Hugging Face's logo

Buckets:
hf-doc-build
/
doc

Files
xet
hf-doc-build/doc / diffusers /main /ko /quantization /overview.html
mishig's picture
mishig
download
raw
35.2 kB
<meta charset="utf-8" /><meta name="hf:doc:metadata" content="{&quot;title&quot;:&quot;시작하기&quot;,&quot;local&quot;:&quot;getting-started&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;Pipeline-level 양자화&quot;,&quot;local&quot;:&quot;pipeline-level-quantization&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;기본 양자화&quot;,&quot;local&quot;:&quot;basic-quantization&quot;,&quot;sections&quot;:[],&quot;depth&quot;:3},{&quot;title&quot;:&quot;고급 양자화&quot;,&quot;local&quot;:&quot;advanced-quantization&quot;,&quot;sections&quot;:[],&quot;depth&quot;:3}],&quot;depth&quot;:2},{&quot;title&quot;:&quot;사전 양자화 체크포인트 로드&quot;,&quot;local&quot;:&quot;loading-prequantized-checkpoints&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;리소스&quot;,&quot;local&quot;:&quot;resources&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2}],&quot;depth&quot;:1}"/>
<link href="/docs/diffusers/main/en/_app/immutable/entry/start.vZc-dOUR.js" rel="modulepreload">
<link href="/docs/diffusers/main/en/_app/immutable/chunks/Bafuik5c.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.B6c3Ml05.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.BQlx092V.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/348.Ba6GVl0h.js" rel="modulepreload">
<link href="/docs/diffusers/main/en/_app/immutable/chunks/B4AMoTAu.js" rel="modulepreload">
<!--cubdu--><meta name="hf:doc:metadata" content="{&quot;title&quot;:&quot;Getting started&quot;,&quot;local&quot;:&quot;getting-started&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;Pipeline-level quantization&quot;,&quot;local&quot;:&quot;pipeline-level-quantization&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;Basic quantization&quot;,&quot;local&quot;:&quot;basic-quantization&quot;,&quot;sections&quot;:[],&quot;depth&quot;:3},{&quot;title&quot;:&quot;Advanced quantization&quot;,&quot;local&quot;:&quot;advanced-quantization&quot;,&quot;sections&quot;:[],&quot;depth&quot;:3}],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Loading prequantized checkpoints&quot;,&quot;local&quot;:&quot;loading-prequantized-checkpoints&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Resources&quot;,&quot;local&quot;:&quot;resources&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="getting-started" 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="#getting-started"><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><!---->시작하기<!----></span></h1><!--]--><!----> <p>Quantization은 원래 데이터의 정밀도를 유지하면서 더 적은 비트로 데이터를 표현하는 데 초점을 맞춥니다. 이는 일반적으로 동일한 정보를 더 적은 비트로 표현하기 위해 데이터 유형을 변환하는 것을 의미합니다. 예를 들어, 모델 가중치가 32비트 부동소수점으로 저장되어 있고 이를 16비트 부동소수점으로 양자화하면 모델 크기가 절반으로 줄어 저장하기가 쉬워지고 메모리 사용량이 감소합니다. 더 낮은 정밀도는 더 적은 비트로 계산을 수행하는 데 시간이 덜 걸리기 때문에 추론 속도도 향상시킬 수 있습니다.</p> <p>Diffusers는 <a href="../api/pipelines/flux">Flux</a>와 같은 대규모 diffusion 모델을 보다 쉽게 사용할 수 있도록 여러 quantization backends를 지원합니다. 이 가이드에서는 사전 학습된 체크포인트나 비양자화 체크포인트에서 파이프라인을 초기화하는 동안 <a href="/docs/diffusers/main/ko/api/quantization#diffusers.PipelineQuantizationConfig">PipelineQuantizationConfig</a> 클래스를 사용하여 파이프라인을 양자화하는 방법을 보여줍니다.</p> <!--[1--><h2 class="relative group"><a id="pipeline-level-quantization" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#pipeline-level-quantization"><span><svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg><!----></span></a> <span><!---->Pipeline-level 양자화<!----></span></h2><!--]--><!----> <p>양자화 설정에 적용할 커스터마이징 수준에 따라 <a href="/docs/diffusers/main/ko/api/quantization#diffusers.PipelineQuantizationConfig">PipelineQuantizationConfig</a>를 사용하는 방법은 두 가지가 있습니다.</p> <ul><li>기본 사용 사례의 경우 <code>quant_backend</code>, <code>quant_kwargs</code>, <code>components_to_quantize</code> 인수를 정의합니다</li> <li>세밀한 양자화 제어를 위해 개별 모델 구성 요소에 대한 양자화 구성을 제공하는 <code>quant_mapping</code>을 정의합니다</li></ul> <!--[2--><h3 class="relative group"><a id="basic-quantization" 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="#basic-quantization"><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><!---->기본 양자화<!----></span></h3><!--]--><!----> <p>다음 매개변수로 <a href="/docs/diffusers/main/ko/api/quantization#diffusers.PipelineQuantizationConfig">PipelineQuantizationConfig</a>를 초기화합니다.</p> <ul><li><code>quant_backend</code>는 사용할 양자화 백엔드를 지정합니다. 현재 지원되는 백엔드에는 <code>bitsandbytes_4bit</code>, <code>bitsandbytes_8bit</code>, <code>gguf</code>, <code>quanto</code>, <code>torchao</code>가 있습니다.</li> <li><code>quant_kwargs</code>는 사용할 양자화 인수를 지정합니다.</li></ul> <blockquote class="tip"><p>이 <code>quant_kwargs</code> 인수는 백엔드마다 다릅니다. 각 백엔드의 인수를 보려면 <a href="../api/quantization">Quantization API</a> 문서를 참조하세요.</p></blockquote> <ul><li><p><code>components_to_quantize</code>는 파이프라인의 어떤 구성 요소를 양자화할지 지정합니다. 일반적으로 Transformer와 같이 계산 집약적인 구성 요소를 양자화해야 합니다. <a href="/docs/diffusers/main/ko/api/pipelines/flux#diffusers.FluxPipeline">FluxPipeline</a>처럼 텍스트 인코더가 여러 개 있는 파이프라인의 경우, 텍스트 인코더도 양자화 대상으로 고려할 수 있습니다. 아래 예시는 <a href="/docs/diffusers/main/ko/api/pipelines/flux#diffusers.FluxPipeline">FluxPipeline</a>에서 T5 텍스트 인코더만 양자화하고 CLIP 모델은 그대로 유지합니다.</p> <p><code>components_to_quantize</code>는 여러 모델을 위한 목록 또는 단일 모델을 위한 문자열을 받아들입니다.</p></li></ul> <p>아래 예제는 <a href="/docs/diffusers/main/ko/api/quantization#diffusers.BitsAndBytesConfig">BitsAndBytesConfig</a>에서 <code>load_in_4bit</code>, <code>bnb_4bit_quant_type</code>, <code>bnb_4bit_compute_dtype</code> 인수를 사용하여 bitsandbytes 백엔드를 로드합니다.</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> 복사됨</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
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>],
)<!----></pre></div><!----> <p><code>pipeline_quant_config</code>를 <a href="/docs/diffusers/main/ko/api/pipelines/overview#diffusers.DiffusionPipeline.from_pretrained">from_pretrained()</a>에 전달하여 파이프라인을 양자화합니다.</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> 복사됨</div><!----></button><!----></div> <pre class="language-py "><!---->pipe = 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>
image = pipe(<span class="hljs-string">&quot;photo of a cute dog&quot;</span>).images[<span class="hljs-number">0</span>]<!----></pre></div><!----> <!--[2--><h3 class="relative group"><a id="advanced-quantization" 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="#advanced-quantization"><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><!---->고급 양자화<!----></span></h3><!--]--><!----> <p><code>quant_mapping</code> 인수는 파이프라인의 각 개별 구성 요소를 양자화하는 방법에 대해 더 많은 옵션을 제공합니다. 예를 들어, 서로 다른 양자화 백엔드를 결합하는 것 등이 있습니다.</p> <p><a href="/docs/diffusers/main/ko/api/quantization#diffusers.PipelineQuantizationConfig">PipelineQuantizationConfig</a>을 초기화하고 <code>quant_mapping</code>을 전달합니다. <code>quant_mapping</code>을 사용하면 transformer, text encoder와 같은 파이프라인의 각 구성 요소에 대한 양자화 옵션을 지정할 수 있습니다.</p> <p>아래 예시에서는 transformer와 text encoder에 각각 두 가지 양자화 백엔드인 <a href="/docs/diffusers/main/ko/api/quantization#diffusers.QuantoConfig">QuantoConfig</a>와 <a href="https://huggingface.co/docs/transformers/main/en/main_classes/quantization#transformers.BitsAndBytesConfig" rel="nofollow">transformers.BitsAndBytesConfig</a>를 사용합니다.</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> 복사됨</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 <span class="hljs-keyword">import</span> BitsAndBytesConfig <span class="hljs-keyword">as</span> DiffusersBitsAndBytesConfig
<span class="hljs-keyword">from</span> diffusers.quantizers.quantization_config <span class="hljs-keyword">import</span> QuantoConfig
<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> BitsAndBytesConfig <span class="hljs-keyword">as</span> TransformersBitsAndBytesConfig
pipeline_quant_config = PipelineQuantizationConfig(
quant_mapping={
<span class="hljs-string">&quot;transformer&quot;</span>: QuantoConfig(weights_dtype=<span class="hljs-string">&quot;int8&quot;</span>),
<span class="hljs-string">&quot;text_encoder_2&quot;</span>: TransformersBitsAndBytesConfig(
load_in_4bit=<span class="hljs-literal">True</span>, compute_dtype=torch.bfloat16
),
}
)<!----></pre></div><!----> <p><a href="https://huggingface.co/docs/transformers/main_classes/quantization#transformers.BitsAndBytesConfig" rel="nofollow">Transformers</a>에는 별도의 bitsandbytes 백엔드가 있습니다. Transformers에서 가져온 구성 요소의 경우 <a href="https://huggingface.co/docs/transformers/main/en/main_classes/quantization#transformers.BitsAndBytesConfig" rel="nofollow">transformers.BitsAndBytesConfig</a>를 가져와서 사용해야 합니다. 예를 들어, <a href="/docs/diffusers/main/ko/api/pipelines/flux#diffusers.FluxPipeline">FluxPipeline</a>의 <code>text_encoder_2</code>는 Transformers의 <a href="https://huggingface.co/docs/transformers/main/en/model_doc/t5#transformers.T5EncoderModel" rel="nofollow">T5EncoderModel</a>이므로 <a href="/docs/diffusers/main/ko/api/quantization#diffusers.BitsAndBytesConfig">diffusers.BitsAndBytesConfig</a> 대신 <a href="https://huggingface.co/docs/transformers/main/en/main_classes/quantization#transformers.BitsAndBytesConfig" rel="nofollow">transformers.BitsAndBytesConfig</a>를 사용해야 합니다.</p> <blockquote class="tip"><p>이러한 개별 import를 관리하지 않거나 각 pipeline 구성 요소의 출처가 불분명한 경우, 위의 <a href="#basic-quantization">기본 양자화</a> 방법을 사용하세요.</p></blockquote> <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> 복사됨</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 <span class="hljs-keyword">import</span> BitsAndBytesConfig <span class="hljs-keyword">as</span> DiffusersBitsAndBytesConfig
<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> BitsAndBytesConfig <span class="hljs-keyword">as</span> TransformersBitsAndBytesConfig
pipeline_quant_config = PipelineQuantizationConfig(
quant_mapping={
<span class="hljs-string">&quot;transformer&quot;</span>: DiffusersBitsAndBytesConfig(load_in_4bit=<span class="hljs-literal">True</span>, bnb_4bit_compute_dtype=torch.bfloat16),
<span class="hljs-string">&quot;text_encoder_2&quot;</span>: TransformersBitsAndBytesConfig(
load_in_4bit=<span class="hljs-literal">True</span>, compute_dtype=torch.bfloat16
),
}
)<!----></pre></div><!----> <p><code>pipeline_quant_config</code>를 <a href="/docs/diffusers/main/ko/api/pipelines/overview#diffusers.DiffusionPipeline.from_pretrained">from_pretrained()</a>에 전달하여 파이프라인을 양자화합니다.</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> 복사됨</div><!----></button><!----></div> <pre class="language-py "><!---->pipe = 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>
image = pipe(<span class="hljs-string">&quot;photo of a cute dog&quot;</span>).images[<span class="hljs-number">0</span>]<!----></pre></div><!----> <!--[1--><h2 class="relative group"><a id="loading-prequantized-checkpoints" 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="#loading-prequantized-checkpoints"><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><!---->사전 양자화 체크포인트 로드<!----></span></h2><!--]--><!----> <p>일부 양자화 백엔드는 이미 양자화되어 Hub에 업로드된 체크포인트를 로드하는 것을 지원합니다. 이 경우 양자화 구성은 모델의 <code>config.json</code>에 저장되며 자동으로 읽힙니다. 따라서 <a href="/docs/diffusers/main/ko/api/quantization#diffusers.PipelineQuantizationConfig">PipelineQuantizationConfig</a>를 생성하거나 <code>quantization_config</code> 인수를 전달할 필요가 없습니다.</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> 복사됨</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
pipe = DiffusionPipeline.from_pretrained(
<span class="hljs-string">&quot;rootonchair/ERNIE-Image-Turbo-nunchaku-lite-nvfp4&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><!----></pre></div><!----> <p>다음 백엔드들은 사전 양자화 체크포인트를 기본적으로 로드할 수 있습니다.</p> <table><thead><tr><th>백엔드</th><th>The input is a single word "Notes".
참고</th></tr></thead><tbody><tr><td><a href="./bitsandbytes">bitsandbytes</a></td><td>Config는 <code>config.json</code>에 저장되며, 추가 인자가 필요하지 않습니다. 온플라이 양자화 및 사전 양자화된 체크포인트 로딩을 모두 지원합니다.</td></tr><tr><td><a href="./torchao">torchao</a></td><td>위와 동일합니다.</td></tr><tr><td><a href="./gguf">GGUF</a></td><td>Model 클래스와 함께 <code>from_single_file</code>를 사용합니다. pipeline 수준의 로딩은 지원되지 않습니다.</td></tr><tr><td><a href="./bfl">BFL</a></td><td>Black Forest Labs의 <code>fp8r</code> 및 <code>nvfp4</code> 단일 파일 체크포인트를 <code>from_single_file</code>로 로드합니다. 파이프라인 수준의 로딩은 지원되지 않습니다.</td></tr><tr><td><a href="./autoround">AutoRound</a></td><td>로드만 지원됩니다. 먼저 AutoRound CLI 또는 Python API를 사용하여 양자화하세요.</td></tr><tr><td><a href="./nunchaku">Nunchaku Lite</a></td><td>Config는 <code>config.json</code>에 저장되며 <code>kernels</code> 패키지가 필요합니다. 로딩만 지원됩니다.</td></tr><tr><td><a href="./modelopt">ModelOpt</a></td><td>모델만 로드하는 것이 지원됩니다.</td></tr><tr><td><a href="./sdnq">SDNQ</a></td><td><code>sdnq</code> 패키지가 필요합니다. 실시간 양자화 및 사전 양자화 모델 로딩을 모두 지원합니다.</td></tr></tbody></table> <!--[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><!---->리소스<!----></span></h2><!--]--><!----> <p>양자화에 대해 자세히 알아보려면 아래 리소스를 확인하세요.</p> <ul><li><p>양자화에 처음 접하시는 경우, DeepLearning.AI와 협력하여 만든 초보자 친화적인 다음 코스를 확인해 보시는 것을 권장합니다.</p> <ul><li><a href="https://www.deeplearning.ai/short-courses/quantization-fundamentals-with-hugging-face/" rel="nofollow">Hugging Face를 활용한 양자화 기초</a></li> <li><a href="https://www.deeplearning.ai/short-courses/quantization-in-depth/" rel="nofollow">양자화 심층 분석</a></li></ul></li> <li><p>새로운 양자화 방법을 추가하는 데 관심이 있다면 <a href="https://huggingface.co/docs/transformers/main/en/quantization/contribute" rel="nofollow">새로운 양자화 방법 기여 가이드</a>를 참조하세요.</p></li> <li><p>Transformers 양자화 <a href="https://huggingface.co/docs/transformers/quantization/overview#when-to-use-what" rel="nofollow">개요</a>는 다양한 양자화 백엔드의 장단점을 요약합니다.</p></li> <li><p>각 양자화 백엔드에 대한 간략한 소개, 백엔드 선택 방법, 그리고 양자화를 다른 메모리 최적화 기법과 결합하는 방법에 대해서는 <a href="https://huggingface.co/blog/diffusers-quantization" rel="nofollow">Exploring Quantization Backends in Diffusers</a> 블로그 포스트를 참고하세요.</p></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/overview.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_1rjuniz = {
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.vZc-dOUR.js"),
import("/docs/diffusers/main/en/_app/immutable/entry/app.B6c3Ml05.js")
]).then(([kit, app]) => {
kit.start(app, element, {
node_ids: [0, 348],
data: [null,null],
form: null,
error: null
});
});
}
</script>

Xet Storage Details

Size:
35.2 kB
·
Xet hash:
2aafe9fb3687fe38c8e0812b8f189c77f44cd6ade97af3d9aa9bcee5f39d6712

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