Buckets:

hf-doc-build
/

doc

Files
xet
hf-doc-build/doc / diffusers /main /fr /optimization /speed-memory-optims.html
mishig
1 day ago
download
raw
28.5 kB
<meta charset="utf-8" /><meta name="hf:doc:metadata" content="{&quot;title&quot;:&quot;Quantifier, compiler et décharger&quot;,&quot;local&quot;:&quot;quantize-compile-and-offload&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;Quantification et torch.compile&quot;,&quot;local&quot;:&quot;quantization-and-torchcompile&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Quantification, torch.compile et offloading&quot;,&quot;local&quot;:&quot;quantization-torchcompile-and-offloading&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2}],&quot;depth&quot;:1}"/>
		<link href="/docs/diffusers/main/en/_app/immutable/entry/start.B6roO9Zr.js" rel="modulepreload">
		<link href="/docs/diffusers/main/en/_app/immutable/chunks/CjYFLaCl.js" rel="modulepreload">
		<link href="/docs/diffusers/main/en/_app/immutable/chunks/Cc4T7sti.js" rel="modulepreload">
		<link href="/docs/diffusers/main/en/_app/immutable/entry/app.CxxEVG8A.js" rel="modulepreload">
		<link href="/docs/diffusers/main/en/_app/immutable/chunks/C8F28shN.js" rel="modulepreload">
		<link href="/docs/diffusers/main/en/_app/immutable/chunks/BTnLXsn-.js" rel="modulepreload">
		<link href="/docs/diffusers/main/en/_app/immutable/chunks/DnVIJfC5.js" rel="modulepreload">
		<link href="/docs/diffusers/main/en/_app/immutable/chunks/DsnmJJEf.js" rel="modulepreload">
		<link href="/docs/diffusers/main/en/_app/immutable/nodes/0.CmHT0Edy.js" rel="modulepreload">
		<link href="/docs/diffusers/main/en/_app/immutable/chunks/C7Ky3Ped.js" rel="modulepreload">
		<link href="/docs/diffusers/main/en/_app/immutable/nodes/11.DQgyMkXt.js" rel="modulepreload">
		<link href="/docs/diffusers/main/en/_app/immutable/chunks/D2YP2_0L.js" rel="modulepreload">
		<link href="/docs/diffusers/main/en/_app/immutable/chunks/i2teRydZ.js" rel="modulepreload">
		<!--qpa0fj--><meta name="hf:doc:metadata" content="{&quot;title&quot;:&quot;Quantize, compile, and offload&quot;,&quot;local&quot;:&quot;quantize-compile-and-offload&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;Quantization and torch.compile&quot;,&quot;local&quot;:&quot;quantization-and-torchcompile&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Quantization, torch.compile, and offloading&quot;,&quot;local&quot;:&quot;quantization-torchcompile-and-offloading&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2}],&quot;depth&quot;:1}"/><!---->
		<link href="/docs/diffusers/main/en/_app/immutable/assets/0.tn0RQdqM.css" rel="modulepreload"> <!--[--><!--[0--><!--[--><!--[0--><!--[--><p></p> <div class="items-center shrink-0 min-w-[100px] max-sm:min-w-[50px] justify-end ml-auto flex" style="float: inline-end; margin-inline-start: 10px; display: inline-flex; position: relative; z-index: 10;"><div dir="ltr" class="inline-flex rounded-md max-sm:rounded-sm"><button class="inline-flex items-center gap-1 h-7 max-sm:h-7 px-2 max-sm:px-1.5 text-sm font-medium text-gray-800 border border-r-0 rounded-l-md max-sm:rounded-l-sm border-gray-200 bg-white hover:shadow-inner dark:border-gray-850 dark:bg-gray-950 dark:text-gray-200 dark:hover:bg-gray-800" aria-live="polite"><span class="inline-flex items-center justify-center rounded-md p-0.5 max-sm:p-0 hover:text-gray-800 dark:hover:text-gray-200"><svg class="sm:size-3.5 size-3" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----></span> <span dir="auto">Copy page</span></button> <button class="inline-flex items-center justify-center w-6 max-sm:w-5 h-7 max-sm:h-7 disabled:pointer-events-none text-sm text-gray-500 hover:text-gray-700 dark:hover:text-white rounded-r-md max-sm:rounded-r-sm border border-l transition border-gray-200 bg-white hover:shadow-inner dark:border-gray-850 dark:bg-gray-950 dark:text-gray-200 dark:hover:bg-gray-800" aria-haspopup="menu" aria-expanded="false" aria-label="Open copy menu"><svg class="transition-transform text-gray-400 overflow-visible sm:size-3.5 size-3 rotate-0" width="1em" height="1em" viewBox="0 0 12 7" fill="none" xmlns="http://www.w3.org/2000/svg"><path d="M1 1L6 6L11 1" stroke="currentColor"></path></svg><!----></button></div> <!--[-1--><!--]--></div><!----> <!--[0--><h1 class="relative group"><a id="quantize-compile-and-offload" 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="#quantize-compile-and-offload"><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><!---->Quantifier, compiler et décharger<!----></span></h1><!--]--><!----> <p>L’optimisation des modèles implique souvent des compromis entre la <a href="./fp16">vitesse d’inférence</a> et la <a href="./memory">consommation mémoire</a>. Par exemple, bien que le <a href="./cache">cache</a> puisse accélérer l’inférence, il augmente également la consommation mémoire car il doit stocker les sorties des couches d’attention intermédiaires. Une stratégie d’optimisation plus équilibrée combine la quantification d’un modèle, <a href="./fp16#torchcompile">torch.compile</a> et diverses <a href="./memory#offloading">méthodes de déchargement</a>.</p> <blockquote class="tip"><p>Consultez le guide <a href="./fp16#torchcompile">torch.compile</a> pour en savoir plus sur la compilation et sur la manière dont elle peut être appliquée ici. Par exemple, la compilation régionale peut réduire considérablement le temps de compilation sans renoncer aux gains de vitesse.</p></blockquote> <p>Pour la génération d'images, la combinaison de la quantification et du <a href="./memory#model-offloading">déchargement du modèle</a> offre souvent le meilleur compromis entre qualité, vitesse et mémoire. Le déchargement par groupes est moins efficace pour la génération d'images, car il n'est généralement pas possible de <em>complètement</em> superposer le transfert de données si le noyau de calcul se termine plus rapidement. Cela entraîne un certain surcoût de communication entre le CPU et le GPU.</p> <p>Pour la génération de vidéos, combiner la quantification et le <a href="./memory#group-offloading">group-offloading</a> est généralement plus efficace, car les modèles de vidéo sont plus limités par la puissance de calcul.</p> <p>Le tableau ci-dessous fournit une comparaison des combinaisons de stratégies d’optimisation et de leur impact sur la latence et l’utilisation de la mémoire pour Flux.</p> <table><thead><tr><th>combinaison</th><th>latence (s)</th><th>utilisation de la mémoire (Go)</th></tr></thead><tbody><tr><td>quantification</td><td>32.602</td><td>14.9453</td></tr><tr><td>quantization, torch.compile</td><td>25.847</td><td>14.9448</td></tr><tr><td>quantization, torch.compile, offloading du modèle sur CPU</td><td>32.312</td><td>12.2369</td></tr></tbody></table> <small>These results are benchmarked on Flux with a RTX 4090. The transformer and text_encoder components are quantized. Refer to the <a href="https://gist.github.com/sayakpaul/0db9d8eeeb3d2a0e5ed7cf0d9ca19b7d">benchmarking script</a> if you're interested in evaluating your own model.</small> <p>Ce guide vous montre comment compiler et décharger un modèle quantifié avec <a href="../quantization/bitsandbytes#torchcompile">bitsandbytes</a>. Assurez-vous d'utiliser <a href="https://pytorch.org/get-started/locally/" rel="nofollow">PyTorch nightly</a> et la dernière version de bitsandbytes.</p> <p>Bien que nous utilisions bitsandbytes dans cet exemple, d'autres backends de quantification tels que <a href="../quantization/torchao">TorchAO</a> prennent également en charge ces fonctionnalités.</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> Copié</div><!----></button><!----></div> <pre class="language-bash "><!---->pip install -U bitsandbytes<!----></pre></div><!----> <!--[1--><h2 class="relative group"><a id="quantization-and-torchcompile" 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="#quantization-and-torchcompile"><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><!---->Quantification et torch.compile<!----></span></h2><!--]--><!----> <p>Commencez par <a href="../quantization/overview">quantiser</a> un modèle pour réduire la mémoire nécessaire au stockage, puis <a href="./fp16#torchcompile">compiler</a> ce modèle pour accélérer l'inférence.</p> <p>Configurez <a href="https://docs.pytorch.org/docs/stable/torch.compiler_dynamo_overview.html" rel="nofollow">Dynamo</a> avec <code>capture_dynamic_output_shape_ops = True</code> pour gérer les sorties dynamiques lors de la compilation des modèles bitsandbytes.</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5   text-gray-600  " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Copié</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.quantizers <span class="hljs-keyword">import</span> PipelineQuantizationConfig

torch._dynamo.config.capture_dynamic_output_shape_ops = <span class="hljs-literal">True</span>

<span class="hljs-comment"># quantize</span>
pipeline_quant_config = PipelineQuantizationConfig(
    quant_backend=<span class="hljs-string">&quot;bitsandbytes_4bit&quot;</span>,
    quant_kwargs={<span class="hljs-string">&quot;load_in_4bit&quot;</span>: <span class="hljs-literal">True</span>, <span class="hljs-string">&quot;bnb_4bit_quant_type&quot;</span>: <span class="hljs-string">&quot;nf4&quot;</span>, <span class="hljs-string">&quot;bnb_4bit_compute_dtype&quot;</span>: torch.bfloat16},
    components_to_quantize=[<span class="hljs-string">&quot;transformer&quot;</span>, <span class="hljs-string">&quot;text_encoder_2&quot;</span>],
)
pipeline = DiffusionPipeline.from_pretrained(
    <span class="hljs-string">&quot;black-forest-labs/FLUX.1-dev&quot;</span>,
    quantization_config=pipeline_quant_config,
    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"># compile</span>
pipeline.transformer.to(memory_format=torch.channels_last)
pipeline.transformer.<span class="hljs-built_in">compile</span>(mode=<span class="hljs-string">&quot;max-autotune&quot;</span>, fullgraph=<span class="hljs-literal">True</span>)
pipeline(<span class="hljs-string">&quot;&quot;&quot;
    cinematic film still of a cat sipping a margarita in a pool in Palm Springs, California
    highly detailed, high budget hollywood movie, cinemascope, moody, epic, gorgeous, film grain
&quot;&quot;&quot;</span>
).images[<span class="hljs-number">0</span>]<!----></pre></div><!----> <!--[1--><h2 class="relative group"><a id="quantization-torchcompile-and-offloading" 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="#quantization-torchcompile-and-offloading"><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><!---->Quantification, torch.compile et offloading<!----></span></h2><!--]--><!----> <p>En plus de la quantification et de torch.compile, essayez le déchargement (offloading) si vous devez réduire davantage l'utilisation de la mémoire. Le déchargement déplace diverses couches ou composants du modèle du CPU vers le GPU au fur et à mesure des besoins de calcul.</p> <p>Configurez <code>cache_size_limit</code> de <a href="https://docs.pytorch.org/docs/stable/torch.compiler_dynamo_overview.html" rel="nofollow">Dynamo</a> lors du déchargement pour éviter une recompilation excessive et définissez <code>capture_dynamic_output_shape_ops = True</code> pour gérer les sorties dynamiques lors de la compilation des modèles bitsandbytes.</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">Déchargement du modèle sur CPU</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">group offloading</div><!--]--></div> <div class="language-select"><div><p><a href="./memory#model-offloading">Le déchargement du modèle sur CPU</a> déplace un composant individuel du pipeline, comme le modèle transformer, sur le GPU lorsqu'il est nécessaire pour le calcul. Sinon, il est déchargé sur le CPU.</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> Copié</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.quantizers <span class="hljs-keyword">import</span> PipelineQuantizationConfig

torch._dynamo.config.cache_size_limit = <span class="hljs-number">1000</span>
torch._dynamo.config.capture_dynamic_output_shape_ops = <span class="hljs-literal">True</span>

<span class="hljs-comment"># quantize</span>
pipeline_quant_config = PipelineQuantizationConfig(
    quant_backend=<span class="hljs-string">&quot;bitsandbytes_4bit&quot;</span>,
    quant_kwargs={<span class="hljs-string">&quot;load_in_4bit&quot;</span>: <span class="hljs-literal">True</span>, <span class="hljs-string">&quot;bnb_4bit_quant_type&quot;</span>: <span class="hljs-string">&quot;nf4&quot;</span>, <span class="hljs-string">&quot;bnb_4bit_compute_dtype&quot;</span>: torch.bfloat16},
    components_to_quantize=[<span class="hljs-string">&quot;transformer&quot;</span>, <span class="hljs-string">&quot;text_encoder_2&quot;</span>],
)
pipeline = DiffusionPipeline.from_pretrained(
    <span class="hljs-string">&quot;black-forest-labs/FLUX.1-dev&quot;</span>,
    quantization_config=pipeline_quant_config,
    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"># model CPU offloading</span>
pipeline.enable_model_cpu_offload()

<span class="hljs-comment"># compile</span>
pipeline.transformer.<span class="hljs-built_in">compile</span>()
pipeline(
    <span class="hljs-string">&quot;cinematic film still of a cat sipping a margarita in a pool in Palm Springs, California, highly detailed, high budget hollywood movie, cinemascope, moody, epic, gorgeous, film grain&quot;</span>
).images[<span class="hljs-number">0</span>]<!----></pre></div><!----><!----></div><!----> <div hidden=""><p><a href="./memory#group-offloading">Group offloading</a> déplace les couches internes d’un composant individuel du pipeline, comme le modèle transformer, vers le GPU pour le calcul, puis les décharge lorsqu’elles ne sont plus nécessaires. En même temps, il utilise la fonctionnalité <a href="./memory#cuda-stream">CUDA stream</a> pour précharger la couche suivante en vue de son exécution.</p> <p>En superposant le calcul et le transfert de données, il est plus rapide que le déchargement du modèle sur CPU tout en économisant de la mémoire.</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> Copié</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-comment"># pip install ftfy</span>
<span class="hljs-keyword">import</span> torch
<span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> AutoModel, DiffusionPipeline
<span class="hljs-keyword">from</span> diffusers.hooks <span class="hljs-keyword">import</span> apply_group_offloading
<span class="hljs-keyword">from</span> diffusers.utils <span class="hljs-keyword">import</span> export_to_video
<span class="hljs-keyword">from</span> diffusers.quantizers <span class="hljs-keyword">import</span> PipelineQuantizationConfig
<span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> UMT5EncoderModel

torch._dynamo.config.cache_size_limit = <span class="hljs-number">1000</span>
torch._dynamo.config.capture_dynamic_output_shape_ops = <span class="hljs-literal">True</span>

<span class="hljs-comment"># quantize</span>
pipeline_quant_config = PipelineQuantizationConfig(
    quant_backend=<span class="hljs-string">&quot;bitsandbytes_4bit&quot;</span>,
    quant_kwargs={<span class="hljs-string">&quot;load_in_4bit&quot;</span>: <span class="hljs-literal">True</span>, <span class="hljs-string">&quot;bnb_4bit_quant_type&quot;</span>: <span class="hljs-string">&quot;nf4&quot;</span>, <span class="hljs-string">&quot;bnb_4bit_compute_dtype&quot;</span>: torch.bfloat16},
    components_to_quantize=[<span class="hljs-string">&quot;transformer&quot;</span>, <span class="hljs-string">&quot;text_encoder&quot;</span>],
)

text_encoder = UMT5EncoderModel.from_pretrained(
    <span class="hljs-string">&quot;Wan-AI/Wan2.1-T2V-14B-Diffusers&quot;</span>, subfolder=<span class="hljs-string">&quot;text_encoder&quot;</span>, dtype=torch.bfloat16
)
pipeline = DiffusionPipeline.from_pretrained(
    <span class="hljs-string">&quot;Wan-AI/Wan2.1-T2V-14B-Diffusers&quot;</span>,
    quantization_config=pipeline_quant_config,
    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"># group offloading</span>
onload_device = torch.device(<span class="hljs-string">&quot;cuda&quot;</span>)
offload_device = torch.device(<span class="hljs-string">&quot;cpu&quot;</span>)

pipeline.transformer.enable_group_offload(
    onload_device=onload_device,
    offload_device=offload_device,
    offload_type=<span class="hljs-string">&quot;leaf_level&quot;</span>,
    use_stream=<span class="hljs-literal">True</span>,
    non_blocking=<span class="hljs-literal">True</span>
)
pipeline.vae.enable_group_offload(
    onload_device=onload_device,
    offload_device=offload_device,
    offload_type=<span class="hljs-string">&quot;leaf_level&quot;</span>,
    use_stream=<span class="hljs-literal">True</span>,
    non_blocking=<span class="hljs-literal">True</span>
)
apply_group_offloading(
    pipeline.text_encoder,
    onload_device=onload_device,
    offload_type=<span class="hljs-string">&quot;leaf_level&quot;</span>,
    use_stream=<span class="hljs-literal">True</span>,
    non_blocking=<span class="hljs-literal">True</span>
)

<span class="hljs-comment"># compile</span>
pipeline.transformer.<span class="hljs-built_in">compile</span>()

prompt = <span class="hljs-string">&quot;&quot;&quot;
The camera rushes from far to near in a low-angle shot, 
revealing a white ferret on a log. It plays, leaps into the water, and emerges, as the camera zooms in 
for a close-up. Water splashes berry bushes nearby, while moss, snow, and leaves blanket the ground. 
Birch trees and a light blue sky frame the scene, with ferns in the foreground. Side lighting casts dynamic 
shadows and warm highlights. Medium composition, front view, low angle, with depth of field.
&quot;&quot;&quot;</span>
negative_prompt = <span class="hljs-string">&quot;&quot;&quot;
Bright tones, overexposed, static, blurred details, subtitles, style, works, paintings, images, static, overall gray, worst quality, 
low quality, JPEG compression residue, ugly, incomplete, extra fingers, poorly drawn hands, poorly drawn faces, deformed, disfigured, 
misshapen limbs, fused fingers, still picture, messy background, three legs, many people in the background, walking backwards
&quot;&quot;&quot;</span>

output = pipeline(
    prompt=prompt,
    negative_prompt=negative_prompt,
    num_frames=<span class="hljs-number">81</span>,
    guidance_scale=<span class="hljs-number">5.0</span>,
).frames[<span class="hljs-number">0</span>]
export_to_video(output, <span class="hljs-string">&quot;output.mp4&quot;</span>, fps=<span class="hljs-number">16</span>)<!----></pre></div><!----><!----></div><!----><!----></div><!----> <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/speed-memory-optims.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_kft2r9 = {
						base: "/docs/diffusers/main/en",
						assets: "/docs/diffusers/main/en"
					};
					const element = document.currentScript.parentElement;
					Promise.all([
						import("/docs/diffusers/main/en/_app/immutable/entry/start.B6roO9Zr.js"),
						import("/docs/diffusers/main/en/_app/immutable/entry/app.CxxEVG8A.js")
					]).then(([kit, app]) => {
						kit.start(app, element, {
							node_ids: [0, 11],
							data: [null,null],
							form: null,
							error: null
						});
					});
				}
			</script>
		
Xet Storage Details

Size:: 28.5 kB
Xet hash:: 3fef4386537d21c336ae9ca395887d7169e94d38810526a5569b0fafefb25f5b
Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.