Hugging Face's logo

Buckets:
hf-doc-build
/
doc

Files
xet
hf-doc-build/doc / diffusers /v0.43.0 /zh /quantization /torchao.html
mishig's picture
mishig
download
raw
32.9 kB
<meta charset="utf-8" /><meta name="hf:doc:metadata" content="{&quot;title&quot;:&quot;torchao&quot;,&quot;local&quot;:&quot;torchao&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;torch.compile&quot;,&quot;local&quot;:&quot;torchcompile&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;支持的量化类型&quot;,&quot;local&quot;:&quot;supported-quantization-types&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;量化模型的序列化与反序列化&quot;,&quot;local&quot;:&quot;serializing-and-deserializing-quantized-models&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;资源&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.0S0_iEoG.js" rel="modulepreload">
<link href="/docs/diffusers/main/en/_app/immutable/chunks/I9eooN77.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.CPdOxCbw.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.DteTvmQZ.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/359.DjT-ulWG.js" rel="modulepreload">
<link href="/docs/diffusers/main/en/_app/immutable/chunks/DSPDl1gz.js" rel="modulepreload">
<!--1q5ej5h--><meta name="hf:doc:metadata" content="{&quot;title&quot;:&quot;torchao&quot;,&quot;local&quot;:&quot;torchao&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;torch.compile&quot;,&quot;local&quot;:&quot;torchcompile&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;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="torchao" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#torchao"><span><svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg><!----></span></a> <span><!---->torchao<!----></span></h1><!--]--><!----> <p><a href="https://github.com/pytorch/ao" rel="nofollow">torchao</a> 提供基于量化和稀疏性的高性能数据类型及优化方案，用于 PyTorch 模型的推理和训练。只要模型支持通过 <a href="https://hf.co/docs/accelerate/index" rel="nofollow">Accelerate</a> 加载，并且包含 <code>torch.nn.Linear</code> 层，即可支持任意模态下的任意模型。</p> <p>请确保使用以下命令安装了 Pytorch 2.5+ 和 torchao。</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> 已复制</div><!----></button><!----></div> <pre class="language-bash "><!---->uv pip install -U torch torchao<!----></pre></div><!----> <p>每种量化数据类型都作为 <a href="https://docs.pytorch.org/ao/main/api_reference/generated/torchao.core.config.AOBaseConfig.html" rel="nofollow">AOBaseConfig</a> 类的一个独立实例提供。通过暴露更多可用参数，这提供了更灵活的配置选项。</p> <p>在 <a href="/docs/diffusers/v0.43.0/zh/api/models/overview#diffusers.ModelMixin.from_pretrained">from_pretrained()</a> 中，将量化数据类型的 <code>AOBaseConfig</code>（例如 <a href="https://docs.pytorch.org/ao/main/api_reference/generated/torchao.quantization.Int4WeightOnlyConfig.html" rel="nofollow">Int4WeightOnlyConfig</a>）传递给 <a href="/docs/diffusers/v0.43.0/zh/api/quantization#diffusers.TorchAoConfig">TorchAoConfig</a>。</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> 已复制</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-keyword">import</span> torch
<span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> DiffusionPipeline, PipelineQuantizationConfig, TorchAoConfig
<span class="hljs-keyword">from</span> torchao.quantization <span class="hljs-keyword">import</span> Int8WeightOnlyConfig
pipeline_quant_config = PipelineQuantizationConfig(
quant_mapping={<span class="hljs-string">&quot;transformer&quot;</span>: TorchAoConfig(Int8WeightOnlyConfig(group_size=<span class="hljs-number">128</span>, version=<span class="hljs-number">2</span>))}
)
pipeline = DiffusionPipeline.from_pretrained(
<span class="hljs-string">&quot;black-forest-labs/FLUX.1-dev&quot;</span>,
quantization_config=pipeline_quant_config,
dtype=torch.bfloat16,
device_map=<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><code>device_map="cuda"</code> 会在加载时对每一层进行 GPU 量化。这种方式速度很快，但会临时占用额外的 GPU 内存来存储原始权重和量化后的权重。如果模型已经占用了大部分 GPU 内存，加载过程可能会因内存不足而失败。在这种情况下，请去掉 <code>device_map</code>，并在加载完成后将 pipeline 移至 GPU：</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> 已复制</div><!----></button><!----></div> <pre class="language-py "><!---->pipeline = DiffusionPipeline.from_pretrained(
<span class="hljs-string">&quot;black-forest-labs/FLUX.1-dev&quot;</span>,
quantization_config=pipeline_quant_config,
torch_dtype=torch.bfloat16,
)
pipeline.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>如果不使用 <code>device_map</code>，Diffusers 会在 CPU 上对层进行量化。这种方式速度较慢，但可避免量化过程中 GPU 内存的临时峰值。若要进一步降低 GPU 内存占用，请改用 <a href="/docs/diffusers/v0.43.0/zh/api/pipelines/overview#diffusers.DiffusionPipeline.enable_model_cpu_offload">enable_model_cpu_offload()</a>。此外，你还可以对文本编码器等其他组件进行量化。</p> <!--[1--><h2 class="relative group"><a id="torchcompile" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#torchcompile"><span><svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg><!----></span></a> <span><!---->torch.compile<!----></span></h2><!--]--><!----> <p>torchao 支持 <a href="../optimization/fp16#torchcompile">torch.compile</a>，只需一行代码即可加速推理。</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> 已复制</div><!----></button><!----></div> <pre class="language-python "><!----><span class="hljs-keyword">import</span> torch
<span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> DiffusionPipeline, PipelineQuantizationConfig, TorchAoConfig
<span class="hljs-keyword">from</span> torchao.quantization <span class="hljs-keyword">import</span> Int4WeightOnlyConfig
pipeline_quant_config = PipelineQuantizationConfig(
quant_mapping={<span class="hljs-string">&quot;transformer&quot;</span>: TorchAoConfig(Int4WeightOnlyConfig(group_size=<span class="hljs-number">128</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,
device_map=<span class="hljs-string">&quot;cuda&quot;</span> <span class="hljs-comment"># or &quot;mps&quot;, &quot;xpu&quot;, &quot;cpu&quot;</span>
)
pipeline.transformer.<span class="hljs-built_in">compile</span>(transformer, mode=<span class="hljs-string">&quot;max-autotune&quot;</span>, fullgraph=<span class="hljs-literal">True</span>)<!----></pre></div><!----> <p>有关 Flux 和 CogVideoX 的推理速度和内存使用基准测试，请参阅此<a href="https://github.com/huggingface/diffusers/pull/10009#issue-2688781450" rel="nofollow">表格</a>。torchao <a href="https://github.com/pytorch/ao/tree/main/torchao/quantization#benchmarks" rel="nofollow">仓库</a>中还提供了各种硬件上的更多基准测试。</p> <blockquote class="tip"><p>torchao 中的 FP8 后训练量化方案适用于计算能力至少为 8.9 的 GPU（如 RTX-4090、Hopper 等）。在生成图像和视频时，FP8 通常能在速度、内存和质量之间提供最佳的平衡。如果您的 GPU 兼容，我们建议将 FP8 与 torch.compile 结合使用。</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><!---->支持的量化类型<!----></span></h2><!--]--><!----> <p>torchao 支持针对 int8、float3-float8 以及 uint1-uint7 的仅权重量化，以及权重与动态激活值量化。</p> <p>仅权重量化以特定的低比特数据类型存储模型权重，但使用更高精度的数据类型（例如 <code>bfloat16</code>）执行计算。这降低了模型权重带来的内存需求，但保留了激活值计算时的内存峰值。</p> <p>动态激活量化将模型权重以低比特数据类型存储，同时在对激活值进行即时量化以节省额外内存。这降低了模型权重带来的内存需求，同时也减少了激活值计算带来的内存开销。不过，这有时可能会带来质量上的权衡，因此建议对不同的模型进行充分测试。</p> <p>请参阅 <a href="https://docs.pytorch.org/ao/stable/index.html" rel="nofollow">torchao 官方文档</a>，以更好地了解可用的量化方法。完整的配置选项列表可在 <a href="https://docs.pytorch.org/ao/main/workflows/inference.html#inference-workflows" rel="nofollow">此处</a> 找到。</p> <p>以下是一些常见的量化配置示例：</p> <table><thead><tr><th><strong>类别</strong></th><th><strong>配置类</strong></th></tr></thead><tbody><tr><td><strong>整数量化</strong></td><td><a href="https://docs.pytorch.org/ao/stable/api_reference/generated/torchao.quantization.Int4WeightOnlyConfig.html" rel="nofollow"><code>Int4WeightOnlyConfig</code></a>, <a href="https://docs.pytorch.org/ao/stable/api_reference/generated/torchao.quantization.Int8WeightOnlyConfig.html" rel="nofollow"><code>Int8WeightOnlyConfig</code></a>, <a href="https://docs.pytorch.org/ao/stable/api_reference/generated/torchao.quantization.Int8DynamicActivationInt8WeightConfig.html" rel="nofollow"><code>Int8DynamicActivationInt8WeightConfig</code></a></td></tr><tr><td><strong>8 位浮点量化</strong></td><td><a href="https://docs.pytorch.org/ao/stable/api_reference/generated/torchao.quantization.Float8WeightOnlyConfig.html" rel="nofollow"><code>Float8WeightOnlyConfig</code></a>, <a href="https://docs.pytorch.org/ao/stable/api_reference/generated/torchao.quantization.Float8DynamicActivationFloat8WeightConfig.html" rel="nofollow"><code>Float8DynamicActivationFloat8WeightConfig</code></a></td></tr><tr><td><strong>无符号整数量化</strong></td><td><a href="https://docs.pytorch.org/ao/stable/api_reference/generated/torchao.quantization.IntxWeightOnlyConfig.html" rel="nofollow"><code>IntxWeightOnlyConfig</code></a></td></tr></tbody></table> <!--[1--><h2 class="relative group"><a id="serializing-and-deserializing-quantized-models" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#serializing-and-deserializing-quantized-models"><span><svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg><!----></span></a> <span><!---->量化模型的序列化与反序列化<!----></span></h2><!--]--><!----> <p>要在指定的 dtype 中序列化量化模型，请先以所需的量化 dtype 加载模型，然后使用 <a href="/docs/diffusers/v0.43.0/zh/api/models/overview#diffusers.ModelMixin.save_pretrained">save_pretrained()</a> 方法将其保存。</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> 已复制</div><!----></button><!----></div> <pre class="language-python "><!----><span class="hljs-keyword">import</span> torch
<span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> AutoModel, TorchAoConfig
<span class="hljs-keyword">from</span> torchao.quantization <span class="hljs-keyword">import</span> Int8WeightOnlyConfig
quantization_config = TorchAoConfig(Int8WeightOnlyConfig())
transformer = AutoModel.from_pretrained(
<span class="hljs-string">&quot;black-forest-labs/Flux.1-Dev&quot;</span>,
subfolder=<span class="hljs-string">&quot;transformer&quot;</span>,
quantization_config=quantization_config,
dtype=torch.bfloat16,
)
transformer.save_pretrained(<span class="hljs-string">&quot;/path/to/flux_int8wo&quot;</span>, safe_serialization=<span class="hljs-literal">False</span>)<!----></pre></div><!----> <p>要加载序列化的量化模型，请使用 <a href="/docs/diffusers/v0.43.0/zh/api/models/overview#diffusers.ModelMixin.from_pretrained">from_pretrained()</a> 方法。</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> 已复制</div><!----></button><!----></div> <pre class="language-python "><!----><span class="hljs-keyword">import</span> torch
<span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> FluxPipeline, AutoModel
transformer = AutoModel.from_pretrained(<span class="hljs-string">&quot;/path/to/flux_int8wo&quot;</span>, dtype=torch.bfloat16, use_safetensors=<span class="hljs-literal">False</span>)
pipe = FluxPipeline.from_pretrained(<span class="hljs-string">&quot;black-forest-labs/Flux.1-Dev&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">30</span>, guidance_scale=<span class="hljs-number">7.0</span>).images[<span class="hljs-number">0</span>]
image.save(<span class="hljs-string">&quot;output.png&quot;</span>)<!----></pre></div><!----> <p>如果您使用的是 <code>torch&lt;=2.6.0</code>，某些量化方法（例如 <code>uint4</code> 仅权重量化）无法直接加载，在尝试加载模型时可能会引发 <code>UnpicklingError</code>，但在保存时表现正常。为了绕过此问题，可以手动将 state dict 加载到模型中。但请注意，这要求在 <code>torch.load</code> 中使用 <code>weights_only=False</code>，因此仅当权重来自可信来源时才应执行此操作。</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> 已复制</div><!----></button><!----></div> <pre class="language-python "><!----><span class="hljs-keyword">import</span> torch
<span class="hljs-keyword">from</span> accelerate <span class="hljs-keyword">import</span> init_empty_weights
<span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> FluxPipeline, AutoModel, TorchAoConfig
<span class="hljs-keyword">from</span> torchao.quantization <span class="hljs-keyword">import</span> IntxWeightOnlyConfig
<span class="hljs-comment"># Serialize the model</span>
transformer = AutoModel.from_pretrained(
<span class="hljs-string">&quot;black-forest-labs/Flux.1-Dev&quot;</span>,
subfolder=<span class="hljs-string">&quot;transformer&quot;</span>,
quantization_config=TorchAoConfig(IntxWeightOnlyConfig(dtype=torch.uint4)),
dtype=torch.bfloat16,
)
transformer.save_pretrained(<span class="hljs-string">&quot;/path/to/flux_uint4wo&quot;</span>, safe_serialization=<span class="hljs-literal">False</span>, max_shard_size=<span class="hljs-string">&quot;50GB&quot;</span>)
<span class="hljs-comment"># ...</span>
<span class="hljs-comment"># Load the model</span>
state_dict = torch.load(<span class="hljs-string">&quot;/path/to/flux_uint4wo/diffusion_pytorch_model.bin&quot;</span>, weights_only=<span class="hljs-literal">False</span>, map_location=<span class="hljs-string">&quot;cpu&quot;</span>)
<span class="hljs-keyword">with</span> init_empty_weights():
transformer = AutoModel.from_config(<span class="hljs-string">&quot;/path/to/flux_uint4wo/config.json&quot;</span>)
transformer.load_state_dict(state_dict, strict=<span class="hljs-literal">True</span>, assign=<span class="hljs-literal">True</span>)<!----></pre></div><!----> <blockquote class="tip"><p>如下示例所示，<a href="/docs/diffusers/v0.43.0/zh/api/models/auto_model#diffusers.AutoModel">AutoModel</a> API 支持 PyTorch &gt;= 2.6。</p></blockquote> <!--[1--><h2 class="relative group"><a id="resources" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#resources"><span><svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg><!----></span></a> <span><!---->资源<!----></span></h2><!--]--><!----> <ul><li><a href="https://docs.pytorch.org/ao/stable/index.html" rel="nofollow">TorchAO 量化 API</a></li> <li><a href="https://github.com/sayakpaul/diffusers-torchao" rel="nofollow">Diffusers-TorchAO 示例</a></li></ul> <a class="!text-gray-400 !no-underline text-sm flex items-center not-prose mt-4" href="https://github.com/huggingface/diffusers/blob/main/docs/source/en/quantization/torchao.md" target="_blank"><svg class="mr-1" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M31,16l-7,7l-1.41-1.41L28.17,16l-5.58-5.59L24,9l7,7z"></path><path d="M1,16l7-7l1.41,1.41L3.83,16l5.58,5.59L8,23l-7-7z"></path><path d="M12.419,25.484L17.639,6.552l1.932,0.518L14.351,26.002z"></path></svg><!----> <span><span class="underline">Update</span> on GitHub</span></a><!----> <p></p><!--]--><!----><!--]--><!--]--><!--]--> <!--[-1--><!--]--><!--]-->
<script>
{
__sveltekit_1qsdbwf = {
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.0S0_iEoG.js"),
import("/docs/diffusers/main/en/_app/immutable/entry/app.CPdOxCbw.js")
]).then(([kit, app]) => {
kit.start(app, element, {
node_ids: [0, 359],
data: [null,null],
form: null,
error: null
});
});
}
</script>

Xet Storage Details

Size:
32.9 kB
·
Xet hash:
4be9e221057ef135c83c5a6521bf9bba4dbf20ddd937f803e8a2bc37f96f3816

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