Buckets:

hf-doc-build
/

doc

Files

xet

hf-doc-build/doc / transformers /main /zh /create_a_model.html

HuggingFaceDocBuilder

1 day ago

download

raw

64.3 kB

	<meta charset="utf-8" /><meta name="hf:doc:metadata" content="{"title":"创建自定义架构","local":"创建自定义架构","sections":[{"title":"配置","local":"配置","sections":[],"depth":2},{"title":"模型","local":"模型","sections":[{"title":"模型头（Model heads）","local":"模型头model-heads","sections":[],"depth":3}],"depth":2},{"title":"分词器","local":"分词器","sections":[],"depth":2},{"title":"图像处理器","local":"图像处理器","sections":[],"depth":2},{"title":"特征提取器","local":"特征提取器","sections":[],"depth":2},{"title":"处理器","local":"处理器","sections":[],"depth":2}],"depth":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.7a5389c5.js">
	<link rel="modulepreload" href="/docs/transformers/main/zh/_app/immutable/chunks/scheduler.70fef3e7.js">
	<link rel="modulepreload" href="/docs/transformers/main/zh/_app/immutable/chunks/singletons.de8a9de6.js">
	<link rel="modulepreload" href="/docs/transformers/main/zh/_app/immutable/chunks/index.048d9726.js">
	<link rel="modulepreload" href="/docs/transformers/main/zh/_app/immutable/chunks/paths.a6fe27ce.js">
	<link rel="modulepreload" href="/docs/transformers/main/zh/_app/immutable/entry/app.ac7d4b06.js">
	<link rel="modulepreload" href="/docs/transformers/main/zh/_app/immutable/chunks/preload-helper.77b1b155.js">
	<link rel="modulepreload" href="/docs/transformers/main/zh/_app/immutable/chunks/index.747f74de.js">
	<link rel="modulepreload" href="/docs/transformers/main/zh/_app/immutable/nodes/0.3408eb2f.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/11.848a3b57.js">
	<link rel="modulepreload" href="/docs/transformers/main/zh/_app/immutable/chunks/Tip.8a10e646.js">
	<link rel="modulepreload" href="/docs/transformers/main/zh/_app/immutable/chunks/CopyLLMTxtMenu.25c10d36.js">
	<link rel="modulepreload" href="/docs/transformers/main/zh/_app/immutable/chunks/MermaidChart.svelte_svelte_type_style_lang.04bd7ab0.js">
	<link rel="modulepreload" href="/docs/transformers/main/zh/_app/immutable/chunks/CodeBlock.15c43204.js"><!-- HEAD_svelte-u9bgzb_START --><meta name="hf:doc:metadata" content="{"title":"创建自定义架构","local":"创建自定义架构","sections":[{"title":"配置","local":"配置","sections":[],"depth":2},{"title":"模型","local":"模型","sections":[{"title":"模型头（Model heads）","local":"模型头model-heads","sections":[],"depth":3}],"depth":2},{"title":"分词器","local":"分词器","sections":[],"depth":2},{"title":"图像处理器","local":"图像处理器","sections":[],"depth":2},{"title":"特征提取器","local":"特征提取器","sections":[],"depth":2},{"title":"处理器","local":"处理器","sections":[],"depth":2}],"depth":1}"><!-- HEAD_svelte-u9bgzb_END --> <p></p> <div class="items-center shrink-0 min-w-[100px] max-sm:min-w-[50px] justify-end ml-auto flex" style="float: right; margin-left: 10px; display: inline-flex; position: relative; z-index: 10;"><div class="inline-flex rounded-md max-sm:rounded-sm"><button class="inline-flex items-center gap-1 h-7 max-sm:h-7 px-2 max-sm:px-1.5 text-sm font-medium text-gray-800 border border-r-0 rounded-l-md max-sm:rounded-l-sm border-gray-200 bg-white hover:shadow-inner dark:border-gray-850 dark:bg-gray-950 dark:text-gray-200 dark:hover:bg-gray-800" aria-live="polite"><span class="inline-flex items-center justify-center rounded-md p-0.5 max-sm:p-0 hover:text-gray-800 dark:hover:text-gray-200"><svg class="sm:size-3.5 size-3" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg></span> <span>Copy page</span></button> <button class="inline-flex items-center justify-center w-6 max-sm:w-5 h-7 max-sm:h-7 disabled:pointer-events-none text-sm text-gray-500 hover:text-gray-700 dark:hover:text-white rounded-r-md max-sm:rounded-r-sm border border-l transition border-gray-200 bg-white hover:shadow-inner dark:border-gray-850 dark:bg-gray-950 dark:text-gray-200 dark:hover:bg-gray-800" aria-haspopup="menu" aria-expanded="false" aria-label="Open copy menu"><svg class="transition-transform text-gray-400 overflow-visible sm:size-3.5 size-3 rotate-0" width="1em" height="1em" viewBox="0 0 12 7" fill="none" xmlns="http://www.w3.org/2000/svg"><path d="M1 1L6 6L11 1" stroke="currentColor"></path></svg></button></div> </div> <h1 class="relative group"><a id="创建自定义架构" 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="#创建自定义架构"><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>创建自定义架构</span></h1> <p data-svelte-h="svelte-163eq9z"><a href="model_doc/auto"><code>AutoClass</code></a> 自动推断模型架构并下载预训练的配置和权重。一般来说，我们建议使用 <code>AutoClass</code> 生成与检查点（checkpoint）无关的代码。希望对特定模型参数有更多控制的用户，可以仅从几个基类创建自定义的 🤗 Transformers 模型。这对于任何有兴趣学习、训练或试验 🤗 Transformers 模型的人可能特别有用。通过本指南，深入了解如何不通过 <code>AutoClass</code> 创建自定义模型。了解如何：</p> <ul data-svelte-h="svelte-1asek4b"><li>加载并自定义模型配置。</li> <li>创建模型架构。</li> <li>为文本创建慢速和快速分词器。</li> <li>为视觉任务创建图像处理器。</li> <li>为音频任务创建特征提取器。</li> <li>为多模态任务创建处理器。</li></ul> <h2 class="relative group"><a id="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="#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>配置</span></h2> <p data-svelte-h="svelte-1nly6ic"><a href="main_classes/configuration">配置</a> 涉及到模型的具体属性。每个模型配置都有不同的属性；例如，所有 NLP 模型都共享 <code>hidden_size</code>、<code>num_attention_heads</code>、 <code>num_hidden_layers</code> 和 <code>vocab_size</code> 属性。这些属性用于指定构建模型时的注意力头数量或隐藏层层数。</p> <p data-svelte-h="svelte-1akqh2h">访问 <code>DistilBertConfig</code> 以更近一步了解 <a href="model_doc/distilbert">DistilBERT</a>，检查它的属性：</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="language-py "><!-- HTML_TAG_START --><span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> DistilBertConfig

	<span class="hljs-meta">>>> </span>config = DistilBertConfig()
	<span class="hljs-meta">>>> </span><span class="hljs-built_in">print</span>(config)
	DistilBertConfig {
	<span class="hljs-string">"activation"</span>: <span class="hljs-string">"gelu"</span>,
	<span class="hljs-string">"attention_dropout"</span>: <span class="hljs-number">0.1</span>,
	<span class="hljs-string">"dim"</span>: <span class="hljs-number">768</span>,
	<span class="hljs-string">"dropout"</span>: <span class="hljs-number">0.1</span>,
	<span class="hljs-string">"hidden_dim"</span>: <span class="hljs-number">3072</span>,
	<span class="hljs-string">"initializer_range"</span>: <span class="hljs-number">0.02</span>,
	<span class="hljs-string">"max_position_embeddings"</span>: <span class="hljs-number">512</span>,
	<span class="hljs-string">"model_type"</span>: <span class="hljs-string">"distilbert"</span>,
	<span class="hljs-string">"n_heads"</span>: <span class="hljs-number">12</span>,
	<span class="hljs-string">"n_layers"</span>: <span class="hljs-number">6</span>,
	<span class="hljs-string">"pad_token_id"</span>: <span class="hljs-number">0</span>,
	<span class="hljs-string">"qa_dropout"</span>: <span class="hljs-number">0.1</span>,
	<span class="hljs-string">"seq_classif_dropout"</span>: <span class="hljs-number">0.2</span>,
	<span class="hljs-string">"sinusoidal_pos_embds"</span>: false,
	<span class="hljs-string">"transformers_version"</span>: <span class="hljs-string">"4.16.2"</span>,
	<span class="hljs-string">"vocab_size"</span>: <span class="hljs-number">30522</span>
	}<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-14v2kvo"><code>DistilBertConfig</code> 显示了构建基础 <code>DistilBertModel</code> 所使用的所有默认属性。所有属性都可以进行自定义，为实验创造了空间。例如，您可以将默认模型自定义为：</p> <ul data-svelte-h="svelte-13r6eur"><li>使用 <code>activation</code> 参数尝试不同的激活函数。</li> <li>使用 <code>attention_dropout</code> 参数为 attention probabilities 使用更高的 dropout ratio。</li></ul> <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="language-py "><!-- HTML_TAG_START --><span class="hljs-meta">>>> </span>my_config = DistilBertConfig(activation=<span class="hljs-string">"relu"</span>, attention_dropout=<span class="hljs-number">0.4</span>)
	<span class="hljs-meta">>>> </span><span class="hljs-built_in">print</span>(my_config)
	DistilBertConfig {
	<span class="hljs-string">"activation"</span>: <span class="hljs-string">"relu"</span>,
	<span class="hljs-string">"attention_dropout"</span>: <span class="hljs-number">0.4</span>,
	<span class="hljs-string">"dim"</span>: <span class="hljs-number">768</span>,
	<span class="hljs-string">"dropout"</span>: <span class="hljs-number">0.1</span>,
	<span class="hljs-string">"hidden_dim"</span>: <span class="hljs-number">3072</span>,
	<span class="hljs-string">"initializer_range"</span>: <span class="hljs-number">0.02</span>,
	<span class="hljs-string">"max_position_embeddings"</span>: <span class="hljs-number">512</span>,
	<span class="hljs-string">"model_type"</span>: <span class="hljs-string">"distilbert"</span>,
	<span class="hljs-string">"n_heads"</span>: <span class="hljs-number">12</span>,
	<span class="hljs-string">"n_layers"</span>: <span class="hljs-number">6</span>,
	<span class="hljs-string">"pad_token_id"</span>: <span class="hljs-number">0</span>,
	<span class="hljs-string">"qa_dropout"</span>: <span class="hljs-number">0.1</span>,
	<span class="hljs-string">"seq_classif_dropout"</span>: <span class="hljs-number">0.2</span>,
	<span class="hljs-string">"sinusoidal_pos_embds"</span>: false,
	<span class="hljs-string">"transformers_version"</span>: <span class="hljs-string">"4.16.2"</span>,
	<span class="hljs-string">"vocab_size"</span>: <span class="hljs-number">30522</span>
	}<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-32qms">预训练模型的属性可以在 <a href="/docs/transformers/main/zh/main_classes/configuration#transformers.PreTrainedConfig.from_pretrained">from_pretrained()</a> 函数中进行修改：</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="language-py "><!-- HTML_TAG_START --><span class="hljs-meta">>>> </span>my_config = DistilBertConfig.from_pretrained(<span class="hljs-string">"distilbert/distilbert-base-uncased"</span>, activation=<span class="hljs-string">"relu"</span>, attention_dropout=<span class="hljs-number">0.4</span>)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-suo1i1">当你对模型配置满意时，可以使用 <a href="/docs/transformers/main/zh/main_classes/configuration#transformers.PreTrainedConfig.save_pretrained">save_pretrained()</a> 来保存配置。你的配置文件将以 JSON 文件的形式存储在指定的保存目录中：</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="language-py "><!-- HTML_TAG_START --><span class="hljs-meta">>>> </span>my_config.save_pretrained(save_directory=<span class="hljs-string">"./your_model_save_path"</span>)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1icd8gt">要重用配置文件，请使用 <a href="/docs/transformers/main/zh/main_classes/configuration#transformers.PreTrainedConfig.from_pretrained">from_pretrained()</a> 进行加载：</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="language-py "><!-- HTML_TAG_START --><span class="hljs-meta">>>> </span>my_config = DistilBertConfig.from_pretrained(<span class="hljs-string">"./your_model_save_path/config.json"</span>)<!-- HTML_TAG_END --></pre></div> <blockquote class="tip"><p data-svelte-h="svelte-k41iwq">你还可以将配置文件保存为字典，甚至只保存自定义配置属性与默认配置属性之间的差异！有关更多详细信息，请参阅 <a href="main_classes/configuration">配置</a> 文档。</p></blockquote> <h2 class="relative group"><a id="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="#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>模型</span></h2> <p data-svelte-h="svelte-15lou6n">接下来，创建一个<a href="main_classes/models">模型</a>。模型，也可泛指架构，定义了每一层网络的行为以及进行的操作。配置中的 <code>num_hidden_layers</code> 等属性用于定义架构。每个模型都共享基类 <a href="/docs/transformers/main/zh/main_classes/model#transformers.PreTrainedModel">PreTrainedModel</a> 和一些常用方法，例如调整输入嵌入的大小和修剪自注意力头。此外，所有模型都是 <a href="https://pytorch.org/docs/stable/generated/torch.nn.Module.html" rel="nofollow"><code>torch.nn.Module</code></a> 的子类。这意味着模型与各自框架的用法兼容。</p> <p data-svelte-h="svelte-1fn6s0p">将自定义配置属性加载到模型中：</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="language-py "><!-- HTML_TAG_START --><span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> DistilBertModel

	<span class="hljs-meta">>>> </span>my_config = DistilBertConfig.from_pretrained(<span class="hljs-string">"./your_model_save_path/config.json"</span>)
	<span class="hljs-meta">>>> </span>model = DistilBertModel(my_config)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1y5tk1f">这段代码创建了一个具有随机参数而不是预训练权重的模型。在训练该模型之前，您还无法将该模型用于任何用途。训练是一项昂贵且耗时的过程。通常来说，最好使用预训练模型来更快地获得更好的结果，同时仅使用训练所需资源的一小部分。</p> <p data-svelte-h="svelte-l9ds4l">使用 <a href="/docs/transformers/main/zh/main_classes/model#transformers.PreTrainedModel.from_pretrained">from_pretrained()</a> 创建预训练模型：</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="language-py "><!-- HTML_TAG_START --><span class="hljs-meta">>>> </span>model = DistilBertModel.from_pretrained(<span class="hljs-string">"distilbert/distilbert-base-uncased"</span>)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-dvjp91">当加载预训练权重时，如果模型是由 🤗 Transformers 提供的，将自动加载默认模型配置。然而，如果你愿意，仍然可以将默认模型配置的某些或者所有属性替换成你自己的配置：</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="language-py "><!-- HTML_TAG_START --><span class="hljs-meta">>>> </span>model = DistilBertModel.from_pretrained(<span class="hljs-string">"distilbert/distilbert-base-uncased"</span>, config=my_config)<!-- HTML_TAG_END --></pre></div> <h3 class="relative group"><a id="模型头model-heads" 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="#模型头model-heads"><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>模型头（Model heads）</span></h3> <p data-svelte-h="svelte-qcf0a1">此时，你已经有了一个输出<em>隐藏状态</em>的基础 DistilBERT 模型。隐藏状态作为输入传递到模型头以生成最终输出。🤗 Transformers 为每个任务提供不同的模型头，只要模型支持该任务（即，您不能使用 DistilBERT 来执行像翻译这样的序列到序列任务）。</p> <p data-svelte-h="svelte-16e0zlv">例如，<code>DistilBertForSequenceClassification</code> 是一个带有序列分类头（sequence classification head）的基础 DistilBERT 模型。序列分类头是池化输出之上的线性层。</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="language-py "><!-- HTML_TAG_START --><span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> DistilBertForSequenceClassification

	<span class="hljs-meta">>>> </span>model = DistilBertForSequenceClassification.from_pretrained(<span class="hljs-string">"distilbert/distilbert-base-uncased"</span>)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-n48y07">通过切换到不同的模型头，可以轻松地将此检查点重复用于其他任务。对于问答任务，你可以使用 <code>DistilBertForQuestionAnswering</code> 模型头。问答头（question answering head）与序列分类头类似，不同点在于它是隐藏状态输出之上的线性层。</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="language-py "><!-- HTML_TAG_START --><span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> DistilBertForQuestionAnswering

	<span class="hljs-meta">>>> </span>model = DistilBertForQuestionAnswering.from_pretrained(<span class="hljs-string">"distilbert/distilbert-base-uncased"</span>)<!-- HTML_TAG_END --></pre></div> <h2 class="relative group"><a id="分词器" 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="#分词器"><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>分词器</span></h2> <p data-svelte-h="svelte-espr00">在将模型用于文本数据之前，你需要的最后一个基类是 <a href="main_classes/tokenizer">tokenizer</a>，它用于将原始文本转换为张量。🤗 Transformers 支持两种类型的分词器：</p> <ul data-svelte-h="svelte-fncuqw"><li><a href="/docs/transformers/main/zh/main_classes/tokenizer#transformers.PythonBackend">PreTrainedTokenizer</a>：分词器的Python实现</li> <li><a href="/docs/transformers/main/zh/main_classes/tokenizer#transformers.TokenizersBackend">PreTrainedTokenizerFast</a>：来自我们基于 Rust 的 <a href="https://huggingface.co/docs/tokenizers/python/latest/" rel="nofollow">🤗 Tokenizer</a> 库的分词器。因为其使用了 Rust 实现，这种分词器类型的速度要快得多，尤其是在批量分词（batch tokenization）的时候。快速分词器还提供其他的方法，例如<em>偏移映射（offset mapping）</em>，它将标记（token）映射到其原始单词或字符。</li></ul> <p data-svelte-h="svelte-51pwdw">这两种分词器都支持常用的方法，如编码和解码、添加新标记以及管理特殊标记。</p> <blockquote class="warning"><p data-svelte-h="svelte-205pqb">并非每个模型都支持快速分词器。参照这张 <a href="index#supported-frameworks">表格</a> 查看模型是否支持快速分词器。</p></blockquote> <p data-svelte-h="svelte-1f3jbei">如果您训练了自己的分词器，则可以从<em>词表</em>文件创建一个分词器：</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="language-py "><!-- HTML_TAG_START --><span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> DistilBertTokenizer

	<span class="hljs-meta">>>> </span>my_tokenizer = DistilBertTokenizer(vocab_file=<span class="hljs-string">"my_vocab_file.txt"</span>, do_lower_case=<span class="hljs-literal">False</span>, padding_side=<span class="hljs-string">"left"</span>)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-15lam1s">请务必记住，自定义分词器生成的词表与预训练模型分词器生成的词表是不同的。如果使用预训练模型，则需要使用预训练模型的词表，否则输入将没有意义。使用 <code>DistilBertTokenizer</code> 类创建具有预训练模型词表的分词器：</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="language-py "><!-- HTML_TAG_START --><span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> DistilBertTokenizer

	<span class="hljs-meta">>>> </span>slow_tokenizer = DistilBertTokenizer.from_pretrained(<span class="hljs-string">"distilbert/distilbert-base-uncased"</span>)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-bg238o">使用 <code>DistilBertTokenizerFast</code> 类创建快速分词器：</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="language-py "><!-- HTML_TAG_START --><span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> DistilBertTokenizerFast

	<span class="hljs-meta">>>> </span>fast_tokenizer = DistilBertTokenizerFast.from_pretrained(<span class="hljs-string">"distilbert/distilbert-base-uncased"</span>)<!-- HTML_TAG_END --></pre></div> <blockquote class="tip"><p data-svelte-h="svelte-12jlto4">默认情况下，<code>AutoTokenizer</code> 将尝试加载快速标记生成器。你可以通过在 <code>from_pretrained</code> 中设置 <code>use_fast=False</code> 以禁用此行为。</p></blockquote> <h2 class="relative group"><a id="图像处理器" 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="#图像处理器"><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>图像处理器</span></h2> <p data-svelte-h="svelte-1w3qxda">图像处理器用于处理视觉输入。它继承自 <a href="/docs/transformers/main/zh/main_classes/image_processor#transformers.ImageProcessingMixin">ImageProcessingMixin</a> 基类。</p> <p data-svelte-h="svelte-1evm4el">要使用它，需要创建一个与你使用的模型关联的图像处理器。例如，如果你使用 <a href="model_doc/vit">ViT</a> 进行图像分类，可以创建一个默认的 <code>ViTImageProcessor</code>：</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg 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="language-py "><!-- HTML_TAG_START --><span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> ViTImageProcessor

	<span class="hljs-meta">>>> </span>vit_extractor = ViTImageProcessor()
	<span class="hljs-meta">>>> </span><span class="hljs-built_in">print</span>(vit_extractor)
	ViTImageProcessor {
	<span class="hljs-string">"do_normalize"</span>: true,
	<span class="hljs-string">"do_resize"</span>: true,
	<span class="hljs-string">"image_processor_type"</span>: <span class="hljs-string">"ViTImageProcessor"</span>,
	<span class="hljs-string">"image_mean"</span>: [
	<span class="hljs-number">0.5</span>,
	<span class="hljs-number">0.5</span>,
	<span class="hljs-number">0.5</span>
	],
	<span class="hljs-string">"image_std"</span>: [
	<span class="hljs-number">0.5</span>,
	<span class="hljs-number">0.5</span>,
	<span class="hljs-number">0.5</span>
	],
	<span class="hljs-string">"resample"</span>: <span class="hljs-number">2</span>,
	<span class="hljs-string">"size"</span>: <span class="hljs-number">224</span>
	}<!-- HTML_TAG_END --></pre></div> <blockquote class="tip"><p data-svelte-h="svelte-1tl7krt">如果您不需要进行任何自定义，只需使用 <code>from_pretrained</code> 方法加载模型的默认图像处理器参数。</p></blockquote> <p data-svelte-h="svelte-y6n2fo">修改任何 <code>ViTImageProcessor</code> 参数以创建自定义图像处理器：</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="language-py "><!-- HTML_TAG_START --><span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> ViTImageProcessor

	<span class="hljs-meta">>>> </span>my_vit_extractor = ViTImageProcessor(resample=<span class="hljs-string">"PIL.Image.BOX"</span>, do_normalize=<span class="hljs-literal">False</span>, image_mean=[<span class="hljs-number">0.3</span>, <span class="hljs-number">0.3</span>, <span class="hljs-number">0.3</span>])
	<span class="hljs-meta">>>> </span><span class="hljs-built_in">print</span>(my_vit_extractor)
	ViTImageProcessor {
	<span class="hljs-string">"do_normalize"</span>: false,
	<span class="hljs-string">"do_resize"</span>: true,
	<span class="hljs-string">"image_processor_type"</span>: <span class="hljs-string">"ViTImageProcessor"</span>,
	<span class="hljs-string">"image_mean"</span>: [
	<span class="hljs-number">0.3</span>,
	<span class="hljs-number">0.3</span>,
	<span class="hljs-number">0.3</span>
	],
	<span class="hljs-string">"image_std"</span>: [
	<span class="hljs-number">0.5</span>,
	<span class="hljs-number">0.5</span>,
	<span class="hljs-number">0.5</span>
	],
	<span class="hljs-string">"resample"</span>: <span class="hljs-string">"PIL.Image.BOX"</span>,
	<span class="hljs-string">"size"</span>: <span class="hljs-number">224</span>
	}<!-- HTML_TAG_END --></pre></div> <h2 class="relative group"><a id="特征提取器" 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="#特征提取器"><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>特征提取器</span></h2> <p data-svelte-h="svelte-k22coh">特征提取器用于处理音频输入。它继承自 <a href="/docs/transformers/main/zh/main_classes/feature_extractor#transformers.FeatureExtractionMixin">FeatureExtractionMixin</a> 基类，亦可继承 <a href="/docs/transformers/main/zh/main_classes/feature_extractor#transformers.SequenceFeatureExtractor">SequenceFeatureExtractor</a> 类来处理音频输入。</p> <p data-svelte-h="svelte-o65i7e">要使用它，创建一个与你使用的模型关联的特征提取器。例如，如果你使用 <a href="model_doc/wav2vec2">Wav2Vec2</a> 进行音频分类，可以创建一个默认的 <code>Wav2Vec2FeatureExtractor</code>：</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg 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="language-py "><!-- HTML_TAG_START --><span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> Wav2Vec2FeatureExtractor

	<span class="hljs-meta">>>> </span>w2v2_extractor = Wav2Vec2FeatureExtractor()
	<span class="hljs-meta">>>> </span><span class="hljs-built_in">print</span>(w2v2_extractor)
	Wav2Vec2FeatureExtractor {
	<span class="hljs-string">"do_normalize"</span>: true,
	<span class="hljs-string">"feature_extractor_type"</span>: <span class="hljs-string">"Wav2Vec2FeatureExtractor"</span>,
	<span class="hljs-string">"feature_size"</span>: <span class="hljs-number">1</span>,
	<span class="hljs-string">"padding_side"</span>: <span class="hljs-string">"right"</span>,
	<span class="hljs-string">"padding_value"</span>: <span class="hljs-number">0.0</span>,
	<span class="hljs-string">"return_attention_mask"</span>: false,
	<span class="hljs-string">"sampling_rate"</span>: <span class="hljs-number">16000</span>
	}<!-- HTML_TAG_END --></pre></div> <blockquote class="tip"><p data-svelte-h="svelte-paor8k">如果您不需要进行任何自定义，只需使用 <code>from_pretrained</code> 方法加载模型的默认特征提取器参数。</p></blockquote> <p data-svelte-h="svelte-wsqmub">修改任何 <code>Wav2Vec2FeatureExtractor</code> 参数以创建自定义特征提取器：</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="language-py "><!-- HTML_TAG_START --><span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> Wav2Vec2FeatureExtractor

	<span class="hljs-meta">>>> </span>w2v2_extractor = Wav2Vec2FeatureExtractor(sampling_rate=<span class="hljs-number">8000</span>, do_normalize=<span class="hljs-literal">False</span>)
	<span class="hljs-meta">>>> </span><span class="hljs-built_in">print</span>(w2v2_extractor)
	Wav2Vec2FeatureExtractor {
	<span class="hljs-string">"do_normalize"</span>: false,
	<span class="hljs-string">"feature_extractor_type"</span>: <span class="hljs-string">"Wav2Vec2FeatureExtractor"</span>,
	<span class="hljs-string">"feature_size"</span>: <span class="hljs-number">1</span>,
	<span class="hljs-string">"padding_side"</span>: <span class="hljs-string">"right"</span>,
	<span class="hljs-string">"padding_value"</span>: <span class="hljs-number">0.0</span>,
	<span class="hljs-string">"return_attention_mask"</span>: false,
	<span class="hljs-string">"sampling_rate"</span>: <span class="hljs-number">8000</span>
	}<!-- HTML_TAG_END --></pre></div> <h2 class="relative group"><a id="处理器" 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="#处理器"><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>处理器</span></h2> <p data-svelte-h="svelte-gboqg0">对于支持多模式任务的模型，🤗 Transformers 提供了一个处理器类，可以方便地将特征提取器和分词器等处理类包装到单个对象中。例如，让我们使用 <code>Wav2Vec2Processor</code> 来执行自动语音识别任务 (ASR)。 ASR 将音频转录为文本，因此您将需要一个特征提取器和一个分词器。</p> <p data-svelte-h="svelte-liph8s">创建一个特征提取器来处理音频输入：</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="language-py "><!-- HTML_TAG_START --><span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> Wav2Vec2FeatureExtractor

	<span class="hljs-meta">>>> </span>feature_extractor = Wav2Vec2FeatureExtractor(padding_value=<span class="hljs-number">1.0</span>, do_normalize=<span class="hljs-literal">True</span>)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-v0jq8i">创建一个分词器来处理文本输入：</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="language-py "><!-- HTML_TAG_START --><span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> Wav2Vec2CTCTokenizer

	<span class="hljs-meta">>>> </span>tokenizer = Wav2Vec2CTCTokenizer(vocab_file=<span class="hljs-string">"my_vocab_file.txt"</span>)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1pme727">将特征提取器和分词器合并到 <code>Wav2Vec2Processor</code> 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="language-py "><!-- HTML_TAG_START --><span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> Wav2Vec2Processor

	<span class="hljs-meta">>>> </span>processor = Wav2Vec2Processor(feature_extractor=feature_extractor, tokenizer=tokenizer)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-11qtpiq">通过两个基类 - 配置类和模型类 - 以及一个附加的预处理类（分词器、图像处理器、特征提取器或处理器），你可以创建 🤗 Transformers 支持的任何模型。每个基类都是可配置的，允许你使用所需的特定属性。你可以轻松设置模型进行训练或修改现有的预训练模型进行微调。</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/create_a_model.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 data-svelte-h="svelte-zjs2n5"><span class="underline">Update</span> on GitHub</span></a> <p></p>

	<script>
	{
	__sveltekit_zpy4e4 = {
	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.7a5389c5.js"),
	import("/docs/transformers/main/zh/_app/immutable/entry/app.ac7d4b06.js")
	]).then(([kit, app]) => {
	kit.start(app, element, {
	node_ids: [0, 11],
	data,
	form: null,
	error: null
	});
	});
	}
	</script>

Xet Storage Details

Size:: 64.3 kB
Xet hash:: 0344fb55af91e0589e0f2587d17416997b20b01eac9cc52b7a4dad50fa329285

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