[<span class="hljs-number">0</span>]<!----></pre></div><!----> <!--[1--><h2 class="relative group"><a id="modular-pipelines" 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="#modular-pipelines"><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><!---->模块化流水线<!----></span></h2><!--]--><!----> <p>模块化流水线不包含 <code>callback_on_step_end</code> 和 <code>interrupt</code> 机制。取而代之的是，它将提供一流的支持流式处理的功能：<a href="https://github.com/huggingface/diffusers/pull/14159" rel="nofollow">huggingface/diffusers#14159</a>（尚未合并）添加了 <code>pipe.stream()</code>，该方法会在每个去噪步骤中返回流水线的状态，并在你停止迭代时结束运行。目前，若要复用现有的回调函数，请在去噪循环中插入一个块：该块会接收包含步骤索引和时间步的循环 <code>BlockState</code>，并且可以读取或编辑 <code>latents</code>。</p> <blockquote class="tip"><p>在调用 <a href="/docs/diffusers/v0.43.0/zh/api/modular_diffusers/pipeline_blocks#diffusers.ModularPipelineBlocks.init_pipeline">init_pipeline()</a> <em>之前</em>编辑 blocks：<code>pipe.blocks</code> 返回的是副本，因此在 pipeline 创建之后对其进行修改不会生效。</p></blockquote> <p>下方的代码块将回调函数作为 pipeline 的输入，并以与上文相同的签名调用它：即 pipeline、步骤索引、时间步以及一个张量字典，其返回值会被写回。<code>walk_denoise_loops</code> 会找到所有块中的去噪循环，无论其属于哪个工作流，因此同一段代码适用于任何模块化 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> 已复制</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-keyword">from</span> typing <span class="hljs-keyword">import</span> <span class="hljs-type">Callable</span>

<span class="hljs-keyword">import</span> torch
<span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> ModularPipeline
<span class="hljs-keyword">from</span> diffusers.modular_pipelines <span class="hljs-keyword">import</span> InputParam, LoopSequentialPipelineBlocks, ModularPipelineBlocks


<span class="hljs-keyword">class</span> <span class="hljs-title class_">CallbackOnStepEnd</span>(<span class="hljs-title class_ inherited__">ModularPipelineBlocks</span>):
<span class="hljs-meta">    @property</span>
    <span class="hljs-keyword">def</span> <span class="hljs-title function_">inputs</span>(<span class="hljs-params">self</span>):
        <span class="hljs-keyword">return</span> [
            InputParam(
                <span class="hljs-string">&quot;callback_on_step_end&quot;</span>,
                type_hint=<span class="hljs-type">Callable</span>,
                description=<span class="hljs-string">&quot;Called after each scheduler step as `callback(pipe, i, t, {&#x27;latents&#x27;: ...})`; the dict it returns is written back.&quot;</span>,
            ),
            InputParam(<span class="hljs-string">&quot;latents&quot;</span>, required=<span class="hljs-literal">True</span>),
        ]

    <span class="hljs-keyword">def</span> <span class="hljs-title function_">__call__</span>(<span class="hljs-params">self, components, block_state, i, t</span>):
        <span class="hljs-keyword">if</span> block_state.callback_on_step_end <span class="hljs-keyword">is</span> <span class="hljs-keyword">not</span> <span class="hljs-literal">None</span>:
            outputs = block_state.callback_on_step_end(components, i, t, {<span class="hljs-string">&quot;latents&quot;</span>: block_state.latents})
            block_state.latents = outputs.get(<span class="hljs-string">&quot;latents&quot;</span>, block_state.latents)
        <span class="hljs-keyword">return</span> components, block_state


<span class="hljs-keyword">def</span> <span class="hljs-title function_">walk_denoise_loops</span>(<span class="hljs-params">block</span>):
    <span class="hljs-string">&quot;&quot;&quot;Every denoising loop in the blocks, whichever workflow it belongs to.&quot;&quot;&quot;</span>
    <span class="hljs-keyword">if</span> <span class="hljs-built_in">isinstance</span>(block, LoopSequentialPipelineBlocks) <span class="hljs-keyword">and</span> <span class="hljs-string">&quot;denoiser&quot;</span> <span class="hljs-keyword">in</span> block.sub_blocks:
        <span class="hljs-keyword">yield</span> block
    <span class="hljs-keyword">for</span> sub <span class="hljs-keyword">in</span> <span class="hljs-built_in">getattr</span>(block, <span class="hljs-string">&quot;sub_blocks&quot;</span>, {}).values():
        <span class="hljs-keyword">yield</span> <span class="hljs-keyword">from</span> walk_denoise_loops(sub)


repo_id = <span class="hljs-string">&quot;stabilityai/stable-diffusion-xl-base-1.0&quot;</span>
blocks = ModularPipeline.from_pretrained(repo_id).blocks
<span class="hljs-keyword">for</span> loop <span class="hljs-keyword">in</span> walk_denoise_loops(blocks):
    loop.sub_blocks[<span class="hljs-string">&quot;callback&quot;</span>] = CallbackOnStepEnd()  <span class="hljs-comment"># appended after the last step of the loop</span>
pipe = blocks.init_pipeline(repo_id)
pipe.load_components(dtype=torch.float16)
pipe.to(<span class="hljs-string">&quot;cuda&quot;</span>)<!----></pre></div><!----> <p><code>callback_on_step_end</code> 现在是 pipeline 的输入参数，为 <code>DiffusionPipeline</code> 编写的回调函数（例如上面的 <code>decode_tensors</code>）可以直接使用：</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> 已复制</div><!----></button><!----></div> <pre class="language-py "><!---->image = pipe(prompt=<span class="hljs-string">&quot;A croissant shaped like a cute bear.&quot;</span>, callback_on_step_end=decode_tensors, output=<span class="hljs-string">&quot;images&quot;</span>)[<span class="hljs-number">0</span>]<!----></pre></div><!----> <p><a href="#early-stopping">Early stopping</a> 的工作方式有所不同：一个 block 无法跳过其所在的循环，因此设置 <code>_interrupt</code> 不会产生任何效果。请改为在回调中抛出异常：该异常会从 <code>pipe(...)</code> 中传播出来（不会解码图像），并且管道在之后仍可继续使用。</p> <p>这适用于所有模块化流水线，但嵌套循环的内层循环除外（例如 Helios 和 Wan-Animate-2 等自回归视频模型，它们在外层循环的一个块内对每个块进行去噪）。在这种情况下，请将块插入到外层循环中：它每个块运行一次，并接收块索引 <code>k</code> 来代替 <code>i</code> 和 <code>t</code>。</p> <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/using-diffusers/callback.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_ywq9g6 = {
						base: "/docs/diffusers/v0.43.0/en",
						assets: "/docs/diffusers/v0.43.0/en"
					};
					const element = document.currentScript.parentElement;
					Promise.all([
						import("/docs/diffusers/v0.43.0/en/_app/immutable/entry/start.1gaXe03k.js"),
						import("/docs/diffusers/v0.43.0/en/_app/immutable/entry/app.BwdYRbCn.js")
					]).then(([kit, app]) => {
						kit.start(app, element, {
							node_ids: [0, 320],
							data: [null,null],
							form: null,
							error: null
						});
					});
				}
			</script>
		
Xet Storage Details

Size:: 34.1 kB
Xet hash:: 139f25ebac974dca664f2f64e3c36d63caa2e5e7748501018566b0477eb76364
Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.