Hugging Face's logo

Buckets:
hf-doc-build
/
doc

Files
xet
hf-doc-build/doc / diffusers /v0.43.0 /id /optimization /cache.html
mishig's picture
mishig
download
raw
39 kB
<meta charset="utf-8" /><meta name="hf:doc:metadata" content="{&quot;title&quot;:&quot;Penyimpanan sementara&quot;,&quot;local&quot;:&quot;caching&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;Pyramid Attention Broadcast&quot;,&quot;local&quot;:&quot;pyramid-attention-broadcast&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;FasterCache&quot;,&quot;local&quot;:&quot;fastercache&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;SeaCache&quot;,&quot;local&quot;:&quot;seacache&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;Cosmos 3&quot;,&quot;local&quot;:&quot;cosmos-3&quot;,&quot;sections&quot;:[],&quot;depth&quot;:3}],&quot;depth&quot;:2},{&quot;title&quot;:&quot;FirstBlockCache&quot;,&quot;local&quot;:&quot;firstblockcache&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Cache TaylorSeer&quot;,&quot;local&quot;:&quot;taylorseer-cache&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;MagCache&quot;,&quot;local&quot;:&quot;magcache&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/308.BXNeqIJG.js" rel="modulepreload">
<link href="/docs/diffusers/v0.43.0/en/_app/immutable/chunks/bKdwVYpU.js" rel="modulepreload">
<!--fs68f1--><meta name="hf:doc:metadata" content="{&quot;title&quot;:&quot;Caching&quot;,&quot;local&quot;:&quot;caching&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;Pyramid Attention Broadcast&quot;,&quot;local&quot;:&quot;pyramid-attention-broadcast&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;FasterCache&quot;,&quot;local&quot;:&quot;fastercache&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;SeaCache&quot;,&quot;local&quot;:&quot;seacache&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;Cosmos 3&quot;,&quot;local&quot;:&quot;cosmos-3&quot;,&quot;sections&quot;:[],&quot;depth&quot;:3}],&quot;depth&quot;:2},{&quot;title&quot;:&quot;FirstBlockCache&quot;,&quot;local&quot;:&quot;firstblockcache&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;TaylorSeer Cache&quot;,&quot;local&quot;:&quot;taylorseer-cache&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;MagCache&quot;,&quot;local&quot;:&quot;magcache&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="caching" 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="#caching"><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><!---->Penyimpanan sementara<!----></span></h1><!--]--><!----> <p>Caching mempercepat inferensi dengan menyimpan dan menggunakan kembali output antara dari berbagai lapisan, seperti lapisan perhatian dan lapisan feedforward, alih-alih melakukan seluruh komputasi pada setiap langkah inferensi. Hal ini secara signifikan meningkatkan kecepatan generasi dengan mengorbankan lebih banyak memori dan tidak memerlukan pelatihan tambahan.</p> <p>Panduan ini menunjukkan cara menggunakan metode caching yang didukung di Diffusers.</p> <!--[1--><h2 class="relative group"><a id="pyramid-attention-broadcast" 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="#pyramid-attention-broadcast"><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><!---->Pyramid Attention Broadcast<!----></span></h2><!--]--><!----> <p><a href="https://huggingface.co/papers/2408.12588" rel="nofollow">Pyramid Attention Broadcast (PAB)</a> didasarkan pada pengamatan bahwa output attention tidak terlalu berbeda antar timestep yang berurutan dalam proses generasi. Perbedaan attention paling kecil terjadi pada lapisan cross attention dan umumnya di-cache pada rentang timestep yang lebih panjang. Hal ini diikuti oleh lapisan temporal attention dan spatial attention.</p> <blockquote class="tip"><p>Tidak semua model video memiliki tiga jenis perhatian (cross, temporal, dan spatial)!</p></blockquote> <p>PAB dapat dikombinasikan dengan teknik lain seperti sequence parallelism dan classifier-free guidance parallelism (data parallelism) untuk menghasilkan video secara hampir real-time.</p> <p>Setel dan teruskan <a href="/docs/diffusers/v0.43.0/id/api/cache#diffusers.PyramidAttentionBroadcastConfig">PyramidAttentionBroadcastConfig</a> ke transformer pipeline untuk mengaktifkannya. <code>spatial_attention_block_skip_range</code> mengontrol seberapa sering perhitungan attention dilewati di blok spatial attention, dan <code>spatial_attention_timestep_skip_range</code> adalah rentang timestep yang akan dilewati. Perhatikan untuk memilih rentang yang tepat karena interval yang lebih kecil dapat menyebabkan kecepatan inferensi yang lebih lambat, sedangkan interval yang lebih besar dapat menghasilkan kualitas generasi yang lebih rendah.</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> Disalin</div><!----></button><!----></div> <pre class="language-python "><!----><span class="hljs-keyword">import</span> torch
<span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> CogVideoXPipeline, PyramidAttentionBroadcastConfig
pipeline = CogVideoXPipeline.from_pretrained(<span class="hljs-string">&quot;THUDM/CogVideoX-5b&quot;</span>, dtype=torch.bfloat16)
pipeline.to(<span class="hljs-string">&quot;cuda&quot;</span>) <span class="hljs-comment"># or &quot;mps&quot;, &quot;xpu&quot;, &quot;cpu&quot;</span>
config = PyramidAttentionBroadcastConfig(
spatial_attention_block_skip_range=<span class="hljs-number">2</span>,
spatial_attention_timestep_skip_range=(<span class="hljs-number">100</span>, <span class="hljs-number">800</span>),
current_timestep_callback=<span class="hljs-keyword">lambda</span>: pipe.current_timestep,
)
pipeline.transformer.enable_cache(config)<!----></pre></div><!----> <!--[1--><h2 class="relative group"><a id="fastercache" 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="#fastercache"><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><!---->FasterCache<!----></span></h2><!--]--><!----> <p><a href="https://huggingface.co/papers/2410.19355" rel="nofollow">FasterCache</a> meng-cache dan menggunakan ulang fitur perhatian serupa dengan <a href="#pyramid-attention-broadcast">PAB</a> karena perbedaan output kecil untuk setiap langkah waktu berikutnya.</p> <p>Metode ini juga dapat memilih untuk melewati prediksi cabang tanpa syarat, ketika menggunakan panduan bebas klasifikator untuk pengambilan sampel (umum pada sebagian besar model dasar), dan memperkirakan dari prediksi cabang bersyarat jika terdapat redundansi yang signifikan pada output laten yang diprediksi antara langkah waktu yang berurutan.</p> <p>Setel dan teruskan <a href="/docs/diffusers/v0.43.0/id/api/cache#diffusers.FasterCacheConfig">FasterCacheConfig</a> ke transformer pipeline untuk mengaktifkannya.</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> Disalin</div><!----></button><!----></div> <pre class="language-python "><!----><span class="hljs-keyword">import</span> torch
<span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> CogVideoXPipeline, FasterCacheConfig
pipe line= CogVideoXPipeline.from_pretrained(<span class="hljs-string">&quot;THUDM/CogVideoX-5b&quot;</span>, dtype=torch.bfloat16)
pipeline.to(<span class="hljs-string">&quot;cuda&quot;</span>) <span class="hljs-comment"># or &quot;mps&quot;, &quot;xpu&quot;, &quot;cpu&quot;</span>
config = FasterCacheConfig(
spatial_attention_block_skip_range=<span class="hljs-number">2</span>,
spatial_attention_timestep_skip_range=(-<span class="hljs-number">1</span>, <span class="hljs-number">681</span>),
current_timestep_callback=<span class="hljs-keyword">lambda</span>: pipe.current_timestep,
attention_weight_callback=<span class="hljs-keyword">lambda</span> _: <span class="hljs-number">0.3</span>,
unconditional_batch_skip_range=<span class="hljs-number">5</span>,
unconditional_batch_timestep_skip_range=(-<span class="hljs-number">1</span>, <span class="hljs-number">781</span>),
tensor_format=<span class="hljs-string">&quot;BFCHW&quot;</span>,
)
pipeline.transformer.enable_cache(config)<!----></pre></div><!----> <!--[1--><h2 class="relative group"><a id="seacache" 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="#seacache"><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><!---->SeaCache<!----></span></h2><!--]--><!----> <p><a href="https://huggingface.co/papers/2602.18993" rel="nofollow">SeaCache</a> membandingkan indikator Spectral-Evolution-Aware (SEA) antara langkah denoising yang berurutan. Ketika perubahan indikator yang terakumulasi tetap di bawah ambang batas, ia melewati tumpukan blok transformer yang mahal dan memprediksi keluarannya dari residual yang di-cache. Bangun indikator dari visual latents yang membentuk output yang dihasilkan. Sertakan frame conditioning yang bersih ketika mereka merupakan bagian dari trajektori output tersebut, seperti dalam generasi image-to-video dan video-to-video. Kecualikan visual hints terpisah yang mengondisikan generasi tetapi bukan bagian dari output. Text conditioning dikecualikan karena bukan merupakan visual latent.</p> <p>Paket Transfer Cosmos 3 mengontrol petunjuk sebagai urutan visual terpisah, sehingga adapter-nya mengecualikannya dari indikator.
Cabang Control-CFG membandingkan lintasan output yang sama sambil mempertahankan residu cache mereka sendiri. Petunjuk kontrol tetap
mengondisikan transformer.</p> <p>Implementasi menyediakan adapter bawaan untuk model-model berikut:</p> <ul><li><strong>Cosmos 3</strong> adalah integrasi utama yang dioptimalkan dan diuji. Ia meng-cache seluruh tumpukan decoder melalui batas normalisasi pasca.</li> <li><strong>Wan T2V</strong> menggunakan jalur blok berulang generik dalam mode eager. Integrasi ini menunjukkan bagaimana transformer video single-stream lainnya dapat menyediakan latens visual mentah ke SeaCache; ini bukan klaim bahwa parameter cache yang sama optimal untuk Wan atau bahwa varian Wan lainnya didukung.</li></ul> <p>Video transformer lain dapat terintegrasi dengan jalur generik ketika mereka menggunakan <code>CacheMixin</code>, mengekspos daftar blok berulang yang dikenali, dan mendaftarkan tata letak input/output blok di <code>TransformerBlockRegistry</code>. Pipeline harus memasuki <code>cache_context</code> untuk setiap panggilan transformer, melampirkan <code>step_index</code>, <code>sigma</code>, dan <code>num_inference_steps</code>, serta menggunakan nama konteks yang terpisah untuk lintasan independen seperti panduan kondisional dan non-kondisional. Sampaikan <code>raw_vision_callback</code> yang mengembalikan latens visual yang membentuk output yang dihasilkan ketika tidak ada adapter bawaan yang tersedia. Validasi kualitas output dan sesuaikan parameter cache untuk setiap model dan scheduler; dukungan dan hasil benchmark tidak dipindahkan secara otomatis dari Cosmos 3.</p> <!--[2--><h3 class="relative group"><a id="cosmos-3" 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="#cosmos-3"><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><!---->Cosmos 3<!----></span></h3><!--]--><!----> <p>SeaCache dinonaktifkan secara default. Aktifkan pada transformer; loop denoising Cosmos 3 menempelkan langkah scheduler aktif, sigma, dan jumlah langkah ke setiap panggilan <code>cache_context</code>, sehingga tidak diperlukan pengkabelan tambahan:</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> Disalin</div><!----></button><!----></div> <pre class="language-python "><!----><span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> Cosmos3OmniPipeline, SeaCacheConfig
pipe = Cosmos3OmniPipeline.from_pretrained(<span class="hljs-string">&quot;nvidia/Cosmos3-Nano&quot;</span>)
pipe.transformer.enable_cache(SeaCacheConfig(threshold=<span class="hljs-number">0.2</span>, max_consecutive_cached=<span class="hljs-number">2</span>))<!----></pre></div><!----> <p>API tingkat model ini bekerja dengan <a href="/docs/diffusers/v0.43.0/id/api/pipelines/cosmos3#diffusers.Cosmos3OmniPipeline">Cosmos3OmniPipeline</a>, <a href="/docs/diffusers/v0.43.0/id/api/pipelines/cosmos3#diffusers.Cosmos3OmniModularPipeline">Cosmos3OmniModularPipeline</a>, dan <a href="/docs/diffusers/v0.43.0/id/api/pipelines/cosmos3#diffusers.Cosmos3DistilledModularPipeline">Cosmos3DistilledModularPipeline</a>. SeaCache adalah optimasi aproksimasi dan dapat mengubah output yang dihasilkan. Panggil <code>pipe.transformer.disable_cache()</code> ketika Anda membutuhkan setiap langkah denoising untuk menjalankan transformer secara penuh.</p> <!--[1--><h2 class="relative group"><a id="firstblockcache" 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="#firstblockcache"><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><!---->FirstBlockCache<!----></span></h2><!--]--><!----> <p><a href="https://huggingface.co/docs/diffusers/main/en/api/cache#diffusers.FirstBlockCacheConfig" rel="nofollow">FirstBlock Cache</a> memeriksa seberapa besar perubahan pada lapisan awal denoiser dari satu timestep ke timestep berikutnya. Jika perubahannya kecil, model akan melewati lapisan-lapisan akhir yang mahal dan menggunakan kembali output sebelumnya.</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> Disalin</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">from</span> diffusers.hooks <span class="hljs-keyword">import</span> apply_first_block_cache, FirstBlockCacheConfig
pipeline = DiffusionPipeline.from_pretrained(
<span class="hljs-string">&quot;Qwen/Qwen-Image&quot;</span>, dtype=torch.bfloat16
)
apply_first_block_cache(pipeline.transformer, FirstBlockCacheConfig(threshold=<span class="hljs-number">0.2</span>))<!----></pre></div><!----> <!--[1--><h2 class="relative group"><a id="taylorseer-cache" 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="#taylorseer-cache"><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><!---->Cache TaylorSeer<!----></span></h2><!--]--><!----> <p><a href="https://huggingface.co/papers/2403.06923" rel="nofollow">TaylorSeer Cache</a> mempercepat inferensi difusi dengan menggunakan ekspansi deret Taylor untuk memperkirakan dan meng-cache aktivasi antara di antara langkah-langkah denoising. Metode ini memprediksi output masa depan berdasarkan komputasi masa lalu, lalu memakainya kembali pada interval tertentu untuk mengurangi perhitungan yang redundan.</p> <p>Mekanisme penyangga ini memberikan hasil yang kuat dengan overhead memori tambahan yang minimal. Untuk analisis kinerja yang lebih rinci, lihat <a href="https://github.com/huggingface/diffusers/pull/12648#issuecomment-3610615080" rel="nofollow">temuan kami di sini</a>.</p> <p>Untuk mengaktifkan TaylorSeer Cache, buat <a href="/docs/diffusers/v0.43.0/id/api/cache#diffusers.TaylorSeerCacheConfig">TaylorSeerCacheConfig</a> dan teruskan ke transformer pipeline Anda:</p> <ul><li><code>cache_interval</code>: Jumlah langkah untuk menggunakan kembali output yang di-cache sebelum melakukan forward pass penuh</li> <li><code>disable_cache_before_step</code>: Langkah-langkah awal yang menggunakan komputasi penuh untuk mengumpulkan data untuk aproksimasi</li> <li><code>max_order</code>: Akurasi aproksimasi (secara teori, nilai yang lebih tinggi meningkatkan kualitas tetapi juga meningkatkan penggunaan memori, namun kami merekomendasikan untuk disetel ke <code>1</code>)</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> Disalin</div><!----></button><!----></div> <pre class="language-python "><!----><span class="hljs-keyword">import</span> torch
<span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> FluxPipeline, TaylorSeerCacheConfig
pipe = FluxPipeline.from_pretrained(
<span class="hljs-string">&quot;black-forest-labs/FLUX.1-dev&quot;</span>,
dtype=torch.bfloat16,
).to(<span class="hljs-string">&quot;cuda&quot;</span>) <span class="hljs-comment"># or &quot;mps&quot;, &quot;xpu&quot;, &quot;cpu&quot;</span>
config = TaylorSeerCacheConfig(
cache_interval=<span class="hljs-number">5</span>,
max_order=<span class="hljs-number">1</span>,
disable_cache_before_step=<span class="hljs-number">10</span>,
taylor_factors_dtype=torch.bfloat16,
)
pipe.transformer.enable_cache(config)<!----></pre></div><!----> <!--[1--><h2 class="relative group"><a id="magcache" 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="#magcache"><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><!---->MagCache<!----></span></h2><!--]--><!----> <p><a href="https://github.com/Zehong-Ma/MagCache" rel="nofollow">MagCache</a> mempercepat inferensi dengan melewati blok transformer berdasarkan magnitudo pembaruan residual. Metode ini mengamati bahwa magnitudo pembaruan (Output - Input) meluruh secara prediktif selama proses difusi. Dengan mengakumulasi “anggaran error” berdasarkan rasio magnitudo yang dihitung sebelumnya, metode ini secara dinamis memutuskan kapan harus melewati komputasi dan menggunakan kembali residual sebelumnya.</p> <p>MagCache bergantung pada <strong>Magnitude Ratios</strong> (<code>mag_ratios</code>) yang menggambarkan kurva peluruhan ini. Rasio-rasio ini spesifik untuk checkpoint model dan scheduler. <code>FLUX_MAG_RATIOS</code> yang disertakan diukur pada FLUX.1; model lain, termasuk Flux2 dan Flux2 Klein, memerlukan kalibrasi tersendiri.</p> <p>Untuk menggunakan MagCache, Anda biasanya mengikuti proses dua langkah: <strong>Kalibrasi</strong> dan <strong>Inferensi</strong>.</p> <ol><li><strong>Kalibrasi</strong>: Jalankan inferensi sekali dengan <code>calibrate=True</code>. Hook akan mengukur magnitudo residual dan mencetak rasio yang dihitung ke konsol.</li> <li><strong>Inferensi</strong>: Teruskan rasio-rasio ini ke <code>MagCacheConfig</code> untuk mengaktifkan akselerasi.</li></ol> <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> Disalin</div><!----></button><!----></div> <pre class="language-python "><!----><span class="hljs-keyword">import</span> torch
<span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> FluxPipeline, MagCacheConfig
pipe = FluxPipeline.from_pretrained(
<span class="hljs-string">&quot;black-forest-labs/FLUX.1-schnell&quot;</span>,
dtype=torch.bfloat16
).to(<span class="hljs-string">&quot;cuda&quot;</span>) <span class="hljs-comment"># or &quot;mps&quot;, &quot;xpu&quot;, &quot;cpu&quot;</span>
<span class="hljs-comment"># 1. Calibration Step</span>
<span class="hljs-comment"># Run full inference to measure model behavior.</span>
calib_config = MagCacheConfig(calibrate=<span class="hljs-literal">True</span>, num_inference_steps=<span class="hljs-number">4</span>)
pipe.transformer.enable_cache(calib_config)
<span class="hljs-comment"># Run a prompt to trigger calibration</span>
pipe(<span class="hljs-string">&quot;A cat playing chess&quot;</span>, num_inference_steps=<span class="hljs-number">4</span>)
<span class="hljs-comment"># Logs will print something like: &quot;MagCache Calibration Results: [1.0, 1.37, 0.97, 0.87]&quot;</span>
<span class="hljs-comment"># 2. Inference Step</span>
<span class="hljs-comment"># Apply the specific ratios obtained from calibration for optimized speed.</span>
<span class="hljs-comment"># Note: For Flux models, you can also import defaults: </span>
<span class="hljs-comment"># from diffusers.hooks.mag_cache import FLUX_MAG_RATIOS</span>
mag_config = MagCacheConfig(
mag_ratios=[<span class="hljs-number">1.0</span>, <span class="hljs-number">1.37</span>, <span class="hljs-number">0.97</span>, <span class="hljs-number">0.87</span>],
num_inference_steps=<span class="hljs-number">4</span>
)
pipe.transformer.enable_cache(mag_config)
image = pipe(<span class="hljs-string">&quot;A cat playing chess&quot;</span>, num_inference_steps=<span class="hljs-number">4</span>).images[<span class="hljs-number">0</span>]<!----></pre></div><!----> <blockquote class="note"><p><code>mag_ratios</code> mewakili kurva peluruhan magnitudo intrinsik model. Rasio yang dikalibrasi untuk jumlah langkah yang tinggi (misalnya, 50) dapat digunakan kembali untuk jumlah langkah yang lebih rendah (misalnya, 20). Implementasi menggunakan interpolasi untuk memetakan kurva ke jumlah langkah inferensi saat ini.</p></blockquote> <blockquote class="tip"><p>Untuk pipeline yang menjalankan Classifier-Free Guidance secara berurutan (seperti Kandinsky 5.0), log kalibrasi mungkin mencetak dua array: satu untuk jalur Conditional dan satu untuk jalur Unconditional. Dalam kebanyakan kasus, Anda harus menggunakan array pertama (Conditional).</p></blockquote> <blockquote class="tip"><p>Untuk pipeline yang menjalankan Classifier-Free Guidance secara <strong>batched</strong> (seperti SDXL atau Flux), <code>hidden_states</code> yang diproses oleh model mengandung cabang kondisional dan non-kondisional yang digabungkan. Proses kalibrasi secara otomatis memperhitungkan hal ini, menghasilkan satu array rasio yang merepresentasikan perilaku gabungan. Anda dapat menggunakan array hasil ini secara langsung tanpa modifikasi.</p></blockquote> <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/optimization/cache.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, 308],
data: [null,null],
form: null,
error: null
});
});
}
</script>

Xet Storage Details

Size:
39 kB
·
Xet hash:
1a4d570d9c14c9cb1e7283bfd43e219573f32ce5e4c5c7d55f5d6893e355c91a

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