Buckets:

hf-doc-build
/

doc-dev

Files

xet

hf-doc-build/doc-dev / diffusers /pr_14039 /en /optimization /tpu.html

HuggingFaceDocBuilder

about 22 hours ago

download

raw

17.4 kB

	<meta charset="utf-8" /><meta name="hf:doc:metadata" content="{"title":"TorchTPU","local":"torchtpu","sections":[{"title":"Eager mode","local":"eager-mode","sections":[],"depth":2},{"title":"Compiled mode","local":"compiled-mode","sections":[],"depth":2}],"depth":1}"/>
	<link href="/docs/diffusers/pr_14039/en/_app/immutable/entry/start.qN1WnII4.js" rel="modulepreload">
	<link href="/docs/diffusers/pr_14039/en/_app/immutable/chunks/20n3gGHO.js" rel="modulepreload">
	<link href="/docs/diffusers/pr_14039/en/_app/immutable/chunks/jDjavuwI.js" rel="modulepreload">
	<link href="/docs/diffusers/pr_14039/en/_app/immutable/entry/app.gUf2cAzk.js" rel="modulepreload">
	<link href="/docs/diffusers/pr_14039/en/_app/immutable/chunks/CMF6C6Mt.js" rel="modulepreload">
	<link href="/docs/diffusers/pr_14039/en/_app/immutable/chunks/DZ_Ssvka.js" rel="modulepreload">
	<link href="/docs/diffusers/pr_14039/en/_app/immutable/chunks/DsnmJJEf.js" rel="modulepreload">
	<link href="/docs/diffusers/pr_14039/en/_app/immutable/chunks/i0LL_sTA.js" rel="modulepreload">
	<link href="/docs/diffusers/pr_14039/en/_app/immutable/nodes/0.BpP2lf8O.js" rel="modulepreload">
	<link href="/docs/diffusers/pr_14039/en/_app/immutable/chunks/9Uo8d8lw.js" rel="modulepreload">
	<link href="/docs/diffusers/pr_14039/en/_app/immutable/nodes/279.B7sCIf6B.js" rel="modulepreload">
	<link href="/docs/diffusers/pr_14039/en/_app/immutable/chunks/BLbO27j4.js" rel="modulepreload">
	<!--vjbfci--><meta name="hf:doc:metadata" content="{"title":"TorchTPU","local":"torchtpu","sections":[{"title":"Eager mode","local":"eager-mode","sections":[],"depth":2},{"title":"Compiled mode","local":"compiled-mode","sections":[],"depth":2}],"depth":1}"/><!---->
	<link href="/docs/diffusers/pr_14039/en/_app/immutable/assets/0.tn0RQdqM.css" rel="modulepreload"> <!--[--><!--[0--><!--[--><!--[0--><!--[--><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> <!--[-1--><!--]--></div><!----> <!--[0--><h1 class="relative group"><a id="torchtpu" 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="#torchtpu"><span><svg 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>TorchTPU</span></h1><!--]--><!----> <p><a href="https://github.com/google-pytorch/torch_tpu/" rel="nofollow">TorchTPU</a> is a PyTorch backend for Google’s Tensor Processing Units (TPUs), which lets you run Diffusers pipelines on Cloud TPUs (v6e, v5p, etc.) with minimal code changes.</p> <p>Two execution modes are available:</p> <table><thead><tr><th>Mode</th><th>Constant</th><th>How to activate</th><th>Notes</th></tr></thead><tbody><tr><td>Strict eager (default)</td><td><code>EagerMode.DEFER_NEVER</code></td><td><code>import torch_tpu</code></td><td>Operations dispatched one at a time, asynchronous</td></tr><tr><td>Compile</td><td>—</td><td><code>pipe.enable_tpu_compile()</code></td><td>AOT compilation with <code>TpuBackend</code></td></tr></tbody></table> <p>Follow the <a href="https://github.com/google-pytorch/torch_tpu/" rel="nofollow">TorchTPU installation guide</a>. After installation, <code>import torch_tpu</code> registers the <code>"tpu"</code> device automatically.</p> <!--[1--><h2 class="relative group"><a id="eager-mode" 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="#eager-mode"><span><svg 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>Eager mode</span></h2><!--]--><!----> <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 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-python "><!----><span class="hljs-keyword">import</span> gc
	<span class="hljs-keyword">import</span> torch
	<span class="hljs-keyword">import</span> torch_tpu <span class="hljs-comment"># noqa: F401</span>

	<span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> FluxPipeline

	pipe = FluxPipeline.from_pretrained(<span class="hljs-string">"black-forest-labs/FLUX.1-schnell"</span>, torch_dtype=torch.bfloat16)

	<span class="hljs-comment"># 1. Encode on TPU.</span>
	pipe.text_encoder.to(<span class="hljs-string">"tpu"</span>)
	pipe.text_encoder_2.to(<span class="hljs-string">"tpu"</span>)
	<span class="hljs-keyword">with</span> torch.no_grad():
	prompt_embeds, pooled_prompt_embeds, _ = pipe.encode_prompt(
	prompt=<span class="hljs-string">"a golden retriever surfing a wave, photorealistic"</span>,
	prompt_2=<span class="hljs-string">"a golden retriever surfing a wave, photorealistic"</span>,
	device=torch.device(<span class="hljs-string">"tpu"</span>),
	max_sequence_length=<span class="hljs-number">512</span>,
	)

	<span class="hljs-comment"># 2. Free the text encoders — nothing below needs them.</span>
	pipe.text_encoder = <span class="hljs-literal">None</span>
	pipe.text_encoder_2 = <span class="hljs-literal">None</span>
	gc.collect()

	<span class="hljs-comment"># 3. Move the transformer and VAE in, then denoise with the precomputed embeddings.</span>
	pipe.transformer.to(<span class="hljs-string">"tpu"</span>)
	pipe.vae.to(<span class="hljs-string">"tpu"</span>)
	image = pipe(
	prompt_embeds=prompt_embeds,
	pooled_prompt_embeds=pooled_prompt_embeds,
	height=<span class="hljs-number">1024</span>,
	width=<span class="hljs-number">1024</span>,
	num_inference_steps=<span class="hljs-number">4</span>,
	guidance_scale=<span class="hljs-number">0.0</span>,
	).images[<span class="hljs-number">0</span>]

	image.save(<span class="hljs-string">"output.png"</span>)<!----></pre></div><!----> <p>If the text encoder alone is too large for a single chip(eg. FLUX.2-dev’s Mistral-3-Small is ~45GB),
	shard it across multiple chips with <a href="/docs/diffusers/pr_14039/en/api/parallel#diffusers.hooks.apply_tensor_parallel">apply_tensor_parallel()</a>, the
	same mechanism <code>enable_parallelism()</code> uses for the transformer (see <a href="../training/distributed_inference#tensor-parallelism">Tensor
	parallelism</a>). It only requires <code>model: torch.nn.Module</code>, so it works directly on a <code>transformers.PreTrainedModel</code> text encoder too, not
	just a diffusers <code>ModelMixin</code>. The text encoder doesn’t define a <code>_tp_plan</code>, so supply one: pair
	each attention/MLP projection that expands the hidden dimension (<code>"colwise"</code>) with the one that
	contracts it back (<code>"rowwise"</code>), matching the <code>transformers</code> model’s actual module names.</p> <!--[1--><h2 class="relative group"><a id="compiled-mode" 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="#compiled-mode"><span><svg 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>Compiled mode</span></h2><!--]--><!----> <p><code>enable_tpu_compile</code> runs <code>torch.compile</code> with <code>TpuBackend</code> on each pipeline module that is already on TPU. The first call (warmup) is slow because it compiles. Later calls reuse the compiled graph. Where it’s supported, it replaces SDP-based attention with <code>AttnProcessor</code> for XLA tracing.</p> <blockquote class="important"><p>TorchTPU requires <strong>static shapes</strong> — <code>torch.compile</code> is called with <code>dynamic=False</code> internally. Every time <code>height</code>, <code>width</code>, or <code>num_inference_steps</code> changes, the graph is
	recompiled from scratch. Keep these values constant across all calls after warmup, or call <code>tpu_warmup</code> again before changing them.</p></blockquote> <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 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-python "><!----><span class="hljs-keyword">import</span> torch
	<span class="hljs-keyword">import</span> torch_tpu <span class="hljs-comment"># noqa: F401</span>

	<span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> FluxPipeline

	pipe = FluxPipeline.from_pretrained(
	<span class="hljs-string">"black-forest-labs/FLUX.1-schnell"</span>,
	torch_dtype=torch.bfloat16,
	)
	pipe.transformer.to(<span class="hljs-string">"tpu"</span>)
	pipe.vae.to(<span class="hljs-string">"tpu"</span>)

	pipe.enable_tpu_compile()

	<span class="hljs-comment"># Warmup — triggers static graph compilation.</span>
	pipe.tpu_warmup(
	prompt=<span class="hljs-string">"warmup"</span>,
	height=<span class="hljs-number">1024</span>,
	width=<span class="hljs-number">1024</span>,
	num_inference_steps=<span class="hljs-number">4</span>,
	guidance_scale=<span class="hljs-number">0.0</span>,
	)

	<span class="hljs-comment"># Timed inference reuses the compiled graph.</span>
	image = pipe(
	prompt=<span class="hljs-string">"a golden retriever surfing a wave, photorealistic"</span>,
	height=<span class="hljs-number">1024</span>,
	width=<span class="hljs-number">1024</span>,
	num_inference_steps=<span class="hljs-number">4</span>,
	guidance_scale=<span class="hljs-number">0.0</span>,
	).images[<span class="hljs-number">0</span>]

	image.save(<span class="hljs-string">"output.png"</span>)<!----></pre></div><!----> <a class="!text-gray-400 !no-underline text-sm flex items-center not-prose mt-4" href="https://github.com/huggingface/diffusers/blob/main/docs/source/en/optimization/tpu.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><span class="underline">Update</span> on GitHub</span></a><!----> <p></p><!--]--><!----><!--]--><!--]--><!--]--> <!--[-1--><!--]--><!--]-->

	<script>
	{
	__sveltekit_1v06nzs = {
	base: "/docs/diffusers/pr_14039/en",
	assets: "/docs/diffusers/pr_14039/en"
	};

	const element = document.currentScript.parentElement;

	Promise.all([
	import("/docs/diffusers/pr_14039/en/_app/immutable/entry/start.qN1WnII4.js"),
	import("/docs/diffusers/pr_14039/en/_app/immutable/entry/app.gUf2cAzk.js")
	]).then(([kit, app]) => {
	kit.start(app, element, {
	node_ids: [0, 279],
	data: [null,null],
	form: null,
	error: null
	});
	});
	}
	</script>

Xet Storage Details

Size:: 17.4 kB
Xet hash:: bc9ff6877e2d5ff69607de00dc9cb015ea7494dc8e60de970e0fcc3b4ba19d64

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