2.5/backends.html



<!DOCTYPE html>
<!--[if IE 8]><html class="no-js lt-ie9" lang="en" > <![endif]-->
<!--[if gt IE 8]><!--> <html class="no-js" lang="en" > <!--<![endif]-->
<head>
  <meta name="robots" content="noindex">
  <meta charset="utf-8">
  
  <meta name="viewport" content="width=device-width, initial-scale=1.0">
  
  <title>torch.backends &mdash; PyTorch 2.5 documentation</title>
  

    <link rel="canonical" href="https://pytorch.org/docs/stable/backends.html"/>
  

  <link rel="stylesheet" href="_static/css/theme.css" type="text/css" />
  <!-- <link rel="stylesheet" href="_static/pygments.css" type="text/css" /> -->
  <link rel="stylesheet" href="_static/pygments.css" type="text/css" />
  <link rel="stylesheet" href="_static/css/theme.css" type="text/css" />
  <link rel="stylesheet" href="_static/copybutton.css" type="text/css" />
  <link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/katex@0.10.0-beta/dist/katex.min.css" type="text/css" />
  <link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/katex@0.13.11/dist/katex.min.css" type="text/css" />
  <link rel="stylesheet" href="_static/katex-math.css" type="text/css" />
  <link rel="stylesheet" href="_static/sphinx-dropdown.css" type="text/css" />
  <link rel="stylesheet" href="_static/panels-bootstrap.min.css" type="text/css" />
  <link rel="stylesheet" href="_static/css/jit.css" type="text/css" />
    <link rel="index" title="Index" href="genindex.html" />
    <link rel="search" title="Search" href="search.html" />
    <link rel="next" title="torch.export" href="export.html" />
    <link rel="prev" title="Meta device" href="meta.html" />


  <!-- Google Tag Manager -->
    <script>(function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':
    new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],
    j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src=
    'https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);
    })(window,document,'script','dataLayer','GTM-T8XT4PS');</script>
    <!-- End Google Tag Manager -->
  

  <script src="_static/js/modernizr.min.js"></script>

  <!-- Preload the theme fonts -->

<link rel="preload" href="_static/fonts/FreightSans/freight-sans-book.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="_static/fonts/FreightSans/freight-sans-medium.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="_static/fonts/IBMPlexMono/IBMPlexMono-Medium.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="_static/fonts/FreightSans/freight-sans-bold.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="_static/fonts/FreightSans/freight-sans-medium-italic.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="_static/fonts/IBMPlexMono/IBMPlexMono-SemiBold.woff2" as="font" type="font/woff2" crossorigin="anonymous">

<!-- Preload the katex fonts -->

<link rel="preload" href="https://cdn.jsdelivr.net/npm/katex@0.10.0/dist/fonts/KaTeX_Math-Italic.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/katex@0.10.0/dist/fonts/KaTeX_Main-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/katex@0.10.0/dist/fonts/KaTeX_Main-Bold.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/katex@0.10.0/dist/fonts/KaTeX_Size1-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/katex@0.10.0/dist/fonts/KaTeX_Size4-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/katex@0.10.0/dist/fonts/KaTeX_Size2-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/katex@0.10.0/dist/fonts/KaTeX_Size3-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/katex@0.10.0/dist/fonts/KaTeX_Caligraphic-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
  <link rel="stylesheet" href="https://use.fontawesome.com/releases/v5.15.2/css/all.css" integrity="sha384-vSIIfh2YWi9wW0r9iZe7RJPrKwp6bG+s9QZMoITbCckVJqGCCRhc+ccxNcdpHuYu" crossorigin="anonymous">
</head>

<div class="container-fluid header-holder tutorials-header" id="header-holder">
  <div class="container">
    <div class="header-container">
      <a class="header-logo" href="https://pytorch.org/" aria-label="PyTorch"></a>

      <div class="main-menu">
        <ul>

          <li class="main-menu-item">
          <div id="resourcesDropdownButton" data-toggle="resources-dropdown" class="resources-dropdown">
              <a class="with-down-arrow">
                Learn
              </a>
              <div class="resources-dropdown-menu">
                <a class="nav-dropdown-item" href="https://pytorch.org/get-started">
                  <span class=dropdown-title>Get Started</span>
                  <p>Run PyTorch locally or get started quickly with one of the supported cloud platforms</p>
                </a>
                <a class="nav-dropdown-item" href="https://pytorch.org/tutorials">
                  <span class="dropdown-title">Tutorials</span>
                  <p>Whats new in PyTorch tutorials</p>
                </a>
                <a class="nav-dropdown-item" href="https://pytorch.org/tutorials/beginner/basics/intro.html">
                  <span class="dropdown-title">Learn the Basics</span>
                  <p>Familiarize yourself with PyTorch concepts and modules</p>
                </a>
                <a class="nav-dropdown-item" href="https://pytorch.org/tutorials/recipes/recipes_index.html">
                  <span class="dropdown-title">PyTorch Recipes</span>
                  <p>Bite-size, ready-to-deploy PyTorch code examples</p>
                </a>
                <a class="nav-dropdown-item" href="https://pytorch.org/tutorials/beginner/introyt.html">
                  <span class="dropdown-title">Intro to PyTorch - YouTube Series</span>
                  <p>Master PyTorch basics with our engaging YouTube tutorial series</p>
                </a>
              </div>
            </div>
          </li>

          <li>
          <div id="resourcesDropdownButton" data-toggle="resources-dropdown" class="resources-dropdown">
              <a class="with-down-arrow">
                Ecosystem
              </a>
              <div class="resources-dropdown-menu">
                <a class="nav-dropdown-item" href="https://pytorch.org/ecosystem">
                  <span class="dropdown-title">Tools</span>
                  <p>Learn about the tools and frameworks in the PyTorch Ecosystem</p>
                </a>
                <a class="nav-dropdown-item" href="https://pytorch.org/#community-module">
                  <span class=dropdown-title>Community</span>
                  <p>Join the PyTorch developer community to contribute, learn, and get your questions answered</p>
                </a>
                <a class="nav-dropdown-item" href="https://discuss.pytorch.org/" target="_blank">
                  <span class=dropdown-title>Forums</span>
                  <p>A place to discuss PyTorch code, issues, install, research</p>
                </a>
                <a class="nav-dropdown-item" href="https://pytorch.org/resources">
                  <span class=dropdown-title>Developer Resources</span>
                  <p>Find resources and get questions answered</p>
                </a>
                <a class="nav-dropdown-item" href="https://pytorch.org/ecosystem/contributor-awards-2023">
                  <span class="dropdown-title">Contributor Awards - 2023</span>
                  <p>Award winners announced at this year's PyTorch Conference</p>
                </a>
              </div>
            </div>
          </li>

          <li>
          <div id="resourcesDropdownButton" data-toggle="resources-dropdown" class="resources-dropdown">
              <a class="with-down-arrow">
                Edge
              </a>
              <div class="resources-dropdown-menu">
                <a class="nav-dropdown-item" href="https://pytorch.org/edge">
                  <span class="dropdown-title">About PyTorch Edge</span>
                  <p>Build innovative and privacy-aware AI experiences for edge devices</p>
                </a>
                <a class="nav-dropdown-item" href="https://pytorch.org/executorch-overview">
                  <span class="dropdown-title">ExecuTorch</span>
                  <p>End-to-end solution for enabling on-device inference capabilities across mobile and edge devices</p>
                </a>
              </div>
            </div>  
          </li>

          <li class="main-menu-item">
            <div id="resourcesDropdownButton" data-toggle="resources-dropdown" class="resources-dropdown">
              <a class="with-down-arrow">
                Docs
              </a>
              <div class="resources-dropdown-menu">
                <a class="nav-dropdown-item" href="https://pytorch.org/docs/stable/index.html">
                  <span class="dropdown-title">PyTorch</span>
                  <p>Explore the documentation for comprehensive guidance on how to use PyTorch</p>
                </a>
                <a class="nav-dropdown-item" href="https://pytorch.org/pytorch-domains">
                  <span class="dropdown-title">PyTorch Domains</span>
                  <p>Read the PyTorch Domains documentation to learn more about domain-specific libraries</p>
                </a>
              </div>
            </div>
          </li>

          <li>
            <div id="resourcesDropdownButton" data-toggle="resources-dropdown" class="resources-dropdown">
              <a class="with-down-arrow">
                Blogs & News 
              </a>
              <div class="resources-dropdown-menu">
                <a class="nav-dropdown-item" href="https://pytorch.org/blog/">
                  <span class="dropdown-title">PyTorch Blog</span>
                  <p>Catch up on the latest technical news and happenings</p>
                </a>
                 <a class="nav-dropdown-item" href="https://pytorch.org/community-blog">
                  <span class="dropdown-title">Community Blog</span>
                  <p>Stories from the PyTorch ecosystem</p>
                </a>
                <a class="nav-dropdown-item" href="https://pytorch.org/videos">
                  <span class="dropdown-title">Videos</span>
                  <p>Learn about the latest PyTorch tutorials, new, and more </p>
                <a class="nav-dropdown-item" href="https://pytorch.org/community-stories">
                  <span class="dropdown-title">Community Stories</span>
                  <p>Learn how our community solves real, everyday machine learning problems with PyTorch</p>
                </a>
                <a class="nav-dropdown-item" href="https://pytorch.org/events">
                  <span class="dropdown-title">Events</span>
                  <p>Find events, webinars, and podcasts</p>
                </a>
            </div>
          </li>

          <li>
            <div id="resourcesDropdownButton" data-toggle="resources-dropdown" class="resources-dropdown">
              <a class="with-down-arrow">
                About
              </a>
              <div class="resources-dropdown-menu">
                <a class="nav-dropdown-item" href="https://pytorch.org/foundation">
                  <span class="dropdown-title">PyTorch Foundation</span>
                  <p>Learn more about the PyTorch Foundation</p>
                </a>
                <a class="nav-dropdown-item" href="https://pytorch.org/governing-board">
                  <span class="dropdown-title">Governing Board</span>
                  <p></p>
                </a>
              </div>
            </div>
          </li>

          <li class="main-menu-item">
            <div class="no-dropdown">
              <a href="https://pytorch.org/join" data-cta="join">
                Become a Member
              </a>
            </div>
          </li>
          <li>
           <div class="main-menu-item">
             <a href="https://github.com/pytorch/pytorch" class="github-icon">
             </a>
           </div>
          </li>
          <!--- TODO: This block adds the search icon to the nav bar. We will enable it later. 
          <li>
            <div class="main-menu-item">
             <a href="https://github.com/pytorch/pytorch" class="search-icon">
             </a>
            </div>
          </li>
          --->
        </ul>
      </div>

      <a class="main-menu-open-button" href="#" data-behavior="open-mobile-menu"></a>
    </div>
  </div>
</div>

<body class="pytorch-body">

   
    <div class="table-of-contents-link-wrapper">
      <span>Table of Contents</span>
      <a href="#" class="toggle-table-of-contents" data-behavior="toggle-table-of-contents"></a>
    </div>

    <nav data-toggle="wy-nav-shift" class="pytorch-left-menu" id="pytorch-left-menu">
      <div class="pytorch-side-scroll">
        <div class="pytorch-menu pytorch-menu-vertical" data-spy="affix" role="navigation" aria-label="main navigation">
          <div class="pytorch-left-menu-search">
            
    <div class="version">
      <a href='https://pytorch.org/docs/versions.html'>2.5 &#x25BC</a>
    </div>
    

<div role="search">
  <form id="rtd-search-form" class="wy-form" action="search.html" method="get">
    <input type="text" name="q" placeholder="Search Docs" />
    <input type="hidden" name="check_keywords" value="yes" />
    <input type="hidden" name="area" value="default" />
  </form>
</div>

          </div>

          
              <p class="caption" role="heading"><span class="caption-text">Community</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="community/build_ci_governance.html">PyTorch Governance | Build + CI</a></li>
<li class="toctree-l1"><a class="reference internal" href="community/contribution_guide.html">PyTorch Contribution Guide</a></li>
<li class="toctree-l1"><a class="reference internal" href="community/design.html">PyTorch Design Philosophy</a></li>
<li class="toctree-l1"><a class="reference internal" href="community/governance.html">PyTorch Governance | Mechanics</a></li>
<li class="toctree-l1"><a class="reference internal" href="community/persons_of_interest.html">PyTorch Governance | Maintainers</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">Developer Notes</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="notes/amp_examples.html">Automatic Mixed Precision examples</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/autograd.html">Autograd mechanics</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/broadcasting.html">Broadcasting semantics</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/cpu_threading_torchscript_inference.html">CPU threading and TorchScript inference</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/cuda.html">CUDA semantics</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/custom_operators.html">PyTorch Custom Operators Landing Page</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/ddp.html">Distributed Data Parallel</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/extending.html">Extending PyTorch</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/extending.func.html">Extending torch.func with autograd.Function</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/faq.html">Frequently Asked Questions</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/fsdp.html">FSDP Notes</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/get_start_xpu.html">Getting Started on Intel GPU</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/gradcheck.html">Gradcheck mechanics</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/hip.html">HIP (ROCm) semantics</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/large_scale_deployments.html">Features for large-scale deployments</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/modules.html">Modules</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/mps.html">MPS backend</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/multiprocessing.html">Multiprocessing best practices</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/numerical_accuracy.html">Numerical accuracy</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/randomness.html">Reproducibility</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/serialization.html">Serialization semantics</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/windows.html">Windows FAQ</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">Language Bindings</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="cpp_index.html">C++</a></li>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/javadoc/">Javadoc</a></li>
<li class="toctree-l1"><a class="reference internal" href="deploy.html">torch::deploy</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">Python API</span></p>
<ul class="current">
<li class="toctree-l1"><a class="reference internal" href="torch.html">torch</a></li>
<li class="toctree-l1"><a class="reference internal" href="nn.html">torch.nn</a></li>
<li class="toctree-l1"><a class="reference internal" href="nn.functional.html">torch.nn.functional</a></li>
<li class="toctree-l1"><a class="reference internal" href="tensors.html">torch.Tensor</a></li>
<li class="toctree-l1"><a class="reference internal" href="tensor_attributes.html">Tensor Attributes</a></li>
<li class="toctree-l1"><a class="reference internal" href="tensor_view.html">Tensor Views</a></li>
<li class="toctree-l1"><a class="reference internal" href="amp.html">torch.amp</a></li>
<li class="toctree-l1"><a class="reference internal" href="autograd.html">torch.autograd</a></li>
<li class="toctree-l1"><a class="reference internal" href="library.html">torch.library</a></li>
<li class="toctree-l1"><a class="reference internal" href="cpu.html">torch.cpu</a></li>
<li class="toctree-l1"><a class="reference internal" href="cuda.html">torch.cuda</a></li>
<li class="toctree-l1"><a class="reference internal" href="torch_cuda_memory.html">Understanding CUDA Memory Usage</a></li>
<li class="toctree-l1"><a class="reference internal" href="torch_cuda_memory.html#generating-a-snapshot">Generating a Snapshot</a></li>
<li class="toctree-l1"><a class="reference internal" href="torch_cuda_memory.html#using-the-visualizer">Using the visualizer</a></li>
<li class="toctree-l1"><a class="reference internal" href="torch_cuda_memory.html#snapshot-api-reference">Snapshot API Reference</a></li>
<li class="toctree-l1"><a class="reference internal" href="mps.html">torch.mps</a></li>
<li class="toctree-l1"><a class="reference internal" href="xpu.html">torch.xpu</a></li>
<li class="toctree-l1"><a class="reference internal" href="mtia.html">torch.mtia</a></li>
<li class="toctree-l1"><a class="reference internal" href="meta.html">Meta device</a></li>
<li class="toctree-l1 current"><a class="current reference internal" href="#">torch.backends</a></li>
<li class="toctree-l1"><a class="reference internal" href="export.html">torch.export</a></li>
<li class="toctree-l1"><a class="reference internal" href="distributed.html">torch.distributed</a></li>
<li class="toctree-l1"><a class="reference internal" href="distributed.tensor.html">torch.distributed.tensor</a></li>
<li class="toctree-l1"><a class="reference internal" href="distributed.algorithms.join.html">torch.distributed.algorithms.join</a></li>
<li class="toctree-l1"><a class="reference internal" href="distributed.elastic.html">torch.distributed.elastic</a></li>
<li class="toctree-l1"><a class="reference internal" href="fsdp.html">torch.distributed.fsdp</a></li>
<li class="toctree-l1"><a class="reference internal" href="distributed.tensor.parallel.html">torch.distributed.tensor.parallel</a></li>
<li class="toctree-l1"><a class="reference internal" href="distributed.optim.html">torch.distributed.optim</a></li>
<li class="toctree-l1"><a class="reference internal" href="distributed.pipelining.html">torch.distributed.pipelining</a></li>
<li class="toctree-l1"><a class="reference internal" href="distributed.checkpoint.html">torch.distributed.checkpoint</a></li>
<li class="toctree-l1"><a class="reference internal" href="distributions.html">torch.distributions</a></li>
<li class="toctree-l1"><a class="reference internal" href="torch.compiler.html">torch.compiler</a></li>
<li class="toctree-l1"><a class="reference internal" href="fft.html">torch.fft</a></li>
<li class="toctree-l1"><a class="reference internal" href="func.html">torch.func</a></li>
<li class="toctree-l1"><a class="reference internal" href="futures.html">torch.futures</a></li>
<li class="toctree-l1"><a class="reference internal" href="fx.html">torch.fx</a></li>
<li class="toctree-l1"><a class="reference internal" href="fx.experimental.html">torch.fx.experimental</a></li>
<li class="toctree-l1"><a class="reference internal" href="hub.html">torch.hub</a></li>
<li class="toctree-l1"><a class="reference internal" href="jit.html">torch.jit</a></li>
<li class="toctree-l1"><a class="reference internal" href="linalg.html">torch.linalg</a></li>
<li class="toctree-l1"><a class="reference internal" href="monitor.html">torch.monitor</a></li>
<li class="toctree-l1"><a class="reference internal" href="signal.html">torch.signal</a></li>
<li class="toctree-l1"><a class="reference internal" href="special.html">torch.special</a></li>
<li class="toctree-l1"><a class="reference internal" href="torch.overrides.html">torch.overrides</a></li>
<li class="toctree-l1"><a class="reference internal" href="package.html">torch.package</a></li>
<li class="toctree-l1"><a class="reference internal" href="profiler.html">torch.profiler</a></li>
<li class="toctree-l1"><a class="reference internal" href="nn.init.html">torch.nn.init</a></li>
<li class="toctree-l1"><a class="reference internal" href="nn.attention.html">torch.nn.attention</a></li>
<li class="toctree-l1"><a class="reference internal" href="onnx.html">torch.onnx</a></li>
<li class="toctree-l1"><a class="reference internal" href="optim.html">torch.optim</a></li>
<li class="toctree-l1"><a class="reference internal" href="complex_numbers.html">Complex Numbers</a></li>
<li class="toctree-l1"><a class="reference internal" href="ddp_comm_hooks.html">DDP Communication Hooks</a></li>
<li class="toctree-l1"><a class="reference internal" href="quantization.html">Quantization</a></li>
<li class="toctree-l1"><a class="reference internal" href="rpc.html">Distributed RPC Framework</a></li>
<li class="toctree-l1"><a class="reference internal" href="random.html">torch.random</a></li>
<li class="toctree-l1"><a class="reference internal" href="masked.html">torch.masked</a></li>
<li class="toctree-l1"><a class="reference internal" href="nested.html">torch.nested</a></li>
<li class="toctree-l1"><a class="reference internal" href="size.html">torch.Size</a></li>
<li class="toctree-l1"><a class="reference internal" href="sparse.html">torch.sparse</a></li>
<li class="toctree-l1"><a class="reference internal" href="storage.html">torch.Storage</a></li>
<li class="toctree-l1"><a class="reference internal" href="testing.html">torch.testing</a></li>
<li class="toctree-l1"><a class="reference internal" href="utils.html">torch.utils</a></li>
<li class="toctree-l1"><a class="reference internal" href="benchmark_utils.html">torch.utils.benchmark</a></li>
<li class="toctree-l1"><a class="reference internal" href="bottleneck.html">torch.utils.bottleneck</a></li>
<li class="toctree-l1"><a class="reference internal" href="checkpoint.html">torch.utils.checkpoint</a></li>
<li class="toctree-l1"><a class="reference internal" href="cpp_extension.html">torch.utils.cpp_extension</a></li>
<li class="toctree-l1"><a class="reference internal" href="data.html">torch.utils.data</a></li>
<li class="toctree-l1"><a class="reference internal" href="deterministic.html">torch.utils.deterministic</a></li>
<li class="toctree-l1"><a class="reference internal" href="jit_utils.html">torch.utils.jit</a></li>
<li class="toctree-l1"><a class="reference internal" href="dlpack.html">torch.utils.dlpack</a></li>
<li class="toctree-l1"><a class="reference internal" href="mobile_optimizer.html">torch.utils.mobile_optimizer</a></li>
<li class="toctree-l1"><a class="reference internal" href="model_zoo.html">torch.utils.model_zoo</a></li>
<li class="toctree-l1"><a class="reference internal" href="tensorboard.html">torch.utils.tensorboard</a></li>
<li class="toctree-l1"><a class="reference internal" href="module_tracker.html">torch.utils.module_tracker</a></li>
<li class="toctree-l1"><a class="reference internal" href="type_info.html">Type Info</a></li>
<li class="toctree-l1"><a class="reference internal" href="named_tensor.html">Named Tensors</a></li>
<li class="toctree-l1"><a class="reference internal" href="name_inference.html">Named Tensors operator coverage</a></li>
<li class="toctree-l1"><a class="reference internal" href="config_mod.html">torch.__config__</a></li>
<li class="toctree-l1"><a class="reference internal" href="future_mod.html">torch.__future__</a></li>
<li class="toctree-l1"><a class="reference internal" href="logging.html">torch._logging</a></li>
<li class="toctree-l1"><a class="reference internal" href="torch_environment_variables.html">Torch Environment Variables</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">Libraries</span></p>
<ul>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/audio/stable">torchaudio</a></li>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/data">TorchData</a></li>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/torchrec">TorchRec</a></li>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/serve">TorchServe</a></li>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/text/stable">torchtext</a></li>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/vision/stable">torchvision</a></li>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/xla/">PyTorch on XLA Devices</a></li>
</ul>

            
        </div>
      </div>
    </nav>

    <div class="pytorch-container">
      <div class="pytorch-page-level-bar" id="pytorch-page-level-bar">
        <div class="pytorch-breadcrumbs-wrapper">
          

<div role="navigation" aria-label="breadcrumbs navigation">

  <ul class="pytorch-breadcrumbs">
    
      <li>
        <a href="index.html">
          
            Docs
          
        </a> &gt;
      </li>

        
      <li>torch.backends</li>
    
    
      <li class="pytorch-breadcrumbs-aside">
        
            
            <a href="_sources/backends.rst.txt" rel="nofollow"><img src="_static/images/view-page-source-icon.svg"></a>
          
        
      </li>
    
  </ul>

  
</div>
        </div>

        <div class="pytorch-shortcuts-wrapper" id="pytorch-shortcuts-wrapper">
          Shortcuts
        </div>
      </div>

      <section data-toggle="wy-nav-shift" id="pytorch-content-wrap" class="pytorch-content-wrap">
        <div class="pytorch-content-left">

        
          <!-- Google Tag Manager (noscript) -->
          <noscript><iframe src="https://www.googletagmanager.com/ns.html?id=GTM-T8XT4PS"
          height="0" width="0" style="display:none;visibility:hidden"></iframe></noscript>
          <!-- End Google Tag Manager (noscript) -->
          
          <div class="rst-content">
          
            <div role="main" class="main-content" itemscope="itemscope" itemtype="http://schema.org/Article">
             <article itemprop="articleBody" id="pytorch-article" class="pytorch-article">
              
  <div class="section" id="module-torch.backends">
<span id="torch-backends"></span><h1>torch.backends<a class="headerlink" href="#module-torch.backends" title="Permalink to this heading">¶</a></h1>
<p><cite>torch.backends</cite> controls the behavior of various backends that PyTorch supports.</p>
<p>These backends include:</p>
<ul class="simple">
<li><p><code class="docutils literal notranslate"><span class="pre">torch.backends.cpu</span></code></p></li>
<li><p><code class="docutils literal notranslate"><span class="pre">torch.backends.cuda</span></code></p></li>
<li><p><code class="docutils literal notranslate"><span class="pre">torch.backends.cudnn</span></code></p></li>
<li><p><code class="docutils literal notranslate"><span class="pre">torch.backends.cusparselt</span></code></p></li>
<li><p><code class="docutils literal notranslate"><span class="pre">torch.backends.mha</span></code></p></li>
<li><p><code class="docutils literal notranslate"><span class="pre">torch.backends.mps</span></code></p></li>
<li><p><code class="docutils literal notranslate"><span class="pre">torch.backends.mkl</span></code></p></li>
<li><p><code class="docutils literal notranslate"><span class="pre">torch.backends.mkldnn</span></code></p></li>
<li><p><code class="docutils literal notranslate"><span class="pre">torch.backends.nnpack</span></code></p></li>
<li><p><code class="docutils literal notranslate"><span class="pre">torch.backends.openmp</span></code></p></li>
<li><p><code class="docutils literal notranslate"><span class="pre">torch.backends.opt_einsum</span></code></p></li>
<li><p><code class="docutils literal notranslate"><span class="pre">torch.backends.xeon</span></code></p></li>
</ul>
<div class="section" id="module-torch.backends.cpu">
<span id="torch-backends-cpu"></span><h2>torch.backends.cpu<a class="headerlink" href="#module-torch.backends.cpu" title="Permalink to this heading">¶</a></h2>
<dl class="py function">
<dt class="sig sig-object py" id="torch.backends.cpu.get_cpu_capability">
<span class="sig-prename descclassname"><span class="pre">torch.backends.cpu.</span></span><span class="sig-name descname"><span class="pre">get_cpu_capability</span></span><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/backends/cpu.html#get_cpu_capability"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.backends.cpu.get_cpu_capability" title="Permalink to this definition">¶</a></dt>
<dd><p>Return cpu capability as a string value.</p>
<p>Possible values:
- “DEFAULT”
- “VSX”
- “Z VECTOR”
- “NO AVX”
- “AVX2”
- “AVX512”</p>
<dl class="field-list simple">
<dt class="field-odd">Return type</dt>
<dd class="field-odd"><p><a class="reference external" href="https://docs.python.org/3/library/stdtypes.html#str" title="(in Python v3.13)">str</a></p>
</dd>
</dl>
</dd></dl>

</div>
<div class="section" id="module-torch.backends.cuda">
<span id="torch-backends-cuda"></span><h2>torch.backends.cuda<a class="headerlink" href="#module-torch.backends.cuda" title="Permalink to this heading">¶</a></h2>
<dl class="py function">
<dt class="sig sig-object py" id="torch.backends.cuda.is_built">
<span class="sig-prename descclassname"><span class="pre">torch.backends.cuda.</span></span><span class="sig-name descname"><span class="pre">is_built</span></span><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/backends/cuda.html#is_built"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.backends.cuda.is_built" title="Permalink to this definition">¶</a></dt>
<dd><p>Return whether PyTorch is built with CUDA support.</p>
<p>Note that this doesn’t necessarily mean CUDA is available; just that if this PyTorch
binary were run on a machine with working CUDA drivers and devices, we would be able to use it.</p>
</dd></dl>

<dl class="py attribute">
<dt class="sig sig-object py" id="torch.backends.cuda.matmul.allow_tf32">
<span class="sig-prename descclassname"><span class="pre">torch.backends.cuda.matmul.</span></span><span class="sig-name descname"><span class="pre">allow_tf32</span></span><a class="headerlink" href="#torch.backends.cuda.matmul.allow_tf32" title="Permalink to this definition">¶</a></dt>
<dd><p>A <a class="reference external" href="https://docs.python.org/3/library/functions.html#bool" title="(in Python v3.13)"><code class="xref py py-class docutils literal notranslate"><span class="pre">bool</span></code></a> that controls whether TensorFloat-32 tensor cores may be used in matrix
multiplications on Ampere or newer GPUs. See <a class="reference internal" href="notes/cuda.html#tf32-on-ampere"><span class="std std-ref">TensorFloat-32 (TF32) on Ampere (and later) devices</span></a>.</p>
</dd></dl>

<dl class="py attribute">
<dt class="sig sig-object py" id="torch.backends.cuda.matmul.allow_fp16_reduced_precision_reduction">
<span class="sig-prename descclassname"><span class="pre">torch.backends.cuda.matmul.</span></span><span class="sig-name descname"><span class="pre">allow_fp16_reduced_precision_reduction</span></span><a class="headerlink" href="#torch.backends.cuda.matmul.allow_fp16_reduced_precision_reduction" title="Permalink to this definition">¶</a></dt>
<dd><p>A <a class="reference external" href="https://docs.python.org/3/library/functions.html#bool" title="(in Python v3.13)"><code class="xref py py-class docutils literal notranslate"><span class="pre">bool</span></code></a> that controls whether reduced precision reductions (e.g., with fp16 accumulation type) are allowed with fp16 GEMMs.</p>
</dd></dl>

<dl class="py attribute">
<dt class="sig sig-object py" id="torch.backends.cuda.matmul.allow_bf16_reduced_precision_reduction">
<span class="sig-prename descclassname"><span class="pre">torch.backends.cuda.matmul.</span></span><span class="sig-name descname"><span class="pre">allow_bf16_reduced_precision_reduction</span></span><a class="headerlink" href="#torch.backends.cuda.matmul.allow_bf16_reduced_precision_reduction" title="Permalink to this definition">¶</a></dt>
<dd><p>A <a class="reference external" href="https://docs.python.org/3/library/functions.html#bool" title="(in Python v3.13)"><code class="xref py py-class docutils literal notranslate"><span class="pre">bool</span></code></a> that controls whether reduced precision reductions are allowed with bf16 GEMMs.</p>
</dd></dl>

<dl class="py attribute">
<dt class="sig sig-object py" id="torch.backends.cuda.cufft_plan_cache">
<span class="sig-prename descclassname"><span class="pre">torch.backends.cuda.</span></span><span class="sig-name descname"><span class="pre">cufft_plan_cache</span></span><a class="headerlink" href="#torch.backends.cuda.cufft_plan_cache" title="Permalink to this definition">¶</a></dt>
<dd><p><code class="docutils literal notranslate"><span class="pre">cufft_plan_cache</span></code> contains the cuFFT plan caches for each CUDA device.
Query a specific device <cite>i</cite>’s cache via <cite>torch.backends.cuda.cufft_plan_cache[i]</cite>.</p>
<dl class="py attribute">
<dt class="sig sig-object py" id="torch.backends.cuda.cufft_plan_cache.size">
<span class="sig-prename descclassname"><span class="pre">torch.backends.cuda.cufft_plan_cache.</span></span><span class="sig-name descname"><span class="pre">size</span></span><a class="headerlink" href="#torch.backends.cuda.cufft_plan_cache.size" title="Permalink to this definition">¶</a></dt>
<dd><p>A readonly <a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.13)"><code class="xref py py-class docutils literal notranslate"><span class="pre">int</span></code></a> that shows the number of plans currently in a cuFFT plan cache.</p>
</dd></dl>

<dl class="py attribute">
<dt class="sig sig-object py" id="torch.backends.cuda.cufft_plan_cache.max_size">
<span class="sig-prename descclassname"><span class="pre">torch.backends.cuda.cufft_plan_cache.</span></span><span class="sig-name descname"><span class="pre">max_size</span></span><a class="headerlink" href="#torch.backends.cuda.cufft_plan_cache.max_size" title="Permalink to this definition">¶</a></dt>
<dd><p>A <a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.13)"><code class="xref py py-class docutils literal notranslate"><span class="pre">int</span></code></a> that controls the capacity of a cuFFT plan cache.</p>
</dd></dl>

<dl class="py method">
<dt class="sig sig-object py" id="torch.backends.cuda.cufft_plan_cache.clear">
<span class="sig-prename descclassname"><span class="pre">torch.backends.cuda.cufft_plan_cache.</span></span><span class="sig-name descname"><span class="pre">clear</span></span><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="headerlink" href="#torch.backends.cuda.cufft_plan_cache.clear" title="Permalink to this definition">¶</a></dt>
<dd><p>Clears a cuFFT plan cache.</p>
</dd></dl>

</dd></dl>

<dl class="py function">
<dt class="sig sig-object py" id="torch.backends.cuda.preferred_blas_library">
<span class="sig-prename descclassname"><span class="pre">torch.backends.cuda.</span></span><span class="sig-name descname"><span class="pre">preferred_blas_library</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">backend</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/backends/cuda.html#preferred_blas_library"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.backends.cuda.preferred_blas_library" title="Permalink to this definition">¶</a></dt>
<dd><p>Override the library PyTorch uses for BLAS operations. Choose between cuBLAS and cuBLASLt.</p>
<div class="admonition warning">
<p class="admonition-title">Warning</p>
<p>This flag is experimental and subject to change.</p>
</div>
<p>When PyTorch runs a CUDA BLAS operation it defaults to cuBLAS even if both cuBLAS and cuBLASLt are available.
For PyTorch built for ROCm, hipBLAS and hipBLASLt may offer different performance.
This flag (a <a class="reference external" href="https://docs.python.org/3/library/stdtypes.html#str" title="(in Python v3.13)"><code class="xref py py-class docutils literal notranslate"><span class="pre">str</span></code></a>) allows overriding which BLAS library to use.</p>
<ul class="simple">
<li><p>If <cite>“cublas”</cite> is set then cuBLAS will be used wherever possible.</p></li>
<li><p>If <cite>“cublaslt”</cite> is set then cuBLASLt will be used wherever possible.</p></li>
<li><p>When no input is given, this function returns the currently preferred library.</p></li>
<li><p>User may use the environment variable TORCH_BLAS_PREFER_CUBLASLT=1 to set the preferred library to cuBLASLt
globally.
This flag only sets the initial value of the preferred library and the preferred library
may still be overridden by this function call later in your script.</p></li>
</ul>
<p>Note: When a library is preferred other libraries may still be used if the preferred library
doesn’t implement the operation(s) called.
This flag may achieve better performance if PyTorch’s library selection is incorrect
for your application’s inputs.</p>
<dl class="field-list simple">
<dt class="field-odd">Return type</dt>
<dd class="field-odd"><p><em>_BlasBackend</em></p>
</dd>
</dl>
</dd></dl>

<dl class="py function">
<dt class="sig sig-object py" id="torch.backends.cuda.preferred_linalg_library">
<span class="sig-prename descclassname"><span class="pre">torch.backends.cuda.</span></span><span class="sig-name descname"><span class="pre">preferred_linalg_library</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">backend</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/backends/cuda.html#preferred_linalg_library"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.backends.cuda.preferred_linalg_library" title="Permalink to this definition">¶</a></dt>
<dd><p>Override the heuristic PyTorch uses to choose between cuSOLVER and MAGMA for CUDA linear algebra operations.</p>
<div class="admonition warning">
<p class="admonition-title">Warning</p>
<p>This flag is experimental and subject to change.</p>
</div>
<p>When PyTorch runs a CUDA linear algebra operation it often uses the cuSOLVER or MAGMA libraries,
and if both are available it decides which to use with a heuristic.
This flag (a <a class="reference external" href="https://docs.python.org/3/library/stdtypes.html#str" title="(in Python v3.13)"><code class="xref py py-class docutils literal notranslate"><span class="pre">str</span></code></a>) allows overriding those heuristics.</p>
<ul class="simple">
<li><p>If <cite>“cusolver”</cite> is set then cuSOLVER will be used wherever possible.</p></li>
<li><p>If <cite>“magma”</cite> is set then MAGMA will be used wherever possible.</p></li>
<li><p>If <cite>“default”</cite> (the default) is set then heuristics will be used to pick between
cuSOLVER and MAGMA if both are available.</p></li>
<li><p>When no input is given, this function returns the currently preferred library.</p></li>
<li><p>User may use the environment variable TORCH_LINALG_PREFER_CUSOLVER=1 to set the preferred library to cuSOLVER
globally.
This flag only sets the initial value of the preferred library and the preferred library
may still be overridden by this function call later in your script.</p></li>
</ul>
<p>Note: When a library is preferred other libraries may still be used if the preferred library
doesn’t implement the operation(s) called.
This flag may achieve better performance if PyTorch’s heuristic library selection is incorrect
for your application’s inputs.</p>
<p>Currently supported linalg operators:</p>
<ul class="simple">
<li><p><a class="reference internal" href="generated/torch.linalg.inv.html#torch.linalg.inv" title="torch.linalg.inv"><code class="xref py py-func docutils literal notranslate"><span class="pre">torch.linalg.inv()</span></code></a></p></li>
<li><p><a class="reference internal" href="generated/torch.linalg.inv_ex.html#torch.linalg.inv_ex" title="torch.linalg.inv_ex"><code class="xref py py-func docutils literal notranslate"><span class="pre">torch.linalg.inv_ex()</span></code></a></p></li>
<li><p><a class="reference internal" href="generated/torch.linalg.cholesky.html#torch.linalg.cholesky" title="torch.linalg.cholesky"><code class="xref py py-func docutils literal notranslate"><span class="pre">torch.linalg.cholesky()</span></code></a></p></li>
<li><p><a class="reference internal" href="generated/torch.linalg.cholesky_ex.html#torch.linalg.cholesky_ex" title="torch.linalg.cholesky_ex"><code class="xref py py-func docutils literal notranslate"><span class="pre">torch.linalg.cholesky_ex()</span></code></a></p></li>
<li><p><a class="reference internal" href="generated/torch.cholesky_solve.html#torch.cholesky_solve" title="torch.cholesky_solve"><code class="xref py py-func docutils literal notranslate"><span class="pre">torch.cholesky_solve()</span></code></a></p></li>
<li><p><a class="reference internal" href="generated/torch.cholesky_inverse.html#torch.cholesky_inverse" title="torch.cholesky_inverse"><code class="xref py py-func docutils literal notranslate"><span class="pre">torch.cholesky_inverse()</span></code></a></p></li>
<li><p><a class="reference internal" href="generated/torch.linalg.lu_factor.html#torch.linalg.lu_factor" title="torch.linalg.lu_factor"><code class="xref py py-func docutils literal notranslate"><span class="pre">torch.linalg.lu_factor()</span></code></a></p></li>
<li><p><a class="reference internal" href="generated/torch.linalg.lu.html#torch.linalg.lu" title="torch.linalg.lu"><code class="xref py py-func docutils literal notranslate"><span class="pre">torch.linalg.lu()</span></code></a></p></li>
<li><p><a class="reference internal" href="generated/torch.linalg.lu_solve.html#torch.linalg.lu_solve" title="torch.linalg.lu_solve"><code class="xref py py-func docutils literal notranslate"><span class="pre">torch.linalg.lu_solve()</span></code></a></p></li>
<li><p><a class="reference internal" href="generated/torch.linalg.qr.html#torch.linalg.qr" title="torch.linalg.qr"><code class="xref py py-func docutils literal notranslate"><span class="pre">torch.linalg.qr()</span></code></a></p></li>
<li><p><a class="reference internal" href="generated/torch.linalg.eigh.html#torch.linalg.eigh" title="torch.linalg.eigh"><code class="xref py py-func docutils literal notranslate"><span class="pre">torch.linalg.eigh()</span></code></a></p></li>
<li><p><code class="xref py py-func docutils literal notranslate"><span class="pre">torch.linalg.eighvals()</span></code></p></li>
<li><p><a class="reference internal" href="generated/torch.linalg.svd.html#torch.linalg.svd" title="torch.linalg.svd"><code class="xref py py-func docutils literal notranslate"><span class="pre">torch.linalg.svd()</span></code></a></p></li>
<li><p><a class="reference internal" href="generated/torch.linalg.svdvals.html#torch.linalg.svdvals" title="torch.linalg.svdvals"><code class="xref py py-func docutils literal notranslate"><span class="pre">torch.linalg.svdvals()</span></code></a></p></li>
</ul>
<dl class="field-list simple">
<dt class="field-odd">Return type</dt>
<dd class="field-odd"><p><em>_LinalgBackend</em></p>
</dd>
</dl>
</dd></dl>

<dl class="py class">
<dt class="sig sig-object py" id="torch.backends.cuda.SDPAParams">
<em class="property"><span class="pre">class</span><span class="w"> </span></em><span class="sig-prename descclassname"><span class="pre">torch.backends.cuda.</span></span><span class="sig-name descname"><span class="pre">SDPAParams</span></span><a class="headerlink" href="#torch.backends.cuda.SDPAParams" title="Permalink to this definition">¶</a></dt>
<dd></dd></dl>

<dl class="py function">
<dt class="sig sig-object py" id="torch.backends.cuda.flash_sdp_enabled">
<span class="sig-prename descclassname"><span class="pre">torch.backends.cuda.</span></span><span class="sig-name descname"><span class="pre">flash_sdp_enabled</span></span><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/backends/cuda.html#flash_sdp_enabled"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.backends.cuda.flash_sdp_enabled" title="Permalink to this definition">¶</a></dt>
<dd><div class="admonition warning">
<p class="admonition-title">Warning</p>
<p>This flag is beta and subject to change.</p>
</div>
<p>Returns whether flash scaled dot product attention is enabled or not.</p>
</dd></dl>

<dl class="py function">
<dt class="sig sig-object py" id="torch.backends.cuda.enable_mem_efficient_sdp">
<span class="sig-prename descclassname"><span class="pre">torch.backends.cuda.</span></span><span class="sig-name descname"><span class="pre">enable_mem_efficient_sdp</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">enabled</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/backends/cuda.html#enable_mem_efficient_sdp"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.backends.cuda.enable_mem_efficient_sdp" title="Permalink to this definition">¶</a></dt>
<dd><div class="admonition warning">
<p class="admonition-title">Warning</p>
<p>This flag is beta and subject to change.</p>
</div>
<p>Enables or disables memory efficient scaled dot product attention.</p>
<dl class="field-list simple">
</dl>
</dd></dl>

<dl class="py function">
<dt class="sig sig-object py" id="torch.backends.cuda.mem_efficient_sdp_enabled">
<span class="sig-prename descclassname"><span class="pre">torch.backends.cuda.</span></span><span class="sig-name descname"><span class="pre">mem_efficient_sdp_enabled</span></span><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/backends/cuda.html#mem_efficient_sdp_enabled"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.backends.cuda.mem_efficient_sdp_enabled" title="Permalink to this definition">¶</a></dt>
<dd><div class="admonition warning">
<p class="admonition-title">Warning</p>
<p>This flag is beta and subject to change.</p>
</div>
<p>Returns whether memory efficient scaled dot product attention is enabled or not.</p>
</dd></dl>

<dl class="py function">
<dt class="sig sig-object py" id="torch.backends.cuda.enable_flash_sdp">
<span class="sig-prename descclassname"><span class="pre">torch.backends.cuda.</span></span><span class="sig-name descname"><span class="pre">enable_flash_sdp</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">enabled</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/backends/cuda.html#enable_flash_sdp"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.backends.cuda.enable_flash_sdp" title="Permalink to this definition">¶</a></dt>
<dd><div class="admonition warning">
<p class="admonition-title">Warning</p>
<p>This flag is beta and subject to change.</p>
</div>
<p>Enables or disables flash scaled dot product attention.</p>
<dl class="field-list simple">
</dl>
</dd></dl>

<dl class="py function">
<dt class="sig sig-object py" id="torch.backends.cuda.math_sdp_enabled">
<span class="sig-prename descclassname"><span class="pre">torch.backends.cuda.</span></span><span class="sig-name descname"><span class="pre">math_sdp_enabled</span></span><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/backends/cuda.html#math_sdp_enabled"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.backends.cuda.math_sdp_enabled" title="Permalink to this definition">¶</a></dt>
<dd><div class="admonition warning">
<p class="admonition-title">Warning</p>
<p>This flag is beta and subject to change.</p>
</div>
<p>Returns whether math scaled dot product attention is enabled or not.</p>
</dd></dl>

<dl class="py function">
<dt class="sig sig-object py" id="torch.backends.cuda.enable_math_sdp">
<span class="sig-prename descclassname"><span class="pre">torch.backends.cuda.</span></span><span class="sig-name descname"><span class="pre">enable_math_sdp</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">enabled</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/backends/cuda.html#enable_math_sdp"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.backends.cuda.enable_math_sdp" title="Permalink to this definition">¶</a></dt>
<dd><div class="admonition warning">
<p class="admonition-title">Warning</p>
<p>This flag is beta and subject to change.</p>
</div>
<p>Enables or disables math scaled dot product attention.</p>
<dl class="field-list simple">
</dl>
</dd></dl>

<dl class="py function">
<dt class="sig sig-object py" id="torch.backends.cuda.fp16_bf16_reduction_math_sdp_allowed">
<span class="sig-prename descclassname"><span class="pre">torch.backends.cuda.</span></span><span class="sig-name descname"><span class="pre">fp16_bf16_reduction_math_sdp_allowed</span></span><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/backends/cuda.html#fp16_bf16_reduction_math_sdp_allowed"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.backends.cuda.fp16_bf16_reduction_math_sdp_allowed" title="Permalink to this definition">¶</a></dt>
<dd><div class="admonition warning">
<p class="admonition-title">Warning</p>
<p>This flag is beta and subject to change.</p>
</div>
<p>Returns whether fp16/bf16 reduction in math scaled dot product attention is enabled or not.</p>
</dd></dl>

<dl class="py function">
<dt class="sig sig-object py" id="torch.backends.cuda.allow_fp16_bf16_reduction_math_sdp">
<span class="sig-prename descclassname"><span class="pre">torch.backends.cuda.</span></span><span class="sig-name descname"><span class="pre">allow_fp16_bf16_reduction_math_sdp</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">enabled</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/backends/cuda.html#allow_fp16_bf16_reduction_math_sdp"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.backends.cuda.allow_fp16_bf16_reduction_math_sdp" title="Permalink to this definition">¶</a></dt>
<dd><div class="admonition warning">
<p class="admonition-title">Warning</p>
<p>This flag is beta and subject to change.</p>
</div>
<p>Enables or disables fp16/bf16 reduction in math scaled dot product attention.</p>
<dl class="field-list simple">
</dl>
</dd></dl>

<dl class="py function">
<dt class="sig sig-object py" id="torch.backends.cuda.cudnn_sdp_enabled">
<span class="sig-prename descclassname"><span class="pre">torch.backends.cuda.</span></span><span class="sig-name descname"><span class="pre">cudnn_sdp_enabled</span></span><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/backends/cuda.html#cudnn_sdp_enabled"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.backends.cuda.cudnn_sdp_enabled" title="Permalink to this definition">¶</a></dt>
<dd><div class="admonition warning">
<p class="admonition-title">Warning</p>
<p>This flag is beta and subject to change.</p>
</div>
<p>Returns whether cuDNN scaled dot product attention is enabled or not.</p>
</dd></dl>

<dl class="py function">
<dt class="sig sig-object py" id="torch.backends.cuda.enable_cudnn_sdp">
<span class="sig-prename descclassname"><span class="pre">torch.backends.cuda.</span></span><span class="sig-name descname"><span class="pre">enable_cudnn_sdp</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">enabled</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/backends/cuda.html#enable_cudnn_sdp"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.backends.cuda.enable_cudnn_sdp" title="Permalink to this definition">¶</a></dt>
<dd><div class="admonition warning">
<p class="admonition-title">Warning</p>
<p>This flag is beta and subject to change.</p>
</div>
<p>Enables or disables cuDNN scaled dot product attention.</p>
<dl class="field-list simple">
</dl>
</dd></dl>

<dl class="py function">
<dt class="sig sig-object py" id="torch.backends.cuda.is_flash_attention_available">
<span class="sig-prename descclassname"><span class="pre">torch.backends.cuda.</span></span><span class="sig-name descname"><span class="pre">is_flash_attention_available</span></span><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/backends/cuda.html#is_flash_attention_available"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.backends.cuda.is_flash_attention_available" title="Permalink to this definition">¶</a></dt>
<dd><p>Check if PyTorch was built with FlashAttention for scaled_dot_product_attention.</p>
<dl class="field-list simple">
<dt class="field-odd">Returns</dt>
<dd class="field-odd"><p>True if FlashAttention is built and available; otherwise, False.</p>
</dd>
<dt class="field-even">Return type</dt>
<dd class="field-even"><p><a class="reference external" href="https://docs.python.org/3/library/functions.html#bool" title="(in Python v3.13)">bool</a></p>
</dd>
</dl>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>This function is dependent on a CUDA-enabled build of PyTorch. It will return False
in non-CUDA environments.</p>
</div>
</dd></dl>

<dl class="py function">
<dt class="sig sig-object py" id="torch.backends.cuda.can_use_flash_attention">
<span class="sig-prename descclassname"><span class="pre">torch.backends.cuda.</span></span><span class="sig-name descname"><span class="pre">can_use_flash_attention</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">params</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">debug</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">False</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/backends/cuda.html#can_use_flash_attention"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.backends.cuda.can_use_flash_attention" title="Permalink to this definition">¶</a></dt>
<dd><p>Check if FlashAttention can be utilized in scaled_dot_product_attention.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><ul class="simple">
<li><p><strong>params</strong> (<a class="reference internal" href="#torch.backends.cuda.SDPAParams" title="torch.backends.cuda._SDPAParams"><em>_SDPAParams</em></a>) – An instance of SDPAParams containing the tensors for query,
key, value, an optional attention mask, dropout rate, and
a flag indicating if the attention is causal.</p></li>
<li><p><strong>debug</strong> (<a class="reference external" href="https://docs.python.org/3/library/functions.html#bool" title="(in Python v3.13)"><em>bool</em></a>) – Whether to logging.warn debug information as to why FlashAttention could not be run.
Defaults to False.</p></li>
</ul>
</dd>
<dt class="field-even">Returns</dt>
<dd class="field-even"><p>True if FlashAttention can be used with the given parameters; otherwise, False.</p>
</dd>
<dt class="field-odd">Return type</dt>
<dd class="field-odd"><p><a class="reference external" href="https://docs.python.org/3/library/functions.html#bool" title="(in Python v3.13)">bool</a></p>
</dd>
</dl>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>This function is dependent on a CUDA-enabled build of PyTorch. It will return False
in non-CUDA environments.</p>
</div>
</dd></dl>

<dl class="py function">
<dt class="sig sig-object py" id="torch.backends.cuda.can_use_efficient_attention">
<span class="sig-prename descclassname"><span class="pre">torch.backends.cuda.</span></span><span class="sig-name descname"><span class="pre">can_use_efficient_attention</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">params</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">debug</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">False</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/backends/cuda.html#can_use_efficient_attention"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.backends.cuda.can_use_efficient_attention" title="Permalink to this definition">¶</a></dt>
<dd><p>Check if efficient_attention can be utilized in scaled_dot_product_attention.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><ul class="simple">
<li><p><strong>params</strong> (<a class="reference internal" href="#torch.backends.cuda.SDPAParams" title="torch.backends.cuda._SDPAParams"><em>_SDPAParams</em></a>) – An instance of SDPAParams containing the tensors for query,
key, value, an optional attention mask, dropout rate, and
a flag indicating if the attention is causal.</p></li>
<li><p><strong>debug</strong> (<a class="reference external" href="https://docs.python.org/3/library/functions.html#bool" title="(in Python v3.13)"><em>bool</em></a>) – Whether to logging.warn with information as to why efficient_attention could not be run.
Defaults to False.</p></li>
</ul>
</dd>
<dt class="field-even">Returns</dt>
<dd class="field-even"><p>True if efficient_attention can be used with the given parameters; otherwise, False.</p>
</dd>
<dt class="field-odd">Return type</dt>
<dd class="field-odd"><p><a class="reference external" href="https://docs.python.org/3/library/functions.html#bool" title="(in Python v3.13)">bool</a></p>
</dd>
</dl>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>This function is dependent on a CUDA-enabled build of PyTorch. It will return False
in non-CUDA environments.</p>
</div>
</dd></dl>

<dl class="py function">
<dt class="sig sig-object py" id="torch.backends.cuda.can_use_cudnn_attention">
<span class="sig-prename descclassname"><span class="pre">torch.backends.cuda.</span></span><span class="sig-name descname"><span class="pre">can_use_cudnn_attention</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">params</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">debug</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">False</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/backends/cuda.html#can_use_cudnn_attention"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.backends.cuda.can_use_cudnn_attention" title="Permalink to this definition">¶</a></dt>
<dd><p>Check if cudnn_attention can be utilized in scaled_dot_product_attention.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><ul class="simple">
<li><p><strong>params</strong> (<a class="reference internal" href="#torch.backends.cuda.SDPAParams" title="torch.backends.cuda._SDPAParams"><em>_SDPAParams</em></a>) – An instance of SDPAParams containing the tensors for query,
key, value, an optional attention mask, dropout rate, and
a flag indicating if the attention is causal.</p></li>
<li><p><strong>debug</strong> (<a class="reference external" href="https://docs.python.org/3/library/functions.html#bool" title="(in Python v3.13)"><em>bool</em></a>) – Whether to logging.warn with information as to why cuDNN attention could not be run.
Defaults to False.</p></li>
</ul>
</dd>
<dt class="field-even">Returns</dt>
<dd class="field-even"><p>True if cuDNN can be used with the given parameters; otherwise, False.</p>
</dd>
<dt class="field-odd">Return type</dt>
<dd class="field-odd"><p><a class="reference external" href="https://docs.python.org/3/library/functions.html#bool" title="(in Python v3.13)">bool</a></p>
</dd>
</dl>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>This function is dependent on a CUDA-enabled build of PyTorch. It will return False
in non-CUDA environments.</p>
</div>
</dd></dl>

<dl class="py function">
<dt class="sig sig-object py" id="torch.backends.cuda.sdp_kernel">
<span class="sig-prename descclassname"><span class="pre">torch.backends.cuda.</span></span><span class="sig-name descname"><span class="pre">sdp_kernel</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">enable_flash</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">True</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">enable_math</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">True</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">enable_mem_efficient</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">True</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">enable_cudnn</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">True</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/backends/cuda.html#sdp_kernel"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.backends.cuda.sdp_kernel" title="Permalink to this definition">¶</a></dt>
<dd><div class="admonition warning">
<p class="admonition-title">Warning</p>
<p>This flag is beta and subject to change.</p>
</div>
<p>This context manager can be used to temporarily enable or disable any of the three backends for scaled dot product attention.
Upon exiting the context manager, the previous state of the flags will be restored.</p>
<dl class="field-list simple">
</dl>
</dd></dl>

</div>
<div class="section" id="module-torch.backends.cudnn">
<span id="torch-backends-cudnn"></span><h2>torch.backends.cudnn<a class="headerlink" href="#module-torch.backends.cudnn" title="Permalink to this heading">¶</a></h2>
<dl class="py function">
<dt class="sig sig-object py" id="torch.backends.cudnn.version">
<span class="sig-prename descclassname"><span class="pre">torch.backends.cudnn.</span></span><span class="sig-name descname"><span class="pre">version</span></span><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/backends/cudnn.html#version"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.backends.cudnn.version" title="Permalink to this definition">¶</a></dt>
<dd><p>Return the version of cuDNN.</p>
</dd></dl>

<dl class="py function">
<dt class="sig sig-object py" id="torch.backends.cudnn.is_available">
<span class="sig-prename descclassname"><span class="pre">torch.backends.cudnn.</span></span><span class="sig-name descname"><span class="pre">is_available</span></span><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/backends/cudnn.html#is_available"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.backends.cudnn.is_available" title="Permalink to this definition">¶</a></dt>
<dd><p>Return a bool indicating if CUDNN is currently available.</p>
</dd></dl>

<dl class="py attribute">
<dt class="sig sig-object py" id="torch.backends.cudnn.enabled">
<span class="sig-prename descclassname"><span class="pre">torch.backends.cudnn.</span></span><span class="sig-name descname"><span class="pre">enabled</span></span><a class="headerlink" href="#torch.backends.cudnn.enabled" title="Permalink to this definition">¶</a></dt>
<dd><p>A <a class="reference external" href="https://docs.python.org/3/library/functions.html#bool" title="(in Python v3.13)"><code class="xref py py-class docutils literal notranslate"><span class="pre">bool</span></code></a> that controls whether cuDNN is enabled.</p>
</dd></dl>

<dl class="py attribute">
<dt class="sig sig-object py" id="torch.backends.cudnn.allow_tf32">
<span class="sig-prename descclassname"><span class="pre">torch.backends.cudnn.</span></span><span class="sig-name descname"><span class="pre">allow_tf32</span></span><a class="headerlink" href="#torch.backends.cudnn.allow_tf32" title="Permalink to this definition">¶</a></dt>
<dd><p>A <a class="reference external" href="https://docs.python.org/3/library/functions.html#bool" title="(in Python v3.13)"><code class="xref py py-class docutils literal notranslate"><span class="pre">bool</span></code></a> that controls where TensorFloat-32 tensor cores may be used in cuDNN
convolutions on Ampere or newer GPUs. See <a class="reference internal" href="notes/cuda.html#tf32-on-ampere"><span class="std std-ref">TensorFloat-32 (TF32) on Ampere (and later) devices</span></a>.</p>
</dd></dl>

<dl class="py attribute">
<dt class="sig sig-object py" id="torch.backends.cudnn.deterministic">
<span class="sig-prename descclassname"><span class="pre">torch.backends.cudnn.</span></span><span class="sig-name descname"><span class="pre">deterministic</span></span><a class="headerlink" href="#torch.backends.cudnn.deterministic" title="Permalink to this definition">¶</a></dt>
<dd><p>A <a class="reference external" href="https://docs.python.org/3/library/functions.html#bool" title="(in Python v3.13)"><code class="xref py py-class docutils literal notranslate"><span class="pre">bool</span></code></a> that, if True, causes cuDNN to only use deterministic convolution algorithms.
See also <a class="reference internal" href="generated/torch.are_deterministic_algorithms_enabled.html#torch.are_deterministic_algorithms_enabled" title="torch.are_deterministic_algorithms_enabled"><code class="xref py py-func docutils literal notranslate"><span class="pre">torch.are_deterministic_algorithms_enabled()</span></code></a> and
<a class="reference internal" href="generated/torch.use_deterministic_algorithms.html#torch.use_deterministic_algorithms" title="torch.use_deterministic_algorithms"><code class="xref py py-func docutils literal notranslate"><span class="pre">torch.use_deterministic_algorithms()</span></code></a>.</p>
</dd></dl>

<dl class="py attribute">
<dt class="sig sig-object py" id="torch.backends.cudnn.benchmark">
<span class="sig-prename descclassname"><span class="pre">torch.backends.cudnn.</span></span><span class="sig-name descname"><span class="pre">benchmark</span></span><a class="headerlink" href="#torch.backends.cudnn.benchmark" title="Permalink to this definition">¶</a></dt>
<dd><p>A <a class="reference external" href="https://docs.python.org/3/library/functions.html#bool" title="(in Python v3.13)"><code class="xref py py-class docutils literal notranslate"><span class="pre">bool</span></code></a> that, if True, causes cuDNN to benchmark multiple convolution algorithms
and select the fastest.</p>
</dd></dl>

<dl class="py attribute">
<dt class="sig sig-object py" id="torch.backends.cudnn.benchmark_limit">
<span class="sig-prename descclassname"><span class="pre">torch.backends.cudnn.</span></span><span class="sig-name descname"><span class="pre">benchmark_limit</span></span><a class="headerlink" href="#torch.backends.cudnn.benchmark_limit" title="Permalink to this definition">¶</a></dt>
<dd><p>A <a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.13)"><code class="xref py py-class docutils literal notranslate"><span class="pre">int</span></code></a> that specifies the maximum number of cuDNN convolution algorithms to try when
<cite>torch.backends.cudnn.benchmark</cite> is True. Set <cite>benchmark_limit</cite> to zero to try every
available algorithm. Note that this setting only affects convolutions dispatched via the
cuDNN v8 API.</p>
</dd></dl>

<span class="target" id="module-torch.backends.cudnn.rnn"></span></div>
<div class="section" id="module-torch.backends.cusparselt">
<span id="torch-backends-cusparselt"></span><h2>torch.backends.cusparselt<a class="headerlink" href="#module-torch.backends.cusparselt" title="Permalink to this heading">¶</a></h2>
<dl class="py function">
<dt class="sig sig-object py" id="torch.backends.cusparselt.version">
<span class="sig-prename descclassname"><span class="pre">torch.backends.cusparselt.</span></span><span class="sig-name descname"><span class="pre">version</span></span><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/backends/cusparselt.html#version"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.backends.cusparselt.version" title="Permalink to this definition">¶</a></dt>
<dd><p>Return the version of cuSPARSELt</p>
<dl class="field-list simple">
<dt class="field-odd">Return type</dt>
<dd class="field-odd"><p><a class="reference external" href="https://docs.python.org/3/library/typing.html#typing.Optional" title="(in Python v3.13)"><em>Optional</em></a>[<a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.13)">int</a>]</p>
</dd>
</dl>
</dd></dl>

<dl class="py function">
<dt class="sig sig-object py" id="torch.backends.cusparselt.is_available">
<span class="sig-prename descclassname"><span class="pre">torch.backends.cusparselt.</span></span><span class="sig-name descname"><span class="pre">is_available</span></span><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/backends/cusparselt.html#is_available"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.backends.cusparselt.is_available" title="Permalink to this definition">¶</a></dt>
<dd><p>Return a bool indicating if cuSPARSELt is currently available.</p>
<dl class="field-list simple">
<dt class="field-odd">Return type</dt>
<dd class="field-odd"><p><a class="reference external" href="https://docs.python.org/3/library/functions.html#bool" title="(in Python v3.13)">bool</a></p>
</dd>
</dl>
</dd></dl>

</div>
<div class="section" id="module-torch.backends.mha">
<span id="torch-backends-mha"></span><h2>torch.backends.mha<a class="headerlink" href="#module-torch.backends.mha" title="Permalink to this heading">¶</a></h2>
<dl class="py function">
<dt class="sig sig-object py" id="torch.backends.mha.get_fastpath_enabled">
<span class="sig-prename descclassname"><span class="pre">torch.backends.mha.</span></span><span class="sig-name descname"><span class="pre">get_fastpath_enabled</span></span><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/backends/mha.html#get_fastpath_enabled"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.backends.mha.get_fastpath_enabled" title="Permalink to this definition">¶</a></dt>
<dd><p>Returns whether fast path for TransformerEncoder and MultiHeadAttention
is enabled, or <code class="docutils literal notranslate"><span class="pre">True</span></code> if jit is scripting.</p>
<dl class="simple">
<dt>..note:</dt><dd><p>The fastpath might not be run even if <code class="docutils literal notranslate"><span class="pre">get_fastpath_enabled</span></code> returns
<code class="docutils literal notranslate"><span class="pre">True</span></code> unless all conditions on inputs are met.</p>
</dd>
</dl>
<dl class="field-list simple">
<dt class="field-odd">Return type</dt>
<dd class="field-odd"><p><a class="reference external" href="https://docs.python.org/3/library/functions.html#bool" title="(in Python v3.13)">bool</a></p>
</dd>
</dl>
</dd></dl>

<dl class="py function">
<dt class="sig sig-object py" id="torch.backends.mha.set_fastpath_enabled">
<span class="sig-prename descclassname"><span class="pre">torch.backends.mha.</span></span><span class="sig-name descname"><span class="pre">set_fastpath_enabled</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">value</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/backends/mha.html#set_fastpath_enabled"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.backends.mha.set_fastpath_enabled" title="Permalink to this definition">¶</a></dt>
<dd><p>Sets whether fast path is enabled</p>
<dl class="field-list simple">
</dl>
</dd></dl>

</div>
<div class="section" id="module-torch.backends.mps">
<span id="torch-backends-mps"></span><h2>torch.backends.mps<a class="headerlink" href="#module-torch.backends.mps" title="Permalink to this heading">¶</a></h2>
<dl class="py function">
<dt class="sig sig-object py" id="torch.backends.mps.is_available">
<span class="sig-prename descclassname"><span class="pre">torch.backends.mps.</span></span><span class="sig-name descname"><span class="pre">is_available</span></span><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/backends/mps.html#is_available"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.backends.mps.is_available" title="Permalink to this definition">¶</a></dt>
<dd><p>Return a bool indicating if MPS is currently available.</p>
<dl class="field-list simple">
<dt class="field-odd">Return type</dt>
<dd class="field-odd"><p><a class="reference external" href="https://docs.python.org/3/library/functions.html#bool" title="(in Python v3.13)">bool</a></p>
</dd>
</dl>
</dd></dl>

<dl class="py function">
<dt class="sig sig-object py" id="torch.backends.mps.is_built">
<span class="sig-prename descclassname"><span class="pre">torch.backends.mps.</span></span><span class="sig-name descname"><span class="pre">is_built</span></span><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/backends/mps.html#is_built"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.backends.mps.is_built" title="Permalink to this definition">¶</a></dt>
<dd><p>Return whether PyTorch is built with MPS support.</p>
<p>Note that this doesn’t necessarily mean MPS is available; just that
if this PyTorch binary were run a machine with working MPS drivers
and devices, we would be able to use it.</p>
<dl class="field-list simple">
<dt class="field-odd">Return type</dt>
<dd class="field-odd"><p><a class="reference external" href="https://docs.python.org/3/library/functions.html#bool" title="(in Python v3.13)">bool</a></p>
</dd>
</dl>
</dd></dl>

</div>
<div class="section" id="module-torch.backends.mkl">
<span id="torch-backends-mkl"></span><h2>torch.backends.mkl<a class="headerlink" href="#module-torch.backends.mkl" title="Permalink to this heading">¶</a></h2>
<dl class="py function">
<dt class="sig sig-object py" id="torch.backends.mkl.is_available">
<span class="sig-prename descclassname"><span class="pre">torch.backends.mkl.</span></span><span class="sig-name descname"><span class="pre">is_available</span></span><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/backends/mkl.html#is_available"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.backends.mkl.is_available" title="Permalink to this definition">¶</a></dt>
<dd><p>Return whether PyTorch is built with MKL support.</p>
</dd></dl>

<dl class="py class">
<dt class="sig sig-object py" id="torch.backends.mkl.verbose">
<em class="property"><span class="pre">class</span><span class="w"> </span></em><span class="sig-prename descclassname"><span class="pre">torch.backends.mkl.</span></span><span class="sig-name descname"><span class="pre">verbose</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">enable</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/backends/mkl.html#verbose"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.backends.mkl.verbose" title="Permalink to this definition">¶</a></dt>
<dd><p>On-demand oneMKL verbosing functionality.</p>
<p>To make it easier to debug performance issues, oneMKL can dump verbose
messages containing execution information like duration while executing
the kernel. The verbosing functionality can be invoked via an environment
variable named <cite>MKL_VERBOSE</cite>. However, this methodology dumps messages in
all steps. Those are a large amount of verbose messages. Moreover, for
investigating the performance issues, generally taking verbose messages
for one single iteration is enough. This on-demand verbosing functionality
makes it possible to control scope for verbose message dumping. In the
following example, verbose messages will be dumped out for the second
inference only.</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="kn">import</span> <span class="nn">torch</span>
<span class="n">model</span><span class="p">(</span><span class="n">data</span><span class="p">)</span>
<span class="k">with</span> <span class="n">torch</span><span class="o">.</span><span class="n">backends</span><span class="o">.</span><span class="n">mkl</span><span class="o">.</span><span class="n">verbose</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">backends</span><span class="o">.</span><span class="n">mkl</span><span class="o">.</span><span class="n">VERBOSE_ON</span><span class="p">):</span>
    <span class="n">model</span><span class="p">(</span><span class="n">data</span><span class="p">)</span>
</pre></div>
</div>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><p><strong>level</strong> – Verbose level
- <code class="docutils literal notranslate"><span class="pre">VERBOSE_OFF</span></code>: Disable verbosing
- <code class="docutils literal notranslate"><span class="pre">VERBOSE_ON</span></code>:  Enable verbosing</p>
</dd>
</dl>
</dd></dl>

</div>
<div class="section" id="module-torch.backends.mkldnn">
<span id="torch-backends-mkldnn"></span><h2>torch.backends.mkldnn<a class="headerlink" href="#module-torch.backends.mkldnn" title="Permalink to this heading">¶</a></h2>
<dl class="py function">
<dt class="sig sig-object py" id="torch.backends.mkldnn.is_available">
<span class="sig-prename descclassname"><span class="pre">torch.backends.mkldnn.</span></span><span class="sig-name descname"><span class="pre">is_available</span></span><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/backends/mkldnn.html#is_available"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.backends.mkldnn.is_available" title="Permalink to this definition">¶</a></dt>
<dd><p>Return whether PyTorch is built with MKL-DNN support.</p>
</dd></dl>

<dl class="py class">
<dt class="sig sig-object py" id="torch.backends.mkldnn.verbose">
<em class="property"><span class="pre">class</span><span class="w"> </span></em><span class="sig-prename descclassname"><span class="pre">torch.backends.mkldnn.</span></span><span class="sig-name descname"><span class="pre">verbose</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">level</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/backends/mkldnn.html#verbose"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.backends.mkldnn.verbose" title="Permalink to this definition">¶</a></dt>
<dd><p>On-demand oneDNN (former MKL-DNN) verbosing functionality.</p>
<p>To make it easier to debug performance issues, oneDNN can dump verbose
messages containing information like kernel size, input data size and
execution duration while executing the kernel. The verbosing functionality
can be invoked via an environment variable named <cite>DNNL_VERBOSE</cite>. However,
this methodology dumps messages in all steps. Those are a large amount of
verbose messages. Moreover, for investigating the performance issues,
generally taking verbose messages for one single iteration is enough.
This on-demand verbosing functionality makes it possible to control scope
for verbose message dumping. In the following example, verbose messages
will be dumped out for the second inference only.</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="kn">import</span> <span class="nn">torch</span>
<span class="n">model</span><span class="p">(</span><span class="n">data</span><span class="p">)</span>
<span class="k">with</span> <span class="n">torch</span><span class="o">.</span><span class="n">backends</span><span class="o">.</span><span class="n">mkldnn</span><span class="o">.</span><span class="n">verbose</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">backends</span><span class="o">.</span><span class="n">mkldnn</span><span class="o">.</span><span class="n">VERBOSE_ON</span><span class="p">):</span>
    <span class="n">model</span><span class="p">(</span><span class="n">data</span><span class="p">)</span>
</pre></div>
</div>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><p><strong>level</strong> – Verbose level
- <code class="docutils literal notranslate"><span class="pre">VERBOSE_OFF</span></code>: Disable verbosing
- <code class="docutils literal notranslate"><span class="pre">VERBOSE_ON</span></code>:  Enable verbosing
- <code class="docutils literal notranslate"><span class="pre">VERBOSE_ON_CREATION</span></code>: Enable verbosing, including oneDNN kernel creation</p>
</dd>
</dl>
</dd></dl>

</div>
<div class="section" id="module-torch.backends.nnpack">
<span id="torch-backends-nnpack"></span><h2>torch.backends.nnpack<a class="headerlink" href="#module-torch.backends.nnpack" title="Permalink to this heading">¶</a></h2>
<dl class="py function">
<dt class="sig sig-object py" id="torch.backends.nnpack.is_available">
<span class="sig-prename descclassname"><span class="pre">torch.backends.nnpack.</span></span><span class="sig-name descname"><span class="pre">is_available</span></span><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/backends/nnpack.html#is_available"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.backends.nnpack.is_available" title="Permalink to this definition">¶</a></dt>
<dd><p>Return whether PyTorch is built with NNPACK support.</p>
</dd></dl>

<dl class="py function">
<dt class="sig sig-object py" id="torch.backends.nnpack.flags">
<span class="sig-prename descclassname"><span class="pre">torch.backends.nnpack.</span></span><span class="sig-name descname"><span class="pre">flags</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">enabled</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">False</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/backends/nnpack.html#flags"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.backends.nnpack.flags" title="Permalink to this definition">¶</a></dt>
<dd><p>Context manager for setting if nnpack is enabled globally</p>
</dd></dl>

<dl class="py function">
<dt class="sig sig-object py" id="torch.backends.nnpack.set_flags">
<span class="sig-prename descclassname"><span class="pre">torch.backends.nnpack.</span></span><span class="sig-name descname"><span class="pre">set_flags</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">_enabled</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/backends/nnpack.html#set_flags"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.backends.nnpack.set_flags" title="Permalink to this definition">¶</a></dt>
<dd><p>Set if nnpack is enabled globally</p>
</dd></dl>

</div>
<div class="section" id="module-torch.backends.openmp">
<span id="torch-backends-openmp"></span><h2>torch.backends.openmp<a class="headerlink" href="#module-torch.backends.openmp" title="Permalink to this heading">¶</a></h2>
<dl class="py function">
<dt class="sig sig-object py" id="torch.backends.openmp.is_available">
<span class="sig-prename descclassname"><span class="pre">torch.backends.openmp.</span></span><span class="sig-name descname"><span class="pre">is_available</span></span><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/backends/openmp.html#is_available"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.backends.openmp.is_available" title="Permalink to this definition">¶</a></dt>
<dd><p>Return whether PyTorch is built with OpenMP support.</p>
</dd></dl>

<span class="target" id="module-torch.backends.quantized"></span><span class="target" id="module-torch.backends.xnnpack"></span></div>
<div class="section" id="module-torch.backends.opt_einsum">
<span id="torch-backends-opt-einsum"></span><h2>torch.backends.opt_einsum<a class="headerlink" href="#module-torch.backends.opt_einsum" title="Permalink to this heading">¶</a></h2>
<dl class="py function">
<dt class="sig sig-object py" id="torch.backends.opt_einsum.is_available">
<span class="sig-prename descclassname"><span class="pre">torch.backends.opt_einsum.</span></span><span class="sig-name descname"><span class="pre">is_available</span></span><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/backends/opt_einsum.html#is_available"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.backends.opt_einsum.is_available" title="Permalink to this definition">¶</a></dt>
<dd><p>Return a bool indicating if opt_einsum is currently available.</p>
<dl class="field-list simple">
<dt class="field-odd">Return type</dt>
<dd class="field-odd"><p><a class="reference external" href="https://docs.python.org/3/library/functions.html#bool" title="(in Python v3.13)">bool</a></p>
</dd>
</dl>
</dd></dl>

<dl class="py function">
<dt class="sig sig-object py" id="torch.backends.opt_einsum.get_opt_einsum">
<span class="sig-prename descclassname"><span class="pre">torch.backends.opt_einsum.</span></span><span class="sig-name descname"><span class="pre">get_opt_einsum</span></span><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/backends/opt_einsum.html#get_opt_einsum"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.backends.opt_einsum.get_opt_einsum" title="Permalink to this definition">¶</a></dt>
<dd><p>Return the opt_einsum package if opt_einsum is currently available, else None.</p>
<dl class="field-list simple">
<dt class="field-odd">Return type</dt>
<dd class="field-odd"><p><a class="reference external" href="https://docs.python.org/3/library/typing.html#typing.Any" title="(in Python v3.13)"><em>Any</em></a></p>
</dd>
</dl>
</dd></dl>

<dl class="py attribute">
<dt class="sig sig-object py" id="torch.backends.opt_einsum.enabled">
<span class="sig-prename descclassname"><span class="pre">torch.backends.opt_einsum.</span></span><span class="sig-name descname"><span class="pre">enabled</span></span><a class="headerlink" href="#torch.backends.opt_einsum.enabled" title="Permalink to this definition">¶</a></dt>
<dd><p>A :class:<code class="docutils literal notranslate"><span class="pre">bool</span></code> that controls whether opt_einsum is enabled (<code class="docutils literal notranslate"><span class="pre">True</span></code> by default). If so,
torch.einsum will use opt_einsum (<a class="reference external" href="https://optimized-einsum.readthedocs.io/en/stable/path_finding.html">https://optimized-einsum.readthedocs.io/en/stable/path_finding.html</a>)
if available to calculate an optimal path of contraction for faster performance.</p>
<p>If opt_einsum is not available, torch.einsum will fall back to the default contraction path
of left to right.</p>
</dd></dl>

<dl class="py attribute">
<dt class="sig sig-object py" id="torch.backends.opt_einsum.strategy">
<span class="sig-prename descclassname"><span class="pre">torch.backends.opt_einsum.</span></span><span class="sig-name descname"><span class="pre">strategy</span></span><a class="headerlink" href="#torch.backends.opt_einsum.strategy" title="Permalink to this definition">¶</a></dt>
<dd><p>A :class:<code class="docutils literal notranslate"><span class="pre">str</span></code> that specifies which strategies to try when <code class="docutils literal notranslate"><span class="pre">torch.backends.opt_einsum.enabled</span></code>
is <code class="docutils literal notranslate"><span class="pre">True</span></code>. By default, torch.einsum will try the “auto” strategy, but the “greedy” and “optimal”
strategies are also supported. Note that the “optimal” strategy is factorial on the number of
inputs as it tries all possible paths. See more details in opt_einsum’s docs
(<a class="reference external" href="https://optimized-einsum.readthedocs.io/en/stable/path_finding.html">https://optimized-einsum.readthedocs.io/en/stable/path_finding.html</a>).</p>
</dd></dl>

</div>
<div class="section" id="module-torch.backends.xeon">
<span id="torch-backends-xeon"></span><h2>torch.backends.xeon<a class="headerlink" href="#module-torch.backends.xeon" title="Permalink to this heading">¶</a></h2>
<span class="target" id="module-torch.backends.xeon.run_cpu"></span></div>
</div>


             </article>
             
            </div>
            <footer>
  
    <div class="rst-footer-buttons" role="navigation" aria-label="footer navigation">
      
        <a href="export.html" class="btn btn-neutral float-right" title="torch.export" accesskey="n" rel="next">Next <img src="_static/images/chevron-right-orange.svg" class="next-page"></a>
      
      
        <a href="meta.html" class="btn btn-neutral" title="Meta device" accesskey="p" rel="prev"><img src="_static/images/chevron-right-orange.svg" class="previous-page"> Previous</a>
      
    </div>
  

    <hr>

  
  <div role="contentinfo">
    <p>
        &copy; Copyright 2023, PyTorch Contributors.

    </p>
  </div>
    
      <div>
        Built with <a href="http://sphinx-doc.org/">Sphinx</a> using a <a href="https://github.com/rtfd/sphinx_rtd_theme">theme</a> provided by <a href="https://readthedocs.org">Read the Docs</a>.
      </div>
     

</footer>

          </div>
<script>

var match = window.location.href.match(/\/_[a-zA-Z0-9_]*.html|_dynamo/gi);
var url = window.location.href.lastIndexOf(match[match.length-1]);

if (url)
  {
    var div = '<div class="admonition note"><p class="admonition-title">Note</p><p><i class="fa fa-exclamation-circle" aria-hidden="true">&nbsp</i> This page describes an internal API which is not intended to be used outside of the PyTorch codebase and can be modified or removed without notice.</p></div>'
    document.getElementById("pytorch-article").insertAdjacentHTML('afterBegin', div)
  }
</script>
        </div>

        <div class="pytorch-content-right" id="pytorch-content-right">
          <div class="pytorch-right-menu" id="pytorch-right-menu">
            <div class="pytorch-side-scroll" id="pytorch-side-scroll-right">
              <ul>
<li><a class="reference internal" href="#">torch.backends</a><ul>
<li><a class="reference internal" href="#module-torch.backends.cpu">torch.backends.cpu</a><ul>
<li><a class="reference internal" href="#torch.backends.cpu.get_cpu_capability"><code class="docutils literal notranslate"><span class="pre">get_cpu_capability()</span></code></a></li>
</ul>
</li>
<li><a class="reference internal" href="#module-torch.backends.cuda">torch.backends.cuda</a><ul>
<li><a class="reference internal" href="#torch.backends.cuda.is_built"><code class="docutils literal notranslate"><span class="pre">is_built()</span></code></a></li>
<li><a class="reference internal" href="#torch.backends.cuda.matmul.allow_tf32"><code class="docutils literal notranslate"><span class="pre">allow_tf32</span></code></a></li>
<li><a class="reference internal" href="#torch.backends.cuda.matmul.allow_fp16_reduced_precision_reduction"><code class="docutils literal notranslate"><span class="pre">allow_fp16_reduced_precision_reduction</span></code></a></li>
<li><a class="reference internal" href="#torch.backends.cuda.matmul.allow_bf16_reduced_precision_reduction"><code class="docutils literal notranslate"><span class="pre">allow_bf16_reduced_precision_reduction</span></code></a></li>
<li><a class="reference internal" href="#torch.backends.cuda.cufft_plan_cache"><code class="docutils literal notranslate"><span class="pre">cufft_plan_cache</span></code></a></li>
<li><a class="reference internal" href="#torch.backends.cuda.cufft_plan_cache.size"><code class="docutils literal notranslate"><span class="pre">size</span></code></a></li>
<li><a class="reference internal" href="#torch.backends.cuda.cufft_plan_cache.max_size"><code class="docutils literal notranslate"><span class="pre">max_size</span></code></a></li>
<li><a class="reference internal" href="#torch.backends.cuda.cufft_plan_cache.clear"><code class="docutils literal notranslate"><span class="pre">clear()</span></code></a></li>
<li><a class="reference internal" href="#torch.backends.cuda.preferred_blas_library"><code class="docutils literal notranslate"><span class="pre">preferred_blas_library()</span></code></a></li>
<li><a class="reference internal" href="#torch.backends.cuda.preferred_linalg_library"><code class="docutils literal notranslate"><span class="pre">preferred_linalg_library()</span></code></a></li>
<li><a class="reference internal" href="#torch.backends.cuda.SDPAParams"><code class="docutils literal notranslate"><span class="pre">SDPAParams</span></code></a></li>
<li><a class="reference internal" href="#torch.backends.cuda.flash_sdp_enabled"><code class="docutils literal notranslate"><span class="pre">flash_sdp_enabled()</span></code></a></li>
<li><a class="reference internal" href="#torch.backends.cuda.enable_mem_efficient_sdp"><code class="docutils literal notranslate"><span class="pre">enable_mem_efficient_sdp()</span></code></a></li>
<li><a class="reference internal" href="#torch.backends.cuda.mem_efficient_sdp_enabled"><code class="docutils literal notranslate"><span class="pre">mem_efficient_sdp_enabled()</span></code></a></li>
<li><a class="reference internal" href="#torch.backends.cuda.enable_flash_sdp"><code class="docutils literal notranslate"><span class="pre">enable_flash_sdp()</span></code></a></li>
<li><a class="reference internal" href="#torch.backends.cuda.math_sdp_enabled"><code class="docutils literal notranslate"><span class="pre">math_sdp_enabled()</span></code></a></li>
<li><a class="reference internal" href="#torch.backends.cuda.enable_math_sdp"><code class="docutils literal notranslate"><span class="pre">enable_math_sdp()</span></code></a></li>
<li><a class="reference internal" href="#torch.backends.cuda.fp16_bf16_reduction_math_sdp_allowed"><code class="docutils literal notranslate"><span class="pre">fp16_bf16_reduction_math_sdp_allowed()</span></code></a></li>
<li><a class="reference internal" href="#torch.backends.cuda.allow_fp16_bf16_reduction_math_sdp"><code class="docutils literal notranslate"><span class="pre">allow_fp16_bf16_reduction_math_sdp()</span></code></a></li>
<li><a class="reference internal" href="#torch.backends.cuda.cudnn_sdp_enabled"><code class="docutils literal notranslate"><span class="pre">cudnn_sdp_enabled()</span></code></a></li>
<li><a class="reference internal" href="#torch.backends.cuda.enable_cudnn_sdp"><code class="docutils literal notranslate"><span class="pre">enable_cudnn_sdp()</span></code></a></li>
<li><a class="reference internal" href="#torch.backends.cuda.is_flash_attention_available"><code class="docutils literal notranslate"><span class="pre">is_flash_attention_available()</span></code></a></li>
<li><a class="reference internal" href="#torch.backends.cuda.can_use_flash_attention"><code class="docutils literal notranslate"><span class="pre">can_use_flash_attention()</span></code></a></li>
<li><a class="reference internal" href="#torch.backends.cuda.can_use_efficient_attention"><code class="docutils literal notranslate"><span class="pre">can_use_efficient_attention()</span></code></a></li>
<li><a class="reference internal" href="#torch.backends.cuda.can_use_cudnn_attention"><code class="docutils literal notranslate"><span class="pre">can_use_cudnn_attention()</span></code></a></li>
<li><a class="reference internal" href="#torch.backends.cuda.sdp_kernel"><code class="docutils literal notranslate"><span class="pre">sdp_kernel()</span></code></a></li>
</ul>
</li>
<li><a class="reference internal" href="#module-torch.backends.cudnn">torch.backends.cudnn</a><ul>
<li><a class="reference internal" href="#torch.backends.cudnn.version"><code class="docutils literal notranslate"><span class="pre">version()</span></code></a></li>
<li><a class="reference internal" href="#torch.backends.cudnn.is_available"><code class="docutils literal notranslate"><span class="pre">is_available()</span></code></a></li>
<li><a class="reference internal" href="#torch.backends.cudnn.enabled"><code class="docutils literal notranslate"><span class="pre">enabled</span></code></a></li>
<li><a class="reference internal" href="#torch.backends.cudnn.allow_tf32"><code class="docutils literal notranslate"><span class="pre">allow_tf32</span></code></a></li>
<li><a class="reference internal" href="#torch.backends.cudnn.deterministic"><code class="docutils literal notranslate"><span class="pre">deterministic</span></code></a></li>
<li><a class="reference internal" href="#torch.backends.cudnn.benchmark"><code class="docutils literal notranslate"><span class="pre">benchmark</span></code></a></li>
<li><a class="reference internal" href="#torch.backends.cudnn.benchmark_limit"><code class="docutils literal notranslate"><span class="pre">benchmark_limit</span></code></a></li>
</ul>
</li>
<li><a class="reference internal" href="#module-torch.backends.cusparselt">torch.backends.cusparselt</a><ul>
<li><a class="reference internal" href="#torch.backends.cusparselt.version"><code class="docutils literal notranslate"><span class="pre">version()</span></code></a></li>
<li><a class="reference internal" href="#torch.backends.cusparselt.is_available"><code class="docutils literal notranslate"><span class="pre">is_available()</span></code></a></li>
</ul>
</li>
<li><a class="reference internal" href="#module-torch.backends.mha">torch.backends.mha</a><ul>
<li><a class="reference internal" href="#torch.backends.mha.get_fastpath_enabled"><code class="docutils literal notranslate"><span class="pre">get_fastpath_enabled()</span></code></a></li>
<li><a class="reference internal" href="#torch.backends.mha.set_fastpath_enabled"><code class="docutils literal notranslate"><span class="pre">set_fastpath_enabled()</span></code></a></li>
</ul>
</li>
<li><a class="reference internal" href="#module-torch.backends.mps">torch.backends.mps</a><ul>
<li><a class="reference internal" href="#torch.backends.mps.is_available"><code class="docutils literal notranslate"><span class="pre">is_available()</span></code></a></li>
<li><a class="reference internal" href="#torch.backends.mps.is_built"><code class="docutils literal notranslate"><span class="pre">is_built()</span></code></a></li>
</ul>
</li>
<li><a class="reference internal" href="#module-torch.backends.mkl">torch.backends.mkl</a><ul>
<li><a class="reference internal" href="#torch.backends.mkl.is_available"><code class="docutils literal notranslate"><span class="pre">is_available()</span></code></a></li>
<li><a class="reference internal" href="#torch.backends.mkl.verbose"><code class="docutils literal notranslate"><span class="pre">verbose</span></code></a></li>
</ul>
</li>
<li><a class="reference internal" href="#module-torch.backends.mkldnn">torch.backends.mkldnn</a><ul>
<li><a class="reference internal" href="#torch.backends.mkldnn.is_available"><code class="docutils literal notranslate"><span class="pre">is_available()</span></code></a></li>
<li><a class="reference internal" href="#torch.backends.mkldnn.verbose"><code class="docutils literal notranslate"><span class="pre">verbose</span></code></a></li>
</ul>
</li>
<li><a class="reference internal" href="#module-torch.backends.nnpack">torch.backends.nnpack</a><ul>
<li><a class="reference internal" href="#torch.backends.nnpack.is_available"><code class="docutils literal notranslate"><span class="pre">is_available()</span></code></a></li>
<li><a class="reference internal" href="#torch.backends.nnpack.flags"><code class="docutils literal notranslate"><span class="pre">flags()</span></code></a></li>
<li><a class="reference internal" href="#torch.backends.nnpack.set_flags"><code class="docutils literal notranslate"><span class="pre">set_flags()</span></code></a></li>
</ul>
</li>
<li><a class="reference internal" href="#module-torch.backends.openmp">torch.backends.openmp</a><ul>
<li><a class="reference internal" href="#torch.backends.openmp.is_available"><code class="docutils literal notranslate"><span class="pre">is_available()</span></code></a></li>
</ul>
</li>
<li><a class="reference internal" href="#module-torch.backends.opt_einsum">torch.backends.opt_einsum</a><ul>
<li><a class="reference internal" href="#torch.backends.opt_einsum.is_available"><code class="docutils literal notranslate"><span class="pre">is_available()</span></code></a></li>
<li><a class="reference internal" href="#torch.backends.opt_einsum.get_opt_einsum"><code class="docutils literal notranslate"><span class="pre">get_opt_einsum()</span></code></a></li>
<li><a class="reference internal" href="#torch.backends.opt_einsum.enabled"><code class="docutils literal notranslate"><span class="pre">enabled</span></code></a></li>
<li><a class="reference internal" href="#torch.backends.opt_einsum.strategy"><code class="docutils literal notranslate"><span class="pre">strategy</span></code></a></li>
</ul>
</li>
<li><a class="reference internal" href="#module-torch.backends.xeon">torch.backends.xeon</a></li>
</ul>
</li>
</ul>

            </div>
          </div>
        </div>
      </section>
    </div>

  
       <script type="text/javascript" id="documentation_options" data-url_root="./" src="_static/documentation_options.js"></script>
         <script data-url_root="./" id="documentation_options" src="_static/documentation_options.js"></script>
         <script src="_static/jquery.js"></script>
         <script src="_static/underscore.js"></script>
         <script src="_static/_sphinx_javascript_frameworks_compat.js"></script>
         <script src="_static/doctools.js"></script>
         <script src="_static/sphinx_highlight.js"></script>
         <script src="_static/clipboard.min.js"></script>
         <script src="_static/copybutton.js"></script>
     

  <script type="text/javascript" src="_static/js/vendor/popper.min.js"></script>
  <script type="text/javascript" src="_static/js/vendor/bootstrap.min.js"></script>
  <script src="https://cdnjs.cloudflare.com/ajax/libs/list.js/1.5.0/list.min.js"></script>
  <script type="text/javascript" src="_static/js/theme.js"></script>

  <script type="text/javascript">
      jQuery(function () {
          SphinxRtdTheme.Navigation.enable(true);
      });
  </script>
 
<script script type="text/javascript">
  var collapsedSections = ['Developer Notes', 'Language Bindings', 'Libraries', 'Community'];
</script>

<img height="1" width="1" style="border-style:none;" alt="" src="https://www.googleadservices.com/pagead/conversion/795629140/?label=txkmCPmdtosBENSssfsC&amp;guid=ON&amp;script=0"/>


  <!-- Begin Footer -->

  <div class="container-fluid docs-tutorials-resources" id="docs-tutorials-resources">
    <div class="container">
      <div class="row">
        <div class="col-md-4 text-center">
          <h2>Docs</h2>
          <p>Access comprehensive developer documentation for PyTorch</p>
          <a class="with-right-arrow" href="https://pytorch.org/docs/stable/index.html">View Docs</a>
        </div>

        <div class="col-md-4 text-center">
          <h2>Tutorials</h2>
          <p>Get in-depth tutorials for beginners and advanced developers</p>
          <a class="with-right-arrow" href="https://pytorch.org/tutorials">View Tutorials</a>
        </div>

        <div class="col-md-4 text-center">
          <h2>Resources</h2>
          <p>Find development resources and get your questions answered</p>
          <a class="with-right-arrow" href="https://pytorch.org/resources">View Resources</a>
        </div>
      </div>
    </div>
  </div>

  <footer class="site-footer">
    <div class="container footer-container">
      <div class="footer-logo-wrapper">
        <a href="https://pytorch.org/" class="footer-logo"></a>
      </div>

      <div class="footer-links-wrapper">
        <div class="footer-links-col">
          <ul>
            <li class="list-title"><a href="https://pytorch.org/">PyTorch</a></li>
            <li><a href="https://pytorch.org/get-started">Get Started</a></li>
            <li><a href="https://pytorch.org/features">Features</a></li>
            <li><a href="https://pytorch.org/ecosystem">Ecosystem</a></li>
            <li><a href="https://pytorch.org/blog/">Blog</a></li>
            <li><a href="https://github.com/pytorch/pytorch/blob/master/CONTRIBUTING.md">Contributing</a></li>
          </ul>
        </div>

        <div class="footer-links-col">
          <ul>
            <li class="list-title"><a href="https://pytorch.org/resources">Resources</a></li>
            <li><a href="https://pytorch.org/tutorials">Tutorials</a></li>
            <li><a href="https://pytorch.org/docs/stable/index.html">Docs</a></li>
            <li><a href="https://discuss.pytorch.org" target="_blank">Discuss</a></li>
            <li><a href="https://github.com/pytorch/pytorch/issues" target="_blank">Github Issues</a></li>
            <li><a href="https://pytorch.org/assets/brand-guidelines/PyTorch-Brand-Guidelines.pdf" target="_blank">Brand Guidelines</a></li>
          </ul>
        </div>

        <div class="footer-links-col">
          <ul>
            <li class="list-title">Stay up to date</li>
            <li><a href="https://www.facebook.com/pytorch" target="_blank">Facebook</a></li>
            <li><a href="https://twitter.com/pytorch" target="_blank">Twitter</a></li>
            <li><a href="https://www.youtube.com/pytorch" target="_blank">YouTube</a></li>
            <li><a href="https://www.linkedin.com/company/pytorch" target="_blank">LinkedIn</a></li>
          </ul>  
          </div>

        <div class="footer-links-col">
          <ul>
            <li class="list-title">PyTorch Podcasts</li>
            <li><a href="https://open.spotify.com/show/6UzHKeiy368jKfQMKKvJY5" target="_blank">Spotify</a></li>
            <li><a href="https://podcasts.apple.com/us/podcast/pytorch-developer-podcast/id1566080008" target="_blank">Apple</a></li>
            <li><a href="https://www.google.com/podcasts?feed=aHR0cHM6Ly9mZWVkcy5zaW1wbGVjYXN0LmNvbS9PQjVGa0lsOA%3D%3D" target="_blank">Google</a></li>
            <li><a href="https://music.amazon.com/podcasts/7a4e6f0e-26c2-49e9-a478-41bd244197d0/PyTorch-Developer-Podcast?" target="_blank">Amazon</a></li>
          </ul>
         </div>
        </div>
        
        <div class="privacy-policy">
          <ul>
            <li class="privacy-policy-links"><a href="https://www.linuxfoundation.org/terms/" target="_blank">Terms</a></li>
            <li class="privacy-policy-links">|</li>
            <li class="privacy-policy-links"><a href="https://www.linuxfoundation.org/privacy-policy/" target="_blank">Privacy</a></li>
          </ul>
        </div>
        <div class="copyright">
        <p>© Copyright The Linux Foundation. The PyTorch Foundation is a project of The Linux Foundation.
          For web site terms of use, trademark policy and other policies applicable to The PyTorch Foundation please see
          <a href="https://www.linuxfoundation.org/policies/">www.linuxfoundation.org/policies/</a>. The PyTorch Foundation supports the PyTorch open source
          project, which has been established as PyTorch Project a Series of LF Projects, LLC. For policies applicable to the PyTorch Project a Series of LF Projects, LLC,
          please see <a href="https://www.lfprojects.org/policies/">www.lfprojects.org/policies/</a>.</p>
      </div>
     </div>

  </footer>

  <div class="cookie-banner-wrapper">
  <div class="container">
    <p class="gdpr-notice">To analyze traffic and optimize your experience, we serve cookies on this site. By clicking or navigating, you agree to allow our usage of cookies. As the current maintainers of this site, Facebook’s Cookies Policy applies. Learn more, including about available controls: <a href="https://www.facebook.com/policies/cookies/">Cookies Policy</a>.</p>
    <img class="close-button" src="_static/images/pytorch-x.svg">
  </div>
</div>

  <!-- End Footer -->

  <!-- Begin Mobile Menu -->

  <div class="mobile-main-menu">
    <div class="container-fluid">
      <div class="container">
        <div class="mobile-main-menu-header-container">
          <a class="header-logo" href="https://pytorch.org/" aria-label="PyTorch"></a>
          <a class="main-menu-close-button" href="#" data-behavior="close-mobile-menu"></a>
        </div>
      </div>
    </div>

    <div class="mobile-main-menu-links-container">
      <div class="main-menu">
        <ul>
           <li class="resources-mobile-menu-title">
             <a>Learn</a>
           </li>
           <ul class="resources-mobile-menu-items">
             <li>
               <a href="https://pytorch.org/get-started">Get Started</a>
             </li>
             <li>
               <a href="https://pytorch.org/tutorials">Tutorials</a>
             </li>
             <li>
               <a href="https://pytorch.org/tutorials/beginner/basics/intro.html">Learn the Basics</a>
             </li>
             <li>
               <a href="https://pytorch.org/tutorials/recipes/recipes_index.html">PyTorch Recipes</a>
             </li>
             <li>
               <a href="https://pytorch.org/tutorials/beginner/introyt.html">Introduction to PyTorch - YouTube Series</a>
             </li>
           </ul>
           <li class="resources-mobile-menu-title">
             <a>Ecosystem</a>
           </li>
           <ul class="resources-mobile-menu-items">
             <li>
               <a href="https://pytorch.org/ecosystem">Tools</a>
             </li>
             <li>
               <a href="https://pytorch.org/#community-module">Community</a>
             </li>
             <li>
               <a href="https://discuss.pytorch.org/">Forums</a>
             </li>
             <li>
               <a href="https://pytorch.org/resources">Developer Resources</a>
             </li>
             <li>
               <a href="https://pytorch.org/ecosystem/contributor-awards-2023">Contributor Awards - 2023</a>
             </li>
           </ul>

           <li class="resources-mobile-menu-title">
             <a>Edge</a>
           </li>

           <ul class="resources-mobile-menu-items">
             <li>
               <a href="https://pytorch.org/edge">About PyTorch Edge</a>
             </li>
             
             <li>
               <a href="https://pytorch.org/executorch-overview">ExecuTorch</a>
             </li>
           </ul>

           <li class="resources-mobile-menu-title">
             <a>Docs</a>
           </li>

           <ul class="resources-mobile-menu-items">
            <li>
              <a href="https://pytorch.org/docs/stable/index.html">PyTorch</a>
            </li>

            <li>
              <a href="https://pytorch.org/pytorch-domains">PyTorch Domains</a>
            </li>
          </ul>

          <li class="resources-mobile-menu-title">
            <a>Blog & News</a>
          </li>
            
           <ul class="resources-mobile-menu-items">
            <li>
              <a href="https://pytorch.org/blog/">PyTorch Blog</a>
            </li>
            <li>
              <a href="https://pytorch.org/community-blog">Community Blog</a>
            </li>

            <li>
              <a href="https://pytorch.org/videos">Videos</a>
            </li>

            <li>
              <a href="https://pytorch.org/community-stories">Community Stories</a>
            </li>
            <li>
              <a href="https://pytorch.org/events">Events</a>
            </li>
          </ul>
          
          <li class="resources-mobile-menu-title">
            <a>About</a>
          </li>

          <ul class="resources-mobile-menu-items">
            <li>
              <a href="https://pytorch.org/foundation">PyTorch Foundation</a>
            </li>
            <li>
              <a href="https://pytorch.org/governing-board">Governing Board</a>
            </li>
          </ul>
        </ul>
      </div>
    </div>
  </div>

  <!-- End Mobile Menu -->

  <script type="text/javascript" src="_static/js/vendor/anchor.min.js"></script>

  <script type="text/javascript">
    $(document).ready(function() {
      mobileMenu.bind();
      mobileTOC.bind();
      pytorchAnchors.bind();
      sideMenus.bind();
      scrollToAnchor.bind();
      highlightNavigation.bind();
      mainMenuDropdown.bind();
      filterTags.bind();

      // Add class to links that have code blocks, since we cannot create links in code blocks
      $("article.pytorch-article a span.pre").each(function(e) {
        $(this).closest("a").addClass("has-code");
      });
    })
  </script>
</body>
</html>