Buckets:

hf-doc-build
/

doc

Files

xet

hf-doc-build/doc / transformers /main /en /experts_interface.html

HuggingFaceDocBuilder

about 9 hours ago

download

raw

26 kB

	<meta charset="utf-8" /><meta name="hf:doc:metadata" content="{"title":"Experts backends","local":"experts-backends","sections":[{"title":"Set an experts backend","local":"set-an-experts-backend","sections":[],"depth":2},{"title":"Backbone-specific experts backend","local":"backbone-specific-experts-backend","sections":[],"depth":2},{"title":"torch.compile","local":"torchcompile","sections":[],"depth":2},{"title":"Benchmarks","local":"benchmarks","sections":[],"depth":2}],"depth":1}">
	<link href="/docs/transformers/main/en/_app/immutable/assets/0.e3b0c442.css" rel="modulepreload">
	<link rel="modulepreload" href="/docs/transformers/main/en/_app/immutable/entry/start.3d6cca8a.js">
	<link rel="modulepreload" href="/docs/transformers/main/en/_app/immutable/chunks/scheduler.31fdf58d.js">
	<link rel="modulepreload" href="/docs/transformers/main/en/_app/immutable/chunks/singletons.6af0ff6e.js">
	<link rel="modulepreload" href="/docs/transformers/main/en/_app/immutable/chunks/index.252883d5.js">
	<link rel="modulepreload" href="/docs/transformers/main/en/_app/immutable/chunks/paths.299a376b.js">
	<link rel="modulepreload" href="/docs/transformers/main/en/_app/immutable/entry/app.b6ccab0d.js">
	<link rel="modulepreload" href="/docs/transformers/main/en/_app/immutable/chunks/preload-helper.c438fa0a.js">
	<link rel="modulepreload" href="/docs/transformers/main/en/_app/immutable/chunks/index.2f76fdf0.js">
	<link rel="modulepreload" href="/docs/transformers/main/en/_app/immutable/nodes/0.f2629ed0.js">
	<link rel="modulepreload" href="/docs/transformers/main/en/_app/immutable/chunks/each.e59479a4.js">
	<link rel="modulepreload" href="/docs/transformers/main/en/_app/immutable/nodes/46.04d0cfb8.js">
	<link rel="modulepreload" href="/docs/transformers/main/en/_app/immutable/chunks/CopyLLMTxtMenu.ad38f6ea.js">
	<link rel="modulepreload" href="/docs/transformers/main/en/_app/immutable/chunks/MermaidChart.svelte_svelte_type_style_lang.fd2f7a8a.js">
	<link rel="modulepreload" href="/docs/transformers/main/en/_app/immutable/chunks/IconCopy.ac192424.js">
	<link rel="modulepreload" href="/docs/transformers/main/en/_app/immutable/chunks/CodeBlock.e52df5d6.js"><!-- HEAD_svelte-u9bgzb_START --><meta name="hf:doc:metadata" content="{"title":"Experts backends","local":"experts-backends","sections":[{"title":"Set an experts backend","local":"set-an-experts-backend","sections":[],"depth":2},{"title":"Backbone-specific experts backend","local":"backbone-specific-experts-backend","sections":[],"depth":2},{"title":"torch.compile","local":"torchcompile","sections":[],"depth":2},{"title":"Benchmarks","local":"benchmarks","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="experts-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="#experts-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>Experts backends</span></h1> <p data-svelte-h="svelte-126nyrc">All Mixture-of-Experts (MoE) implementations perform the same high-level computation. For each token, a router selects <em>k</em> experts. The token hidden state is then projected through the selected experts’ parameters and aggregated with routing weights. The difference between experts backends is <em>how</em> those expert matrix multiplications execute.</p> <p data-svelte-h="svelte-12gq5xs">The <code>ExpertsInterface</code> provides optimized experts backends. It decouples the experts implementation from the model code to simplify experimentation with different functions. Add new backends through the same interface.</p> <table data-svelte-h="svelte-1xyh7br"><thead><tr><th>experts backend</th> <th>description</th> <th>GPU</th> <th>CPU</th></tr></thead> <tbody><tr><td><code>"eager"</code></td> <td>Reference implementation that loops over selected experts and applies projections on their tokens.</td> <td>Reasonable baseline performance without requiring compilation.</td> <td>Slower than <code>grouped_mm</code> but faster than <code>batched_mm</code>.</td></tr> <tr><td><code>"batched_mm"</code></td> <td>Duplicates selected expert parameters for each token and projects all tokens in a single batched GEMM using <a href="https://docs.pytorch.org/docs/stable/generated/torch.bmm.html" rel="nofollow"><code>torch.bmm</code></a>.</td> <td>Fastest for small inputs, especially with compilation. Uses more memory due to parameter duplication.</td> <td>Not recommended (significantly slower than other backends).</td></tr> <tr><td><code>"grouped_mm"</code></td> <td>Orders tokens by selected experts and uses <a href="https://docs.pytorch.org/docs/stable/generated/torch.nn.functional.grouped_mm.html" rel="nofollow"><code>torch.nn.functional.grouped_mm</code></a> to project all tokens in a single grouped GEMM (requires PyTorch 2.9+).</td> <td>Best for larger inputs and more memory efficient as it avoids duplicating expert parameters. Fast with compilation.</td> <td>Most efficient backend for all input sizes.</td></tr></tbody></table> <blockquote class="note" data-svelte-h="svelte-11kzsgw"><p>When using <code>experts_implementation="grouped_mm"</code> on GPU, the model automatically switches to <code>"batched_mm"</code> during the decode stage of generation (after prefill). This is because <code>batched_mm</code> is significantly faster on lower token count during autoregressive decoding on GPU. On CPU, <code>grouped_mm</code> remains active throughout generation as it is more efficient for all input sizes.</p></blockquote> <h2 class="relative group"><a id="set-an-experts-backend" 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="#set-an-experts-backend"><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>Set an experts backend</span></h2> <p data-svelte-h="svelte-1hb7nsg">Use the <code>experts_implementation</code> argument in <a href="/docs/transformers/main/en/main_classes/model#transformers.PreTrainedModel.from_pretrained">from_pretrained()</a> to instantiate a model with a specific experts backend.</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-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoModelForCausalLM

	model = AutoModelForCausalLM.from_pretrained(
	<span class="hljs-string">"Qwen/Qwen1.5-MoE-A2.7B"</span>,
	dtype=<span class="hljs-string">"bfloat16"</span>,
	experts_implementation=<span class="hljs-string">"batched_mm"</span>,
	)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1u9skn2">Switch between experts backends at runtime without reloading the model using <a href="/docs/transformers/main/en/main_classes/model#transformers.PreTrainedModel.set_experts_implementation">set_experts_implementation()</a>.</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg 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 -->model.set_experts_implementation(<span class="hljs-string">"eager"</span>)<!-- HTML_TAG_END --></pre></div> <h2 class="relative group"><a id="backbone-specific-experts-backend" 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="#backbone-specific-experts-backend"><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>Backbone-specific experts backend</span></h2> <p data-svelte-h="svelte-17bh5xl">Multimodal models can have multiple sub-configs (for example, different backbones). You can set a different experts backend per sub-config by passing a <code>dict</code> to <code>experts_implementation</code> at load time.</p> <p data-svelte-h="svelte-pdltj3">Keys in the mapping must match sub-config names.</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-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoModelForImageTextToText

	experts_implementation_per_backbone = {
	<span class="hljs-string">"text_config"</span>: <span class="hljs-string">"grouped_mm"</span>,
	<span class="hljs-string">"vision_config"</span>: <span class="hljs-string">"eager"</span>,
	}

	model = AutoModelForImageTextToText.from_pretrained(
	<span class="hljs-string">"Qwen/Qwen3-VL-Moe"</span>,
	experts_implementation=experts_implementation_per_backbone,
	)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-5rug0z">Set the experts backend globally with an empty key.</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 -->model = AutoModelForCausalLM.from_pretrained(
	<span class="hljs-string">"Qwen/Qwen1.5-MoE-A2.7B"</span>,
	experts_implementation={<span class="hljs-string">""</span>: <span class="hljs-string">"batched_mm"</span>},
	)<!-- HTML_TAG_END --></pre></div> <h2 class="relative group"><a id="torchcompile" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#torchcompile"><span><svg 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>torch.compile</span></h2> <p data-svelte-h="svelte-1n7nl87">All three backends (<code>"eager"</code>, <code>"batched_mm"</code>, <code>"grouped_mm"</code>) are compatible with <code>torch.compile</code> to certain extents. The following table summarizes compatibility:</p> <table data-svelte-h="svelte-18ojk73"><thead><tr><th>Implementation</th> <th>compilation modes</th> <th>dtypes</th> <th><code>fullgraph=True</code></th></tr></thead> <tbody><tr><td><code>grouped_mm</code></td> <td><code>None</code>, <code>max-autotune-no-cudagraphs</code></td> <td><code>bfloat16</code></td> <td>Yes</td></tr> <tr><td><code>grouped_mm</code> (fallback)</td> <td><code>None</code>, <code>max-autotune-no-cudagraphs</code></td> <td><code>bfloat16</code>, <code>float16</code>, <code>float32</code></td> <td>Yes</td></tr> <tr><td><code>batched_mm</code></td> <td>all</td> <td><code>bfloat16</code>, <code>float16</code>, <code>float32</code></td> <td>Yes</td></tr> <tr><td><code>eager</code></td> <td>all</td> <td><code>bfloat16</code>, <code>float16</code>, <code>float32</code></td> <td>No</td></tr></tbody></table> <p data-svelte-h="svelte-1biq3pv">Notes:</p> <ul data-svelte-h="svelte-1hsback"><li>The <code>grouped_mm</code> experts backend currently only supports <code>bfloat16</code> when compiled with <code>torch.compile</code>. Additionally, it is not compatible with CUDA graphs, so you must use <code>mode=None</code> or <code>mode="max-autotune-no-cudagraphs"</code> when compiling.</li> <li>The <code>eager</code> experts backend uses a data-dependent operation to find which experts are used in a forward pass. This operation is not compatible with full graph compilation (<code>fullgraph=True</code>).</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-keyword">import</span> torch
	<span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoModelForCausalLM

	model = AutoModelForCausalLM.from_pretrained(
	<span class="hljs-string">"Qwen/Qwen1.5-MoE-A2.7B"</span>,
	dtype=<span class="hljs-string">"bfloat16"</span>,
	experts_implementation=<span class="hljs-string">"grouped_mm"</span>,
	).<span class="hljs-built_in">eval</span>().cuda()

	<span class="hljs-comment"># Works for grouped_mm (no CUDA graphs)</span>
	model.forward = torch.<span class="hljs-built_in">compile</span>(model.forward, mode=<span class="hljs-string">"max-autotune-no-cudagraphs"</span>)<!-- HTML_TAG_END --></pre></div> <h2 class="relative group"><a id="benchmarks" 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="#benchmarks"><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>Benchmarks</span></h2> <p data-svelte-h="svelte-1c6q5vt">This <a href="https://github.com/user-attachments/files/24125816/bench.py" rel="nofollow">benchmark</a> compares different input sizes and experts implementations with and without <code>torch.compile</code>.</p> <iframe src="https://huggingface-moe-experts-benchmarks.static.hf.space" frameborder="0" width="100%" height="800"></iframe> <iframe src="https://huggingface.co/datasets/docs-benchmarks/experts-backends/embed/viewer/bs1-seq16/train" frameborder="0" width="100%" height="560px"></iframe> <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/experts_interface.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_rv114u = {
	assets: "/docs/transformers/main/en",
	base: "/docs/transformers/main/en",
	env: {}
	};

	const element = document.currentScript.parentElement;

	const data = [null,null];

	Promise.all([
	import("/docs/transformers/main/en/_app/immutable/entry/start.3d6cca8a.js"),
	import("/docs/transformers/main/en/_app/immutable/entry/app.b6ccab0d.js")
	]).then(([kit, app]) => {
	kit.start(app, element, {
	node_ids: [0, 46],
	data,
	form: null,
	error: null
	});
	});
	}
	</script>

Xet Storage Details

Size:: 26 kB
Xet hash:: 5c11cc92b59d9298004e37666d6747a3fe365480d7a2bde992863c6f5b0b5788

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