Buckets:
| <meta charset="utf-8" /><meta name="hf:doc:metadata" content="{"title":"Tokenizers","local":"tokenizers","sections":[{"title":"Load a tokenizer","local":"load-a-tokenizer","sections":[],"depth":2},{"title":"Encode and decode","local":"encode-and-decode","sections":[],"depth":2},{"title":"Special tokens","local":"special-tokens","sections":[],"depth":2},{"title":"Batch processing","local":"batch-processing","sections":[{"title":"Padding","local":"padding","sections":[],"depth":3},{"title":"Truncation","local":"truncation","sections":[],"depth":3}],"depth":2},{"title":"Backends","local":"backends","sections":[],"depth":2},{"title":"Inspect the tokenizer architecture","local":"inspect-the-tokenizer-architecture","sections":[],"depth":2},{"title":"Resources","local":"resources","sections":[],"depth":2}],"depth":1}"> | |
| <link href="/docs/transformers/pr_26617/en/_app/immutable/assets/0.e3b0c442.css" rel="modulepreload"> | |
| <link rel="modulepreload" href="/docs/transformers/pr_26617/en/_app/immutable/entry/start.b5ae2c21.js"> | |
| <link rel="modulepreload" href="/docs/transformers/pr_26617/en/_app/immutable/chunks/scheduler.31fdf58d.js"> | |
| <link rel="modulepreload" href="/docs/transformers/pr_26617/en/_app/immutable/chunks/singletons.512cdb48.js"> | |
| <link rel="modulepreload" href="/docs/transformers/pr_26617/en/_app/immutable/chunks/index.252883d5.js"> | |
| <link rel="modulepreload" href="/docs/transformers/pr_26617/en/_app/immutable/chunks/paths.81255c3b.js"> | |
| <link rel="modulepreload" href="/docs/transformers/pr_26617/en/_app/immutable/entry/app.9acf2c3e.js"> | |
| <link rel="modulepreload" href="/docs/transformers/pr_26617/en/_app/immutable/chunks/preload-helper.bb442aeb.js"> | |
| <link rel="modulepreload" href="/docs/transformers/pr_26617/en/_app/immutable/chunks/index.2f76fdf0.js"> | |
| <link rel="modulepreload" href="/docs/transformers/pr_26617/en/_app/immutable/nodes/0.da6b3909.js"> | |
| <link rel="modulepreload" href="/docs/transformers/pr_26617/en/_app/immutable/chunks/each.e59479a4.js"> | |
| <link rel="modulepreload" href="/docs/transformers/pr_26617/en/_app/immutable/nodes/40.8a5f97d6.js"> | |
| <link rel="modulepreload" href="/docs/transformers/pr_26617/en/_app/immutable/chunks/CopyLLMTxtMenu.a69e059a.js"> | |
| <link rel="modulepreload" href="/docs/transformers/pr_26617/en/_app/immutable/chunks/MermaidChart.svelte_svelte_type_style_lang.e4c7f916.js"> | |
| <link rel="modulepreload" href="/docs/transformers/pr_26617/en/_app/immutable/chunks/IconCopy.ac192424.js"> | |
| <link rel="modulepreload" href="/docs/transformers/pr_26617/en/_app/immutable/chunks/CodeBlock.ab12f8e1.js"> | |
| <link rel="modulepreload" href="/docs/transformers/pr_26617/en/_app/immutable/chunks/HfOption.fb051768.js"><!-- HEAD_svelte-u9bgzb_START --><meta name="hf:doc:metadata" content="{"title":"Tokenizers","local":"tokenizers","sections":[{"title":"Load a tokenizer","local":"load-a-tokenizer","sections":[],"depth":2},{"title":"Encode and decode","local":"encode-and-decode","sections":[],"depth":2},{"title":"Special tokens","local":"special-tokens","sections":[],"depth":2},{"title":"Batch processing","local":"batch-processing","sections":[{"title":"Padding","local":"padding","sections":[],"depth":3},{"title":"Truncation","local":"truncation","sections":[],"depth":3}],"depth":2},{"title":"Backends","local":"backends","sections":[],"depth":2},{"title":"Inspect the tokenizer architecture","local":"inspect-the-tokenizer-architecture","sections":[],"depth":2},{"title":"Resources","local":"resources","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="tokenizers" 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="#tokenizers"><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>Tokenizers</span></h1> <p data-svelte-h="svelte-ezsjwy">A tokenizer converts text into tensors, which are the inputs to a model. It normalizes and splits text, applies the tokenization algorithm, adds special tokens, and decodes output ids back into text.</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> AutoTokenizer | |
| tokenizer = AutoTokenizer.from_pretrained(<span class="hljs-string">"google/gemma-2-2b"</span>) | |
| tokenizer(<span class="hljs-string">"Sphinx of black quartz, judge my vow."</span>, return_tensors=<span class="hljs-string">"pt"</span>) | |
| { | |
| <span class="hljs-string">'input_ids'</span>: tensor([[ <span class="hljs-number">2</span>, <span class="hljs-number">235277</span>, <span class="hljs-number">82913</span>, <span class="hljs-number">576</span>, <span class="hljs-number">2656</span>, <span class="hljs-number">30407</span>, <span class="hljs-number">235269</span>, <span class="hljs-number">11490</span>, <span class="hljs-number">970</span>, <span class="hljs-number">29871</span>, <span class="hljs-number">235265</span>]]), | |
| <span class="hljs-string">'attention_mask'</span>: tensor([[<span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>]]) | |
| }<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-m21ffs">This guide covers loading, encoding, decoding, batch processing, and the available tokenizer backends.</p> <h2 class="relative group"><a id="load-a-tokenizer" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#load-a-tokenizer"><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>Load a tokenizer</span></h2> <p data-svelte-h="svelte-ke93oc">Load a tokenizer with the <a href="/docs/transformers/pr_26617/en/model_doc/auto#transformers.AutoTokenizer">AutoTokenizer</a> class or a model-specific tokenizer class.</p> <div class="flex space-x-2 items-center my-1.5 mr-8 h-7 !pl-0 -mx-3 md:mx-0"><div class="flex items-center border rounded-lg px-1.5 py-1 leading-none select-none text-smd border-gray-800 bg-black dark:bg-gray-700 text-white">AutoTokenizer </div><div class="flex items-center border rounded-lg px-1.5 py-1 leading-none select-none text-smd text-gray-500 cursor-pointer opacity-90 hover:text-gray-700 dark:hover:text-gray-200 hover:shadow-sm">model-specific tokenizer </div></div> <div class="language-select"><p data-svelte-h="svelte-19q568c"><a href="/docs/transformers/pr_26617/en/model_doc/auto#transformers.AutoTokenizer.from_pretrained">AutoTokenizer.from_pretrained()</a> reads the model config, resolves the correct tokenizer class, and returns an instance of it. You don’t need to know the tokenizer class beforehand. Most tokenizers resolve to a subclass of <a href="/docs/transformers/pr_26617/en/main_classes/tokenizer#transformers.TokenizersBackend">TokenizersBackend</a>, a fast Rust-based tokenizer from the <a href="https://huggingface.co/docs/tokenizers/index" rel="nofollow">Tokenizers</a> library.</p> <p data-svelte-h="svelte-arjqni">Loading with <a href="/docs/transformers/pr_26617/en/model_doc/auto#transformers.AutoTokenizer">AutoTokenizer</a> is the recommended approach.</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> AutoTokenizer | |
| tokenizer = AutoTokenizer.from_pretrained(<span class="hljs-string">"google/gemma-2-2b"</span>)<!-- HTML_TAG_END --></pre></div> </div> <h2 class="relative group"><a id="encode-and-decode" 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="#encode-and-decode"><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>Encode and decode</span></h2> <p data-svelte-h="svelte-ek9spo">The <a href="/docs/transformers/pr_26617/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.__call__">TokenizersBackend.<strong>call</strong>()</a> method encodes text or a batch of text into <code>input_ids</code>, <code>attention_mask</code>, and other model inputs. It also controls padding, truncation, and special token insertion.</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> AutoTokenizer | |
| tokenizer = AutoTokenizer.from_pretrained(<span class="hljs-string">"google/gemma-2-2b"</span>) | |
| tokenizer(<span class="hljs-string">"Sphinx of black quartz, judge my vow."</span>, return_tensors=<span class="hljs-string">"pt"</span>) | |
| { | |
| <span class="hljs-string">'input_ids'</span>: tensor([[ <span class="hljs-number">2</span>, <span class="hljs-number">235277</span>, <span class="hljs-number">82913</span>, <span class="hljs-number">576</span>, <span class="hljs-number">2656</span>, <span class="hljs-number">30407</span>, <span class="hljs-number">235269</span>, <span class="hljs-number">11490</span>, <span class="hljs-number">970</span>, <span class="hljs-number">29871</span>, <span class="hljs-number">235265</span>]]), | |
| <span class="hljs-string">'attention_mask'</span>: tensor([[<span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>]]) | |
| }<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1fqbhb2"><a href="/docs/transformers/pr_26617/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.encode">TokenizersBackend.encode()</a> is similar but only returns the <code>input_ids</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=""><!-- HTML_TAG_START --><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoTokenizer | |
| tokenizer = AutoTokenizer.from_pretrained(<span class="hljs-string">"google/gemma-2-2b"</span>) | |
| tokenizer.encode(<span class="hljs-string">"Sphinx of black quartz, judge my vow."</span>) | |
| [<span class="hljs-number">2</span>, <span class="hljs-number">235277</span>, <span class="hljs-number">82913</span>, <span class="hljs-number">576</span>, <span class="hljs-number">2656</span>, <span class="hljs-number">30407</span>, <span class="hljs-number">235269</span>, <span class="hljs-number">11490</span>, <span class="hljs-number">970</span>, <span class="hljs-number">29871</span>, <span class="hljs-number">235265</span>]<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1bsoml2"><a href="/docs/transformers/pr_26617/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.decode">TokenizersBackend.decode()</a> converts a single sequence or batch of tokenized <code>input_ids</code> back to text.</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 -->tokenizer.decode(outputs[<span class="hljs-string">"input_ids"</span>]) | |
| [<span class="hljs-string">'<bos>Sphinx of black quartz, judge my vow.'</span>]<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1atfrau"><a href="/docs/transformers/pr_26617/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.decode">TokenizersBackend.decode()</a> preserves the exact tokenization spacing. Set <code>clean_up_tokenization_spaces</code> to remove spaces before punctuation, and <code>skip_special_tokens</code> to strip special tokens from the output.</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 -->tokenizer.decode(outputs[<span class="hljs-string">"input_ids"</span>], skip_special_tokens=<span class="hljs-literal">True</span>) | |
| [<span class="hljs-string">'Sphinx of black quartz, judge my vow.'</span>]<!-- HTML_TAG_END --></pre></div> <h2 class="relative group"><a id="special-tokens" 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="#special-tokens"><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>Special tokens</span></h2> <p data-svelte-h="svelte-owpjlb">Special tokens mark structural boundaries in a sequence, like the beginning-of-sequence or padding positions. Each model defines its own set of special tokens. The tokenizer adds them when you call it.</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 -->tokenizer.encode(<span class="hljs-string">"Sphinx of black quartz, judge my vow."</span>) | |
| [<span class="hljs-number">2</span>, <span class="hljs-number">235277</span>, <span class="hljs-number">82913</span>, <span class="hljs-number">576</span>, <span class="hljs-number">2656</span>, <span class="hljs-number">30407</span>, <span class="hljs-number">235269</span>, <span class="hljs-number">11490</span>, <span class="hljs-number">970</span>, <span class="hljs-number">29871</span>, <span class="hljs-number">235265</span>] | |
| tokenizer.decode(outputs[<span class="hljs-string">"input_ids"</span>]) | |
| [<span class="hljs-string">'<bos>Sphinx of black quartz, judge my vow.'</span>]<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1l2gsfq">Register additional named special tokens with the <code>extra_special_tokens</code> argument. Multimodal models use them as placeholders for images, video, or audio.</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> AutoTokenizer | |
| tokenizer = AutoTokenizer.from_pretrained( | |
| <span class="hljs-string">"google/gemma-3-4b-pt"</span>, | |
| extra_special_tokens={<span class="hljs-string">"image_token"</span>: <span class="hljs-string">"<image>"</span>} | |
| )<!-- HTML_TAG_END --></pre></div> <h2 class="relative group"><a id="batch-processing" 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="#batch-processing"><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>Batch processing</span></h2> <p data-svelte-h="svelte-1cshkxz">Batch processing tokenizes multiple sequences in a single call. <a href="/docs/transformers/pr_26617/en/main_classes/tokenizer#transformers.TokenizersBackend">TokenizersBackend</a> handles large batches faster because its Rust-based backend parallelizes work across threads.</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> AutoTokenizer | |
| tokenizer = AutoTokenizer.from_pretrained(<span class="hljs-string">"google/gemma-2-2b"</span>) | |
| tokenizer( | |
| [ | |
| <span class="hljs-string">"Sphinx of black quartz, judge my vow."</span>, | |
| <span class="hljs-string">"Pack my box with five dozen liquor jugs."</span>, | |
| <span class="hljs-string">"How vexingly quick daft zebras jump!"</span> | |
| ], | |
| return_tensors=<span class="hljs-string">"pt"</span> | |
| )<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-15vs1bl">Batch processing requires all sequences to share the same length. Padding and truncation are strategies to handle varying-length sequences.</p> <h3 class="relative group"><a id="padding" 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="#padding"><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>Padding</span></h3> <p data-svelte-h="svelte-1mmj2ia">Padding appends special tokens so shorter sequences match the longest sequence in a batch. The attention mask marks padding positions as <code>0</code> so the model ignores them. Set <code>padding=True</code> to pad to the longest sequence or pass <code>max_length</code> to pad to a fixed size.</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> AutoTokenizer | |
| tokenizer = AutoTokenizer.from_pretrained(<span class="hljs-string">"google/gemma-2-2b"</span>) | |
| tokenizer( | |
| [ | |
| <span class="hljs-string">"Sphinx of black quartz, judge my vow."</span>, | |
| <span class="hljs-string">"Pack my box with five dozen liquor jugs."</span>, | |
| <span class="hljs-string">"How vexingly quick daft zebras jump!"</span> | |
| ], | |
| return_tensors=<span class="hljs-string">"pt"</span>, | |
| padding=<span class="hljs-literal">True</span>, | |
| ) | |
| { | |
| <span class="hljs-string">'input_ids'</span>: tensor([ | |
| [ <span class="hljs-number">2</span>, <span class="hljs-number">235277</span>, <span class="hljs-number">82913</span>, <span class="hljs-number">576</span>, <span class="hljs-number">2656</span>, <span class="hljs-number">30407</span>, <span class="hljs-number">235269</span>, <span class="hljs-number">11490</span>, <span class="hljs-number">970</span>, <span class="hljs-number">29871</span>, <span class="hljs-number">235265</span>], | |
| [ <span class="hljs-number">0</span>, <span class="hljs-number">2</span>, <span class="hljs-number">6519</span>, <span class="hljs-number">970</span>, <span class="hljs-number">3741</span>, <span class="hljs-number">675</span>, <span class="hljs-number">4105</span>, <span class="hljs-number">25955</span>, <span class="hljs-number">42184</span>, <span class="hljs-number">225789</span>, <span class="hljs-number">235265</span>], | |
| [ <span class="hljs-number">0</span>, <span class="hljs-number">2</span>, <span class="hljs-number">2299</span>, <span class="hljs-number">73378</span>, <span class="hljs-number">17844</span>, <span class="hljs-number">4320</span>, <span class="hljs-number">224463</span>, <span class="hljs-number">4949</span>, <span class="hljs-number">48977</span>, <span class="hljs-number">9902</span>, <span class="hljs-number">235341</span>] | |
| ]), | |
| <span class="hljs-string">'attention_mask'</span>: tensor([ | |
| [<span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>], | |
| [<span class="hljs-number">0</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>], | |
| [<span class="hljs-number">0</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>] | |
| ]) | |
| }<!-- HTML_TAG_END --></pre></div> <blockquote class="note" data-svelte-h="svelte-1dbjo9j"><p>Large language models pad on the <em>left</em> side to avoid disrupting generation, which predicts the next token from the <em>right</em> side.</p></blockquote> <h3 class="relative group"><a id="truncation" 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="#truncation"><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>Truncation</span></h3> <p data-svelte-h="svelte-1ft2uza">Truncation clips tokens so a sequence fits within a maximum length. Set <code>truncation=True</code> and specify <code>max_length</code> to enable it.</p> <p data-svelte-h="svelte-1nuqojs">Padding and truncation work together. Short sequences gain padding tokens while long sequences lose trailing tokens. Together, they produce a packed rectangular tensor.</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> AutoTokenizer | |
| tokenizer = AutoTokenizer.from_pretrained(<span class="hljs-string">"google/gemma-2-2b"</span>) | |
| tokenizer( | |
| [ | |
| <span class="hljs-string">"Sphinx of black quartz, judge my vow."</span>, | |
| <span class="hljs-string">"Pack my box with five dozen liquor jugs."</span>, | |
| <span class="hljs-string">"How vexingly quick daft zebras jump!"</span> | |
| ], | |
| return_tensors=<span class="hljs-string">"pt"</span>, | |
| padding=<span class="hljs-literal">True</span>, | |
| truncation=<span class="hljs-literal">True</span>, | |
| max_length=<span class="hljs-number">5</span> | |
| ) | |
| { | |
| <span class="hljs-string">'input_ids'</span>: tensor([ | |
| [ <span class="hljs-number">2</span>, <span class="hljs-number">235277</span>, <span class="hljs-number">82913</span>, <span class="hljs-number">576</span>, <span class="hljs-number">2656</span>], | |
| [ <span class="hljs-number">2</span>, <span class="hljs-number">6519</span>, <span class="hljs-number">970</span>, <span class="hljs-number">3741</span>, <span class="hljs-number">675</span>], | |
| [ <span class="hljs-number">2</span>, <span class="hljs-number">2299</span>, <span class="hljs-number">73378</span>, <span class="hljs-number">17844</span>, <span class="hljs-number">4320</span>] | |
| ]), | |
| <span class="hljs-string">'attention_mask'</span>: tensor([ | |
| [<span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>], | |
| [<span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>], | |
| [<span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>] | |
| ]) | |
| }<!-- HTML_TAG_END --></pre></div> <h2 class="relative group"><a id="backends" 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="#backends"><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>Backends</span></h2> <p data-svelte-h="svelte-1km33xh">Each model tokenizer is defined in a single file and supports four tokenization backends.</p> <table data-svelte-h="svelte-q7zkis"><thead><tr><th>backend</th> <th>implementation</th> <th>description</th></tr></thead> <tbody><tr><td><a href="/docs/transformers/pr_26617/en/main_classes/tokenizer#transformers.TokenizersBackend">TokenizersBackend</a></td> <td><a href="https://huggingface.co/docs/tokenizers" rel="nofollow">Tokenizers</a></td> <td>default for most models</td></tr> <tr><td><a href="/docs/transformers/pr_26617/en/main_classes/tokenizer#transformers.SentencePieceBackend">SentencePieceBackend</a></td> <td><a href="https://github.com/google/sentencepiece" rel="nofollow">SentencePiece</a></td> <td>models requiring SentencePiece</td></tr> <tr><td><a href="/docs/transformers/pr_26617/en/main_classes/tokenizer#transformers.PythonBackend">PythonBackend</a></td> <td>Python</td> <td>models requiring specialized custom tokenizers</td></tr> <tr><td><a href="/docs/transformers/pr_26617/en/model_doc/mixtral#transformers.MistralCommonBackend">MistralCommonBackend</a></td> <td><a href="https://mistralai.github.io/mistral-common/" rel="nofollow">mistral-common</a></td> <td>Mistral and Pixtral models</td></tr></tbody></table> <p data-svelte-h="svelte-6u5vzs">All backends inherit from <a href="/docs/transformers/pr_26617/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase">PreTrainedTokenizerBase</a> and share the same APIs for encoding, decoding, padding, truncation, saving, and loading. The difference is which tokenization pipeline runs underneath.</p> <p data-svelte-h="svelte-1snn7h7"><a href="/docs/transformers/pr_26617/en/model_doc/auto#transformers.AutoTokenizer">AutoTokenizer</a> selects the best available backend when you call <a href="/docs/transformers/pr_26617/en/model_doc/auto#transformers.AutoTokenizer.from_pretrained">from_pretrained()</a>.</p> <ol><li data-svelte-h="svelte-199ri7u"><p>It reads the <code>tokenizer_config.json</code> file for the <code>tokenizer_class</code> field.</p></li> <li><p data-svelte-h="svelte-fz0heg">The registry matches <code>tokenizer_class</code> to a class name. The resolved class inherits from one of the four backends. For example, <a href="/docs/transformers/pr_26617/en/model_doc/gemma#transformers.GemmaTokenizer">GemmaTokenizer</a> inherits from <a href="/docs/transformers/pr_26617/en/main_classes/tokenizer#transformers.TokenizersBackend">TokenizersBackend</a>, and <a href="/docs/transformers/pr_26617/en/model_doc/siglip#transformers.SiglipTokenizer">SiglipTokenizer</a> inherits from <a href="/docs/transformers/pr_26617/en/main_classes/tokenizer#transformers.SentencePieceBackend">SentencePieceBackend</a>.</p> <p data-svelte-h="svelte-1xkl0dk">Some models, like GLM, map directly to <a href="/docs/transformers/pr_26617/en/main_classes/tokenizer#transformers.TokenizersBackend">TokenizersBackend</a> because the <code>tokenizer.json</code> file fully describes the pipeline. <a href="/docs/transformers/pr_26617/en/model_doc/gemma#transformers.GemmaTokenizer">GemmaTokenizer</a> exists as a subclass since it defines additional model-specific settings in Python that <code>tokenizer.json</code> doesn’t capture.</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 -->TOKENIZER_MAPPING_NAMES = OrderedDict([ | |
| (<span class="hljs-string">"gemma2"</span>, <span class="hljs-string">"GemmaTokenizer"</span> <span class="hljs-keyword">if</span> is_tokenizers_available() <span class="hljs-keyword">else</span> <span class="hljs-literal">None</span>), | |
| (<span class="hljs-string">"glm"</span>, <span class="hljs-string">"TokenizersBackend"</span> <span class="hljs-keyword">if</span> is_tokenizers_available() <span class="hljs-keyword">else</span> <span class="hljs-literal">None</span>), | |
| ( | |
| <span class="hljs-string">"mistral"</span>, | |
| <span class="hljs-string">"MistralCommonBackend"</span> | |
| <span class="hljs-keyword">if</span> is_mistral_common_available() | |
| <span class="hljs-keyword">else</span> (<span class="hljs-string">"TokenizersBackend"</span> <span class="hljs-keyword">if</span> is_tokenizers_available() <span class="hljs-keyword">else</span> <span class="hljs-literal">None</span>), | |
| ), | |
| (<span class="hljs-string">"siglip"</span>, <span class="hljs-string">"SiglipTokenizer"</span> <span class="hljs-keyword">if</span> is_sentencepiece_available() <span class="hljs-keyword">else</span> <span class="hljs-literal">None</span>), | |
| ... | |
| ]<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1bd1sx1">When a backend like mistral-common isn’t installed, <a href="/docs/transformers/pr_26617/en/model_doc/auto#transformers.AutoTokenizer">AutoTokenizer</a> falls back to <a href="/docs/transformers/pr_26617/en/main_classes/tokenizer#transformers.TokenizersBackend">TokenizersBackend</a>.</p></li></ol> <p data-svelte-h="svelte-1fakyua">Check which backend a tokenizer is using with the <code>backend</code> property.</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> AutoTokenizer | |
| tokenizer = AutoTokenizer.from_pretrained(<span class="hljs-string">"google/gemma-2-2b"</span>) | |
| tokenizer.backend | |
| <span class="hljs-string">'tokenizers'</span><!-- HTML_TAG_END --></pre></div> <h2 class="relative group"><a id="inspect-the-tokenizer-architecture" 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="#inspect-the-tokenizer-architecture"><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>Inspect the tokenizer architecture</span></h2> <p data-svelte-h="svelte-1n8qt5">Inspect a tokenizer’s internal components (normalizer, pre-tokenizer, model, decoder) with the <code>_tokenizer</code> attribute.</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> AutoTokenizer | |
| tokenizer = AutoTokenizer.from_pretrained(<span class="hljs-string">"google/gemma-2-2b"</span>) | |
| <span class="hljs-built_in">print</span>(tokenizer._tokenizer.normalizer) | |
| <span class="hljs-built_in">print</span>(tokenizer._tokenizer.pre_tokenizer) | |
| <span class="hljs-built_in">print</span>(tokenizer._tokenizer.model) | |
| <span class="hljs-built_in">print</span>(tokenizer._tokenizer.decoder)<!-- HTML_TAG_END --></pre></div> <h2 class="relative group"><a id="resources" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#resources"><span><svg 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>Resources</span></h2> <ul data-svelte-h="svelte-u99gy7"><li>The <a href="https://huggingface.co/blog/tokenizers" rel="nofollow">Tokenization in Transformers v5</a> post discusses the motivation behind the new tokenization backends.</li> <li>Review the <a href="https://github.com/huggingface/transformers/blob/main/MIGRATION_GUIDE_V5.md#tokenization" rel="nofollow">migration guide</a> for an overview of the tokenization changes.</li></ul> <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/en/fast_tokenizers.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_1x0t0ja = { | |
| assets: "/docs/transformers/pr_26617/en", | |
| base: "/docs/transformers/pr_26617/en", | |
| env: {} | |
| }; | |
| const element = document.currentScript.parentElement; | |
| const data = [null,null]; | |
| Promise.all([ | |
| import("/docs/transformers/pr_26617/en/_app/immutable/entry/start.b5ae2c21.js"), | |
| import("/docs/transformers/pr_26617/en/_app/immutable/entry/app.9acf2c3e.js") | |
| ]).then(([kit, app]) => { | |
| kit.start(app, element, { | |
| node_ids: [0, 40], | |
| data, | |
| form: null, | |
| error: null | |
| }); | |
| }); | |
| } | |
| </script> | |
Xet Storage Details
- Size:
- 58.6 kB
- Xet hash:
- d940923eb620187c39757378c70e42e338bcc376c546e14f45d43e5abd878785
·
Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.