Buckets:

hf-doc-build
/

doc-dev

Files
xet
hf-doc-build/doc-dev / diffusers /pr_15232 /en /optimization /speed-memory-optims.html
HuggingFaceDocBuilder
about 24 hours ago
download
raw
27.9 kB
<meta charset="utf-8" /><meta name="hf:doc:metadata" content="{&quot;title&quot;:&quot;Compiling and offloading quantized models&quot;,&quot;local&quot;:&quot;compiling-and-offloading-quantized-models&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;Quantization and torch.compile&quot;,&quot;local&quot;:&quot;quantization-and-torchcompile&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Quantization, torch.compile, and offloading&quot;,&quot;local&quot;:&quot;quantization-torchcompile-and-offloading&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/339.dJh-4ujy.js" rel="modulepreload">
		<link href="/docs/diffusers/main/en/_app/immutable/chunks/B4AMoTAu.js" rel="modulepreload">
		<link href="/docs/diffusers/main/en/_app/immutable/chunks/B_Tpb3TG.js" rel="modulepreload">
		<!--qpa0fj--><meta name="hf:doc:metadata" content="{&quot;title&quot;:&quot;Compiling and offloading quantized models&quot;,&quot;local&quot;:&quot;compiling-and-offloading-quantized-models&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;Quantization and torch.compile&quot;,&quot;local&quot;:&quot;quantization-and-torchcompile&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Quantization, torch.compile, and offloading&quot;,&quot;local&quot;:&quot;quantization-torchcompile-and-offloading&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="compiling-and-offloading-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="#compiling-and-offloading-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><!---->Compiling and offloading quantized models<!----></span></h1><!--]--><!----> <p>Optimizing models often involves trade-offs between <a href="./fp16">inference speed</a> and <a href="./memory">memory-usage</a>. For instance, while <a href="./cache">caching</a> can boost inference speed, it also increases memory consumption since it needs to store the outputs of intermediate attention layers. A more balanced optimization strategy combines quantizing a model, <a href="./fp16#torchcompile">torch.compile</a> and various <a href="./memory#offloading">offloading methods</a>.</p> <blockquote class="tip"><p>Check the <a href="./fp16#torchcompile">torch.compile</a> guide to learn more about compilation and how they can be applied here. For example, regional compilation can significantly reduce compilation time without giving up any speedups.</p></blockquote> <p>For image generation, combining quantization and <a href="./memory#model-offloading">model offloading</a> can often give the best trade-off between quality, speed, and memory. Group offloading is not as effective for image generation because it is usually not possible to <em>fully</em> overlap data transfer if the compute kernel finishes faster. This results in some communication overhead between the CPU and GPU.</p> <p>For video generation, combining quantization and <a href="./memory#group-offloading">group-offloading</a> tends to be better because video models are more compute-bound.</p> <p>The table below provides a comparison of optimization strategy combinations and their impact on latency and memory-usage for Flux.</p> <table><thead><tr><th>combination</th><th>latency (s)</th><th>memory-usage (GB)</th></tr></thead><tbody><tr><td>quantization</td><td>32.602</td><td>14.9453</td></tr><tr><td>quantization, torch.compile</td><td>25.847</td><td>14.9448</td></tr><tr><td>quantization, torch.compile, model CPU offloading</td><td>32.312</td><td>12.2369</td></tr></tbody></table> <small>These results are benchmarked on Flux with a RTX 4090. The transformer and text_encoder components are quantized. Refer to the <a href="https://gist.github.com/sayakpaul/0db9d8eeeb3d2a0e5ed7cf0d9ca19b7d">benchmarking script</a> if you're interested in evaluating your own model.</small> <p>This guide will show you how to compile and offload a quantized model with <a href="../quantization/bitsandbytes#torchcompile">bitsandbytes</a>. Make sure you are using <a href="https://pytorch.org/get-started/locally/" rel="nofollow">PyTorch nightly</a> and the latest version of bitsandbytes.</p> <p>While we use bitsandbytes in this example, other quantization backends such as <a href="../quantization/torchao">TorchAO</a> also support these features.</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 bitsandbytes<!----></pre></div><!----> <!--[1--><h2 class="relative group"><a id="quantization-and-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="#quantization-and-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><!---->Quantization and torch.compile<!----></span></h2><!--]--><!----> <p>Start by <a href="../quantization/overview">quantizing</a> a model to reduce the memory required for storage and <a href="./fp16#torchcompile">compiling</a> it to accelerate inference.</p> <p>Configure the <a href="https://docs.pytorch.org/docs/stable/torch.compiler_dynamo_overview.html" rel="nofollow">Dynamo</a> <code>capture_dynamic_output_shape_ops = True</code> to handle dynamic outputs when compiling bitsandbytes models.</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-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

torch._dynamo.config.capture_dynamic_output_shape_ops = <span class="hljs-literal">True</span>

<span class="hljs-comment"># quantize</span>
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>],
)
pipeline = 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>

<span class="hljs-comment"># compile</span>
pipeline.transformer.to(memory_format=torch.channels_last)
pipeline.transformer.<span class="hljs-built_in">compile</span>(mode=<span class="hljs-string">&quot;max-autotune&quot;</span>, fullgraph=<span class="hljs-literal">True</span>)
pipeline(<span class="hljs-string">&quot;&quot;&quot;
    cinematic film still of a cat sipping a margarita in a pool in Palm Springs, California
    highly detailed, high budget hollywood movie, cinemascope, moody, epic, gorgeous, film grain
&quot;&quot;&quot;</span>
).images[<span class="hljs-number">0</span>]<!----></pre></div><!----> <!--[1--><h2 class="relative group"><a id="quantization-torchcompile-and-offloading" 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="#quantization-torchcompile-and-offloading"><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><!---->Quantization, torch.compile, and offloading<!----></span></h2><!--]--><!----> <p>In addition to quantization and torch.compile, try offloading if you need to reduce memory-usage further. Offloading moves various layers or model components from the CPU to the GPU as needed for computations.</p> <p>Configure the <a href="https://docs.pytorch.org/docs/stable/torch.compiler_dynamo_overview.html" rel="nofollow">Dynamo</a> <code>cache_size_limit</code> during offloading to avoid excessive recompilation and set <code>capture_dynamic_output_shape_ops = True</code> to handle dynamic outputs when compiling bitsandbytes models.</p> <div class="flex space-x-2 items-center my-1.5 mr-8 h-7 !pl-0 -mx-3 md:mx-0"><!--[--><div class="flex items-center border rounded-lg px-1.5 py-1 leading-none select-none text-smd border-gray-800 bg-black dark:bg-gray-700 text-white">model CPU offloading</div><div class="flex items-center border rounded-lg px-1.5 py-1 leading-none select-none text-smd text-gray-500 cursor-pointer opacity-90 hover:text-gray-700 dark:hover:text-gray-200 hover:shadow-sm">group offloading</div><!--]--></div> <div class="language-select"><div><p><a href="./memory#model-offloading">Model CPU offloading</a> moves an individual pipeline component, like the transformer model, to the GPU when it is needed for computation. Otherwise, it is offloaded to the CPU.</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-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

torch._dynamo.config.cache_size_limit = <span class="hljs-number">1000</span>
torch._dynamo.config.capture_dynamic_output_shape_ops = <span class="hljs-literal">True</span>

<span class="hljs-comment"># quantize</span>
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>],
)
pipeline = 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>

<span class="hljs-comment"># model CPU offloading</span>
pipeline.enable_model_cpu_offload()

<span class="hljs-comment"># compile</span>
pipeline.transformer.<span class="hljs-built_in">compile</span>()
pipeline(
    <span class="hljs-string">&quot;cinematic film still of a cat sipping a margarita in a pool in Palm Springs, California, highly detailed, high budget hollywood movie, cinemascope, moody, epic, gorgeous, film grain&quot;</span>
).images[<span class="hljs-number">0</span>]<!----></pre></div><!----><!----></div><!----> <div hidden=""><p><a href="./memory#group-offloading">Group offloading</a> moves the internal layers of an individual pipeline component, like the transformer model, to the GPU for computation and offloads it when it’s not required. At the same time, it uses the <a href="./memory#cuda-stream">CUDA stream</a> feature to prefetch the next layer for execution.</p> <p>By overlapping computation and data transfer, it is faster than model CPU offloading while also saving memory.</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-py "><!----><span class="hljs-comment"># pip install ftfy</span>
<span class="hljs-keyword">import</span> torch
<span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> AutoModel, DiffusionPipeline
<span class="hljs-keyword">from</span> diffusers.hooks <span class="hljs-keyword">import</span> apply_group_offloading
<span class="hljs-keyword">from</span> diffusers.utils <span class="hljs-keyword">import</span> export_to_video
<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> UMT5EncoderModel

torch._dynamo.config.cache_size_limit = <span class="hljs-number">1000</span>
torch._dynamo.config.capture_dynamic_output_shape_ops = <span class="hljs-literal">True</span>

<span class="hljs-comment"># quantize</span>
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&quot;</span>],
)

text_encoder = UMT5EncoderModel.from_pretrained(
    <span class="hljs-string">&quot;Wan-AI/Wan2.1-T2V-14B-Diffusers&quot;</span>, subfolder=<span class="hljs-string">&quot;text_encoder&quot;</span>, dtype=torch.bfloat16
)
pipeline = DiffusionPipeline.from_pretrained(
    <span class="hljs-string">&quot;Wan-AI/Wan2.1-T2V-14B-Diffusers&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>

<span class="hljs-comment"># group offloading</span>
onload_device = torch.device(<span class="hljs-string">&quot;cuda&quot;</span>)
offload_device = torch.device(<span class="hljs-string">&quot;cpu&quot;</span>)

pipeline.transformer.enable_group_offload(
    onload_device=onload_device,
    offload_device=offload_device,
    offload_type=<span class="hljs-string">&quot;leaf_level&quot;</span>,
    use_stream=<span class="hljs-literal">True</span>,
    non_blocking=<span class="hljs-literal">True</span>
)
pipeline.vae.enable_group_offload(
    onload_device=onload_device,
    offload_device=offload_device,
    offload_type=<span class="hljs-string">&quot;leaf_level&quot;</span>,
    use_stream=<span class="hljs-literal">True</span>,
    non_blocking=<span class="hljs-literal">True</span>
)
apply_group_offloading(
    pipeline.text_encoder,
    onload_device=onload_device,
    offload_type=<span class="hljs-string">&quot;leaf_level&quot;</span>,
    use_stream=<span class="hljs-literal">True</span>,
    non_blocking=<span class="hljs-literal">True</span>
)

<span class="hljs-comment"># compile</span>
pipeline.transformer.<span class="hljs-built_in">compile</span>()

prompt = <span class="hljs-string">&quot;&quot;&quot;
The camera rushes from far to near in a low-angle shot, 
revealing a white ferret on a log. It plays, leaps into the water, and emerges, as the camera zooms in 
for a close-up. Water splashes berry bushes nearby, while moss, snow, and leaves blanket the ground. 
Birch trees and a light blue sky frame the scene, with ferns in the foreground. Side lighting casts dynamic 
shadows and warm highlights. Medium composition, front view, low angle, with depth of field.
&quot;&quot;&quot;</span>
negative_prompt = <span class="hljs-string">&quot;&quot;&quot;
Bright tones, overexposed, static, blurred details, subtitles, style, works, paintings, images, static, overall gray, worst quality, 
low quality, JPEG compression residue, ugly, incomplete, extra fingers, poorly drawn hands, poorly drawn faces, deformed, disfigured, 
misshapen limbs, fused fingers, still picture, messy background, three legs, many people in the background, walking backwards
&quot;&quot;&quot;</span>

output = pipeline(
    prompt=prompt,
    negative_prompt=negative_prompt,
    num_frames=<span class="hljs-number">81</span>,
    guidance_scale=<span class="hljs-number">5.0</span>,
).frames[<span class="hljs-number">0</span>]
export_to_video(output, <span class="hljs-string">&quot;output.mp4&quot;</span>, fps=<span class="hljs-number">16</span>)<!----></pre></div><!----><!----></div><!----><!----></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/optimization/speed-memory-optims.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, 339],
							data: [null,null],
							form: null,
							error: null
						});
					});
				}
			</script>
		
Xet Storage Details

Size:: 27.9 kB
Xet hash:: f87f25cc28b4d97ae78a2966d59f29b32eecb307faf674c1e7becfbb65cfe0ad
Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.