Hugging Face's logo

Buckets:
hf-doc-build
/
doc-dev

Files
xet
hf-doc-build/doc-dev / transformers /main /zh /custom_models.html
rtrm's picture
rtrm
download
raw
53.7 kB
<meta charset="utf-8" /><meta name="hf:doc:metadata" content="{&quot;title&quot;:&quot;共享自定义模型&quot;,&quot;local&quot;:&quot;共享自定义模型&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;编写自定义配置&quot;,&quot;local&quot;:&quot;编写自定义配置&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;编写自定义模型&quot;,&quot;local&quot;:&quot;编写自定义模型&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;将代码发送到 Hub&quot;,&quot;local&quot;:&quot;将代码发送到-hub&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;使用带有自定义代码的模型&quot;,&quot;local&quot;:&quot;使用带有自定义代码的模型&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;将自定义代码的模型注册到自动类&quot;,&quot;local&quot;:&quot;将自定义代码的模型注册到自动类&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2}],&quot;depth&quot;:1}">
 <link href="/docs/transformers/main/zh/_app/immutable/assets/0.e3b0c442.css" rel="modulepreload">
 <link rel="modulepreload" href="/docs/transformers/main/zh/_app/immutable/entry/start.a61b9c50.js">
 <link rel="modulepreload" href="/docs/transformers/main/zh/_app/immutable/chunks/scheduler.9991993c.js">
 <link rel="modulepreload" href="/docs/transformers/main/zh/_app/immutable/chunks/singletons.2822fe91.js">
 <link rel="modulepreload" href="/docs/transformers/main/zh/_app/immutable/chunks/index.02cfeb18.js">
 <link rel="modulepreload" href="/docs/transformers/main/zh/_app/immutable/chunks/paths.d66588b4.js">
 <link rel="modulepreload" href="/docs/transformers/main/zh/_app/immutable/entry/app.99775688.js">
 <link rel="modulepreload" href="/docs/transformers/main/zh/_app/immutable/chunks/index.7fc9a5e7.js">
 <link rel="modulepreload" href="/docs/transformers/main/zh/_app/immutable/nodes/0.f4c5a5c1.js">
 <link rel="modulepreload" href="/docs/transformers/main/zh/_app/immutable/chunks/each.e59479a4.js">
 <link rel="modulepreload" href="/docs/transformers/main/zh/_app/immutable/nodes/9.c01e9457.js">
 <link rel="modulepreload" href="/docs/transformers/main/zh/_app/immutable/chunks/Tip.9de92fc6.js">
 <link rel="modulepreload" href="/docs/transformers/main/zh/_app/immutable/chunks/CodeBlock.e11cba92.js">
 <link rel="modulepreload" href="/docs/transformers/main/zh/_app/immutable/chunks/EditOnGithub.84ab7f0e.js"><!-- HEAD_svelte-u9bgzb_START --><meta name="hf:doc:metadata" content="{&quot;title&quot;:&quot;共享自定义模型&quot;,&quot;local&quot;:&quot;共享自定义模型&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;编写自定义配置&quot;,&quot;local&quot;:&quot;编写自定义配置&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;编写自定义模型&quot;,&quot;local&quot;:&quot;编写自定义模型&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;将代码发送到 Hub&quot;,&quot;local&quot;:&quot;将代码发送到-hub&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;使用带有自定义代码的模型&quot;,&quot;local&quot;:&quot;使用带有自定义代码的模型&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;将自定义代码的模型注册到自动类&quot;,&quot;local&quot;:&quot;将自定义代码的模型注册到自动类&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2}],&quot;depth&quot;:1}"><!-- HEAD_svelte-u9bgzb_END --> <p></p> <h1 class="relative group"><a id="共享自定义模型" 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="#共享自定义模型"><span><svg class="" 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>共享自定义模型</span></h1> <p data-svelte-h="svelte-9nxend">🤗 Transformers 库设计得易于扩展。每个模型的代码都在仓库给定的子文件夹中，没有进行抽象，因此你可以轻松复制模型代码文件并根据需要进行调整。</p> <p data-svelte-h="svelte-19a103f">如果你要编写全新的模型，从头开始可能更容易。在本教程中，我们将向你展示如何编写自定义模型及其配置，以便可以在 Transformers 中使用它；以及如何与社区共享它（及其依赖的代码），以便任何人都可以使用，即使它不在 🤗 Transformers 库中。</p> <p data-svelte-h="svelte-1b61zv2">我们将以 ResNet 模型为例，通过将 <a href="https://github.com/rwightman/pytorch-image-models" rel="nofollow">timm 库</a> 的 ResNet 类封装到 <a href="/docs/transformers/main/zh/main_classes/model#transformers.PreTrainedModel">PreTrainedModel</a> 中来进行说明。</p> <h2 class="relative group"><a id="编写自定义配置" 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="#编写自定义配置"><span><svg class="" 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>编写自定义配置</span></h2> <p data-svelte-h="svelte-1227xwf">在深入研究模型之前，让我们首先编写其配置。模型的配置是一个对象，其中包含构建模型所需的所有信息。我们将在下一节中看到，模型只能接受一个 <code>config</code> 来进行初始化，因此我们很需要使该对象尽可能完整。</p> <p data-svelte-h="svelte-i8hsxz">我们将采用一些我们可能想要调整的 ResNet 类的参数举例。不同的配置将为我们提供不同类型可能的 ResNet 模型。在确认其中一些参数的有效性后，我们只需存储这些参数。</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 class="" 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=""><!-- HTML_TAG_START --><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> PretrainedConfig
<span class="hljs-keyword">from</span> typing <span class="hljs-keyword">import</span> <span class="hljs-type">List</span>
<span class="hljs-keyword">class</span> <span class="hljs-title class_">ResnetConfig</span>(<span class="hljs-title class_ inherited__">PretrainedConfig</span>):
 model_type = <span class="hljs-string">&quot;resnet&quot;</span>
<span class="hljs-keyword">def</span> <span class="hljs-title function_">__init__</span>(<span class="hljs-params">
 self,
 block_type=<span class="hljs-string">&quot;bottleneck&quot;</span>,
 layers: <span class="hljs-type">List</span>[<span class="hljs-built_in">int</span>] = [<span class="hljs-number">3</span>, <span class="hljs-number">4</span>, <span class="hljs-number">6</span>, <span class="hljs-number">3</span>],
 num_classes: <span class="hljs-built_in">int</span> = <span class="hljs-number">1000</span>,
 input_channels: <span class="hljs-built_in">int</span> = <span class="hljs-number">3</span>,
 cardinality: <span class="hljs-built_in">int</span> = <span class="hljs-number">1</span>,
 base_width: <span class="hljs-built_in">int</span> = <span class="hljs-number">64</span>,
 stem_width: <span class="hljs-built_in">int</span> = <span class="hljs-number">64</span>,
 stem_type: <span class="hljs-built_in">str</span> = <span class="hljs-string">&quot;&quot;</span>,
 avg_down: <span class="hljs-built_in">bool</span> = <span class="hljs-literal">False</span>,
 **kwargs,
 </span>):
 <span class="hljs-keyword">if</span> block_type <span class="hljs-keyword">not</span> <span class="hljs-keyword">in</span> [<span class="hljs-string">&quot;basic&quot;</span>, <span class="hljs-string">&quot;bottleneck&quot;</span>]:
 <span class="hljs-keyword">raise</span> ValueError(<span class="hljs-string">f&quot;`block_type` must be &#x27;basic&#x27; or bottleneck&#x27;, got <span class="hljs-subst">{block_type}</span>.&quot;</span>)
 <span class="hljs-keyword">if</span> stem_type <span class="hljs-keyword">not</span> <span class="hljs-keyword">in</span> [<span class="hljs-string">&quot;&quot;</span>, <span class="hljs-string">&quot;deep&quot;</span>, <span class="hljs-string">&quot;deep-tiered&quot;</span>]:
 <span class="hljs-keyword">raise</span> ValueError(<span class="hljs-string">f&quot;`stem_type` must be &#x27;&#x27;, &#x27;deep&#x27; or &#x27;deep-tiered&#x27;, got <span class="hljs-subst">{stem_type}</span>.&quot;</span>)
self.block_type = block_type
 self.layers = layers
 self.num_classes = num_classes
 self.input_channels = input_channels
 self.cardinality = cardinality
 self.base_width = base_width
 self.stem_width = stem_width
 self.stem_type = stem_type
 self.avg_down = avg_down
 <span class="hljs-built_in">super</span>().__init__(**kwargs)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1a7oeqv">编写自定义配置时需要记住的三个重要事项如下：</p> <ul data-svelte-h="svelte-1s6yo4g"><li>必须继承自 <code>PretrainedConfig</code>，</li> <li><code>PretrainedConfig</code> 的 <code>__init__</code> 方法必须接受任何 kwargs，</li> <li>这些 <code>kwargs</code> 需要传递给超类的 <code>__init__</code> 方法。</li></ul> <p data-svelte-h="svelte-tl0pjo">继承是为了确保你获得来自 🤗 Transformers 库的所有功能，而另外两个约束源于 <code>PretrainedConfig</code> 的字段比你设置的字段多。在使用 <code>from_pretrained</code> 方法重新加载配置时，这些字段需要被你的配置接受，然后传递给超类。</p> <p data-svelte-h="svelte-1atr0z9">为你的配置定义 <code>model_type</code>（此处为 <code>model_type=&quot;resnet&quot;</code>）不是必须的，除非你想使用自动类注册你的模型（请参阅最后一节）。</p> <p data-svelte-h="svelte-1p70ds2">做完这些以后，就可以像使用库里任何其他模型配置一样，轻松地创建和保存配置。以下代码展示了如何创建并保存 resnet50d 配置：</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 class="" 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=""><!-- HTML_TAG_START -->resnet50d_config = ResnetConfig(block_type=<span class="hljs-string">&quot;bottleneck&quot;</span>, stem_width=<span class="hljs-number">32</span>, stem_type=<span class="hljs-string">&quot;deep&quot;</span>, avg_down=<span class="hljs-literal">True</span>)
resnet50d_config.save_pretrained(<span class="hljs-string">&quot;custom-resnet&quot;</span>)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-o9rk46">这行代码将在 <code>custom-resnet</code> 文件夹内保存一个名为 <code>config.json</code> 的文件。然后，你可以使用 <code>from_pretrained</code> 方法重新加载配置：</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 class="" 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=""><!-- HTML_TAG_START -->resnet50d_config = ResnetConfig.from_pretrained(<span class="hljs-string">&quot;custom-resnet&quot;</span>)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-ujmhd4">你还可以使用 <a href="/docs/transformers/main/zh/main_classes/configuration#transformers.PretrainedConfig">PretrainedConfig</a> 类的任何其他方法，例如 <a href="/docs/transformers/main/zh/main_classes/model#transformers.utils.PushToHubMixin.push_to_hub">push_to_hub()</a>，直接将配置上传到 Hub。</p> <h2 class="relative group"><a id="编写自定义模型" 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="#编写自定义模型"><span><svg class="" 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>编写自定义模型</span></h2> <p data-svelte-h="svelte-bmcwbb">有了 ResNet 配置后，就可以继续编写模型了。实际上，我们将编写两个模型：一个模型用于从一批图像中提取隐藏特征（类似于 <code>BertModel</code>），另一个模型适用于图像分类（类似于 <code>BertForSequenceClassification</code>）。</p> <p data-svelte-h="svelte-1lbtc6j">正如之前提到的，我们只会编写一个松散的模型包装，以使示例保持简洁。在编写此类之前，只需要建立起块类型（block types）与实际块类（block classes）之间的映射。然后，通过将所有内容传递给ResNet类，从配置中定义模型：</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 class="" 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=""><!-- HTML_TAG_START --><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> PreTrainedModel
<span class="hljs-keyword">from</span> timm.models.resnet <span class="hljs-keyword">import</span> BasicBlock, Bottleneck, ResNet
<span class="hljs-keyword">from</span> .configuration_resnet <span class="hljs-keyword">import</span> ResnetConfig
BLOCK_MAPPING = {<span class="hljs-string">&quot;basic&quot;</span>: BasicBlock, <span class="hljs-string">&quot;bottleneck&quot;</span>: Bottleneck}
<span class="hljs-keyword">class</span> <span class="hljs-title class_">ResnetModel</span>(<span class="hljs-title class_ inherited__">PreTrainedModel</span>):
 config_class = ResnetConfig
<span class="hljs-keyword">def</span> <span class="hljs-title function_">__init__</span>(<span class="hljs-params">self, config</span>):
 <span class="hljs-built_in">super</span>().__init__(config)
 block_layer = BLOCK_MAPPING[config.block_type]
 self.model = ResNet(
 block_layer,
 config.layers,
 num_classes=config.num_classes,
 in_chans=config.input_channels,
 cardinality=config.cardinality,
 base_width=config.base_width,
 stem_width=config.stem_width,
 stem_type=config.stem_type,
 avg_down=config.avg_down,
 )
<span class="hljs-keyword">def</span> <span class="hljs-title function_">forward</span>(<span class="hljs-params">self, tensor</span>):
 <span class="hljs-keyword">return</span> self.model.forward_features(tensor)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-16341pc">对用于进行图像分类的模型，我们只需更改前向方法：</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 class="" 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=""><!-- HTML_TAG_START --><span class="hljs-keyword">import</span> torch
<span class="hljs-keyword">class</span> <span class="hljs-title class_">ResnetModelForImageClassification</span>(<span class="hljs-title class_ inherited__">PreTrainedModel</span>):
 config_class = ResnetConfig
<span class="hljs-keyword">def</span> <span class="hljs-title function_">__init__</span>(<span class="hljs-params">self, config</span>):
 <span class="hljs-built_in">super</span>().__init__(config)
 block_layer = BLOCK_MAPPING[config.block_type]
 self.model = ResNet(
 block_layer,
 config.layers,
 num_classes=config.num_classes,
 in_chans=config.input_channels,
 cardinality=config.cardinality,
 base_width=config.base_width,
 stem_width=config.stem_width,
 stem_type=config.stem_type,
 avg_down=config.avg_down,
 )
<span class="hljs-keyword">def</span> <span class="hljs-title function_">forward</span>(<span class="hljs-params">self, tensor, labels=<span class="hljs-literal">None</span></span>):
 logits = self.model(tensor)
 <span class="hljs-keyword">if</span> labels <span class="hljs-keyword">is</span> <span class="hljs-keyword">not</span> <span class="hljs-literal">None</span>:
 loss = torch.nn.functional.cross_entropy(logits, labels)
 <span class="hljs-keyword">return</span> {<span class="hljs-string">&quot;loss&quot;</span>: loss, <span class="hljs-string">&quot;logits&quot;</span>: logits}
 <span class="hljs-keyword">return</span> {<span class="hljs-string">&quot;logits&quot;</span>: logits}<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-18rfxea">在这两种情况下，请注意我们如何继承 <code>PreTrainedModel</code> 并使用 <code>config</code> 调用了超类的初始化（有点像编写常规的torch.nn.Module）。设置 <code>config_class</code> 的那行代码不是必须的，除非你想使用自动类注册你的模型（请参阅最后一节）。</p> <div class="course-tip bg-gradient-to-br dark:bg-gradient-to-r before:border-green-500 dark:before:border-green-800 from-green-50 dark:from-gray-900 to-white dark:to-gray-950 border border-green-50 text-green-700 dark:text-gray-400"><p data-svelte-h="svelte-1t5zcat">如果你的模型与库中的某个模型非常相似，你可以重用与该模型相同的配置。</p></div> <p data-svelte-h="svelte-od54k8">你可以让模型返回任何你想要的内容，但是像我们为 <code>ResnetModelForImageClassification</code> 做的那样返回一个字典，并在传递标签时包含loss，可以使你的模型能够在 <code>Trainer</code> 类中直接使用。只要你计划使用自己的训练循环或其他库进行训练，也可以使用其他输出格式。</p> <p data-svelte-h="svelte-ym2h95">现在我们已经有了模型类，让我们创建一个：</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 class="" 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=""><!-- HTML_TAG_START -->resnet50d = ResnetModelForImageClassification(resnet50d_config)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1eqzw6j">同样的，你可以使用 <a href="/docs/transformers/main/zh/main_classes/model#transformers.PreTrainedModel">PreTrainedModel</a> 的任何方法，比如 <a href="/docs/transformers/main/zh/main_classes/model#transformers.PreTrainedModel.save_pretrained">save_pretrained()</a> 或者 <a href="/docs/transformers/main/zh/main_classes/model#transformers.utils.PushToHubMixin.push_to_hub">push_to_hub()</a>。我们将在下一节中使用第二种方法，并了解如何如何使用我们的模型的代码推送模型权重。但首先，让我们在模型内加载一些预训练权重。</p> <p data-svelte-h="svelte-36p84q">在你自己的用例中，你可能会在自己的数据上训练自定义模型。为了快速完成本教程，我们将使用 resnet50d 的预训练版本。由于我们的模型只是它的包装，转移这些权重将会很容易：</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 class="" 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=""><!-- HTML_TAG_START --><span class="hljs-keyword">import</span> timm
pretrained_model = timm.create_model(<span class="hljs-string">&quot;resnet50d&quot;</span>, pretrained=<span class="hljs-literal">True</span>)
resnet50d.model.load_state_dict(pretrained_model.state_dict())<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1wai86">现在让我们看看，如何确保在执行 <a href="/docs/transformers/main/zh/main_classes/model#transformers.PreTrainedModel.save_pretrained">save_pretrained()</a> 或 <a href="/docs/transformers/main/zh/main_classes/model#transformers.utils.PushToHubMixin.push_to_hub">push_to_hub()</a> 时，模型的代码被保存。</p> <h2 class="relative group"><a id="将代码发送到-hub" 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="#将代码发送到-hub"><span><svg class="" 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>将代码发送到 Hub</span></h2> <div class="course-tip course-tip-orange bg-gradient-to-br dark:bg-gradient-to-r before:border-orange-500 dark:before:border-orange-800 from-orange-50 dark:from-gray-900 to-white dark:to-gray-950 border border-orange-50 text-orange-700 dark:text-gray-400"><p data-svelte-h="svelte-k4mbd2">此 API 是实验性的，未来的发布中可能会有一些轻微的不兼容更改。</p></div> <p data-svelte-h="svelte-3w8bak">首先，确保你的模型在一个 <code>.py</code> 文件中完全定义。只要所有文件都位于同一目录中，它就可以依赖于某些其他文件的相对导入（目前我们还不为子模块支持此功能）。对于我们的示例，我们将在当前工作目录中名为 <code>resnet_model</code> 的文件夹中定义一个 <code>modeling_resnet.py</code> 文件和一个 <code>configuration_resnet.py</code> 文件。 配置文件包含 <code>ResnetConfig</code> 的代码，模型文件包含 <code>ResnetModel</code> 和 <code>ResnetModelForImageClassification</code> 的代码。</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 class="" 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=""><!-- HTML_TAG_START -->.
└── resnet_model
 ├── __init__.<span class="hljs-keyword">py</span>
 ├── configuration_resnet.<span class="hljs-keyword">py</span>
 └── modeling_resnet.<span class="hljs-keyword">py</span><!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-n806b8"><code>__init__.py</code> 可以为空，它的存在只是为了让 Python 检测到 <code>resnet_model</code> 可以用作模块。</p> <div class="course-tip course-tip-orange bg-gradient-to-br dark:bg-gradient-to-r before:border-orange-500 dark:before:border-orange-800 from-orange-50 dark:from-gray-900 to-white dark:to-gray-950 border border-orange-50 text-orange-700 dark:text-gray-400"><p data-svelte-h="svelte-3nrr3e">如果从库中复制模型文件，你需要将文件顶部的所有相对导入替换为从 <code>transformers</code> 包中的导入。</p></div> <p data-svelte-h="svelte-18wuvr2">请注意，你可以重用（或子类化）现有的配置/模型。</p> <p data-svelte-h="svelte-7gjgjm">要与社区共享您的模型，请参照以下步骤：首先从新创建的文件中导入ResNet模型和配置：</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 class="" 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=""><!-- HTML_TAG_START --><span class="hljs-keyword">from</span> resnet_model.configuration_resnet <span class="hljs-keyword">import</span> ResnetConfig
<span class="hljs-keyword">from</span> resnet_model.modeling_resnet <span class="hljs-keyword">import</span> ResnetModel, ResnetModelForImageClassification<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-nsc87p">接下来，你需要告诉库，当使用 <code>save_pretrained</code> 方法时，你希望复制这些对象的代码文件，并将它们正确注册到给定的 Auto 类（特别是对于模型），只需要运行以下代码：</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 class="" 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=""><!-- HTML_TAG_START -->ResnetConfig.register_for_auto_class()
ResnetModel.register_for_auto_class(<span class="hljs-string">&quot;AutoModel&quot;</span>)
ResnetModelForImageClassification.register_for_auto_class(<span class="hljs-string">&quot;AutoModelForImageClassification&quot;</span>)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1dt591c">请注意，对于配置（只有一个自动类 <code>AutoConfig</code>），不需要指定自动类，但对于模型来说情况不同。 你的自定义模型可能适用于许多不同的任务，因此你必须指定哪一个自动类适合你的模型。</p> <p data-svelte-h="svelte-1vunet5">接下来，让我们像之前一样创建配置和模型：</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 class="" 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=""><!-- HTML_TAG_START -->resnet50d_config = ResnetConfig(block_type=<span class="hljs-string">&quot;bottleneck&quot;</span>, stem_width=<span class="hljs-number">32</span>, stem_type=<span class="hljs-string">&quot;deep&quot;</span>, avg_down=<span class="hljs-literal">True</span>)
resnet50d = ResnetModelForImageClassification(resnet50d_config)
pretrained_model = timm.create_model(<span class="hljs-string">&quot;resnet50d&quot;</span>, pretrained=<span class="hljs-literal">True</span>)
resnet50d.model.load_state_dict(pretrained_model.state_dict())<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-11g63u9">现在要将模型推送到集线器，请确保你已登录。你看可以在终端中运行以下命令：</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 class="" 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=""><!-- HTML_TAG_START -->huggingface-cli login<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1emymh6">或者在笔记本中运行以下代码：</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 class="" 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=""><!-- HTML_TAG_START --><span class="hljs-keyword">from</span> huggingface_hub <span class="hljs-keyword">import</span> notebook_login
notebook_login()<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1xi8u2o">然后，可以这样将模型推送到自己的命名空间（或你所属的组织）：</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 class="" 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=""><!-- HTML_TAG_START -->resnet50d.push_to_hub(<span class="hljs-string">&quot;custom-resnet50d&quot;</span>)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1qq9kbx">除了模型权重和 JSON 格式的配置外，这行代码也会复制 <code>custom-resnet50d</code> 文件夹内的模型以及配置的 <code>.py</code> 文件并将结果上传至 Hub。你可以在此<a href="https://huggingface.co/sgugger/custom-resnet50d" rel="nofollow">模型仓库</a>中查看结果。</p> <p data-svelte-h="svelte-1e6fo38">有关推推送至 Hub 方法的更多信息，请参阅<a href="model_sharing">共享教程</a>。</p> <h2 class="relative group"><a id="使用带有自定义代码的模型" 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="#使用带有自定义代码的模型"><span><svg class="" 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>使用带有自定义代码的模型</span></h2> <p data-svelte-h="svelte-xah23j">可以使用自动类（auto-classes）和 <code>from_pretrained</code> 方法，使用模型仓库里带有自定义代码的配置、模型或分词器文件。所有上传到 Hub 的文件和代码都会进行恶意软件扫描（有关更多信息，请参阅 <a href="https://huggingface.co/docs/hub/security#malware-scanning" rel="nofollow">Hub 安全</a> 文档）, 但你仍应查看模型代码和作者，以避免在你的计算机上执行恶意代码。 设置 <code>trust_remote_code=True</code> 以使用带有自定义代码的模型：</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 class="" 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=""><!-- HTML_TAG_START --><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoModelForImageClassification
model = AutoModelForImageClassification.from_pretrained(<span class="hljs-string">&quot;sgugger/custom-resnet50d&quot;</span>, trust_remote_code=<span class="hljs-literal">True</span>)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1dvm79h">我们强烈建议为 <code>revision</code> 参数传递提交哈希（commit hash），以确保模型的作者没有使用一些恶意的代码行更新了代码（除非您完全信任模型的作者）。</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 class="" 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=""><!-- HTML_TAG_START -->commit_hash = <span class="hljs-string">&quot;ed94a7c6247d8aedce4647f00f20de6875b5b292&quot;</span>
model = AutoModelForImageClassification.from_pretrained(
 <span class="hljs-string">&quot;sgugger/custom-resnet50d&quot;</span>, trust_remote_code=<span class="hljs-literal">True</span>, revision=commit_hash
)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-8xcaeg">在 Hub 上浏览模型仓库的提交历史时，有一个按钮可以轻松复制任何提交的提交哈希。</p> <h2 class="relative group"><a id="将自定义代码的模型注册到自动类" 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="#将自定义代码的模型注册到自动类"><span><svg class="" 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>将自定义代码的模型注册到自动类</span></h2> <p data-svelte-h="svelte-1y2zt0l">如果你在编写一个扩展 🤗 Transformers 的库，你可能想要扩展自动类以包含您自己的模型。这与将代码推送到 Hub 不同，因为用户需要导入你的库才能获取自定义模型（与从 Hub 自动下载模型代码相反）。</p> <p data-svelte-h="svelte-l7gnnd">只要你的配置 <code>model_type</code> 属性与现有模型类型不同，并且你的模型类有正确的 <code>config_class</code> 属性，你可以像这样将它们添加到自动类中：</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 class="" 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=""><!-- HTML_TAG_START --><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoConfig, AutoModel, AutoModelForImageClassification
AutoConfig.register(<span class="hljs-string">&quot;resnet&quot;</span>, ResnetConfig)
AutoModel.register(ResnetConfig, ResnetModel)
AutoModelForImageClassification.register(ResnetConfig, ResnetModelForImageClassification)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-zrvvjw">请注意，将自定义配置注册到 <code>AutoConfig</code> 时，使用的第一个参数需要与自定义配置的 <code>model_type</code> 匹配；而将自定义模型注册到任何自动模型类时，使用的第一个参数需要与 <code>config_class</code> 匹配。</p> <a class="!text-gray-400 !no-underline text-sm flex items-center not-prose mt-4" href="https://github.com/huggingface/transformers/blob/main/docs/source/zh/custom_models.md" target="_blank"><span data-svelte-h="svelte-1kd6by1">&lt;</span> <span data-svelte-h="svelte-x0xyl0">&gt;</span> <span data-svelte-h="svelte-1dajgef"><span class="underline ml-1.5">Update</span> on GitHub</span></a> <p></p>
<script>
 {
 __sveltekit_173uja2 = {
 assets: "/docs/transformers/main/zh",
 base: "/docs/transformers/main/zh",
 env: {}
 };
 const element = document.currentScript.parentElement;
 const data = [null,null];
 Promise.all([
 import("/docs/transformers/main/zh/_app/immutable/entry/start.a61b9c50.js"),
 import("/docs/transformers/main/zh/_app/immutable/entry/app.99775688.js")
 ]).then(([kit, app]) => {
 kit.start(app, element, {
 node_ids: [0, 9],
 data,
 form: null,
 error: null
 });
 });
 }
 </script>

Xet Storage Details

Size:
53.7 kB
·
Xet hash:
87f95b796d755d8c1ad57f2a091bfb6893e869c1ccc0fd627ee86eecc7efc2c2

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