Buckets:

hf-doc-build
/

doc

Files
xet
hf-doc-build/doc / diffusers /main /ko /quantization /modelopt.html
mishig
about 15 hours ago
download
raw
26.4 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;지원되는 양자화 유형&quot;,&quot;local&quot;:&quot;supported-quantization-types&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;양자화 모델 직렬화 및 역직렬화&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.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/346.DhQ_G8bn.js" rel="modulepreload">
		<link href="/docs/diffusers/main/en/_app/immutable/chunks/B4AMoTAu.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>는 양자화, 프루닝, 증류, 추측 디코딩 등 최첨단 모델 최적화 기법을 통합한 라이브러리입니다. TensorRT-LLM 또는 TensorRT와 같은 다운스트림 배포 프레임워크에서 딥러닝 모델을 압축하여 추론 속도를 최적화합니다.</p> <p>시작하기 전에 nvidia_modelopt가 설치되어 있는지 확인하세요.</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-bash "><!---->pip install -U <span class="hljs-string">&quot;nvidia_modelopt[hf]&quot;</span><!----></pre></div><!----> <p><code>NVIDIAModelOptConfig</code>를 <a href="/docs/diffusers/main/ko/api/models/overview#diffusers.ModelMixin.from_pretrained">from_pretrained()</a>에 전달하여 모델을 양자화합니다(사전 양자화된 모델을 로드할 수도 있습니다). 이는 <a href="https://hf.co/docs/accelerate/index" rel="nofollow">Accelerate</a>로 로딩을 지원하고 <code>torch.nn.Linear</code> 레이어를 포함하는 모든 모달리티의 모든 모델에 적용됩니다.</p> <p>…</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-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>참고:</strong></p> <p>NVIDIA-ModelOpt의 양자화 방법은 모델 성능을 유지하면서 다양한 QAT(Quantization-Aware Training) 및 PTQ(Post-Training Quantization) 기법을 활용하여 모델 가중치의 메모리 사용량을 줄이는 데 설계되었습니다. 그러나 추론 시 실제 성능 향상은 배포 프레임워크(예: TRT-LLM, TensorRT)와 특정 하드웨어 구성에 따라 달라집니다.</p> <p>자세한 내용은 <a href="https://github.com/NVIDIA/Model-Optimizer/tree/main/examples" rel="nofollow">여기</a>에서 확인할 수 있습니다.</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><code>NVIDIAModelOptConfig</code> 클래스는 세 가지 파라미터를 받습니다:</p> <ul><li><code>quant_type</code>: 아래 양자화 유형 중 하나를 나타내는 문자열 값.</li> <li><code>modules_to_not_convert</code>: 양자화를 수행하지 않을 모듈의 전체/부분 모듈 이름 목록입니다. 예를 들어, <a href="/docs/diffusers/main/ko/api/models/sd3_transformer2d#diffusers.SD3Transformer2DModel">SD3Transformer2DModel</a>의 pos_embed 투영 블록에 양자화를 수행하지 않으려면 다음과 같이 지정합니다: <code>modules_to_not_convert=["pos_embed.proj.weight"]</code>.</li> <li><code>disable_conv_quantization</code>: 모델의 모든 컨볼루션 레이어에 대한 양자화를 비활성화하려면 <code>True</code>로 설정하는 불리언 값입니다. 채널 및 블록 양자화는 일반적으로 컨볼루션 레이어(INT4, NF4, NVFP4와 함께 사용)와 잘 동작하지 않으므로 유용합니다. 특정 컨볼루션 레이어에 대해서만 양자화를 비활성화하려면 대신 <code>modules_to_not_convert</code>를 사용하세요.</li> <li><code>algorithm</code>: 스케일을 결정하는 데 사용할 알고리즘입니다. 기본값은 <code>"max"</code>입니다. 추가 알고리즘 및 세부 사항은 modelopt 문서를 확인하세요.</li> <li><code>forward_loop</code>: 양자화 과정에서 활성화 값을 보정하는 데 사용할 forward loop 함수입니다. 제공되지 않으면 가중치만 사용하여 계산된 정적 스케일 값에 의존합니다.</li> <li><code>kwargs</code>: <code>quant_type</code>에 따라 호출되는 기본 양자화 메서드에 전달할 키워드 인자의 딕셔너리.</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><!---->지원되는 양자화 유형<!----></span></h2><!--]--><!----> <p>ModelOpt는 weight-only, channel 및 block 양자화를 지원하며, int8, fp8, int4, nf4, nvfp4 포맷을 사용할 수 있습니다. 이러한 양자화 방식은 추론 시 모델의 성능을 유지하면서 모델 가중치의 메모리 사용량을 줄이는 것을 목표로 설계되었습니다.</p> <p>Weight-only quantization은 모델 가중치를 특정 저비트 데이터 형식으로 저장하지만, <code>bfloat16</code>와 같은 더 높은 정밀도의 데이터 형식으로 계산을 수행합니다. 이를 통해 모델 가중치에 따른 메모리 요구량을 줄이되, 활성화 계산에 따른 메모리 피크는 유지됩니다.</p> <p>지원되는 양자화 방법은 다음과 같습니다:</p> <table><thead><tr><th><strong>양자화 유형</strong></th><th><strong>지원되는 스키마</strong></th><th><strong>필수 Kwargs</strong></th><th><strong>추가 참고 사항</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>사용 가능한 양자화 방법과 사용 가능한 구성 옵션의 전체 목록에 대한 더 자세한 이해를 위해 <a href="https://nvidia.github.io/Model-Optimizer/" rel="nofollow">공식 modelopt 문서</a>를 참조하세요.</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><!---->양자화 모델 직렬화 및 역직렬화<!----></span></h2><!--]--><!----> <p>주어진 dtype로 양자화된 모델을 직렬화하려면, 먼저 원하는 양자화 dtype로 모델을 로드한 다음 <a href="/docs/diffusers/main/ko/api/models/overview#diffusers.ModelMixin.save_pretrained">save_pretrained()</a> 메서드를 사용하여 저장합니다.</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-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>시리얼라이즈된 양자화 모델을 로드하려면 <a href="/docs/diffusers/main/ko/api/models/overview#diffusers.ModelMixin.from_pretrained">from_pretrained()</a> 메서드를 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-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_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, 346],
							data: [null,null],
							form: null,
							error: null
						});
					});
				}
			</script>
		
Xet Storage Details

Size:: 26.4 kB
Xet hash:: 0fe35bed855ff7dd04cfc16725a7475babcb5d6e8cafd893fecf48162355f34e
Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.