Buckets:

hf-doc-build
/

doc

Files

xet

hf-doc-build/doc / diffusers /main /en /quantization /nunchaku.html

HuggingFaceDocBuilder

about 17 hours ago

download

raw

28.4 kB

	<meta charset="utf-8" /><meta name="hf:doc:metadata" content="{"title":"Nunchaku Lite","local":"nunchaku-lite","sections":[{"title":"Install the CUDA kernels","local":"install-the-cuda-kernels","sections":[],"depth":2},{"title":"Load a quantized pipeline","local":"load-a-quantized-pipeline","sections":[],"depth":2},{"title":"Supported quantization types","local":"supported-quantization-types","sections":[],"depth":2},{"title":"NunchakuLiteQuantizationConfig","local":"nunchakulitequantizationconfig","sections":[],"depth":2},{"title":"torch.compile","local":"torchcompile","sections":[],"depth":2},{"title":"Resources","local":"resources","sections":[],"depth":2}],"depth":1}"/>
	<link href="/docs/diffusers/main/en/_app/immutable/entry/start.cTElinSc.js" rel="modulepreload">
	<link href="/docs/diffusers/main/en/_app/immutable/chunks/BnlWIvxx.js" rel="modulepreload">
	<link href="/docs/diffusers/main/en/_app/immutable/chunks/DK803DsY.js" rel="modulepreload">
	<link href="/docs/diffusers/main/en/_app/immutable/entry/app.BpZzzFww.js" rel="modulepreload">
	<link href="/docs/diffusers/main/en/_app/immutable/chunks/DTwaC60R.js" rel="modulepreload">
	<link href="/docs/diffusers/main/en/_app/immutable/chunks/BTASUwav.js" rel="modulepreload">
	<link href="/docs/diffusers/main/en/_app/immutable/chunks/C1r7BcyV.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.wv1K_KEY.js" rel="modulepreload">
	<link href="/docs/diffusers/main/en/_app/immutable/chunks/BzOvRKAw.js" rel="modulepreload">
	<link href="/docs/diffusers/main/en/_app/immutable/nodes/312.DhhhNO2h.js" rel="modulepreload">
	<link href="/docs/diffusers/main/en/_app/immutable/chunks/CmJXCtRL.js" rel="modulepreload">
	<!--p4df5g--><meta name="hf:doc:metadata" content="{"title":"Nunchaku Lite","local":"nunchaku-lite","sections":[{"title":"Install the CUDA kernels","local":"install-the-cuda-kernels","sections":[],"depth":2},{"title":"Load a quantized pipeline","local":"load-a-quantized-pipeline","sections":[],"depth":2},{"title":"Supported quantization types","local":"supported-quantization-types","sections":[],"depth":2},{"title":"NunchakuLiteQuantizationConfig","local":"nunchakulitequantizationconfig","sections":[],"depth":2},{"title":"torch.compile","local":"torchcompile","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: right; margin-left: 10px; display: inline-flex; position: relative; z-index: 10;"><div 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>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="nunchaku-lite" 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="#nunchaku-lite"><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>Nunchaku Lite</span></h1><!--]--><!----> <p>Nunchaku Lite is a quantization backend for loading prequantized checkpoints in Diffusers. Create compatible checkpoints with <a href="https://github.com/rootonchair/diffuse-compressor" rel="nofollow">diffuse-compressor</a>. It quantizes and exports a transformer, then packages it as a Diffusers pipeline.</p> <p>Nunchaku Lite builds on the original <a href="https://github.com/nunchaku-ai/nunchaku" rel="nofollow">Nunchaku</a> inference engine, <a href="https://github.com/nunchaku-ai/deepcompressor" rel="nofollow">DeepCompressor</a> quantization library, and <a href="https://arxiv.org/abs/2411.05007" rel="nofollow">SVDQuant paper</a>.</p> <!--[1--><h2 class="relative group"><a id="install-the-cuda-kernels" 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="#install-the-cuda-kernels"><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>Install the CUDA kernels</span></h2><!--]--><!----> <p>The kernels package supplies the optimized CUDA kernels, which load automatically. Install it first.</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> Copied</div><!----></button><!----></div> <pre class="language-bash "><!---->pip install -U kernels<!----></pre></div><!----> <p>Nunchaku Lite loads its kernels from the <a href="https://huggingface.co/rootonchair/nunchaku-lite-kernels" rel="nofollow"><code>rootonchair/nunchaku-lite-kernels</code></a> repository, whose publisher is not a trusted kernel publisher on the Hub. Loading it downloads and executes code from the Hub, so Diffusers requires you to explicitly opt in by setting <code>DIFFUSERS_TRUST_REMOTE_KERNELS=true</code>. See <a href="../optimization/attention_backends#trusting-remote-kernels">Trusting remote kernels</a> for details.</p> <!--[1--><h2 class="relative group"><a id="load-a-quantized-pipeline" 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="#load-a-quantized-pipeline"><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>Load a quantized pipeline</span></h2><!--]--><!----> <p>Load the prequantized pipeline with <a href="/docs/diffusers/main/en/api/pipelines/overview#diffusers.DiffusionPipeline.from_pretrained">from_pretrained()</a>, which reads the quantization
	config from <code>config.json</code>.</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> Copied</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

	model_id = <span class="hljs-string">"rootonchair/ERNIE-Image-Turbo-nunchaku-lite-nvfp4"</span>

	pipe = DiffusionPipeline.from_pretrained(
	model_id, dtype=torch.bfloat16,
	).to(<span class="hljs-string">"cuda"</span>)

	prompt = <span class="hljs-string">"A modern red armchair in a quiet studio, soft window light, realistic product photography"</span>
	image = pipe(
	prompt=prompt,
	height=<span class="hljs-number">1024</span>,
	width=<span class="hljs-number">1024</span>,
	num_inference_steps=<span class="hljs-number">8</span>,
	guidance_scale=<span class="hljs-number">1.0</span>,
	).images[<span class="hljs-number">0</span>]
	image.save(<span class="hljs-string">"ernie-image-turbo-nunchaku-lite.png"</span>)<!----></pre></div><!----> <blockquote class="note"><p>The exported state dict must match the target Diffusers model architecture exactly. For example, a checkpoint
	quantized with fused QKV projections won’t load into a model config that expects separate Q, K, and V projection
	modules.</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>Supported quantization types</span></h2><!--]--><!----> <p>Nunchaku Lite supports the following quantized linear layer formats.</p> <blockquote class="tip"><p>Use <code>nvfp4</code> on Blackwell GPUs. Running <code>int4</code> checkpoints on Blackwell can be slower than <code>nvfp4</code>.</p></blockquote> <p>The CUDA kernels currently support the following NVIDIA GPU architectures:</p> <ul><li><code>sm_75</code> (Turing, for example RTX 2080)</li> <li><code>sm_80</code> (Ampere, for example A100)</li> <li><code>sm_86</code> (Ampere, for example RTX 3090 and RTX A6000)</li> <li><code>sm_89</code> (Ada, for example RTX 4090)</li> <li><code>sm_120</code> (Blackwell, for example RTX 5090)</li></ul> <blockquote class="note"><p>Hopper GPUs, such as <code>sm_90</code> H100 and H200, are not currently supported.</p></blockquote> <p><code>nvfp4</code> checkpoints require a Blackwell or newer NVIDIA GPU. On Blackwell GPUs, use PyTorch >= 2.7 with CUDA >= 12.8. <code>int4</code> checkpoints require a Turing or newer NVIDIA GPU.</p> <table><thead><tr><th>Method</th><th align="right">Precision</th><th align="right">Group size</th><th>Notes</th></tr></thead><tbody><tr><td><code>svdq_w4a4</code></td><td align="right"><code>nvfp4</code></td><td align="right">16</td><td>Uses NVFP4 runtime kernels with SVDQ low-rank correction.</td></tr><tr><td><code>svdq_w4a4</code></td><td align="right"><code>int4</code></td><td align="right">64</td><td>Uses INT4 W4A4 kernels with SVDQ low-rank correction.</td></tr><tr><td><code>awq_w4a16</code></td><td align="right"><code>int4</code></td><td align="right">64</td><td>Uses INT4 weight-only AWQ-style kernels.</td></tr></tbody></table> <!--[1--><h2 class="relative group"><a id="nunchakulitequantizationconfig" 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="#nunchakulitequantizationconfig"><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>NunchakuLiteQuantizationConfig</span></h2><!--]--><!----> <p>The <code>config.json</code> file must include a <a href="/docs/diffusers/main/en/api/quantization#diffusers.NunchakuLiteQuantizationConfig">NunchakuLiteQuantizationConfig</a>. It defines the runtime <code>compute_dtype</code> and the target modules for each Nunchaku Lite quantization method.</p> <ul><li><code>compute_dtype</code>: runtime dtype for floating-point buffers in quantized modules, typically <code>torch.bfloat16</code>.</li> <li><code>svdq_w4a4</code>: SVDQ W4A4 target config with <code>precision</code>, <code>group_size</code>, <code>rank</code>, and <code>targets</code>.</li> <li><code>awq_w4a16</code>: AWQ W4A16 target config with <code>precision</code>, <code>group_size</code>, and <code>targets</code>.</li></ul> <p>Each entry in <code>targets</code> must point to a linear layer. Diffusers swaps each <code>svdq_w4a4</code> target for an SVDQ W4A4 layer and each <code>awq_w4a16</code> target for an AWQ W4A16 layer. The example below shows the
	expected shape with shortened target lists.</p> <p>List each module you want to quantize under <code>svdq_w4a4</code> or <code>awq_w4a16</code>. A module can only use one method, so don’t list the same target under both.</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> Copied</div><!----></button><!----></div> <pre class="language-json "><!----><span class="hljs-punctuation">{</span>
	<span class="hljs-attr">"_class_name"</span><span class="hljs-punctuation">:</span> <span class="hljs-string">"ErnieImageTransformer2DModel"</span><span class="hljs-punctuation">,</span>
	<span class="hljs-attr">"quantization_config"</span><span class="hljs-punctuation">:</span> <span class="hljs-punctuation">{</span>
	<span class="hljs-attr">"quant_method"</span><span class="hljs-punctuation">:</span> <span class="hljs-string">"nunchaku_lite"</span><span class="hljs-punctuation">,</span>
	<span class="hljs-attr">"compute_dtype"</span><span class="hljs-punctuation">:</span> <span class="hljs-string">"bfloat16"</span><span class="hljs-punctuation">,</span>
	<span class="hljs-attr">"svdq_w4a4"</span><span class="hljs-punctuation">:</span> <span class="hljs-punctuation">{</span>
	<span class="hljs-attr">"precision"</span><span class="hljs-punctuation">:</span> <span class="hljs-string">"nvfp4"</span><span class="hljs-punctuation">,</span>
	<span class="hljs-attr">"group_size"</span><span class="hljs-punctuation">:</span> <span class="hljs-number">16</span><span class="hljs-punctuation">,</span>
	<span class="hljs-attr">"rank"</span><span class="hljs-punctuation">:</span> <span class="hljs-number">32</span><span class="hljs-punctuation">,</span>
	<span class="hljs-attr">"targets"</span><span class="hljs-punctuation">:</span> <span class="hljs-punctuation">[</span><span class="hljs-string">"layers.0.self_attention.to_q"</span><span class="hljs-punctuation">]</span>
	<span class="hljs-punctuation">}</span><span class="hljs-punctuation">,</span>
	<span class="hljs-attr">"awq_w4a16"</span><span class="hljs-punctuation">:</span> <span class="hljs-punctuation">{</span>
	<span class="hljs-attr">"precision"</span><span class="hljs-punctuation">:</span> <span class="hljs-string">"int4"</span><span class="hljs-punctuation">,</span>
	<span class="hljs-attr">"group_size"</span><span class="hljs-punctuation">:</span> <span class="hljs-number">64</span><span class="hljs-punctuation">,</span>
	<span class="hljs-attr">"targets"</span><span class="hljs-punctuation">:</span> <span class="hljs-punctuation">[</span><span class="hljs-string">"final_linear"</span><span class="hljs-punctuation">]</span>
	<span class="hljs-punctuation">}</span>
	<span class="hljs-punctuation">}</span>
	<span class="hljs-punctuation">}</span><!----></pre></div><!----> <!--[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>Nunchaku Lite kernels and quantized linear layers are compatible with <a href="../optimization/fp16#torchcompile"><code>torch.compile</code></a>.
	Compile the quantized transformer after loading the pipeline for faster inference.</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> Copied</div><!----></button><!----></div> <pre class="language-python "><!---->pipe.transformer = torch.<span class="hljs-built_in">compile</span>(pipe.transformer, mode=<span class="hljs-string">"default"</span>, fullgraph=<span class="hljs-literal">True</span>)<!----></pre></div><!----> <p>The compiled Nunchaku Lite NVFP4 pipeline runs 1.8x faster than the original BF16 pipeline (2.271s → 1.675s on an RTX PRO 6000).</p> <!--[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>Resources</span></h2><!--]--><!----> <ul><li><a href="https://github.com/rootonchair/diffuse-compressor" rel="nofollow">diffuse-compressor</a></li> <li><a href="https://nunchaku.tech/docs/nunchaku/installation/installation.html" rel="nofollow">Nunchaku installation requirements</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/nunchaku.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_acxue5 = {
	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.cTElinSc.js"),
	import("/docs/diffusers/main/en/_app/immutable/entry/app.BpZzzFww.js")
	]).then(([kit, app]) => {
	kit.start(app, element, {
	node_ids: [0, 312],
	data: [null,null],
	form: null,
	error: null
	});
	});
	}
	</script>

Xet Storage Details

Size:: 28.4 kB
Xet hash:: cc5389ae94c1f21ce78b7ca2aa202b174e792a4261e65f6e407837a82266127c

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