Buckets:
| <meta charset="utf-8" /><meta name="hf:doc:metadata" content="{"title":"Assisted decoding","local":"assisted-decoding","sections":[{"title":"Speculative decoding","local":"speculative-decoding","sections":[],"depth":2},{"title":"Prompt lookup decoding","local":"prompt-lookup-decoding","sections":[],"depth":2},{"title":"Self-speculative decoding","local":"self-speculative-decoding","sections":[],"depth":2},{"title":"Universal assisted decoding","local":"universal-assisted-decoding","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/6.f41493df.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":"Assisted decoding","local":"assisted-decoding","sections":[{"title":"Speculative decoding","local":"speculative-decoding","sections":[],"depth":2},{"title":"Prompt lookup decoding","local":"prompt-lookup-decoding","sections":[],"depth":2},{"title":"Self-speculative decoding","local":"self-speculative-decoding","sections":[],"depth":2},{"title":"Universal assisted decoding","local":"universal-assisted-decoding","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="assisted-decoding" 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="#assisted-decoding"><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>Assisted decoding</span></h1> <p data-svelte-h="svelte-ek9f3n">Assisted decoding speeds up text generation by allowing a helper propose candidate tokens before the main model commits to them. The main model verifies the candidate tokens in one forward pass. The helper is fast and cheap and can replace dozens of more expensive forward passes by the main model.</p> <p data-svelte-h="svelte-xpnht6">This guide covers assisted decoding methods in Transformers.</p> <h2 class="relative group"><a id="speculative-decoding" 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="#speculative-decoding"><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>Speculative decoding</span></h2> <p data-svelte-h="svelte-j2ora4"><a href="https://hf.co/papers/2211.17192" rel="nofollow">Speculative decoding</a> uses a smaller assistant model to draft candidate tokens. The main model checks these tokens in one pass. Validated tokens enter the final output and rejected tokens trigger standard sampling. Generation is faster because the main model runs fewer expensive forward passes.</p> <p data-svelte-h="svelte-18e7qvf">The method works best when the assistant model is significantly smaller than the main model and uses the same tokenizer. Speculative decoding supports greedy search and sampling but not batched inputs.</p> <p data-svelte-h="svelte-1on40z8">Pass <code>assistant_model</code> to <a href="/docs/transformers/pr_26617/en/main_classes/text_generation#transformers.GenerationMixin.generate">generate()</a>. Set <code>do_sample=True</code> to resample if token validation fails.</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">greedy search </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">sampling </div></div> <div class="language-select"><div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START --><span class="hljs-keyword">import</span> torch | |
| <span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoModelForCausalLM, AutoTokenizer | |
| tokenizer = AutoTokenizer.from_pretrained(<span class="hljs-string">"HuggingFaceTB/SmolLM-1.7B"</span>) | |
| model = AutoModelForCausalLM.from_pretrained(<span class="hljs-string">"HuggingFaceTB/SmolLM-1.7B"</span>, dtype=<span class="hljs-string">"auto"</span>) | |
| assistant_model = AutoModelForCausalLM.from_pretrained(<span class="hljs-string">"HuggingFaceTB/SmolLM-135M"</span>, dtype=<span class="hljs-string">"auto"</span>) | |
| inputs = tokenizer(<span class="hljs-string">"Hugging Face is an open-source company"</span>, return_tensors=<span class="hljs-string">"pt"</span>) | |
| outputs = model.generate(**inputs, assistant_model=assistant_model) | |
| tokenizer.batch_decode(outputs, skip_special_tokens=<span class="hljs-literal">True</span>) | |
| <span class="hljs-string">'Hugging Face is an open-source company that provides a platform for developers to build and deploy machine'</span><!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-q4tn19">The <code>assistant_model</code> argument is also available in the <a href="/docs/transformers/pr_26617/en/main_classes/pipelines#transformers.Pipeline">Pipeline</a> API.</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START --><span class="hljs-keyword">import</span> torch | |
| <span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> pipeline | |
| pipeline = pipeline( | |
| <span class="hljs-string">"text-generation"</span>, | |
| model=<span class="hljs-string">"meta-llama/Llama-3.1-8B"</span>, | |
| assistant_model=<span class="hljs-string">"meta-llama/Llama-3.2-1B"</span>, | |
| dtype=<span class="hljs-string">"auto"</span> | |
| ) | |
| pipeline(<span class="hljs-string">"Hugging Face is an open-source company, "</span>, max_new_tokens=<span class="hljs-number">50</span>, do_sample=<span class="hljs-literal">False</span>)<!-- HTML_TAG_END --></pre></div> </div> <h2 class="relative group"><a id="prompt-lookup-decoding" 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="#prompt-lookup-decoding"><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>Prompt lookup decoding</span></h2> <p data-svelte-h="svelte-1n32ox5">Prompt lookup decoding doesn’t need an assistant model. It finds overlapping n-grams in the prompt to propose candidate tokens. If no match exists, it falls back to normal autoregressive decoding. This suits input-grounded tasks like summarization and translation because candidate tokens often mirror local patterns in the source text.</p> <p data-svelte-h="svelte-uwxgck">Pass <code>prompt_lookup_num_tokens</code> to <a href="/docs/transformers/pr_26617/en/main_classes/text_generation#transformers.GenerationMixin.generate">generate()</a>. This sets how many tokens the algorithm tries to copy from earlier in the prompt when it detects a repeated pattern.</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">greedy decoding </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">sampling </div></div> <div class="language-select"><div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START --><span class="hljs-keyword">import</span> torch | |
| <span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoModelForCausalLM, AutoTokenizer | |
| tokenizer = AutoTokenizer.from_pretrained(<span class="hljs-string">"HuggingFaceTB/SmolLM-1.7B"</span>) | |
| model = AutoModelForCausalLM.from_pretrained(<span class="hljs-string">"HuggingFaceTB/SmolLM-1.7B"</span>, dtype=<span class="hljs-string">"auto"</span>) | |
| inputs = tokenizer(<span class="hljs-string">"Hugging Face is an open-source company"</span>, return_tensors=<span class="hljs-string">"pt"</span>) | |
| outputs = model.generate(**inputs, prompt_lookup_num_tokens=<span class="hljs-number">5</span>) | |
| tokenizer.batch_decode(outputs, skip_special_tokens=<span class="hljs-literal">True</span>) | |
| <span class="hljs-string">'Hugging Face is an open-source company that provides a platform for developers to build and deploy machine learning models. It offers a variety of tools'</span><!-- HTML_TAG_END --></pre></div> </div> <h2 class="relative group"><a id="self-speculative-decoding" 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="#self-speculative-decoding"><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>Self-speculative decoding</span></h2> <p data-svelte-h="svelte-1mbgfut">Self-speculative decoding uses a model’s intermediate layers as the assistant to propose candidate tokens. If the proposal matches, the model exits early and the remaining layers verify or correct the tokens.</p> <p data-svelte-h="svelte-11qnsbv">Because it’s all one model, weights and caches are shared, which boosts speed without extra memory overhead. This technique only works for models trained to support early-exit logits from intermediate layers.</p> <p data-svelte-h="svelte-1uzz97i">Pass <code>assistant_early_exit</code> to <a href="/docs/transformers/pr_26617/en/main_classes/text_generation#transformers.GenerationMixin.generate">generate()</a> to set the exit layer.</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START --><span class="hljs-keyword">import</span> torch | |
| <span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoModelForCausalLM, AutoTokenizer | |
| tokenizer = AutoTokenizer.from_pretrained(<span class="hljs-string">"facebook/layerskip-llama3.2-1B"</span>) | |
| model = AutoModelForCausalLM.from_pretrained(<span class="hljs-string">"facebook/layerskip-llama3.2-1B"</span>, dtype=<span class="hljs-string">"auto"</span>) | |
| inputs = tokenizer(<span class="hljs-string">"Hugging Face is an open-source company"</span>, return_tensors=<span class="hljs-string">"pt"</span>) | |
| outputs = model.generate(**inputs, assistant_early_exit=<span class="hljs-number">4</span>, do_sample=<span class="hljs-literal">False</span>, max_new_tokens=<span class="hljs-number">20</span>) | |
| tokenizer.batch_decode(outputs, skip_special_tokens=<span class="hljs-literal">True</span>)<!-- HTML_TAG_END --></pre></div> <h2 class="relative group"><a id="universal-assisted-decoding" 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="#universal-assisted-decoding"><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>Universal assisted decoding</span></h2> <p data-svelte-h="svelte-gy00t3">Universal assisted decoding (UAD) makes speculative decoding possible even when the main and assistant models have different tokenizers. It lets you pair any small assistant model with the main model. Candidate tokens are re-encoded and the algorithm computes the longest common subsequence so the continuation stays aligned.</p> <p data-svelte-h="svelte-m4gjtu">Pass <code>tokenizer</code>, <code>assistant_tokenizer</code>, and <code>assistant_model</code> to <a href="/docs/transformers/pr_26617/en/main_classes/text_generation#transformers.GenerationMixin.generate">generate()</a> to enable UAD.</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START --><span class="hljs-keyword">import</span> torch | |
| <span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoModelForCausalLM, AutoTokenizer | |
| assistant_tokenizer = AutoTokenizer.from_pretrained(<span class="hljs-string">"double7/vicuna-68m"</span>) | |
| tokenizer = AutoTokenizer.from_pretrained(<span class="hljs-string">"google/gemma-2-9b"</span>) | |
| model = AutoModelForCausalLM.from_pretrained(<span class="hljs-string">"google/gemma-2-9b"</span>, dtype=<span class="hljs-string">"auto"</span>) | |
| assistant_model = AutoModelForCausalLM.from_pretrained(<span class="hljs-string">"double7/vicuna-68m"</span>, dtype=<span class="hljs-string">"auto"</span>) | |
| inputs = tokenizer(<span class="hljs-string">"Hugging Face is an open-source company"</span>, return_tensors=<span class="hljs-string">"pt"</span>) | |
| outputs = model.generate(**inputs, assistant_model=assistant_model, tokenizer=tokenizer, assistant_tokenizer=assistant_tokenizer) | |
| tokenizer.batch_decode(outputs, skip_special_tokens=<span class="hljs-literal">True</span>) | |
| <span class="hljs-string">'Hugging Face is an open-source company that is dedicated to creating a better world through technology.'</span><!-- 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-1up2o7y"><li>Read the <a href="https://huggingface.co/blog/assisted-generation" rel="nofollow">Assisted Generation: a new direction toward low-latency text generation</a> blog post for more context about text generation latency and assisted generation.</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/assisted_decoding.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, 6], | |
| data, | |
| form: null, | |
| error: null | |
| }); | |
| }); | |
| } | |
| </script> | |
Xet Storage Details
- Size:
- 29.6 kB
- Xet hash:
- cbad63f61c5f20aaf73a031852ba214bb9b6e20e633b2c760f6ae4e02d758e90
·
Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.