Buckets:
| <meta charset="utf-8" /><meta name="hf:doc:metadata" content="{"title":"torchao","local":"torchao","sections":[{"title":"torch.compile","local":"torchcompile","sections":[],"depth":2},{"title":"Desteklenen kuantizasyon türleri","local":"supported-quantization-types","sections":[],"depth":2},{"title":"Kantize modellerin seri hale getirilmesi ve dizide çıkarılması","local":"serializing-and-deserializing-quantized-models","sections":[],"depth":2},{"title":"Kaynaklar","local":"resources","sections":[],"depth":2}],"depth":1}"/> | |
| <link href="/docs/diffusers/main/en/_app/immutable/entry/start.CM-2NerT.js" rel="modulepreload"> | |
| <link href="/docs/diffusers/main/en/_app/immutable/chunks/CTWHa_kr.js" rel="modulepreload"> | |
| <link href="/docs/diffusers/main/en/_app/immutable/chunks/BLi2et3c.js" rel="modulepreload"> | |
| <link href="/docs/diffusers/main/en/_app/immutable/entry/app.DrYcqW7m.js" rel="modulepreload"> | |
| <link href="/docs/diffusers/main/en/_app/immutable/chunks/BeiZzCMo.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/chunks/t03vT2vk.js" rel="modulepreload"> | |
| <link href="/docs/diffusers/main/en/_app/immutable/nodes/0.DdeITYyU.js" rel="modulepreload"> | |
| <link href="/docs/diffusers/main/en/_app/immutable/chunks/DAwShlVY.js" rel="modulepreload"> | |
| <link href="/docs/diffusers/main/en/_app/immutable/nodes/12.5axJKIfg.js" rel="modulepreload"> | |
| <link href="/docs/diffusers/main/en/_app/immutable/chunks/8Bi71p98.js" rel="modulepreload"> | |
| <!--1q5ej5h--><meta name="hf:doc:metadata" content="{"title":"torchao","local":"torchao","sections":[{"title":"torch.compile","local":"torchcompile","sections":[],"depth":2},{"title":"Supported quantization types","local":"supported-quantization-types","sections":[],"depth":2},{"title":"Serializing and Deserializing quantized models","local":"serializing-and-deserializing-quantized-models","sections":[],"depth":2},{"title":"Resources","local":"resources","sections":[],"depth":2}],"depth":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="torchao" 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="#torchao"><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><!---->torchao<!----></span></h1><!--]--><!----> <p><a href="https://github.com/pytorch/ao" rel="nofollow">torchao</a>, PyTorch modellerinin çıkarım ve eğitiminde kullanılmak üzere, kuantizasyon ve seyreklik tabanlı yüksek performanslı dtypes ve optimizasyonlar sağlar. Herhangi bir modalitedeki herhangi bir model için desteklenir; yeter ki <a href="https://hf.co/docs/accelerate/index" rel="nofollow">Accelerate</a> ile yüklenmeyi desteklesin ve <code>torch.nn.Linear</code> katmanları içersin.</p> <p>Aşağıdaki komutla Pytorch 2.5+ ve torchao'nun yüklendiğinden emin olun.</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> Kopyalandı</div><!----></button><!----></div> <pre class="language-bash "><!---->uv pip install -U torch torchao<!----></pre></div><!----> <p>Her kuantizasyon dtype'ı, <a href="https://docs.pytorch.org/ao/main/api_reference/generated/torchao.core.config.AOBaseConfig.html" rel="nofollow">AOBaseConfig</a> sınıfının ayrı bir örneği olarak kullanılabilir. Bu, daha fazla kullanılabilir argüman açığa çıkararak daha esnek yapılandırma seçenekleri sunar.</p> <p><code>AOBaseConfig</code>'ini, <a href="https://docs.pytorch.org/ao/main/api_reference/generated/torchao.quantization.Int4WeightOnlyConfig.html" rel="nofollow">Int4WeightOnlyConfig</a> gibi bir kuantizasyon veri türü olarak, <a href="/docs/diffusers/main/tr/api/models/overview#diffusers.ModelMixin.from_pretrained">from_pretrained()</a> içindeki <a href="/docs/diffusers/main/tr/api/quantization#diffusers.TorchAoConfig">TorchAoConfig</a>'ye iletin.</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> Kopyalandı</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, PipelineQuantizationConfig, TorchAoConfig | |
| <span class="hljs-keyword">from</span> torchao.quantization <span class="hljs-keyword">import</span> Int8WeightOnlyConfig | |
| pipeline_quant_config = PipelineQuantizationConfig( | |
| quant_mapping={<span class="hljs-string">"transformer"</span>: TorchAoConfig(Int8WeightOnlyConfig(group_size=<span class="hljs-number">128</span>, version=<span class="hljs-number">2</span>))} | |
| ) | |
| pipeline = DiffusionPipeline.from_pretrained( | |
| <span class="hljs-string">"black-forest-labs/FLUX.1-dev"</span>, | |
| quantization_config=pipeline_quant_config, | |
| dtype=torch.bfloat16, | |
| device_map=<span class="hljs-string">"cuda"</span> <span class="hljs-comment"># or "mps", "xpu", "cpu"</span> | |
| )<!----></pre></div><!----> <p><code>device_map="cuda"</code>, her katmanı yüklenirken GPU üzerinde kuantize eder. Bu hızlıdır, ancak orijinal ve kuantize ağırlıklar için geçici olarak ek GPU belleği gerektirir. Model zaten GPU belleğinizin çoğunu kullanıyorsa, yükleme bellek yetersizliği (out-of-memory) hatasıyla başarısız olabilir. Bu durumda, <code>device_map</code> parametresini kaldırın ve pipeline'ı yükleme işleminden sonra GPU'ya taşıyın:</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> Kopyalandı</div><!----></button><!----></div> <pre class="language-py "><!---->pipeline = DiffusionPipeline.from_pretrained( | |
| <span class="hljs-string">"black-forest-labs/FLUX.1-dev"</span>, | |
| quantization_config=pipeline_quant_config, | |
| torch_dtype=torch.bfloat16, | |
| ) | |
| pipeline.to(<span class="hljs-string">"cuda"</span>) <span class="hljs-comment"># or "mps", "xpu", "cpu"</span><!----></pre></div><!----> <p><code>device_map</code> olmadan Diffusers, katmanları CPU üzerinde kuantize eder. Bu daha yavaştır ancak kuantizasyon sırasında geçici GPU bellek artışını önler. GPU bellek kullanımını daha da azaltmak için bunun yerine <a href="/docs/diffusers/main/tr/api/pipelines/overview#diffusers.DiffusionPipeline.enable_model_cpu_offload">enable_model_cpu_offload()</a> kullanın. Metin kodlayıcı gibi ek bileşenleri de kuantize edebilirsiniz.</p> <!--[1--><h2 class="relative group"><a id="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="#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><!---->torch.compile<!----></span></h2><!--]--><!----> <p>torchao, tek satır kodla çıkarımı hızlandırabilen <a href="../optimization/fp16#torchcompile">torch.compile</a> özelliğini destekler.</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> Kopyalandı</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> DiffusionPipeline, PipelineQuantizationConfig, TorchAoConfig | |
| <span class="hljs-keyword">from</span> torchao.quantization <span class="hljs-keyword">import</span> Int4WeightOnlyConfig | |
| pipeline_quant_config = PipelineQuantizationConfig( | |
| quant_mapping={<span class="hljs-string">"transformer"</span>: TorchAoConfig(Int4WeightOnlyConfig(group_size=<span class="hljs-number">128</span>))} | |
| ) | |
| pipeline = DiffusionPipeline.from_pretrained( | |
| <span class="hljs-string">"black-forest-labs/FLUX.1-dev"</span>, | |
| quantization_config=pipeline_quant_config, | |
| dtype=torch.bfloat16, | |
| device_map=<span class="hljs-string">"cuda"</span> <span class="hljs-comment"># or "mps", "xpu", "cpu"</span> | |
| ) | |
| pipeline.transformer.<span class="hljs-built_in">compile</span>(transformer, mode=<span class="hljs-string">"max-autotune"</span>, fullgraph=<span class="hljs-literal">True</span>)<!----></pre></div><!----> <p>Flux ve CogVideoX için çıkarım hızı ve bellek kullanımı ölçümlerine ilişkin bu <a href="https://github.com/huggingface/diffusers/pull/10009#issue-2688781450" rel="nofollow">tabloya</a> bakın. Çeşitli donanımlara ait daha fazla ölçüm, torchao <a href="https://github.com/pytorch/ao/tree/main/torchao/quantization#benchmarks" rel="nofollow">deposunda</a> da mevcuttur.</p> <blockquote class="tip"><p>torchao içindeki FP8 eğitim sonrası nicelleştirme şemaları, hesaplama yeteneği en az 8.9 olan GPU'lar (RTX-4090, Hopper vb.) için etkilidir. Görsel ve video üretirken FP8 genellikle hız, bellek ve kalite arasında en iyi dengeyi sağlar. GPU'nuz uyumluysa FP8'i torch.compile ile birleştirmenizi öneririz.</p></blockquote> <!--[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><!---->Desteklenen kuantizasyon türleri<!----></span></h2><!--]--><!----> <p>torchao, int8, float3-float8 ve uint1-uint7 için yalnızca ağırlık kuantizasyonu ile ağırlık ve dinamik aktivasyon kuantizasyonunu destekler.</p> <p>Sadece ağırlık kuantizasyonu, model ağırlıklarını belirli bir düşük bit veri türünde saklar ancak <code>bfloat16</code> gibi daha yüksek hassasiyetli bir veri türüyle hesaplama yapar. Bu, model ağırlıklarından kaynaklanan bellek gereksinimini azaltır ancak aktivasyon hesaplama için bellek zirvelerini korur.</p> <p>Dinamik aktivasyon kuantizasyonu, model ağırlıklarını düşük bitlik bir veri türünde (dtype) saklarken, ek bellek tasarrufu sağlamak için aktivasyonları da anlık olarak kuantize eder. Bu, model ağırlıklarından kaynaklanan bellek gereksinimini azaltırken, aktivasyon hesaplamalarından kaynaklanan bellek yükünü de düşürür. Ancak bu, bazen kalite açısından bir ödün gerektirebilir; bu nedenle farklı modelleri kapsamlı bir şekilde test etmeniz önerilir.</p> <p>Mevcut kuantizasyon yöntemlerini daha iyi anlamak için <a href="https://docs.pytorch.org/ao/stable/index.html" rel="nofollow">resmi torchao dokümantasyonuna</a> başvurun. Yapılandırma seçeneklerinin kapsamlı bir listesi <a href="https://docs.pytorch.org/ao/main/workflows/inference.html#inference-workflows" rel="nofollow">burada</a> mevcuttur.</p> <p>Bazı popüler örnek kuantizasyon yapılandırmaları aşağıdaki gibidir:</p> <table><thead><tr><th><strong>Kategori</strong></th><th><strong>Yapılandırma Sınıfları</strong></th></tr></thead><tbody><tr><td><strong>Tamsayıl kuantizasyonu</strong></td><td><a href="https://docs.pytorch.org/ao/stable/api_reference/generated/torchao.quantization.Int4WeightOnlyConfig.html" rel="nofollow"><code>Int4WeightOnlyConfig</code></a>, <a href="https://docs.pytorch.org/ao/stable/api_reference/generated/torchao.quantization.Int8WeightOnlyConfig.html" rel="nofollow"><code>Int8WeightOnlyConfig</code></a>, <a href="https://docs.pytorch.org/ao/stable/api_reference/generated/torchao.quantization.Int8DynamicActivationInt8WeightConfig.html" rel="nofollow"><code>Int8DynamicActivationInt8WeightConfig</code></a></td></tr><tr><td><strong>Yüzen noktalı 8-bit kuantizasyon</strong></td><td><a href="https://docs.pytorch.org/ao/stable/api_reference/generated/torchao.quantization.Float8WeightOnlyConfig.html" rel="nofollow"><code>Float8WeightOnlyConfig</code></a>, <a href="https://docs.pytorch.org/ao/stable/api_reference/generated/torchao.quantization.Float8DynamicActivationFloat8WeightConfig.html" rel="nofollow"><code>Float8DynamicActivationFloat8WeightConfig</code></a></td></tr><tr><td><strong>İşaretsiz tamsayı kuantizasyonu</strong></td><td><a href="https://docs.pytorch.org/ao/stable/api_reference/generated/torchao.quantization.IntxWeightOnlyConfig.html" rel="nofollow"><code>IntxWeightOnlyConfig</code></a></td></tr></tbody></table> <!--[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><!---->Kantize modellerin seri hale getirilmesi ve dizide çıkarılması<!----></span></h2><!--]--><!----> <p>Belirtilen bir dtype ile kuantize edilmiş bir modeli serialize etmek için, önce modeli istenen kuantizasyon dtype'i ile yükleyin ve ardından <a href="/docs/diffusers/main/tr/api/models/overview#diffusers.ModelMixin.save_pretrained">save_pretrained()</a> yöntemini kullanarak kaydedin.</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> Kopyalandı</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, TorchAoConfig | |
| <span class="hljs-keyword">from</span> torchao.quantization <span class="hljs-keyword">import</span> Int8WeightOnlyConfig | |
| quantization_config = TorchAoConfig(Int8WeightOnlyConfig()) | |
| transformer = AutoModel.from_pretrained( | |
| <span class="hljs-string">"black-forest-labs/Flux.1-Dev"</span>, | |
| subfolder=<span class="hljs-string">"transformer"</span>, | |
| quantization_config=quantization_config, | |
| dtype=torch.bfloat16, | |
| ) | |
| transformer.save_pretrained(<span class="hljs-string">"/path/to/flux_int8wo"</span>, safe_serialization=<span class="hljs-literal">False</span>)<!----></pre></div><!----> <p>Serileştirilmiş bir kuantize modeli yüklemek için <a href="/docs/diffusers/main/tr/api/models/overview#diffusers.ModelMixin.from_pretrained">from_pretrained()</a> yöntemini kullanın.</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> Kopyalandı</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, AutoModel | |
| transformer = AutoModel.from_pretrained(<span class="hljs-string">"/path/to/flux_int8wo"</span>, dtype=torch.bfloat16, use_safetensors=<span class="hljs-literal">False</span>) | |
| pipe = FluxPipeline.from_pretrained(<span class="hljs-string">"black-forest-labs/Flux.1-Dev"</span>, transformer=transformer, dtype=torch.bfloat16) | |
| pipe.to(<span class="hljs-string">"cuda"</span>) <span class="hljs-comment"># or "mps", "xpu", "cpu"</span> | |
| prompt = <span class="hljs-string">"A cat holding a sign that says hello world"</span> | |
| image = pipe(prompt, num_inference_steps=<span class="hljs-number">30</span>, guidance_scale=<span class="hljs-number">7.0</span>).images[<span class="hljs-number">0</span>] | |
| image.save(<span class="hljs-string">"output.png"</span>)<!----></pre></div><!----> <p><code>torch<=2.6.0</code> kullanıyorsanız, <code>uint4</code> weight-only gibi bazı kuantizasyon yöntemleri doğrudan yüklenemez ve modelleri yüklemeye çalışırken <code>UnpicklingError</code> ile sonuçlanabilir, ancak kaydetme sırasında beklenen şekilde çalışır. Bu durumu aşmak için state dict'i modele manuel olarak yükleyebilirsiniz. Bununla birlikte, bunun <code>torch.load</code> içinde <code>weights_only=False</code> kullanılmasını gerektirdiğini unutmayın; bu nedenle, ağırlıkların güvenilir bir kaynaktan elde edildiğinden emin olmanız durumunda çalıştırılmalıdır.</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> Kopyalandı</div><!----></button><!----></div> <pre class="language-python "><!----><span class="hljs-keyword">import</span> torch | |
| <span class="hljs-keyword">from</span> accelerate <span class="hljs-keyword">import</span> init_empty_weights | |
| <span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> FluxPipeline, AutoModel, TorchAoConfig | |
| <span class="hljs-keyword">from</span> torchao.quantization <span class="hljs-keyword">import</span> IntxWeightOnlyConfig | |
| <span class="hljs-comment"># Serialize the model</span> | |
| transformer = AutoModel.from_pretrained( | |
| <span class="hljs-string">"black-forest-labs/Flux.1-Dev"</span>, | |
| subfolder=<span class="hljs-string">"transformer"</span>, | |
| quantization_config=TorchAoConfig(IntxWeightOnlyConfig(dtype=torch.uint4)), | |
| dtype=torch.bfloat16, | |
| ) | |
| transformer.save_pretrained(<span class="hljs-string">"/path/to/flux_uint4wo"</span>, safe_serialization=<span class="hljs-literal">False</span>, max_shard_size=<span class="hljs-string">"50GB"</span>) | |
| <span class="hljs-comment"># ...</span> | |
| <span class="hljs-comment"># Load the model</span> | |
| state_dict = torch.load(<span class="hljs-string">"/path/to/flux_uint4wo/diffusion_pytorch_model.bin"</span>, weights_only=<span class="hljs-literal">False</span>, map_location=<span class="hljs-string">"cpu"</span>) | |
| <span class="hljs-keyword">with</span> init_empty_weights(): | |
| transformer = AutoModel.from_config(<span class="hljs-string">"/path/to/flux_uint4wo/config.json"</span>) | |
| transformer.load_state_dict(state_dict, strict=<span class="hljs-literal">True</span>, assign=<span class="hljs-literal">True</span>)<!----></pre></div><!----> <blockquote class="tip"><p>Aşağıdaki örneklerde gösterildiği gibi <a href="/docs/diffusers/main/tr/api/models/auto_model#diffusers.AutoModel">AutoModel</a> API'si PyTorch >= 2.6 için desteklenmektedir.</p></blockquote> <!--[1--><h2 class="relative group"><a id="resources" 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="#resources"><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><!---->Kaynaklar<!----></span></h2><!--]--><!----> <ul><li><a href="https://docs.pytorch.org/ao/stable/index.html" rel="nofollow">TorchAO Niceliklendirme API'si</a></li> <li><a href="https://github.com/sayakpaul/diffusers-torchao" rel="nofollow">Diffusers-TorchAO örnekleri</a></li></ul> <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/torchao.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_1u9qigp = { | |
| 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.CM-2NerT.js"), | |
| import("/docs/diffusers/main/en/_app/immutable/entry/app.DrYcqW7m.js") | |
| ]).then(([kit, app]) => { | |
| kit.start(app, element, { | |
| node_ids: [0, 12], | |
| data: [null,null], | |
| form: null, | |
| error: null | |
| }); | |
| }); | |
| } | |
| </script> | |
Xet Storage Details
- Size:
- 34 kB
- Xet hash:
- 34a5bee9010556c024dd07d083df00f454ee5bccbb81a23c00911c200b7213d4
·
Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.