Buckets:

hf-doc-build
/

doc

Files
xet
hf-doc-build/doc / diffusers /main /fr /quantization /modelopt.html
mishig
3 days ago
download
raw
26.7 kB
<meta charset="utf-8" /><meta name="hf:doc:metadata" content="{&quot;title&quot;:&quot;NVIDIA ModelOpt&quot;,&quot;local&quot;:&quot;nvidia-modelopt&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;NVIDIAModelOptConfig&quot;,&quot;local&quot;:&quot;nvidiamodeloptconfig&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Types de quantification pris en charge&quot;,&quot;local&quot;:&quot;supported-quantization-types&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Sérialisation et désérialisation des modèles quantifiés&quot;,&quot;local&quot;:&quot;serializing-and-deserializing-quantized-models&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2}],&quot;depth&quot;:1}"/>
		<link href="/docs/diffusers/main/en/_app/immutable/entry/start.0S0_iEoG.js" rel="modulepreload">
		<link href="/docs/diffusers/main/en/_app/immutable/chunks/I9eooN77.js" rel="modulepreload">
		<link href="/docs/diffusers/main/en/_app/immutable/chunks/Bupkjr9f.js" rel="modulepreload">
		<link href="/docs/diffusers/main/en/_app/immutable/entry/app.CPdOxCbw.js" rel="modulepreload">
		<link href="/docs/diffusers/main/en/_app/immutable/chunks/CyvZ3FS5.js" rel="modulepreload">
		<link href="/docs/diffusers/main/en/_app/immutable/chunks/l0QnnOED.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.DteTvmQZ.js" rel="modulepreload">
		<link href="/docs/diffusers/main/en/_app/immutable/chunks/CoSrdQax.js" rel="modulepreload">
		<link href="/docs/diffusers/main/en/_app/immutable/nodes/354.Da0b8rJZ.js" rel="modulepreload">
		<link href="/docs/diffusers/main/en/_app/immutable/chunks/DSPDl1gz.js" rel="modulepreload">
		<!--v66wdz--><meta name="hf:doc:metadata" content="{&quot;title&quot;:&quot;NVIDIA ModelOpt&quot;,&quot;local&quot;:&quot;nvidia-modelopt&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;NVIDIAModelOptConfig&quot;,&quot;local&quot;:&quot;nvidiamodeloptconfig&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Supported quantization types&quot;,&quot;local&quot;:&quot;supported-quantization-types&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Serializing and Deserializing quantized models&quot;,&quot;local&quot;:&quot;serializing-and-deserializing-quantized-models&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="nvidia-modelopt" 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="#nvidia-modelopt"><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><!---->NVIDIA ModelOpt<!----></span></h1><!--]--><!----> <p><a href="https://github.com/NVIDIA/Model-Optimizer" rel="nofollow">NVIDIA-ModelOpt</a> est une bibliothèque unifiée de techniques d’optimisation de modèles de pointe, telles que la quantification, l’élagage, la distillation, le décodage spéculatif, etc. Elle compresse les modèles d’apprentissage profond pour les frameworks de déploiement en aval tels que TensorRT-LLM ou TensorRT afin d’optimiser la vitesse d’inférence.</p> <p>Avant de commencer, assurez-vous que nvidia_modelopt est installé.</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 <span class="hljs-string">&quot;nvidia_modelopt[hf]&quot;</span><!----></pre></div><!----> <p>Quantifier un modèle en passant <code>NVIDIAModelOptConfig</code> à <a href="/docs/diffusers/main/fr/api/models/overview#diffusers.ModelMixin.from_pretrained">from_pretrained()</a> (vous pouvez également charger des modèles pré-quantifiés). Cela fonctionne pour tout modèle de toute modalité, à condition qu’il prenne en charge le chargement avec <a href="https://hf.co/docs/accelerate/index" rel="nofollow">Accelerate</a> et qu’il contienne des couches <code>torch.nn.Linear</code>.</p> <p>L'exemple ci-dessous ne quantise que les poids en FP8.</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-python "><!----><span class="hljs-keyword">import</span> torch
<span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> AutoModel, SanaPipeline, NVIDIAModelOptConfig

model_id = <span class="hljs-string">&quot;Efficient-Large-Model/Sana_600M_1024px_diffusers&quot;</span>
dtype = torch.bfloat16

quantization_config = NVIDIAModelOptConfig(quant_type=<span class="hljs-string">&quot;FP8&quot;</span>, quant_method=<span class="hljs-string">&quot;modelopt&quot;</span>)
transformer = AutoModel.from_pretrained(
    model_id,
    subfolder=<span class="hljs-string">&quot;transformer&quot;</span>,
    quantization_config=quantization_config,
    dtype=dtype,
)
pipe = SanaPipeline.from_pretrained(
    model_id,
    transformer=transformer,
    dtype=dtype,
)
pipe.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-built_in">print</span>(<span class="hljs-string">f&quot;Pipeline memory usage: <span class="hljs-subst">{torch.cuda.max_memory_reserved() / <span class="hljs-number">1024</span>**<span class="hljs-number">3</span>:<span class="hljs-number">.3</span>f}</span> GB&quot;</span>)

prompt = <span class="hljs-string">&quot;A cat holding a sign that says hello world&quot;</span>
image = pipe(
    prompt, num_inference_steps=<span class="hljs-number">50</span>, guidance_scale=<span class="hljs-number">4.5</span>, max_sequence_length=<span class="hljs-number">512</span>
).images[<span class="hljs-number">0</span>]
image.save(<span class="hljs-string">&quot;output.png&quot;</span>)<!----></pre></div><!----> <blockquote><p><strong>Remarque :</strong></p> <p>Les méthodes de quantification de NVIDIA-ModelOpt sont conçues pour réduire l’empreinte mémoire des poids du modèle à l’aide de diverses techniques de QAT (Quantization-Aware Training) et de PTQ (Post-Training Quantization), tout en préservant les performances du modèle. Toutefois, le gain de performance réel lors de l’inférence dépend du framework de déploiement (par exemple, TRT-LLM, TensorRT) et de la configuration matérielle spécifique.</p> <p>Plus de détails peuvent être trouvés <a href="https://github.com/NVIDIA/Model-Optimizer/tree/main/examples" rel="nofollow">ici</a>.</p></blockquote> <!--[1--><h2 class="relative group"><a id="nvidiamodeloptconfig" 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="#nvidiamodeloptconfig"><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><!---->NVIDIAModelOptConfig<!----></span></h2><!--]--><!----> <p>La classe <code>NVIDIAModelOptConfig</code> accepte trois paramètres :</p> <ul><li><code>quant_type</code> : une valeur de type chaîne mentionnant l'un des types de quantification ci-dessous.</li> <li><code>modules_to_not_convert</code> : une liste de noms complets ou partiels de modules pour lesquels la quantification ne doit pas être effectuée. Par exemple, pour ne quantifier aucun des blocs de projection pos_embed du <a href="/docs/diffusers/main/fr/api/models/sd3_transformer2d#diffusers.SD3Transformer2DModel">SD3Transformer2DModel</a>, on spécifierait : <code>modules_to_not_convert=["pos_embed.proj.weight"]</code>.</li> <li><code>disable_conv_quantization</code> : une valeur booléenne qui, lorsqu’elle est définie sur <code>True</code>, désactive la quantification pour toutes les couches convolutives du modèle. Cela est utile car la quantification par canal et par bloc ne fonctionne généralement pas bien avec les couches convolutives (utilisée avec INT4, NF4, NVFP4). Si vous souhaitez désactiver la quantification pour des couches convolutives spécifiques, utilisez plutôt <code>modules_to_not_convert</code>.</li> <li><code>algorithm</code> : l’algorithme à utiliser pour déterminer l’échelle, par défaut <code>"max"</code>. Consultez la documentation de modelopt pour plus d’algorithmes et de détails.</li> <li><code>forward_loop</code> : fonction de la boucle avant à utiliser pour calibrer l’activation lors de la quantification. Si elle n’est pas fournie, elle s’appuie sur des valeurs d’échelle statiques calculées à partir des poids uniquement.</li> <li><code>kwargs</code> : un dictionnaire d’arguments nommés à passer à la méthode de quantification sous-jacente, qui sera appelée en fonction de <code>quant_type</code>.</li></ul> <!--[1--><h2 class="relative group"><a id="supported-quantization-types" 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="#supported-quantization-types"><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><!---->Types de quantification pris en charge<!----></span></h2><!--]--><!----> <p>ModelOpt prend en charge la quantification int8, fp8, int4, nf4 et nvfp4, au niveau des poids uniquement, par canal et par bloc. Les méthodes de quantification sont conçues pour réduire l’empreinte mémoire des poids du modèle tout en maintenant ses performances lors de l’inférence.</p> <p>La quantification des poids uniquement stocke les poids du modèle dans un type de données à faible nombre de bits, mais effectue les calculs avec un type de données de plus haute précision, comme <code>bfloat16</code>. Cela réduit les besoins en mémoire liés aux poids du modèle, mais conserve les pics de mémoire pour le calcul des activations.</p> <p>Les méthodes de quantisation prises en charge sont les suivantes :</p> <table><thead><tr><th><strong>Type de quantification</strong></th><th><strong>Schémas pris en charge</strong></th><th><strong>Arguments obligatoires</strong></th><th><strong>Notes supplémentaires</strong></th></tr></thead><tbody><tr><td><strong>INT8</strong></td><td><code>int8 weight only</code>, <code>int8 channel quantization</code>, <code>int8 block quantization</code></td><td><code>quant_type</code>, <code>quant_type + channel_quantize</code>, <code>quant_type + channel_quantize + block_quantize</code></td><td></td></tr><tr><td><strong>FP8</strong></td><td><code>fp8 weight only</code>, <code>fp8 channel quantization</code>, <code>fp8 block quantization</code></td><td><code>quant_type</code>, <code>quant_type + channel_quantize</code>, <code>quant_type + channel_quantize + block_quantize</code></td><td></td></tr><tr><td><strong>INT4</strong></td><td><code>int4 weight only</code>, <code>int4 block quantization</code></td><td><code>quant_type</code>, <code>quant_type + channel_quantize + block_quantize</code></td><td><code>channel_quantize = -1 is only supported for now</code></td></tr><tr><td><strong>NF4</strong></td><td><code>nf4 weight only</code>, <code>nf4 double block quantization</code></td><td><code>quant_type</code>, <code>quant_type + channel_quantize + block_quantize + scale_channel_quantize</code> + <code>scale_block_quantize</code></td><td><code>channel_quantize = -1 and scale_channel_quantize = -1 are only supported for now</code></td></tr><tr><td><strong>NVFP4</strong></td><td><code>nvfp4 weight only</code>, <code>nvfp4 block quantization</code></td><td><code>quant_type</code>, <code>quant_type + channel_quantize + block_quantize</code></td><td><code>channel_quantize = -1 is only supported for now</code></td></tr></tbody></table> <p>Reportez-vous à la <a href="https://nvidia.github.io/Model-Optimizer/" rel="nofollow">documentation officielle de modelopt</a> pour une meilleure compréhension des méthodes de quantification disponibles et de la liste exhaustive des options de configuration.</p> <!--[1--><h2 class="relative group"><a id="serializing-and-deserializing-quantized-models" 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="#serializing-and-deserializing-quantized-models"><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><!---->Sérialisation et désérialisation des modèles quantifiés<!----></span></h2><!--]--><!----> <p>Pour sérialiser un modèle quantifié dans un dtype donné, chargez d'abord le modèle avec le dtype de quantification souhaité, puis enregistrez-le à l'aide de la méthode <a href="/docs/diffusers/main/fr/api/models/overview#diffusers.ModelMixin.save_pretrained">save_pretrained()</a>.</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5   text-gray-600  " title="code excerpt" type="button"><svg 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-python "><!----><span class="hljs-keyword">import</span> torch
<span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> AutoModel, NVIDIAModelOptConfig
<span class="hljs-keyword">from</span> modelopt.torch.opt <span class="hljs-keyword">import</span> enable_huggingface_checkpointing

enable_huggingface_checkpointing()

model_id = <span class="hljs-string">&quot;Efficient-Large-Model/Sana_600M_1024px_diffusers&quot;</span>
quant_config_fp8 = {<span class="hljs-string">&quot;quant_type&quot;</span>: <span class="hljs-string">&quot;FP8&quot;</span>, <span class="hljs-string">&quot;quant_method&quot;</span>: <span class="hljs-string">&quot;modelopt&quot;</span>}
quant_config_fp8 = NVIDIAModelOptConfig(**quant_config_fp8)
model = AutoModel.from_pretrained(
    model_id,
    subfolder=<span class="hljs-string">&quot;transformer&quot;</span>,
    quantization_config=quant_config_fp8,
    dtype=torch.bfloat16,
)
model.save_pretrained(<span class="hljs-string">&#x27;path/to/sana_fp8&#x27;</span>, safe_serialization=<span class="hljs-literal">False</span>)<!----></pre></div><!----> <p>Pour charger un modèle quantifié sérialisé, utilisez la méthode <a href="/docs/diffusers/main/fr/api/models/overview#diffusers.ModelMixin.from_pretrained">from_pretrained()</a>.</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5   text-gray-600  " title="code excerpt" type="button"><svg 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-python "><!----><span class="hljs-keyword">import</span> torch
<span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> AutoModel, NVIDIAModelOptConfig, SanaPipeline
<span class="hljs-keyword">from</span> modelopt.torch.opt <span class="hljs-keyword">import</span> enable_huggingface_checkpointing

enable_huggingface_checkpointing()

quantization_config = NVIDIAModelOptConfig(quant_type=<span class="hljs-string">&quot;FP8&quot;</span>, quant_method=<span class="hljs-string">&quot;modelopt&quot;</span>)
transformer = AutoModel.from_pretrained(
    <span class="hljs-string">&quot;path/to/sana_fp8&quot;</span>,
    subfolder=<span class="hljs-string">&quot;transformer&quot;</span>,
    quantization_config=quantization_config,
    dtype=torch.bfloat16,
)
pipe = SanaPipeline.from_pretrained(
    <span class="hljs-string">&quot;Efficient-Large-Model/Sana_600M_1024px_diffusers&quot;</span>,
    transformer=transformer,
    dtype=torch.bfloat16,
)
pipe.to(<span class="hljs-string">&quot;cuda&quot;</span>)  <span class="hljs-comment"># or &quot;mps&quot;, &quot;xpu&quot;, &quot;cpu&quot;</span>
prompt = <span class="hljs-string">&quot;A cat holding a sign that says hello world&quot;</span>
image = pipe(
    prompt, num_inference_steps=<span class="hljs-number">50</span>, guidance_scale=<span class="hljs-number">4.5</span>, max_sequence_length=<span class="hljs-number">512</span>
).images[<span class="hljs-number">0</span>]
image.save(<span class="hljs-string">&quot;output.png&quot;</span>)<!----></pre></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/quantization/modelopt.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_1qsdbwf = {
						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.0S0_iEoG.js"),
						import("/docs/diffusers/main/en/_app/immutable/entry/app.CPdOxCbw.js")
					]).then(([kit, app]) => {
						kit.start(app, element, {
							node_ids: [0, 354],
							data: [null,null],
							form: null,
							error: null
						});
					});
				}
			</script>
		
Xet Storage Details

Size:: 26.7 kB
Xet hash:: 050e1851f60e79fac1eac001760c0ad26a8bdf3972a20968797212513472efc4
Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.