Buckets:

hf-doc-build
/

doc-dev

Files

xet

hf-doc-build/doc-dev / transformers /main /en /perf_torch_compile.html

rtrm

about 1 month ago

download

raw

67.8 kB

	<meta charset="utf-8" /><meta name="hf:doc:metadata" content="{"title":"Optimize inference using torch.compile()","local":"optimize-inference-using-torchcompile","sections":[{"title":"Benefits of torch.compile","local":"benefits-of-torchcompile","sections":[],"depth":2},{"title":"Benchmarking code","local":"benchmarking-code","sections":[{"title":"Image Classification with ViT","local":"image-classification-with-vit","sections":[{"title":"Object Detection with DETR","local":"object-detection-with-detr","sections":[],"depth":4},{"title":"Image Segmentation with Segformer","local":"image-segmentation-with-segformer","sections":[],"depth":4}],"depth":3},{"title":"A100 (batch size: 1)","local":"a100-batch-size-1","sections":[],"depth":3},{"title":"A100 (batch size: 4)","local":"a100-batch-size-4","sections":[],"depth":3},{"title":"A100 (batch size: 16)","local":"a100-batch-size-16","sections":[],"depth":3},{"title":"V100 (batch size: 1)","local":"v100-batch-size-1","sections":[],"depth":3},{"title":"V100 (batch size: 4)","local":"v100-batch-size-4","sections":[],"depth":3},{"title":"V100 (batch size: 16)","local":"v100-batch-size-16","sections":[],"depth":3},{"title":"T4 (batch size: 1)","local":"t4-batch-size-1","sections":[],"depth":3},{"title":"T4 (batch size: 4)","local":"t4-batch-size-4","sections":[],"depth":3},{"title":"T4 (batch size: 16)","local":"t4-batch-size-16","sections":[],"depth":3}],"depth":2},{"title":"PyTorch Nightly","local":"pytorch-nightly","sections":[{"title":"A100","local":"a100","sections":[],"depth":3},{"title":"T4","local":"t4","sections":[],"depth":3},{"title":"V100","local":"v100","sections":[],"depth":3}],"depth":2},{"title":"Reduce Overhead","local":"reduce-overhead","sections":[{"title":"A100","local":"a100","sections":[],"depth":3},{"title":"T4","local":"t4","sections":[],"depth":3}],"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.2135b7e6.js">
	<link rel="modulepreload" href="/docs/transformers/main/en/_app/immutable/chunks/scheduler.25b97de1.js">
	<link rel="modulepreload" href="/docs/transformers/main/en/_app/immutable/chunks/singletons.0f2b7d5f.js">
	<link rel="modulepreload" href="/docs/transformers/main/en/_app/immutable/chunks/index.e188933d.js">
	<link rel="modulepreload" href="/docs/transformers/main/en/_app/immutable/chunks/paths.3d04d2c6.js">
	<link rel="modulepreload" href="/docs/transformers/main/en/_app/immutable/entry/app.24372c84.js">
	<link rel="modulepreload" href="/docs/transformers/main/en/_app/immutable/chunks/index.d9030fc9.js">
	<link rel="modulepreload" href="/docs/transformers/main/en/_app/immutable/nodes/0.026d2fdd.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/368.1e7055c9.js">
	<link rel="modulepreload" href="/docs/transformers/main/en/_app/immutable/chunks/CodeBlock.e6cd0d95.js">
	<link rel="modulepreload" href="/docs/transformers/main/en/_app/immutable/chunks/EditOnGithub.91d95064.js"><!-- HEAD_svelte-u9bgzb_START --><meta name="hf:doc:metadata" content="{"title":"Optimize inference using torch.compile()","local":"optimize-inference-using-torchcompile","sections":[{"title":"Benefits of torch.compile","local":"benefits-of-torchcompile","sections":[],"depth":2},{"title":"Benchmarking code","local":"benchmarking-code","sections":[{"title":"Image Classification with ViT","local":"image-classification-with-vit","sections":[{"title":"Object Detection with DETR","local":"object-detection-with-detr","sections":[],"depth":4},{"title":"Image Segmentation with Segformer","local":"image-segmentation-with-segformer","sections":[],"depth":4}],"depth":3},{"title":"A100 (batch size: 1)","local":"a100-batch-size-1","sections":[],"depth":3},{"title":"A100 (batch size: 4)","local":"a100-batch-size-4","sections":[],"depth":3},{"title":"A100 (batch size: 16)","local":"a100-batch-size-16","sections":[],"depth":3},{"title":"V100 (batch size: 1)","local":"v100-batch-size-1","sections":[],"depth":3},{"title":"V100 (batch size: 4)","local":"v100-batch-size-4","sections":[],"depth":3},{"title":"V100 (batch size: 16)","local":"v100-batch-size-16","sections":[],"depth":3},{"title":"T4 (batch size: 1)","local":"t4-batch-size-1","sections":[],"depth":3},{"title":"T4 (batch size: 4)","local":"t4-batch-size-4","sections":[],"depth":3},{"title":"T4 (batch size: 16)","local":"t4-batch-size-16","sections":[],"depth":3}],"depth":2},{"title":"PyTorch Nightly","local":"pytorch-nightly","sections":[{"title":"A100","local":"a100","sections":[],"depth":3},{"title":"T4","local":"t4","sections":[],"depth":3},{"title":"V100","local":"v100","sections":[],"depth":3}],"depth":2},{"title":"Reduce Overhead","local":"reduce-overhead","sections":[{"title":"A100","local":"a100","sections":[],"depth":3},{"title":"T4","local":"t4","sections":[],"depth":3}],"depth":2}],"depth":1}"><!-- HEAD_svelte-u9bgzb_END --> <p></p> <h1 class="relative group"><a id="optimize-inference-using-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="#optimize-inference-using-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>Optimize inference using torch.compile()</span></h1> <p data-svelte-h="svelte-a1xmpp">This guide aims to provide a benchmark on the inference speed-ups introduced with <a href="https://pytorch.org/tutorials/intermediate/torch_compile_tutorial.html" rel="nofollow"><code>torch.compile()</code></a> for <a href="https://huggingface.co/models?pipeline_tag=image-classification&library=transformers&sort=trending" rel="nofollow">computer vision models in 🤗 Transformers</a>.</p> <h2 class="relative group"><a id="benefits-of-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="#benefits-of-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>Benefits of torch.compile</span></h2> <p data-svelte-h="svelte-108lixw">Depending on the model and the GPU, <code>torch.compile()</code> yields up to 30% speed-up during inference. To use <code>torch.compile()</code>, simply install any version of <code>torch</code> above 2.0.</p> <p data-svelte-h="svelte-1b66kne">Compiling a model takes time, so it’s useful if you are compiling the model only once instead of every time you infer.
	To compile any computer vision model of your choice, call <code>torch.compile()</code> on the model as shown below:</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 -->from transformers import AutoModelForImageClassification

	model = AutoModelForImageClassification.from_pretrained(MODEL_ID).to("cuda")
	<span class="hljs-addition">+ model = torch.compile(model)</span><!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-13c27qn"><code>compile()</code> comes with multiple modes for compiling, which essentially differ in compilation time and inference overhead. <code>max-autotune</code> takes longer than <code>reduce-overhead</code> but results in faster inference. Default mode is fastest for compilation but is not as efficient compared to <code>reduce-overhead</code> for inference time. In this guide, we used the default mode. You can learn more about it <a href="https://pytorch.org/get-started/pytorch-2.0/#user-experience" rel="nofollow">here</a>.</p> <p data-svelte-h="svelte-116qe4s">We benchmarked <code>torch.compile</code> with different computer vision models, tasks, types of hardware, and batch sizes on <code>torch</code> version 2.0.1.</p> <h2 class="relative group"><a id="benchmarking-code" 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="#benchmarking-code"><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>Benchmarking code</span></h2> <p data-svelte-h="svelte-2dld2n">Below you can find the benchmarking code for each task. We warm up the GPU before inference and take the mean time of 300 inferences, using the same image each time.</p> <h3 class="relative group"><a id="image-classification-with-vit" 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="#image-classification-with-vit"><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>Image Classification with ViT</span></h3> <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> PIL <span class="hljs-keyword">import</span> Image
	<span class="hljs-keyword">import</span> requests
	<span class="hljs-keyword">import</span> numpy <span class="hljs-keyword">as</span> np
	<span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoImageProcessor, AutoModelForImageClassification

	url = <span class="hljs-string">'http://images.cocodataset.org/val2017/000000039769.jpg'</span>
	image = Image.<span class="hljs-built_in">open</span>(requests.get(url, stream=<span class="hljs-literal">True</span>).raw)

	processor = AutoImageProcessor.from_pretrained(<span class="hljs-string">"google/vit-base-patch16-224"</span>)
	model = AutoModelForImageClassification.from_pretrained(<span class="hljs-string">"google/vit-base-patch16-224"</span>).to(<span class="hljs-string">"cuda"</span>)
	model = torch.<span class="hljs-built_in">compile</span>(model)

	processed_input = processor(image, return_tensors=<span class="hljs-string">'pt'</span>).to(device=<span class="hljs-string">"cuda"</span>)

	<span class="hljs-keyword">with</span> torch.no_grad():
	_ = model(**processed_input)
	<!-- HTML_TAG_END --></pre></div> <h4 class="relative group"><a id="object-detection-with-detr" 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="#object-detection-with-detr"><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>Object Detection with DETR</span></h4> <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> AutoImageProcessor, AutoModelForObjectDetection

	processor = AutoImageProcessor.from_pretrained(<span class="hljs-string">"facebook/detr-resnet-50"</span>)
	model = AutoModelForObjectDetection.from_pretrained(<span class="hljs-string">"facebook/detr-resnet-50"</span>).to(<span class="hljs-string">"cuda"</span>)
	model = torch.<span class="hljs-built_in">compile</span>(model)

	texts = [<span class="hljs-string">"a photo of a cat"</span>, <span class="hljs-string">"a photo of a dog"</span>]
	inputs = processor(text=texts, images=image, return_tensors=<span class="hljs-string">"pt"</span>).to(<span class="hljs-string">"cuda"</span>)

	<span class="hljs-keyword">with</span> torch.no_grad():
	_ = model(**inputs)<!-- HTML_TAG_END --></pre></div> <h4 class="relative group"><a id="image-segmentation-with-segformer" 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="#image-segmentation-with-segformer"><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>Image Segmentation with Segformer</span></h4> <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> SegformerImageProcessor, SegformerForSemanticSegmentation

	processor = SegformerImageProcessor.from_pretrained(<span class="hljs-string">"nvidia/segformer-b0-finetuned-ade-512-512"</span>)
	model = SegformerForSemanticSegmentation.from_pretrained(<span class="hljs-string">"nvidia/segformer-b0-finetuned-ade-512-512"</span>).to(<span class="hljs-string">"cuda"</span>)
	model = torch.<span class="hljs-built_in">compile</span>(model)
	seg_inputs = processor(images=image, return_tensors=<span class="hljs-string">"pt"</span>).to(<span class="hljs-string">"cuda"</span>)

	<span class="hljs-keyword">with</span> torch.no_grad():
	_ = model(**seg_inputs)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-fbvml8">Below you can find the list of the models we benchmarked.</p> <p data-svelte-h="svelte-1pzh5ag"><strong>Image Classification</strong></p> <ul data-svelte-h="svelte-s6kdqe"><li><a href="https://huggingface.co/google/vit-base-patch16-224" rel="nofollow">google/vit-base-patch16-224</a></li> <li><a href="https://huggingface.co/microsoft/beit-base-patch16-224-pt22k-ft22k" rel="nofollow">microsoft/beit-base-patch16-224-pt22k-ft22k</a></li> <li><a href="https://huggingface.co/facebook/convnext-large-224" rel="nofollow">facebook/convnext-large-224</a></li> <li><a href="https://huggingface.co/microsoft/resnet-50" rel="nofollow">microsoft/resnet-50</a></li></ul> <p data-svelte-h="svelte-vq25eq"><strong>Image Segmentation</strong></p> <ul data-svelte-h="svelte-1vcuz7e"><li><a href="https://huggingface.co/nvidia/segformer-b0-finetuned-ade-512-512" rel="nofollow">nvidia/segformer-b0-finetuned-ade-512-512</a></li> <li><a href="https://huggingface.co/facebook/mask2former-swin-tiny-coco-panoptic" rel="nofollow">facebook/mask2former-swin-tiny-coco-panoptic</a></li> <li><a href="https://huggingface.co/facebook/maskformer-swin-base-ade" rel="nofollow">facebook/maskformer-swin-base-ade</a></li> <li><a href="https://huggingface.co/google/deeplabv3_mobilenet_v2_1.0_513" rel="nofollow">google/deeplabv3_mobilenet_v2_1.0_513</a></li></ul> <p data-svelte-h="svelte-1e8jpwt"><strong>Object Detection</strong></p> <ul data-svelte-h="svelte-yhf9tm"><li><a href="https://huggingface.co/google/owlvit-base-patch32" rel="nofollow">google/owlvit-base-patch32</a></li> <li><a href="https://huggingface.co/facebook/detr-resnet-101" rel="nofollow">facebook/detr-resnet-101</a></li> <li><a href="https://huggingface.co/microsoft/conditional-detr-resnet-50" rel="nofollow">microsoft/conditional-detr-resnet-50</a></li></ul> <p data-svelte-h="svelte-7yg5eu">Below you can find visualization of inference durations with and without <code>torch.compile()</code> and percentage improvements for each model in different hardware and batch sizes.</p> <div class="flex" data-svelte-h="svelte-1jw9wmi"><div><img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/torch_compile/a100_batch_comp.png"></div> <div><img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/torch_compile/v100_batch_comp.png"></div> <div><img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/torch_compile/t4_batch_comp.png"></div></div> <div class="flex" data-svelte-h="svelte-nlzsqo"><div><img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/torch_compile/A100_1_duration.png"></div> <div><img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/torch_compile/A100_1_percentage.png"></div></div> <p data-svelte-h="svelte-gdeipd"><img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/torch_compile/v100_1_duration.png" alt="Duration Comparison on V100 with Batch Size of 1"></p> <p data-svelte-h="svelte-1cusdpa"><img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/torch_compile/T4_4_percentage.png" alt="Percentage Improvement on T4 with Batch Size of 4"></p> <p data-svelte-h="svelte-16crmf3">Below you can find inference durations in milliseconds for each model with and without <code>compile()</code>. Note that OwlViT results in OOM in larger batch sizes.</p> <h3 class="relative group"><a id="a100-batch-size-1" 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="#a100-batch-size-1"><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>A100 (batch size: 1)</span></h3> <table data-svelte-h="svelte-6uvhqg"><thead><tr><th align="center"><strong>Task/Model</strong></th> <th align="center"><strong>torch 2.0 - <br>no compile</strong></th> <th align="center"><strong>torch 2.0 - <br>compile</strong></th></tr></thead> <tbody><tr><td align="center">Image Classification/ViT</td> <td align="center">9.325</td> <td align="center">7.584</td></tr> <tr><td align="center">Image Segmentation/Segformer</td> <td align="center">11.759</td> <td align="center">10.500</td></tr> <tr><td align="center">Object Detection/OwlViT</td> <td align="center">24.978</td> <td align="center">18.420</td></tr> <tr><td align="center">Image Classification/BeiT</td> <td align="center">11.282</td> <td align="center">8.448</td></tr> <tr><td align="center">Object Detection/DETR</td> <td align="center">34.619</td> <td align="center">19.040</td></tr> <tr><td align="center">Image Classification/ConvNeXT</td> <td align="center">10.410</td> <td align="center">10.208</td></tr> <tr><td align="center">Image Classification/ResNet</td> <td align="center">6.531</td> <td align="center">4.124</td></tr> <tr><td align="center">Image Segmentation/Mask2former</td> <td align="center">60.188</td> <td align="center">49.117</td></tr> <tr><td align="center">Image Segmentation/Maskformer</td> <td align="center">75.764</td> <td align="center">59.487</td></tr> <tr><td align="center">Image Segmentation/MobileNet</td> <td align="center">8.583</td> <td align="center">3.974</td></tr> <tr><td align="center">Object Detection/Resnet-101</td> <td align="center">36.276</td> <td align="center">18.197</td></tr> <tr><td align="center">Object Detection/Conditional-DETR</td> <td align="center">31.219</td> <td align="center">17.993</td></tr></tbody></table> <h3 class="relative group"><a id="a100-batch-size-4" 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="#a100-batch-size-4"><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>A100 (batch size: 4)</span></h3> <table data-svelte-h="svelte-f4zjoc"><thead><tr><th align="center"><strong>Task/Model</strong></th> <th align="center"><strong>torch 2.0 - <br>no compile</strong></th> <th align="center"><strong>torch 2.0 - <br>compile</strong></th></tr></thead> <tbody><tr><td align="center">Image Classification/ViT</td> <td align="center">14.832</td> <td align="center">14.499</td></tr> <tr><td align="center">Image Segmentation/Segformer</td> <td align="center">18.838</td> <td align="center">16.476</td></tr> <tr><td align="center">Image Classification/BeiT</td> <td align="center">13.205</td> <td align="center">13.048</td></tr> <tr><td align="center">Object Detection/DETR</td> <td align="center">48.657</td> <td align="center">32.418</td></tr> <tr><td align="center">Image Classification/ConvNeXT</td> <td align="center">22.940</td> <td align="center">21.631</td></tr> <tr><td align="center">Image Classification/ResNet</td> <td align="center">6.657</td> <td align="center">4.268</td></tr> <tr><td align="center">Image Segmentation/Mask2former</td> <td align="center">74.277</td> <td align="center">61.781</td></tr> <tr><td align="center">Image Segmentation/Maskformer</td> <td align="center">180.700</td> <td align="center">159.116</td></tr> <tr><td align="center">Image Segmentation/MobileNet</td> <td align="center">14.174</td> <td align="center">8.515</td></tr> <tr><td align="center">Object Detection/Resnet-101</td> <td align="center">68.101</td> <td align="center">44.998</td></tr> <tr><td align="center">Object Detection/Conditional-DETR</td> <td align="center">56.470</td> <td align="center">35.552</td></tr></tbody></table> <h3 class="relative group"><a id="a100-batch-size-16" 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="#a100-batch-size-16"><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>A100 (batch size: 16)</span></h3> <table data-svelte-h="svelte-9ju0ii"><thead><tr><th align="center"><strong>Task/Model</strong></th> <th align="center"><strong>torch 2.0 - <br>no compile</strong></th> <th align="center"><strong>torch 2.0 - <br>compile</strong></th></tr></thead> <tbody><tr><td align="center">Image Classification/ViT</td> <td align="center">40.944</td> <td align="center">40.010</td></tr> <tr><td align="center">Image Segmentation/Segformer</td> <td align="center">37.005</td> <td align="center">31.144</td></tr> <tr><td align="center">Image Classification/BeiT</td> <td align="center">41.854</td> <td align="center">41.048</td></tr> <tr><td align="center">Object Detection/DETR</td> <td align="center">164.382</td> <td align="center">161.902</td></tr> <tr><td align="center">Image Classification/ConvNeXT</td> <td align="center">82.258</td> <td align="center">75.561</td></tr> <tr><td align="center">Image Classification/ResNet</td> <td align="center">7.018</td> <td align="center">5.024</td></tr> <tr><td align="center">Image Segmentation/Mask2former</td> <td align="center">178.945</td> <td align="center">154.814</td></tr> <tr><td align="center">Image Segmentation/Maskformer</td> <td align="center">638.570</td> <td align="center">579.826</td></tr> <tr><td align="center">Image Segmentation/MobileNet</td> <td align="center">51.693</td> <td align="center">30.310</td></tr> <tr><td align="center">Object Detection/Resnet-101</td> <td align="center">232.887</td> <td align="center">155.021</td></tr> <tr><td align="center">Object Detection/Conditional-DETR</td> <td align="center">180.491</td> <td align="center">124.032</td></tr></tbody></table> <h3 class="relative group"><a id="v100-batch-size-1" 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="#v100-batch-size-1"><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>V100 (batch size: 1)</span></h3> <table data-svelte-h="svelte-18ncoxq"><thead><tr><th align="center"><strong>Task/Model</strong></th> <th align="center"><strong>torch 2.0 - <br>no compile</strong></th> <th align="center"><strong>torch 2.0 - <br>compile</strong></th></tr></thead> <tbody><tr><td align="center">Image Classification/ViT</td> <td align="center">10.495</td> <td align="center">6.00</td></tr> <tr><td align="center">Image Segmentation/Segformer</td> <td align="center">13.321</td> <td align="center">5.862</td></tr> <tr><td align="center">Object Detection/OwlViT</td> <td align="center">25.769</td> <td align="center">22.395</td></tr> <tr><td align="center">Image Classification/BeiT</td> <td align="center">11.347</td> <td align="center">7.234</td></tr> <tr><td align="center">Object Detection/DETR</td> <td align="center">33.951</td> <td align="center">19.388</td></tr> <tr><td align="center">Image Classification/ConvNeXT</td> <td align="center">11.623</td> <td align="center">10.412</td></tr> <tr><td align="center">Image Classification/ResNet</td> <td align="center">6.484</td> <td align="center">3.820</td></tr> <tr><td align="center">Image Segmentation/Mask2former</td> <td align="center">64.640</td> <td align="center">49.873</td></tr> <tr><td align="center">Image Segmentation/Maskformer</td> <td align="center">95.532</td> <td align="center">72.207</td></tr> <tr><td align="center">Image Segmentation/MobileNet</td> <td align="center">9.217</td> <td align="center">4.753</td></tr> <tr><td align="center">Object Detection/Resnet-101</td> <td align="center">52.818</td> <td align="center">28.367</td></tr> <tr><td align="center">Object Detection/Conditional-DETR</td> <td align="center">39.512</td> <td align="center">20.816</td></tr></tbody></table> <h3 class="relative group"><a id="v100-batch-size-4" 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="#v100-batch-size-4"><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>V100 (batch size: 4)</span></h3> <table data-svelte-h="svelte-15udyd3"><thead><tr><th align="center"><strong>Task/Model</strong></th> <th align="center"><strong>torch 2.0 - <br>no compile</strong></th> <th align="center"><strong>torch 2.0 - <br>compile</strong></th></tr></thead> <tbody><tr><td align="center">Image Classification/ViT</td> <td align="center">15.181</td> <td align="center">14.501</td></tr> <tr><td align="center">Image Segmentation/Segformer</td> <td align="center">16.787</td> <td align="center">16.188</td></tr> <tr><td align="center">Image Classification/BeiT</td> <td align="center">15.171</td> <td align="center">14.753</td></tr> <tr><td align="center">Object Detection/DETR</td> <td align="center">88.529</td> <td align="center">64.195</td></tr> <tr><td align="center">Image Classification/ConvNeXT</td> <td align="center">29.574</td> <td align="center">27.085</td></tr> <tr><td align="center">Image Classification/ResNet</td> <td align="center">6.109</td> <td align="center">4.731</td></tr> <tr><td align="center">Image Segmentation/Mask2former</td> <td align="center">90.402</td> <td align="center">76.926</td></tr> <tr><td align="center">Image Segmentation/Maskformer</td> <td align="center">234.261</td> <td align="center">205.456</td></tr> <tr><td align="center">Image Segmentation/MobileNet</td> <td align="center">24.623</td> <td align="center">14.816</td></tr> <tr><td align="center">Object Detection/Resnet-101</td> <td align="center">134.672</td> <td align="center">101.304</td></tr> <tr><td align="center">Object Detection/Conditional-DETR</td> <td align="center">97.464</td> <td align="center">69.739</td></tr></tbody></table> <h3 class="relative group"><a id="v100-batch-size-16" 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="#v100-batch-size-16"><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>V100 (batch size: 16)</span></h3> <table data-svelte-h="svelte-rw07j7"><thead><tr><th align="center"><strong>Task/Model</strong></th> <th align="center"><strong>torch 2.0 - <br>no compile</strong></th> <th align="center"><strong>torch 2.0 - <br>compile</strong></th></tr></thead> <tbody><tr><td align="center">Image Classification/ViT</td> <td align="center">52.209</td> <td align="center">51.633</td></tr> <tr><td align="center">Image Segmentation/Segformer</td> <td align="center">61.013</td> <td align="center">55.499</td></tr> <tr><td align="center">Image Classification/BeiT</td> <td align="center">53.938</td> <td align="center">53.581</td></tr> <tr><td align="center">Object Detection/DETR</td> <td align="center">OOM</td> <td align="center">OOM</td></tr> <tr><td align="center">Image Classification/ConvNeXT</td> <td align="center">109.682</td> <td align="center">100.771</td></tr> <tr><td align="center">Image Classification/ResNet</td> <td align="center">14.857</td> <td align="center">12.089</td></tr> <tr><td align="center">Image Segmentation/Mask2former</td> <td align="center">249.605</td> <td align="center">222.801</td></tr> <tr><td align="center">Image Segmentation/Maskformer</td> <td align="center">831.142</td> <td align="center">743.645</td></tr> <tr><td align="center">Image Segmentation/MobileNet</td> <td align="center">93.129</td> <td align="center">55.365</td></tr> <tr><td align="center">Object Detection/Resnet-101</td> <td align="center">482.425</td> <td align="center">361.843</td></tr> <tr><td align="center">Object Detection/Conditional-DETR</td> <td align="center">344.661</td> <td align="center">255.298</td></tr></tbody></table> <h3 class="relative group"><a id="t4-batch-size-1" 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="#t4-batch-size-1"><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>T4 (batch size: 1)</span></h3> <table data-svelte-h="svelte-37x5jw"><thead><tr><th align="center"><strong>Task/Model</strong></th> <th align="center"><strong>torch 2.0 - <br>no compile</strong></th> <th align="center"><strong>torch 2.0 - <br>compile</strong></th></tr></thead> <tbody><tr><td align="center">Image Classification/ViT</td> <td align="center">16.520</td> <td align="center">15.786</td></tr> <tr><td align="center">Image Segmentation/Segformer</td> <td align="center">16.116</td> <td align="center">14.205</td></tr> <tr><td align="center">Object Detection/OwlViT</td> <td align="center">53.634</td> <td align="center">51.105</td></tr> <tr><td align="center">Image Classification/BeiT</td> <td align="center">16.464</td> <td align="center">15.710</td></tr> <tr><td align="center">Object Detection/DETR</td> <td align="center">73.100</td> <td align="center">53.99</td></tr> <tr><td align="center">Image Classification/ConvNeXT</td> <td align="center">32.932</td> <td align="center">30.845</td></tr> <tr><td align="center">Image Classification/ResNet</td> <td align="center">6.031</td> <td align="center">4.321</td></tr> <tr><td align="center">Image Segmentation/Mask2former</td> <td align="center">79.192</td> <td align="center">66.815</td></tr> <tr><td align="center">Image Segmentation/Maskformer</td> <td align="center">200.026</td> <td align="center">188.268</td></tr> <tr><td align="center">Image Segmentation/MobileNet</td> <td align="center">18.908</td> <td align="center">11.997</td></tr> <tr><td align="center">Object Detection/Resnet-101</td> <td align="center">106.622</td> <td align="center">82.566</td></tr> <tr><td align="center">Object Detection/Conditional-DETR</td> <td align="center">77.594</td> <td align="center">56.984</td></tr></tbody></table> <h3 class="relative group"><a id="t4-batch-size-4" 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="#t4-batch-size-4"><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>T4 (batch size: 4)</span></h3> <table data-svelte-h="svelte-1mc5027"><thead><tr><th align="center"><strong>Task/Model</strong></th> <th align="center"><strong>torch 2.0 - <br>no compile</strong></th> <th align="center"><strong>torch 2.0 - <br>compile</strong></th></tr></thead> <tbody><tr><td align="center">Image Classification/ViT</td> <td align="center">43.653</td> <td align="center">43.626</td></tr> <tr><td align="center">Image Segmentation/Segformer</td> <td align="center">45.327</td> <td align="center">42.445</td></tr> <tr><td align="center">Image Classification/BeiT</td> <td align="center">52.007</td> <td align="center">51.354</td></tr> <tr><td align="center">Object Detection/DETR</td> <td align="center">277.850</td> <td align="center">268.003</td></tr> <tr><td align="center">Image Classification/ConvNeXT</td> <td align="center">119.259</td> <td align="center">105.580</td></tr> <tr><td align="center">Image Classification/ResNet</td> <td align="center">13.039</td> <td align="center">11.388</td></tr> <tr><td align="center">Image Segmentation/Mask2former</td> <td align="center">201.540</td> <td align="center">184.670</td></tr> <tr><td align="center">Image Segmentation/Maskformer</td> <td align="center">764.052</td> <td align="center">711.280</td></tr> <tr><td align="center">Image Segmentation/MobileNet</td> <td align="center">74.289</td> <td align="center">48.677</td></tr> <tr><td align="center">Object Detection/Resnet-101</td> <td align="center">421.859</td> <td align="center">357.614</td></tr> <tr><td align="center">Object Detection/Conditional-DETR</td> <td align="center">289.002</td> <td align="center">226.945</td></tr></tbody></table> <h3 class="relative group"><a id="t4-batch-size-16" 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="#t4-batch-size-16"><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>T4 (batch size: 16)</span></h3> <table data-svelte-h="svelte-10eiin7"><thead><tr><th align="center"><strong>Task/Model</strong></th> <th align="center"><strong>torch 2.0 - <br>no compile</strong></th> <th align="center"><strong>torch 2.0 - <br>compile</strong></th></tr></thead> <tbody><tr><td align="center">Image Classification/ViT</td> <td align="center">163.914</td> <td align="center">160.907</td></tr> <tr><td align="center">Image Segmentation/Segformer</td> <td align="center">192.412</td> <td align="center">163.620</td></tr> <tr><td align="center">Image Classification/BeiT</td> <td align="center">188.978</td> <td align="center">187.976</td></tr> <tr><td align="center">Object Detection/DETR</td> <td align="center">OOM</td> <td align="center">OOM</td></tr> <tr><td align="center">Image Classification/ConvNeXT</td> <td align="center">422.886</td> <td align="center">388.078</td></tr> <tr><td align="center">Image Classification/ResNet</td> <td align="center">44.114</td> <td align="center">37.604</td></tr> <tr><td align="center">Image Segmentation/Mask2former</td> <td align="center">756.337</td> <td align="center">695.291</td></tr> <tr><td align="center">Image Segmentation/Maskformer</td> <td align="center">2842.940</td> <td align="center">2656.88</td></tr> <tr><td align="center">Image Segmentation/MobileNet</td> <td align="center">299.003</td> <td align="center">201.942</td></tr> <tr><td align="center">Object Detection/Resnet-101</td> <td align="center">1619.505</td> <td align="center">1262.758</td></tr> <tr><td align="center">Object Detection/Conditional-DETR</td> <td align="center">1137.513</td> <td align="center">897.390</td></tr></tbody></table> <h2 class="relative group"><a id="pytorch-nightly" 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="#pytorch-nightly"><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>PyTorch Nightly</span></h2> <p data-svelte-h="svelte-ov5yx8">We also benchmarked on PyTorch nightly (2.1.0dev, find the wheel <a href="https://download.pytorch.org/whl/nightly/cu118" rel="nofollow">here</a>) and observed improvement in latency both for uncompiled and compiled models.</p> <h3 class="relative group"><a id="a100" 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="#a100"><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>A100</span></h3> <table data-svelte-h="svelte-1cg5nyy"><thead><tr><th align="center"><strong>Task/Model</strong></th> <th align="center"><strong>Batch Size</strong></th> <th align="center"><strong>torch 2.0 - no compile</strong></th> <th align="center"><strong>torch 2.0 -<br> compile</strong></th></tr></thead> <tbody><tr><td align="center">Image Classification/BeiT</td> <td align="center">Unbatched</td> <td align="center">12.462</td> <td align="center">6.954</td></tr> <tr><td align="center">Image Classification/BeiT</td> <td align="center">4</td> <td align="center">14.109</td> <td align="center">12.851</td></tr> <tr><td align="center">Image Classification/BeiT</td> <td align="center">16</td> <td align="center">42.179</td> <td align="center">42.147</td></tr> <tr><td align="center">Object Detection/DETR</td> <td align="center">Unbatched</td> <td align="center">30.484</td> <td align="center">15.221</td></tr> <tr><td align="center">Object Detection/DETR</td> <td align="center">4</td> <td align="center">46.816</td> <td align="center">30.942</td></tr> <tr><td align="center">Object Detection/DETR</td> <td align="center">16</td> <td align="center">163.749</td> <td align="center">163.706</td></tr></tbody></table> <h3 class="relative group"><a id="t4" 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="#t4"><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>T4</span></h3> <table data-svelte-h="svelte-1nlzppe"><thead><tr><th align="center"><strong>Task/Model</strong></th> <th align="center"><strong>Batch Size</strong></th> <th align="center"><strong>torch 2.0 - <br>no compile</strong></th> <th align="center"><strong>torch 2.0 - <br>compile</strong></th></tr></thead> <tbody><tr><td align="center">Image Classification/BeiT</td> <td align="center">Unbatched</td> <td align="center">14.408</td> <td align="center">14.052</td></tr> <tr><td align="center">Image Classification/BeiT</td> <td align="center">4</td> <td align="center">47.381</td> <td align="center">46.604</td></tr> <tr><td align="center">Image Classification/BeiT</td> <td align="center">16</td> <td align="center">42.179</td> <td align="center">42.147</td></tr> <tr><td align="center">Object Detection/DETR</td> <td align="center">Unbatched</td> <td align="center">68.382</td> <td align="center">53.481</td></tr> <tr><td align="center">Object Detection/DETR</td> <td align="center">4</td> <td align="center">269.615</td> <td align="center">204.785</td></tr> <tr><td align="center">Object Detection/DETR</td> <td align="center">16</td> <td align="center">OOM</td> <td align="center">OOM</td></tr></tbody></table> <h3 class="relative group"><a id="v100" 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="#v100"><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>V100</span></h3> <table data-svelte-h="svelte-ok1p6e"><thead><tr><th align="center"><strong>Task/Model</strong></th> <th align="center"><strong>Batch Size</strong></th> <th align="center"><strong>torch 2.0 - <br>no compile</strong></th> <th align="center"><strong>torch 2.0 - <br>compile</strong></th></tr></thead> <tbody><tr><td align="center">Image Classification/BeiT</td> <td align="center">Unbatched</td> <td align="center">13.477</td> <td align="center">7.926</td></tr> <tr><td align="center">Image Classification/BeiT</td> <td align="center">4</td> <td align="center">15.103</td> <td align="center">14.378</td></tr> <tr><td align="center">Image Classification/BeiT</td> <td align="center">16</td> <td align="center">52.517</td> <td align="center">51.691</td></tr> <tr><td align="center">Object Detection/DETR</td> <td align="center">Unbatched</td> <td align="center">28.706</td> <td align="center">19.077</td></tr> <tr><td align="center">Object Detection/DETR</td> <td align="center">4</td> <td align="center">88.402</td> <td align="center">62.949</td></tr> <tr><td align="center">Object Detection/DETR</td> <td align="center">16</td> <td align="center">OOM</td> <td align="center">OOM</td></tr></tbody></table> <h2 class="relative group"><a id="reduce-overhead" 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="#reduce-overhead"><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>Reduce Overhead</span></h2> <p data-svelte-h="svelte-1axz5ve">We benchmarked <code>reduce-overhead</code> compilation mode for A100 and T4 in Nightly.</p> <h3 class="relative group"><a id="a100" 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="#a100"><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>A100</span></h3> <table data-svelte-h="svelte-13rnx0"><thead><tr><th align="center"><strong>Task/Model</strong></th> <th align="center"><strong>Batch Size</strong></th> <th align="center"><strong>torch 2.0 - <br>no compile</strong></th> <th align="center"><strong>torch 2.0 - <br>compile</strong></th></tr></thead> <tbody><tr><td align="center">Image Classification/ConvNeXT</td> <td align="center">Unbatched</td> <td align="center">11.758</td> <td align="center">7.335</td></tr> <tr><td align="center">Image Classification/ConvNeXT</td> <td align="center">4</td> <td align="center">23.171</td> <td align="center">21.490</td></tr> <tr><td align="center">Image Classification/ResNet</td> <td align="center">Unbatched</td> <td align="center">7.435</td> <td align="center">3.801</td></tr> <tr><td align="center">Image Classification/ResNet</td> <td align="center">4</td> <td align="center">7.261</td> <td align="center">2.187</td></tr> <tr><td align="center">Object Detection/Conditional-DETR</td> <td align="center">Unbatched</td> <td align="center">32.823</td> <td align="center">11.627</td></tr> <tr><td align="center">Object Detection/Conditional-DETR</td> <td align="center">4</td> <td align="center">50.622</td> <td align="center">33.831</td></tr> <tr><td align="center">Image Segmentation/MobileNet</td> <td align="center">Unbatched</td> <td align="center">9.869</td> <td align="center">4.244</td></tr> <tr><td align="center">Image Segmentation/MobileNet</td> <td align="center">4</td> <td align="center">14.385</td> <td align="center">7.946</td></tr></tbody></table> <h3 class="relative group"><a id="t4" 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="#t4"><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>T4</span></h3> <table data-svelte-h="svelte-oh2zql"><thead><tr><th align="center"><strong>Task/Model</strong></th> <th align="center"><strong>Batch Size</strong></th> <th align="center"><strong>torch 2.0 - <br>no compile</strong></th> <th align="center"><strong>torch 2.0 - <br>compile</strong></th></tr></thead> <tbody><tr><td align="center">Image Classification/ConvNeXT</td> <td align="center">Unbatched</td> <td align="center">32.137</td> <td align="center">31.84</td></tr> <tr><td align="center">Image Classification/ConvNeXT</td> <td align="center">4</td> <td align="center">120.944</td> <td align="center">110.209</td></tr> <tr><td align="center">Image Classification/ResNet</td> <td align="center">Unbatched</td> <td align="center">9.761</td> <td align="center">7.698</td></tr> <tr><td align="center">Image Classification/ResNet</td> <td align="center">4</td> <td align="center">15.215</td> <td align="center">13.871</td></tr> <tr><td align="center">Object Detection/Conditional-DETR</td> <td align="center">Unbatched</td> <td align="center">72.150</td> <td align="center">57.660</td></tr> <tr><td align="center">Object Detection/Conditional-DETR</td> <td align="center">4</td> <td align="center">301.494</td> <td align="center">247.543</td></tr> <tr><td align="center">Image Segmentation/MobileNet</td> <td align="center">Unbatched</td> <td align="center">22.266</td> <td align="center">19.339</td></tr> <tr><td align="center">Image Segmentation/MobileNet</td> <td align="center">4</td> <td align="center">78.311</td> <td align="center">50.983</td></tr></tbody></table> <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/perf_torch_compile.md" target="_blank"><span data-svelte-h="svelte-1kd6by1"><</span> <span data-svelte-h="svelte-x0xyl0">></span> <span data-svelte-h="svelte-1dajgef"><span class="underline ml-1.5">Update</span> on GitHub</span></a> <p></p>

	<script>
	{
	__sveltekit_1xexzbk = {
	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.2135b7e6.js"),
	import("/docs/transformers/main/en/_app/immutable/entry/app.24372c84.js")
	]).then(([kit, app]) => {
	kit.start(app, element, {
	node_ids: [0, 368],
	data,
	form: null,
	error: null
	});
	});
	}
	</script>

Xet Storage Details

Size:: 67.8 kB
Xet hash:: 046cce2aa34424e45f16f1fa01bdbdc4be519adebc37b610d00b55d19e647683

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