2.2/quantization.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 charset="utf-8">
  
  <meta name="viewport" content="width=device-width, initial-scale=1.0">
  
  <title>Quantization &mdash; PyTorch 2.2 documentation</title>
  

    <link rel="canonical" href="https://pytorch.org/docs/stable/quantization.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="Quantization API Reference" href="quantization-support.html" />
    <link rel="prev" title="Pipeline Parallelism" href="pipeline.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>
            <a href="https://pytorch.org/get-started">Get Started</a>
          </li>

          <li>
            <a href="https://pytorch.org/ecosystem">Ecosystem</a>
          </li>

          <li>
          <div id="resourcesDropdownButton" data-toggle="resources-dropdown" class="resources-dropdown">
              <a class="resource-option 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>
                </a>
                <a class="nav-dropdown-item" href="https://pytorch.org/executorch">
                  <span class="dropdown-title">ExecuTorch</span>
                </a>
              </div>
            </div>  
          </li>

          <li>
            <a href="https://pytorch.org/blog/">Blog</a>
          </li>

          <li>
            <a href="https://pytorch.org/tutorials">Tutorials</a>
          </li>

          <li class="active docs-active">
            <div id="resourcesDropdownButton" data-toggle="resources-dropdown" class="resources-dropdown">
              <a class="resource-option with-down-orange-arrow">
                Docs
              </a>
              <div class="resources-dropdown-menu">
                <a class="doc-dropdown-option nav-dropdown-item" href="https://pytorch.org/docs/stable/index.html">
                  <span class="dropdown-title">PyTorch</span>
                  <p></p>
                </a>
                <a class="doc-dropdown-option nav-dropdown-item" href="https://pytorch.org/audio/stable/index.html">
                  <span class="dropdown-title">torchaudio</span>
                  <p></p>
                </a>
                <a class="doc-dropdown-option nav-dropdown-item" href="https://pytorch.org/text/stable/index.html">
                  <span class="dropdown-title">torchtext</span>
                  <p></p>
                </a>
                <a class="doc-dropdown-option nav-dropdown-item" href="https://pytorch.org/vision/stable/index.html">
                  <span class="dropdown-title">torchvision</span>
                  <p></p>
                </a>
                <a class="doc-dropdown-option nav-dropdown-item" href="https://pytorch.org/torcharrow">
                  <span class="dropdown-title">torcharrow</span>
                  <p></p>
                </a>
                <a class="doc-dropdown-option nav-dropdown-item" href="https://pytorch.org/data">
                  <span class="dropdown-title">TorchData</span>
                  <p></p>
                </a>
                <a class="doc-dropdown-option nav-dropdown-item" href="https://pytorch.org/torchrec">
                  <span class="dropdown-title">TorchRec</span>
                  <p></p>
                </a>
                <a class="doc-dropdown-option nav-dropdown-item" href="https://pytorch.org/serve/">
                  <span class="dropdown-title">TorchServe</span>
                  <p></p>
                </a>
                <a class="doc-dropdown-option nav-dropdown-item" href="https://pytorch.org/torchx/">
                  <span class="dropdown-title">TorchX</span>
                  <p></p>
                </a>
                <a class="doc-dropdown-option nav-dropdown-item" href="https://pytorch.org/xla">
                  <span class="dropdown-title">PyTorch on XLA Devices</span>
                  <p></p>
                </a>
            </div>
          </li>

          <li>
            <div id="resourcesDropdownButton" data-toggle="resources-dropdown" class="resources-dropdown">
              <a class="resource-option with-down-arrow">
                Resources
              </a>
              <div class="resources-dropdown-menu">
                <a class="nav-dropdown-item" href="https://pytorch.org/features">
                  <span class="dropdown-title">About</span>
                  <p>Learn about PyTorch’s features and capabilities</p>
                </a>
                <a class="nav-dropdown-item" href="https://pytorch.org/foundation">
                  <span class="dropdown-title">PyTorch Foundation</span>
                  <p>Learn about the PyTorch foundation</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://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/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/events">
                  <span class="dropdown-title">Events</span>
                  <p>Find events, webinars, and podcasts</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/hub">
                  <span class="dropdown-title">Models (Beta)</span>
                  <p>Discover, publish, and reuse pre-trained models</p>
                </a>
              </div>
            </div>
          </li>

          <li>
            <a href="https://github.com/pytorch/pytorch">GitHub</a>
          </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'>2.2 &#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">CUDA 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/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/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="backends.html">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.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.optim.html">torch.distributed.optim</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.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="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="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="pipeline.html">Pipeline Parallelism</a></li>
<li class="toctree-l1 current"><a class="current reference internal" href="#">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="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="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="logging.html">torch._logging</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>Quantization</li>
    
    
      <li class="pytorch-breadcrumbs-aside">
        
            
            <a href="_sources/quantization.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.ao.quantization">
<span id="quantization"></span><span id="quantization-doc"></span><h1>Quantization<a class="headerlink" href="#module-torch.ao.quantization" title="Permalink to this heading">¶</a></h1>
<span class="target" id="module-torch.ao.quantization.fx"></span><div class="admonition warning">
<p class="admonition-title">Warning</p>
<p>Quantization is in beta and subject to change.</p>
</div>
<div class="section" id="introduction-to-quantization">
<h2>Introduction to Quantization<a class="headerlink" href="#introduction-to-quantization" title="Permalink to this heading">¶</a></h2>
<p>Quantization refers to techniques for performing computations and storing
tensors at lower bitwidths than floating point precision. A quantized model
executes some or all of the operations on tensors with reduced precision rather than
full precision (floating point) values. This allows for a more compact model representation and
the use of high performance vectorized operations on many hardware platforms.
PyTorch supports INT8 quantization compared to typical FP32 models allowing for
a 4x reduction in the model size and a 4x reduction in memory bandwidth
requirements. Hardware support for INT8 computations is typically 2 to 4
times faster compared to FP32 compute. Quantization is primarily a technique to
speed up inference and only the forward pass is supported for quantized
operators.</p>
<p>PyTorch supports multiple approaches to quantizing a deep learning model. In
most cases the model is trained in FP32 and then the model is converted to
INT8. In addition, PyTorch also supports quantization aware training, which
models quantization errors in both the forward and backward passes using
fake-quantization modules. Note that the entire computation is carried out in
floating point. At the end of quantization aware training, PyTorch provides
conversion functions to convert the trained model into lower precision.</p>
<p>At lower level, PyTorch provides a way to represent quantized tensors and
perform operations with them. They can be used to directly construct models
that perform all or part of the computation in lower precision. Higher-level
APIs are provided that incorporate typical workflows of converting FP32 model
to lower precision with minimal accuracy loss.</p>
</div>
<div class="section" id="quantization-api-summary">
<h2>Quantization API Summary<a class="headerlink" href="#quantization-api-summary" title="Permalink to this heading">¶</a></h2>
<p>PyTorch provides three different modes of quantization: Eager Mode Quantization, FX Graph Mode Quantization (maintainence) and PyTorch 2 Export Quantization.</p>
<p>Eager Mode Quantization is a beta feature. User needs to do fusion and specify where quantization and dequantization happens manually, also it only supports modules and not functionals.</p>
<p>FX Graph Mode Quantization is an automated quantization workflow in PyTorch, and currently it’s a prototype feature, it is in maintainence mode since we have PyTorch 2 Export Quantization. It improves upon Eager Mode Quantization by adding support for functionals and automating the quantization process, although people might need to refactor the model to make the model compatible with FX Graph Mode Quantization (symbolically traceable with <code class="docutils literal notranslate"><span class="pre">torch.fx</span></code>). Note that FX Graph Mode Quantization is not expected to work on arbitrary models since the model might not be symbolically traceable, we will integrate it into domain libraries like torchvision and users will be able to quantize models similar to the ones in supported domain libraries with FX Graph Mode Quantization. For arbitrary models we’ll provide general guidelines, but to actually make it work, users might need to be familiar with <code class="docutils literal notranslate"><span class="pre">torch.fx</span></code>, especially on how to make a model symbolically traceable.</p>
<p>PyTorch 2 Export Quantization is the new full graph mode quantization workflow, released as prototype feature in PyTorch 2.1. With PyTorch 2, we are moving to a better solution for full program capture (torch.export) since it can capture a higher percentage (88.8% on 14K models) of models compared to torch.fx.symbolic_trace (72.7% on 14K models), the program capture solution used by FX Graph Mode Quantization. torch.export still has limitations around some python constructs and requires user involvement to support dynamism in the exported model, but overall it is an improvement over the previous program capture solution. PyTorch 2 Export Quantization is built for models captured by torch.export, with flexibility and productivity of both modeling users and backend developers in mind. The main features are
(1). Programmable API for configuring how a model is quantized that can scale to many more use cases
(2). Simplified UX for modeling users and backend developers since they only need to interact with a single object (Quantizer) for expressing user’s intention about how to quantize a model and what the backend support.
(3). Optional reference quantized model representation that can represent quantized computation with integer operations that maps closer to actual quantized computations that happens in hardware.</p>
<p>New users of quantization are encouraged to try out PyTorch 2 Export Quantization first, if it does not work well, user can try eager mode quantization.</p>
<p>The following table compares the differences between Eager Mode Quantization, FX Graph Mode Quantization and PyTorch 2 Export Quantization:</p>
<table class="docutils colwidths-auto align-default">
<tbody>
<tr class="row-odd"><td></td>
<td><p>Eager Mode
Quantization</p></td>
<td><p>FX Graph
Mode
Quantization</p></td>
<td><p>PyTorch 2 Export
Quantization</p></td>
</tr>
<tr class="row-even"><td><p>Release
Status</p></td>
<td><p>beta</p></td>
<td><p>prototype
(maintainence)</p></td>
<td><p>prototype</p></td>
</tr>
<tr class="row-odd"><td><p>Operator
Fusion</p></td>
<td><p>Manual</p></td>
<td><p>Automatic</p></td>
<td><p>Automatic</p></td>
</tr>
<tr class="row-even"><td><p>Quant/DeQuant
Placement</p></td>
<td><p>Manual</p></td>
<td><p>Automatic</p></td>
<td><p>Automatic</p></td>
</tr>
<tr class="row-odd"><td><p>Quantizing
Modules</p></td>
<td><p>Supported</p></td>
<td><p>Supported</p></td>
<td><p>Supported</p></td>
</tr>
<tr class="row-even"><td><p>Quantizing
Functionals/Torch
Ops</p></td>
<td><p>Manual</p></td>
<td><p>Automatic</p></td>
<td><p>Supported</p></td>
</tr>
<tr class="row-odd"><td><p>Support for
Customization</p></td>
<td><p>Limited Support</p></td>
<td><p>Fully
Supported</p></td>
<td><p>Fully Supported</p></td>
</tr>
<tr class="row-even"><td><p>Quantization Mode
Support</p></td>
<td><p>Post Training
Quantization:
Static, Dynamic,
Weight Only</p>
<p>Quantization Aware
Training:
Static</p>
</td>
<td><p>Post Training
Quantization:
Static, Dynamic,
Weight Only</p>
<p>Quantization Aware
Training:
Static</p>
</td>
<td><p>Defined by
Backend Specific
Quantizer</p></td>
</tr>
<tr class="row-odd"><td><p>Input/Output
Model Type</p></td>
<td><p><code class="docutils literal notranslate"><span class="pre">torch.nn.Module</span></code></p></td>
<td><p><code class="docutils literal notranslate"><span class="pre">torch.nn.Module</span></code>
(May need some
refactors to make
the model
compatible with FX
Graph Mode
Quantization)</p></td>
<td><p><code class="docutils literal notranslate"><span class="pre">torch.fx.GraphModule</span></code>
(captured by
<code class="docutils literal notranslate"><span class="pre">torch.export</span></code></p></td>
</tr>
</tbody>
</table>
<p>There are three types of quantization supported:</p>
<ol class="arabic simple">
<li><p>dynamic quantization (weights quantized with activations read/stored in
floating point and quantized for compute)</p></li>
<li><p>static quantization (weights quantized, activations quantized, calibration
required post training)</p></li>
<li><p>static quantization aware training (weights quantized, activations quantized,
quantization numerics modeled during training)</p></li>
</ol>
<p>Please see our <a class="reference external" href="https://pytorch.org/blog/introduction-to-quantization-on-pytorch/">Introduction to Quantization on PyTorch</a> blog post
for a more comprehensive overview of the tradeoffs between these quantization
types.</p>
<p>Operator coverage varies between dynamic and static quantization and is captured in the table below.</p>
<table class="docutils colwidths-auto align-default">
<tbody>
<tr class="row-odd"><td></td>
<td><p>Static
Quantization</p></td>
<td><p>Dynamic
Quantization</p></td>
</tr>
<tr class="row-even"><td><div class="line-block">
<div class="line">nn.Linear</div>
<div class="line">nn.Conv1d/2d/3d</div>
</div>
</td>
<td><div class="line-block">
<div class="line">Y</div>
<div class="line">Y</div>
</div>
</td>
<td><div class="line-block">
<div class="line">Y</div>
<div class="line">N</div>
</div>
</td>
</tr>
<tr class="row-odd"><td><div class="line-block">
<div class="line">nn.LSTM</div>
<div class="line"><br /></div>
<div class="line">nn.GRU</div>
</div>
</td>
<td><div class="line-block">
<div class="line">Y (through</div>
<div class="line">custom modules)</div>
<div class="line">N</div>
</div>
</td>
<td><div class="line-block">
<div class="line">Y</div>
<div class="line"><br /></div>
<div class="line">Y</div>
</div>
</td>
</tr>
<tr class="row-even"><td><div class="line-block">
<div class="line">nn.RNNCell</div>
<div class="line">nn.GRUCell</div>
<div class="line">nn.LSTMCell</div>
</div>
</td>
<td><div class="line-block">
<div class="line">N</div>
<div class="line">N</div>
<div class="line">N</div>
</div>
</td>
<td><div class="line-block">
<div class="line">Y</div>
<div class="line">Y</div>
<div class="line">Y</div>
</div>
</td>
</tr>
<tr class="row-odd"><td><p>nn.EmbeddingBag</p></td>
<td><p>Y (activations
are in fp32)</p></td>
<td><p>Y</p></td>
</tr>
<tr class="row-even"><td><p>nn.Embedding</p></td>
<td><p>Y</p></td>
<td><p>Y</p></td>
</tr>
<tr class="row-odd"><td><p>nn.MultiheadAttention</p></td>
<td><p>Y (through
custom modules)</p></td>
<td><p>Not supported</p></td>
</tr>
<tr class="row-even"><td><p>Activations</p></td>
<td><p>Broadly supported</p></td>
<td><p>Un-changed,
computations
stay in fp32</p></td>
</tr>
</tbody>
</table>
<div class="section" id="eager-mode-quantization">
<h3>Eager Mode Quantization<a class="headerlink" href="#eager-mode-quantization" title="Permalink to this heading">¶</a></h3>
<p>For a general introduction to the quantization flow, including different types of quantization, please take a look at <a class="reference internal" href="#general-quantization-flow">General Quantization Flow</a>.</p>
<div class="section" id="post-training-dynamic-quantization">
<h4>Post Training Dynamic Quantization<a class="headerlink" href="#post-training-dynamic-quantization" title="Permalink to this heading">¶</a></h4>
<p>This is the simplest to apply form of quantization where the weights are
quantized ahead of time but the activations are dynamically quantized
during inference. This is used for situations where the model execution time
is dominated by loading weights from memory rather than computing the matrix
multiplications. This is true for LSTM and Transformer type models with
small batch size.</p>
<p>Diagram:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="c1"># original model</span>
<span class="c1"># all tensors and computations are in floating point</span>
<span class="n">previous_layer_fp32</span> <span class="o">--</span> <span class="n">linear_fp32</span> <span class="o">--</span> <span class="n">activation_fp32</span> <span class="o">--</span> <span class="n">next_layer_fp32</span>
                 <span class="o">/</span>
<span class="n">linear_weight_fp32</span>

<span class="c1"># dynamically quantized model</span>
<span class="c1"># linear and LSTM weights are in int8</span>
<span class="n">previous_layer_fp32</span> <span class="o">--</span> <span class="n">linear_int8_w_fp32_inp</span> <span class="o">--</span> <span class="n">activation_fp32</span> <span class="o">--</span> <span class="n">next_layer_fp32</span>
                     <span class="o">/</span>
   <span class="n">linear_weight_int8</span>
</pre></div>
</div>
<p>PTDQ API Example:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="kn">import</span> <span class="nn">torch</span>

<span class="c1"># define a floating point model</span>
<span class="k">class</span> <span class="nc">M</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">nn</span><span class="o">.</span><span class="n">Module</span><span class="p">):</span>
    <span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
        <span class="nb">super</span><span class="p">()</span><span class="o">.</span><span class="fm">__init__</span><span class="p">()</span>
        <span class="bp">self</span><span class="o">.</span><span class="n">fc</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">nn</span><span class="o">.</span><span class="n">Linear</span><span class="p">(</span><span class="mi">4</span><span class="p">,</span> <span class="mi">4</span><span class="p">)</span>

    <span class="k">def</span> <span class="nf">forward</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">x</span><span class="p">):</span>
        <span class="n">x</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">fc</span><span class="p">(</span><span class="n">x</span><span class="p">)</span>
        <span class="k">return</span> <span class="n">x</span>

<span class="c1"># create a model instance</span>
<span class="n">model_fp32</span> <span class="o">=</span> <span class="n">M</span><span class="p">()</span>
<span class="c1"># create a quantized model instance</span>
<span class="n">model_int8</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">ao</span><span class="o">.</span><span class="n">quantization</span><span class="o">.</span><span class="n">quantize_dynamic</span><span class="p">(</span>
    <span class="n">model_fp32</span><span class="p">,</span>  <span class="c1"># the original model</span>
    <span class="p">{</span><span class="n">torch</span><span class="o">.</span><span class="n">nn</span><span class="o">.</span><span class="n">Linear</span><span class="p">},</span>  <span class="c1"># a set of layers to dynamically quantize</span>
    <span class="n">dtype</span><span class="o">=</span><span class="n">torch</span><span class="o">.</span><span class="n">qint8</span><span class="p">)</span>  <span class="c1"># the target dtype for quantized weights</span>

<span class="c1"># run the model</span>
<span class="n">input_fp32</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">randn</span><span class="p">(</span><span class="mi">4</span><span class="p">,</span> <span class="mi">4</span><span class="p">,</span> <span class="mi">4</span><span class="p">,</span> <span class="mi">4</span><span class="p">)</span>
<span class="n">res</span> <span class="o">=</span> <span class="n">model_int8</span><span class="p">(</span><span class="n">input_fp32</span><span class="p">)</span>
</pre></div>
</div>
<p>To learn more about dynamic quantization please see our <a class="reference external" href="https://pytorch.org/tutorials/recipes/recipes/dynamic_quantization.html">dynamic quantization tutorial</a>.</p>
</div>
<div class="section" id="post-training-static-quantization">
<h4>Post Training Static Quantization<a class="headerlink" href="#post-training-static-quantization" title="Permalink to this heading">¶</a></h4>
<p>Post Training Static Quantization (PTQ static) quantizes the weights and activations of the model.  It
fuses activations into preceding layers where possible.  It requires
calibration with a representative dataset to determine optimal quantization
parameters for activations. Post Training Static Quantization is typically used when
both memory bandwidth and compute savings are important with CNNs being a
typical use case.</p>
<p>We may need to modify the model before applying post training static quantization. Please see <a class="reference internal" href="#model-preparation-for-eager-mode-static-quantization">Model Preparation for Eager Mode Static Quantization</a>.</p>
<p>Diagram:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="c1"># original model</span>
<span class="c1"># all tensors and computations are in floating point</span>
<span class="n">previous_layer_fp32</span> <span class="o">--</span> <span class="n">linear_fp32</span> <span class="o">--</span> <span class="n">activation_fp32</span> <span class="o">--</span> <span class="n">next_layer_fp32</span>
                    <span class="o">/</span>
    <span class="n">linear_weight_fp32</span>

<span class="c1"># statically quantized model</span>
<span class="c1"># weights and activations are in int8</span>
<span class="n">previous_layer_int8</span> <span class="o">--</span> <span class="n">linear_with_activation_int8</span> <span class="o">--</span> <span class="n">next_layer_int8</span>
                    <span class="o">/</span>
  <span class="n">linear_weight_int8</span>
</pre></div>
</div>
<p>PTSQ API Example:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="kn">import</span> <span class="nn">torch</span>

<span class="c1"># define a floating point model where some layers could be statically quantized</span>
<span class="k">class</span> <span class="nc">M</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">nn</span><span class="o">.</span><span class="n">Module</span><span class="p">):</span>
    <span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
        <span class="nb">super</span><span class="p">()</span><span class="o">.</span><span class="fm">__init__</span><span class="p">()</span>
        <span class="c1"># QuantStub converts tensors from floating point to quantized</span>
        <span class="bp">self</span><span class="o">.</span><span class="n">quant</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">ao</span><span class="o">.</span><span class="n">quantization</span><span class="o">.</span><span class="n">QuantStub</span><span class="p">()</span>
        <span class="bp">self</span><span class="o">.</span><span class="n">conv</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">nn</span><span class="o">.</span><span class="n">Conv2d</span><span class="p">(</span><span class="mi">1</span><span class="p">,</span> <span class="mi">1</span><span class="p">,</span> <span class="mi">1</span><span class="p">)</span>
        <span class="bp">self</span><span class="o">.</span><span class="n">relu</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">nn</span><span class="o">.</span><span class="n">ReLU</span><span class="p">()</span>
        <span class="c1"># DeQuantStub converts tensors from quantized to floating point</span>
        <span class="bp">self</span><span class="o">.</span><span class="n">dequant</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">ao</span><span class="o">.</span><span class="n">quantization</span><span class="o">.</span><span class="n">DeQuantStub</span><span class="p">()</span>

    <span class="k">def</span> <span class="nf">forward</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">x</span><span class="p">):</span>
        <span class="c1"># manually specify where tensors will be converted from floating</span>
        <span class="c1"># point to quantized in the quantized model</span>
        <span class="n">x</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">quant</span><span class="p">(</span><span class="n">x</span><span class="p">)</span>
        <span class="n">x</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">conv</span><span class="p">(</span><span class="n">x</span><span class="p">)</span>
        <span class="n">x</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">relu</span><span class="p">(</span><span class="n">x</span><span class="p">)</span>
        <span class="c1"># manually specify where tensors will be converted from quantized</span>
        <span class="c1"># to floating point in the quantized model</span>
        <span class="n">x</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">dequant</span><span class="p">(</span><span class="n">x</span><span class="p">)</span>
        <span class="k">return</span> <span class="n">x</span>

<span class="c1"># create a model instance</span>
<span class="n">model_fp32</span> <span class="o">=</span> <span class="n">M</span><span class="p">()</span>

<span class="c1"># model must be set to eval mode for static quantization logic to work</span>
<span class="n">model_fp32</span><span class="o">.</span><span class="n">eval</span><span class="p">()</span>

<span class="c1"># attach a global qconfig, which contains information about what kind</span>
<span class="c1"># of observers to attach. Use &#39;x86&#39; for server inference and &#39;qnnpack&#39;</span>
<span class="c1"># for mobile inference. Other quantization configurations such as selecting</span>
<span class="c1"># symmetric or asymmetric quantization and MinMax or L2Norm calibration techniques</span>
<span class="c1"># can be specified here.</span>
<span class="c1"># Note: the old &#39;fbgemm&#39; is still available but &#39;x86&#39; is the recommended default</span>
<span class="c1"># for server inference.</span>
<span class="c1"># model_fp32.qconfig = torch.ao.quantization.get_default_qconfig(&#39;fbgemm&#39;)</span>
<span class="n">model_fp32</span><span class="o">.</span><span class="n">qconfig</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">ao</span><span class="o">.</span><span class="n">quantization</span><span class="o">.</span><span class="n">get_default_qconfig</span><span class="p">(</span><span class="s1">&#39;x86&#39;</span><span class="p">)</span>

<span class="c1"># Fuse the activations to preceding layers, where applicable.</span>
<span class="c1"># This needs to be done manually depending on the model architecture.</span>
<span class="c1"># Common fusions include `conv + relu` and `conv + batchnorm + relu`</span>
<span class="n">model_fp32_fused</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">ao</span><span class="o">.</span><span class="n">quantization</span><span class="o">.</span><span class="n">fuse_modules</span><span class="p">(</span><span class="n">model_fp32</span><span class="p">,</span> <span class="p">[[</span><span class="s1">&#39;conv&#39;</span><span class="p">,</span> <span class="s1">&#39;relu&#39;</span><span class="p">]])</span>

<span class="c1"># Prepare the model for static quantization. This inserts observers in</span>
<span class="c1"># the model that will observe activation tensors during calibration.</span>
<span class="n">model_fp32_prepared</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">ao</span><span class="o">.</span><span class="n">quantization</span><span class="o">.</span><span class="n">prepare</span><span class="p">(</span><span class="n">model_fp32_fused</span><span class="p">)</span>

<span class="c1"># calibrate the prepared model to determine quantization parameters for activations</span>
<span class="c1"># in a real world setting, the calibration would be done with a representative dataset</span>
<span class="n">input_fp32</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">randn</span><span class="p">(</span><span class="mi">4</span><span class="p">,</span> <span class="mi">1</span><span class="p">,</span> <span class="mi">4</span><span class="p">,</span> <span class="mi">4</span><span class="p">)</span>
<span class="n">model_fp32_prepared</span><span class="p">(</span><span class="n">input_fp32</span><span class="p">)</span>

<span class="c1"># Convert the observed model to a quantized model. This does several things:</span>
<span class="c1"># quantizes the weights, computes and stores the scale and bias value to be</span>
<span class="c1"># used with each activation tensor, and replaces key operators with quantized</span>
<span class="c1"># implementations.</span>
<span class="n">model_int8</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">ao</span><span class="o">.</span><span class="n">quantization</span><span class="o">.</span><span class="n">convert</span><span class="p">(</span><span class="n">model_fp32_prepared</span><span class="p">)</span>

<span class="c1"># run the model, relevant calculations will happen in int8</span>
<span class="n">res</span> <span class="o">=</span> <span class="n">model_int8</span><span class="p">(</span><span class="n">input_fp32</span><span class="p">)</span>
</pre></div>
</div>
<p>To learn more about static quantization, please see the <a class="reference external" href="https://pytorch.org/tutorials/advanced/static_quantization_tutorial.html">static quantization tutorial</a>.</p>
</div>
<div class="section" id="quantization-aware-training-for-static-quantization">
<h4>Quantization Aware Training for Static Quantization<a class="headerlink" href="#quantization-aware-training-for-static-quantization" title="Permalink to this heading">¶</a></h4>
<p>Quantization Aware Training (QAT) models the effects of quantization during training
allowing for higher accuracy compared to other quantization methods. We can do QAT for static, dynamic or weight only quantization.  During
training, all calculations are done in floating point, with fake_quant modules
modeling the effects of quantization by clamping and rounding to simulate the
effects of INT8.  After model conversion, weights and
activations are quantized, and activations are fused into the preceding layer
where possible.  It is commonly used with CNNs and yields a higher accuracy
compared to static quantization.</p>
<p>We may need to modify the model before applying post training static quantization. Please see <a class="reference internal" href="#model-preparation-for-eager-mode-static-quantization">Model Preparation for Eager Mode Static Quantization</a>.</p>
<p>Diagram:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="c1"># original model</span>
<span class="c1"># all tensors and computations are in floating point</span>
<span class="n">previous_layer_fp32</span> <span class="o">--</span> <span class="n">linear_fp32</span> <span class="o">--</span> <span class="n">activation_fp32</span> <span class="o">--</span> <span class="n">next_layer_fp32</span>
                      <span class="o">/</span>
    <span class="n">linear_weight_fp32</span>

<span class="c1"># model with fake_quants for modeling quantization numerics during training</span>
<span class="n">previous_layer_fp32</span> <span class="o">--</span> <span class="n">fq</span> <span class="o">--</span> <span class="n">linear_fp32</span> <span class="o">--</span> <span class="n">activation_fp32</span> <span class="o">--</span> <span class="n">fq</span> <span class="o">--</span> <span class="n">next_layer_fp32</span>
                           <span class="o">/</span>
   <span class="n">linear_weight_fp32</span> <span class="o">--</span> <span class="n">fq</span>

<span class="c1"># quantized model</span>
<span class="c1"># weights and activations are in int8</span>
<span class="n">previous_layer_int8</span> <span class="o">--</span> <span class="n">linear_with_activation_int8</span> <span class="o">--</span> <span class="n">next_layer_int8</span>
                     <span class="o">/</span>
   <span class="n">linear_weight_int8</span>
</pre></div>
</div>
<p>QAT API Example:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="kn">import</span> <span class="nn">torch</span>

<span class="c1"># define a floating point model where some layers could benefit from QAT</span>
<span class="k">class</span> <span class="nc">M</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">nn</span><span class="o">.</span><span class="n">Module</span><span class="p">):</span>
    <span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
        <span class="nb">super</span><span class="p">()</span><span class="o">.</span><span class="fm">__init__</span><span class="p">()</span>
        <span class="c1"># QuantStub converts tensors from floating point to quantized</span>
        <span class="bp">self</span><span class="o">.</span><span class="n">quant</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">ao</span><span class="o">.</span><span class="n">quantization</span><span class="o">.</span><span class="n">QuantStub</span><span class="p">()</span>
        <span class="bp">self</span><span class="o">.</span><span class="n">conv</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">nn</span><span class="o">.</span><span class="n">Conv2d</span><span class="p">(</span><span class="mi">1</span><span class="p">,</span> <span class="mi">1</span><span class="p">,</span> <span class="mi">1</span><span class="p">)</span>
        <span class="bp">self</span><span class="o">.</span><span class="n">bn</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">nn</span><span class="o">.</span><span class="n">BatchNorm2d</span><span class="p">(</span><span class="mi">1</span><span class="p">)</span>
        <span class="bp">self</span><span class="o">.</span><span class="n">relu</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">nn</span><span class="o">.</span><span class="n">ReLU</span><span class="p">()</span>
        <span class="c1"># DeQuantStub converts tensors from quantized to floating point</span>
        <span class="bp">self</span><span class="o">.</span><span class="n">dequant</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">ao</span><span class="o">.</span><span class="n">quantization</span><span class="o">.</span><span class="n">DeQuantStub</span><span class="p">()</span>

    <span class="k">def</span> <span class="nf">forward</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">x</span><span class="p">):</span>
        <span class="n">x</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">quant</span><span class="p">(</span><span class="n">x</span><span class="p">)</span>
        <span class="n">x</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">conv</span><span class="p">(</span><span class="n">x</span><span class="p">)</span>
        <span class="n">x</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">bn</span><span class="p">(</span><span class="n">x</span><span class="p">)</span>
        <span class="n">x</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">relu</span><span class="p">(</span><span class="n">x</span><span class="p">)</span>
        <span class="n">x</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">dequant</span><span class="p">(</span><span class="n">x</span><span class="p">)</span>
        <span class="k">return</span> <span class="n">x</span>

<span class="c1"># create a model instance</span>
<span class="n">model_fp32</span> <span class="o">=</span> <span class="n">M</span><span class="p">()</span>

<span class="c1"># model must be set to eval for fusion to work</span>
<span class="n">model_fp32</span><span class="o">.</span><span class="n">eval</span><span class="p">()</span>

<span class="c1"># attach a global qconfig, which contains information about what kind</span>
<span class="c1"># of observers to attach. Use &#39;x86&#39; for server inference and &#39;qnnpack&#39;</span>
<span class="c1"># for mobile inference. Other quantization configurations such as selecting</span>
<span class="c1"># symmetric or asymmetric quantization and MinMax or L2Norm calibration techniques</span>
<span class="c1"># can be specified here.</span>
<span class="c1"># Note: the old &#39;fbgemm&#39; is still available but &#39;x86&#39; is the recommended default</span>
<span class="c1"># for server inference.</span>
<span class="c1"># model_fp32.qconfig = torch.ao.quantization.get_default_qconfig(&#39;fbgemm&#39;)</span>
<span class="n">model_fp32</span><span class="o">.</span><span class="n">qconfig</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">ao</span><span class="o">.</span><span class="n">quantization</span><span class="o">.</span><span class="n">get_default_qat_qconfig</span><span class="p">(</span><span class="s1">&#39;x86&#39;</span><span class="p">)</span>

<span class="c1"># fuse the activations to preceding layers, where applicable</span>
<span class="c1"># this needs to be done manually depending on the model architecture</span>
<span class="n">model_fp32_fused</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">ao</span><span class="o">.</span><span class="n">quantization</span><span class="o">.</span><span class="n">fuse_modules</span><span class="p">(</span><span class="n">model_fp32</span><span class="p">,</span>
    <span class="p">[[</span><span class="s1">&#39;conv&#39;</span><span class="p">,</span> <span class="s1">&#39;bn&#39;</span><span class="p">,</span> <span class="s1">&#39;relu&#39;</span><span class="p">]])</span>

<span class="c1"># Prepare the model for QAT. This inserts observers and fake_quants in</span>
<span class="c1"># the model needs to be set to train for QAT logic to work</span>
<span class="c1"># the model that will observe weight and activation tensors during calibration.</span>
<span class="n">model_fp32_prepared</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">ao</span><span class="o">.</span><span class="n">quantization</span><span class="o">.</span><span class="n">prepare_qat</span><span class="p">(</span><span class="n">model_fp32_fused</span><span class="o">.</span><span class="n">train</span><span class="p">())</span>

<span class="c1"># run the training loop (not shown)</span>
<span class="n">training_loop</span><span class="p">(</span><span class="n">model_fp32_prepared</span><span class="p">)</span>

<span class="c1"># Convert the observed model to a quantized model. This does several things:</span>
<span class="c1"># quantizes the weights, computes and stores the scale and bias value to be</span>
<span class="c1"># used with each activation tensor, fuses modules where appropriate,</span>
<span class="c1"># and replaces key operators with quantized implementations.</span>
<span class="n">model_fp32_prepared</span><span class="o">.</span><span class="n">eval</span><span class="p">()</span>
<span class="n">model_int8</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">ao</span><span class="o">.</span><span class="n">quantization</span><span class="o">.</span><span class="n">convert</span><span class="p">(</span><span class="n">model_fp32_prepared</span><span class="p">)</span>

<span class="c1"># run the model, relevant calculations will happen in int8</span>
<span class="n">res</span> <span class="o">=</span> <span class="n">model_int8</span><span class="p">(</span><span class="n">input_fp32</span><span class="p">)</span>
</pre></div>
</div>
<p>To learn more about quantization aware training, please see the <a class="reference external" href="https://pytorch.org/tutorials/advanced/static_quantization_tutorial.html">QAT
tutorial</a>.</p>
</div>
<div class="section" id="model-preparation-for-eager-mode-static-quantization">
<h4>Model Preparation for Eager Mode Static Quantization<a class="headerlink" href="#model-preparation-for-eager-mode-static-quantization" title="Permalink to this heading">¶</a></h4>
<p>It is necessary to currently make some modifications to the model definition
prior to Eager mode quantization. This is because currently quantization works on a module
by module basis. Specifically, for all quantization techniques, the user needs to:</p>
<ol class="arabic simple">
<li><p>Convert any operations that require output requantization (and thus have
additional parameters) from functionals to module form (for example,
using <code class="docutils literal notranslate"><span class="pre">torch.nn.ReLU</span></code> instead of <code class="docutils literal notranslate"><span class="pre">torch.nn.functional.relu</span></code>).</p></li>
<li><p>Specify which parts of the model need to be quantized either by assigning
<code class="docutils literal notranslate"><span class="pre">.qconfig</span></code> attributes on submodules or by specifying <code class="docutils literal notranslate"><span class="pre">qconfig_mapping</span></code>.
For example, setting <code class="docutils literal notranslate"><span class="pre">model.conv1.qconfig</span> <span class="pre">=</span> <span class="pre">None</span></code> means that the
<code class="docutils literal notranslate"><span class="pre">model.conv</span></code> layer will not be quantized, and setting
<code class="docutils literal notranslate"><span class="pre">model.linear1.qconfig</span> <span class="pre">=</span> <span class="pre">custom_qconfig</span></code> means that the quantization
settings for <code class="docutils literal notranslate"><span class="pre">model.linear1</span></code> will be using <code class="docutils literal notranslate"><span class="pre">custom_qconfig</span></code> instead
of the global qconfig.</p></li>
</ol>
<p>For static quantization techniques which quantize activations, the user needs
to do the following in addition:</p>
<ol class="arabic simple">
<li><p>Specify where activations are quantized and de-quantized. This is done using
<a class="reference internal" href="generated/torch.ao.quantization.QuantStub.html#torch.ao.quantization.QuantStub" title="torch.ao.quantization.QuantStub"><code class="xref py py-class docutils literal notranslate"><span class="pre">QuantStub</span></code></a> and
<a class="reference internal" href="generated/torch.ao.quantization.DeQuantStub.html#torch.ao.quantization.DeQuantStub" title="torch.ao.quantization.DeQuantStub"><code class="xref py py-class docutils literal notranslate"><span class="pre">DeQuantStub</span></code></a> modules.</p></li>
<li><p>Use <a class="reference internal" href="generated/torch.ao.nn.quantized.FloatFunctional.html#torch.ao.nn.quantized.FloatFunctional" title="torch.ao.nn.quantized.FloatFunctional"><code class="xref py py-class docutils literal notranslate"><span class="pre">FloatFunctional</span></code></a> to wrap tensor operations
that require special handling for quantization into modules. Examples
are operations like <code class="docutils literal notranslate"><span class="pre">add</span></code> and <code class="docutils literal notranslate"><span class="pre">cat</span></code> which require special handling to
determine output quantization parameters.</p></li>
<li><p>Fuse modules: combine operations/modules into a single module to obtain
higher accuracy and performance. This is done using the
<a class="reference internal" href="generated/torch.ao.quantization.fuse_modules.fuse_modules.html#torch.ao.quantization.fuse_modules.fuse_modules" title="torch.ao.quantization.fuse_modules.fuse_modules"><code class="xref py py-func docutils literal notranslate"><span class="pre">fuse_modules()</span></code></a> API, which takes in lists of modules
to be fused. We currently support the following fusions:
[Conv, Relu], [Conv, BatchNorm], [Conv, BatchNorm, Relu], [Linear, Relu]</p></li>
</ol>
</div>
</div>
<div class="section" id="prototype-maintaince-mode-fx-graph-mode-quantization">
<h3>(Prototype - maintaince mode) FX Graph Mode Quantization<a class="headerlink" href="#prototype-maintaince-mode-fx-graph-mode-quantization" title="Permalink to this heading">¶</a></h3>
<p>There are multiple quantization types in post training quantization (weight only, dynamic and static) and the configuration is done through <cite>qconfig_mapping</cite> (an argument of the <cite>prepare_fx</cite> function).</p>
<p>FXPTQ API Example:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="kn">import</span> <span class="nn">torch</span>
<span class="kn">from</span> <span class="nn">torch.ao.quantization</span> <span class="kn">import</span> <span class="p">(</span>
  <span class="n">get_default_qconfig_mapping</span><span class="p">,</span>
  <span class="n">get_default_qat_qconfig_mapping</span><span class="p">,</span>
  <span class="n">QConfigMapping</span><span class="p">,</span>
<span class="p">)</span>
<span class="kn">import</span> <span class="nn">torch.ao.quantization.quantize_fx</span> <span class="k">as</span> <span class="nn">quantize_fx</span>
<span class="kn">import</span> <span class="nn">copy</span>

<span class="n">model_fp</span> <span class="o">=</span> <span class="n">UserModel</span><span class="p">()</span>

<span class="c1">#</span>
<span class="c1"># post training dynamic/weight_only quantization</span>
<span class="c1">#</span>

<span class="c1"># we need to deepcopy if we still want to keep model_fp unchanged after quantization since quantization apis change the input model</span>
<span class="n">model_to_quantize</span> <span class="o">=</span> <span class="n">copy</span><span class="o">.</span><span class="n">deepcopy</span><span class="p">(</span><span class="n">model_fp</span><span class="p">)</span>
<span class="n">model_to_quantize</span><span class="o">.</span><span class="n">eval</span><span class="p">()</span>
<span class="n">qconfig_mapping</span> <span class="o">=</span> <span class="n">QConfigMapping</span><span class="p">()</span><span class="o">.</span><span class="n">set_global</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">ao</span><span class="o">.</span><span class="n">quantization</span><span class="o">.</span><span class="n">default_dynamic_qconfig</span><span class="p">)</span>
<span class="c1"># a tuple of one or more example inputs are needed to trace the model</span>
<span class="n">example_inputs</span> <span class="o">=</span> <span class="p">(</span><span class="n">input_fp32</span><span class="p">)</span>
<span class="c1"># prepare</span>
<span class="n">model_prepared</span> <span class="o">=</span> <span class="n">quantize_fx</span><span class="o">.</span><span class="n">prepare_fx</span><span class="p">(</span><span class="n">model_to_quantize</span><span class="p">,</span> <span class="n">qconfig_mapping</span><span class="p">,</span> <span class="n">example_inputs</span><span class="p">)</span>
<span class="c1"># no calibration needed when we only have dynamic/weight_only quantization</span>
<span class="c1"># quantize</span>
<span class="n">model_quantized</span> <span class="o">=</span> <span class="n">quantize_fx</span><span class="o">.</span><span class="n">convert_fx</span><span class="p">(</span><span class="n">model_prepared</span><span class="p">)</span>

<span class="c1">#</span>
<span class="c1"># post training static quantization</span>
<span class="c1">#</span>

<span class="n">model_to_quantize</span> <span class="o">=</span> <span class="n">copy</span><span class="o">.</span><span class="n">deepcopy</span><span class="p">(</span><span class="n">model_fp</span><span class="p">)</span>
<span class="n">qconfig_mapping</span> <span class="o">=</span> <span class="n">get_default_qconfig_mapping</span><span class="p">(</span><span class="s2">&quot;qnnpack&quot;</span><span class="p">)</span>
<span class="n">model_to_quantize</span><span class="o">.</span><span class="n">eval</span><span class="p">()</span>
<span class="c1"># prepare</span>
<span class="n">model_prepared</span> <span class="o">=</span> <span class="n">quantize_fx</span><span class="o">.</span><span class="n">prepare_fx</span><span class="p">(</span><span class="n">model_to_quantize</span><span class="p">,</span> <span class="n">qconfig_mapping</span><span class="p">,</span> <span class="n">example_inputs</span><span class="p">)</span>
<span class="c1"># calibrate (not shown)</span>
<span class="c1"># quantize</span>
<span class="n">model_quantized</span> <span class="o">=</span> <span class="n">quantize_fx</span><span class="o">.</span><span class="n">convert_fx</span><span class="p">(</span><span class="n">model_prepared</span><span class="p">)</span>

<span class="c1">#</span>
<span class="c1"># quantization aware training for static quantization</span>
<span class="c1">#</span>

<span class="n">model_to_quantize</span> <span class="o">=</span> <span class="n">copy</span><span class="o">.</span><span class="n">deepcopy</span><span class="p">(</span><span class="n">model_fp</span><span class="p">)</span>
<span class="n">qconfig_mapping</span> <span class="o">=</span> <span class="n">get_default_qat_qconfig_mapping</span><span class="p">(</span><span class="s2">&quot;qnnpack&quot;</span><span class="p">)</span>
<span class="n">model_to_quantize</span><span class="o">.</span><span class="n">train</span><span class="p">()</span>
<span class="c1"># prepare</span>
<span class="n">model_prepared</span> <span class="o">=</span> <span class="n">quantize_fx</span><span class="o">.</span><span class="n">prepare_qat_fx</span><span class="p">(</span><span class="n">model_to_quantize</span><span class="p">,</span> <span class="n">qconfig_mapping</span><span class="p">,</span> <span class="n">example_inputs</span><span class="p">)</span>
<span class="c1"># training loop (not shown)</span>
<span class="c1"># quantize</span>
<span class="n">model_quantized</span> <span class="o">=</span> <span class="n">quantize_fx</span><span class="o">.</span><span class="n">convert_fx</span><span class="p">(</span><span class="n">model_prepared</span><span class="p">)</span>

<span class="c1">#</span>
<span class="c1"># fusion</span>
<span class="c1">#</span>
<span class="n">model_to_quantize</span> <span class="o">=</span> <span class="n">copy</span><span class="o">.</span><span class="n">deepcopy</span><span class="p">(</span><span class="n">model_fp</span><span class="p">)</span>
<span class="n">model_fused</span> <span class="o">=</span> <span class="n">quantize_fx</span><span class="o">.</span><span class="n">fuse_fx</span><span class="p">(</span><span class="n">model_to_quantize</span><span class="p">)</span>
</pre></div>
</div>
<p>Please follow the tutorials below to learn more about FX Graph Mode Quantization:</p>
<ul class="simple">
<li><p><a class="reference external" href="https://pytorch.org/tutorials/prototype/fx_graph_mode_quant_guide.html">User Guide on Using FX Graph Mode Quantization</a></p></li>
<li><p><a class="reference external" href="https://pytorch.org/tutorials/prototype/fx_graph_mode_ptq_static.html">FX Graph Mode Post Training Static Quantization</a></p></li>
<li><p><a class="reference external" href="https://pytorch.org/tutorials/prototype/fx_graph_mode_ptq_dynamic.html">FX Graph Mode Post Training Dynamic Quantization</a></p></li>
</ul>
</div>
<div class="section" id="prototype-pytorch-2-export-quantization">
<h3>(Prototype) PyTorch 2 Export Quantization<a class="headerlink" href="#prototype-pytorch-2-export-quantization" title="Permalink to this heading">¶</a></h3>
<p>API Example:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="kn">import</span> <span class="nn">torch</span>
<span class="kn">from</span> <span class="nn">torch.ao.quantization.quantize_pt2e</span> <span class="kn">import</span> <span class="n">prepare_pt2e</span>
<span class="kn">from</span> <span class="nn">torch._export</span> <span class="kn">import</span> <span class="n">capture_pre_autograd_graph</span>
<span class="kn">from</span> <span class="nn">torch.ao.quantization.quantizer</span> <span class="kn">import</span> <span class="p">(</span>
    <span class="n">XNNPACKQuantizer</span><span class="p">,</span>
    <span class="n">get_symmetric_quantization_config</span><span class="p">,</span>
<span class="p">)</span>

<span class="k">class</span> <span class="nc">M</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">nn</span><span class="o">.</span><span class="n">Module</span><span class="p">):</span>
    <span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
        <span class="nb">super</span><span class="p">()</span><span class="o">.</span><span class="fm">__init__</span><span class="p">()</span>
        <span class="bp">self</span><span class="o">.</span><span class="n">linear</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">nn</span><span class="o">.</span><span class="n">Linear</span><span class="p">(</span><span class="mi">5</span><span class="p">,</span> <span class="mi">10</span><span class="p">)</span>

   <span class="k">def</span> <span class="nf">forward</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">x</span><span class="p">):</span>
       <span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">linear</span><span class="p">(</span><span class="n">x</span><span class="p">)</span>

<span class="c1"># initialize a floating point model</span>
<span class="n">float_model</span> <span class="o">=</span> <span class="n">M</span><span class="p">()</span><span class="o">.</span><span class="n">eval</span><span class="p">()</span>

<span class="c1"># define calibration function</span>
<span class="k">def</span> <span class="nf">calibrate</span><span class="p">(</span><span class="n">model</span><span class="p">,</span> <span class="n">data_loader</span><span class="p">):</span>
    <span class="n">model</span><span class="o">.</span><span class="n">eval</span><span class="p">()</span>
    <span class="k">with</span> <span class="n">torch</span><span class="o">.</span><span class="n">no_grad</span><span class="p">():</span>
        <span class="k">for</span> <span class="n">image</span><span class="p">,</span> <span class="n">target</span> <span class="ow">in</span> <span class="n">data_loader</span><span class="p">:</span>
            <span class="n">model</span><span class="p">(</span><span class="n">image</span><span class="p">)</span>

<span class="c1"># Step 1. program capture</span>
<span class="c1"># NOTE: this API will be updated to torch.export API in the future, but the captured</span>
<span class="c1"># result shoud mostly stay the same</span>
<span class="n">m</span> <span class="o">=</span> <span class="n">capture_pre_autograd_graph</span><span class="p">(</span><span class="n">m</span><span class="p">,</span> <span class="o">*</span><span class="n">example_inputs</span><span class="p">)</span>
<span class="c1"># we get a model with aten ops</span>

<span class="c1"># Step 2. quantization</span>
<span class="c1"># backend developer will write their own Quantizer and expose methods to allow</span>
<span class="c1"># users to express how they</span>
<span class="c1"># want the model to be quantized</span>
<span class="n">quantizer</span> <span class="o">=</span> <span class="n">XNNPACKQuantizer</span><span class="p">()</span><span class="o">.</span><span class="n">set_global</span><span class="p">(</span><span class="n">get_symmetric_quantization_config</span><span class="p">())</span>
<span class="c1"># or prepare_qat_pt2e for Quantization Aware Training</span>
<span class="n">m</span> <span class="o">=</span> <span class="n">prepare_pt2e</span><span class="p">(</span><span class="n">m</span><span class="p">,</span> <span class="n">quantizer</span><span class="p">)</span>

<span class="c1"># run calibration</span>
<span class="c1"># calibrate(m, sample_inference_data)</span>
<span class="n">m</span> <span class="o">=</span> <span class="n">convert_pt2e</span><span class="p">(</span><span class="n">m</span><span class="p">)</span>

<span class="c1"># Step 3. lowering</span>
<span class="c1"># lower to target backend</span>
</pre></div>
</div>
<p>Please follow these tutorials to get started on PyTorch 2 Export Quantization:</p>
<p>Modeling Users:</p>
<ul class="simple">
<li><p><a class="reference external" href="https://pytorch.org/tutorials/prototype/pt2e_quant_ptq.html">PyTorch 2 Export Post Training Quantization</a></p></li>
<li><p><a class="reference external" href="https://pytorch.org/tutorials/prototype/pt2e_quant_ptq_x86_inductor.html">PyTorch 2 Export Post Training Quantization with X86 Backend through Inductor</a></p></li>
<li><p><a class="reference external" href="https://pytorch.org/tutorials/prototype/pt2e_quant_qat.html">PyTorch 2 Export Quantization Aware Training</a></p></li>
</ul>
<p>Backend Developers (please check out all Modeling Users docs as well):</p>
<ul class="simple">
<li><p><a class="reference external" href="https://pytorch.org/tutorials/prototype/pt2e_quantizer.html">How to Write a Quantizer for PyTorch 2 Export Quantization</a></p></li>
</ul>
</div>
</div>
<div class="section" id="quantization-stack">
<h2>Quantization Stack<a class="headerlink" href="#quantization-stack" title="Permalink to this heading">¶</a></h2>
<p>Quantization is the process to convert a floating point model to a quantized model. So at high level the quantization stack can be split into two parts: 1). The building blocks or abstractions for a quantized model 2). The building blocks or abstractions for the quantization flow that converts a floating point model to a quantized model</p>
<div class="section" id="quantized-model">
<h3>Quantized Model<a class="headerlink" href="#quantized-model" title="Permalink to this heading">¶</a></h3>
<div class="section" id="quantized-tensor">
<h4>Quantized Tensor<a class="headerlink" href="#quantized-tensor" title="Permalink to this heading">¶</a></h4>
<p>In order to do quantization in PyTorch, we need to be able to represent
quantized data in Tensors. A Quantized Tensor allows for storing
quantized data (represented as int8/uint8/int32) along with quantization
parameters like scale and zero_point. Quantized Tensors allow for many
useful operations making quantized arithmetic easy, in addition to
allowing for serialization of data in a quantized format.</p>
<p>PyTorch supports both per tensor and per channel symmetric and asymmetric quantization. Per tensor means that all the values within the tensor are quantized the same way with the same quantization parameters. Per channel means that for each dimension, typically the channel dimension of a tensor, the values in the tensor are quantized with different quantization parameters. This allows for less error in converting tensors to quantized values since outlier values would only impact the channel it was in, instead of the entire Tensor.</p>
<p>The mapping is performed by converting the floating point tensors using</p>
<a class="reference internal image-reference" href="_images/math-quantizer-equation.png"><img alt="_images/math-quantizer-equation.png" src="_images/math-quantizer-equation.png" style="width: 40%;" /></a>
<p>Note that, we ensure that zero in floating point is represented with no error
after quantization, thereby ensuring that operations like padding do not cause
additional quantization error.</p>
<p>Here are a few key attributes for quantized Tensor:</p>
<ul class="simple">
<li><p>QScheme (torch.qscheme): a enum that specifies the way we quantize the Tensor</p>
<ul>
<li><p>torch.per_tensor_affine</p></li>
<li><p>torch.per_tensor_symmetric</p></li>
<li><p>torch.per_channel_affine</p></li>
<li><p>torch.per_channel_symmetric</p></li>
</ul>
</li>
<li><p>dtype (torch.dtype): data type of the quantized Tensor</p>
<ul>
<li><p>torch.quint8</p></li>
<li><p>torch.qint8</p></li>
<li><p>torch.qint32</p></li>
<li><p>torch.float16</p></li>
</ul>
</li>
<li><p>quantization parameters (varies based on QScheme): parameters for the chosen way of quantization</p>
<ul>
<li><p>torch.per_tensor_affine would have quantization parameters of</p>
<ul>
<li><p>scale (float)</p></li>
<li><p>zero_point (int)</p></li>
</ul>
</li>
<li><p>torch.per_channel_affine would have quantization parameters of</p>
<ul>
<li><p>per_channel_scales (list of float)</p></li>
<li><p>per_channel_zero_points (list of int)</p></li>
<li><p>axis (int)</p></li>
</ul>
</li>
</ul>
</li>
</ul>
</div>
<div class="section" id="quantize-and-dequantize">
<h4>Quantize and Dequantize<a class="headerlink" href="#quantize-and-dequantize" title="Permalink to this heading">¶</a></h4>
<p>The input and output of a model are floating point Tensors, but activations in the quantized model are quantized, so we need operators to convert between floating point and quantized Tensors.</p>
<ul class="simple">
<li><p>Quantize (float -&gt; quantized)</p>
<ul>
<li><p>torch.quantize_per_tensor(x, scale, zero_point, dtype)</p></li>
<li><p>torch.quantize_per_channel(x, scales, zero_points, axis, dtype)</p></li>
<li><p>torch.quantize_per_tensor_dynamic(x, dtype, reduce_range)</p></li>
<li><p>to(torch.float16)</p></li>
</ul>
</li>
<li><p>Dequantize (quantized -&gt; float)</p>
<ul>
<li><p>quantized_tensor.dequantize() - calling dequantize on a torch.float16 Tensor will convert the Tensor back to torch.float</p></li>
<li><p>torch.dequantize(x)</p></li>
</ul>
</li>
</ul>
</div>
<div class="section" id="quantized-operators-modules">
<h4>Quantized Operators/Modules<a class="headerlink" href="#quantized-operators-modules" title="Permalink to this heading">¶</a></h4>
<ul class="simple">
<li><p>Quantized Operator are the operators that takes quantized Tensor as inputs, and outputs a quantized Tensor.</p></li>
<li><p>Quantized Modules are PyTorch Modules that performs quantized operations. They are typically defined for weighted operations like linear and conv.</p></li>
</ul>
</div>
<div class="section" id="quantized-engine">
<h4>Quantized Engine<a class="headerlink" href="#quantized-engine" title="Permalink to this heading">¶</a></h4>
<p>When a quantized model is executed, the qengine (torch.backends.quantized.engine) specifies which backend is to be used for execution. It is important to ensure that the qengine is compatible with the quantized model in terms of value range of quantized activation and weights.</p>
</div>
</div>
<div class="section" id="quantization-flow">
<h3>Quantization Flow<a class="headerlink" href="#quantization-flow" title="Permalink to this heading">¶</a></h3>
<div class="section" id="observer-and-fakequantize">
<h4>Observer and FakeQuantize<a class="headerlink" href="#observer-and-fakequantize" title="Permalink to this heading">¶</a></h4>
<ul class="simple">
<li><p>Observer are PyTorch Modules used to:</p>
<ul>
<li><p>collect tensor statistics like min value and max value of the Tensor passing through the observer</p></li>
<li><p>and calculate quantization parameters based on the collected tensor statistics</p></li>
</ul>
</li>
<li><p>FakeQuantize are PyTorch Modules used to:</p>
<ul>
<li><p>simulate quantization (performing quantize/dequantize) for a Tensor in the network</p></li>
<li><p>it can calculate quantization parameters based on the collected statistics from observer, or it can learn the quantization parameters as well</p></li>
</ul>
</li>
</ul>
</div>
<div class="section" id="qconfig">
<h4>QConfig<a class="headerlink" href="#qconfig" title="Permalink to this heading">¶</a></h4>
<ul class="simple">
<li><p>QConfig is a namedtuple of Observer or FakeQuantize Module class that can are configurable with qscheme, dtype etc. it is used to configure how an operator should be observed</p>
<ul>
<li><p>Quantization configuration for an operator/module</p>
<ul>
<li><p>different types of Observer/FakeQuantize</p></li>
<li><p>dtype</p></li>
<li><p>qscheme</p></li>
<li><p>quant_min/quant_max: can be used to simulate lower precision Tensors</p></li>
</ul>
</li>
<li><p>Currently supports configuration for activation and weight</p></li>
<li><p>We insert input/weight/output observer based on the qconfig that is configured for a given operator or module</p></li>
</ul>
</li>
</ul>
</div>
<div class="section" id="general-quantization-flow">
<h4>General Quantization Flow<a class="headerlink" href="#general-quantization-flow" title="Permalink to this heading">¶</a></h4>
<p>In general, the flow is the following</p>
<ul class="simple">
<li><p>prepare</p>
<ul>
<li><p>insert Observer/FakeQuantize modules based on user specified qconfig</p></li>
</ul>
</li>
<li><p>calibrate/train (depending on post training quantization or quantization aware training)</p>
<ul>
<li><p>allow Observers to collect statistics or FakeQuantize modules to learn the quantization parameters</p></li>
</ul>
</li>
<li><p>convert</p>
<ul>
<li><p>convert a calibrated/trained model to a quantized model</p></li>
</ul>
</li>
</ul>
<p>There are different modes of quantization, they can be classified in two ways:</p>
<p>In terms of where we apply the quantization flow, we have:</p>
<ol class="arabic simple">
<li><p>Post Training Quantization (apply quantization after training, quantization parameters are calculated based on sample calibration data)</p></li>
<li><p>Quantization Aware Training (simulate quantization during training so that the quantization parameters can be learned together with the model using training data)</p></li>
</ol>
<p>And in terms of how we quantize the operators, we can have:</p>
<ul class="simple">
<li><p>Weight Only Quantization (only weight is statically quantized)</p></li>
<li><p>Dynamic Quantization (weight is statically quantized, activation is dynamically quantized)</p></li>
<li><p>Static Quantization (both weight and activations are statically quantized)</p></li>
</ul>
<p>We can mix different ways of quantizing operators in the same quantization flow. For example, we can have post training quantization that has both statically and dynamically quantized operators.</p>
</div>
</div>
</div>
<div class="section" id="quantization-support-matrix">
<h2>Quantization Support Matrix<a class="headerlink" href="#quantization-support-matrix" title="Permalink to this heading">¶</a></h2>
<div class="section" id="quantization-mode-support">
<h3>Quantization Mode Support<a class="headerlink" href="#quantization-mode-support" title="Permalink to this heading">¶</a></h3>
<table class="docutils colwidths-auto align-default">
<tbody>
<tr class="row-odd"><td></td>
<td colspan="2"><p>Quantization
Mode</p></td>
<td><p>Dataset
Requirement</p></td>
<td><p>Works Best For</p></td>
<td><p>Accuracy</p></td>
<td><p>Notes</p></td>
</tr>
<tr class="row-even"><td rowspan="2"><p>Post Training Quantization</p></td>
<td><p>Dynamic/Weight Only Quantization</p></td>
<td><p>activation
dynamically
quantized (fp16,
int8) or not
quantized, weight
statically quantized
(fp16, int8, in4)</p></td>
<td><p>None</p></td>
<td><p>LSTM, MLP,
Embedding,
Transformer</p></td>
<td><p>good</p></td>
<td><p>Easy to use,
close to static
quantization when
performance is
compute or memory
bound due to
weights</p></td>
</tr>
<tr class="row-odd"><td><p>Static Quantization</p></td>
<td><p>activation and
weights statically
quantized (int8)</p></td>
<td><p>calibration
dataset</p></td>
<td><p>CNN</p></td>
<td><p>good</p></td>
<td><p>Provides best
perf, may have
big impact on
accuracy, good
for hardwares
that only support
int8 computation</p></td>
</tr>
<tr class="row-even"><td rowspan="2"><p>Quantization Aware Training</p></td>
<td><p>Dynamic Quantization</p></td>
<td><p>activation and
weight are fake
quantized</p></td>
<td><p>fine-tuning
dataset</p></td>
<td><p>MLP, Embedding</p></td>
<td><p>best</p></td>
<td><p>Limited support
for now</p></td>
</tr>
<tr class="row-odd"><td><p>Static Quantization</p></td>
<td><p>activation and
weight are fake
quantized</p></td>
<td><p>fine-tuning
dataset</p></td>
<td><p>CNN, MLP,
Embedding</p></td>
<td><p>best</p></td>
<td><p>Typically used
when static
quantization
leads to bad
accuracy, and
used to close the
accuracy gap</p></td>
</tr>
</tbody>
</table>
<p>Please see our <a class="reference external" href="https://pytorch.org/blog/introduction-to-quantization-on-pytorch/">Introduction to Quantization on Pytorch</a> blog post
for a more comprehensive overview of the tradeoffs between these quantization
types.</p>
</div>
<div class="section" id="quantization-flow-support">
<h3>Quantization Flow Support<a class="headerlink" href="#quantization-flow-support" title="Permalink to this heading">¶</a></h3>
<p>PyTorch provides two modes of quantization: Eager Mode Quantization and FX Graph Mode Quantization.</p>
<p>Eager Mode Quantization is a beta feature. User needs to do fusion and specify where quantization and dequantization happens manually, also it only supports modules and not functionals.</p>
<p>FX Graph Mode Quantization is an automated quantization framework in PyTorch, and currently it’s a prototype feature. It improves upon Eager Mode Quantization by adding support for functionals and automating the quantization process, although people might need to refactor the model to make the model compatible with FX Graph Mode Quantization (symbolically traceable with <code class="docutils literal notranslate"><span class="pre">torch.fx</span></code>). Note that FX Graph Mode Quantization is not expected to work on arbitrary models since the model might not be symbolically traceable, we will integrate it into domain libraries like torchvision and users will be able to quantize models similar to the ones in supported domain libraries with FX Graph Mode Quantization. For arbitrary models we’ll provide general guidelines, but to actually make it work, users might need to be familiar with <code class="docutils literal notranslate"><span class="pre">torch.fx</span></code>, especially on how to make a model symbolically traceable.</p>
<p>New users of quantization are encouraged to try out FX Graph Mode Quantization first, if it does not work, user may try to follow the guideline of <a class="reference external" href="https://pytorch.org/tutorials/prototype/fx_graph_mode_quant_guide.html">using FX Graph Mode Quantization</a> or fall back to eager mode quantization.</p>
<p>The following table compares the differences between Eager Mode Quantization and FX Graph Mode Quantization:</p>
<table class="docutils colwidths-auto align-default">
<tbody>
<tr class="row-odd"><td></td>
<td><p>Eager Mode
Quantization</p></td>
<td><p>FX Graph
Mode
Quantization</p></td>
</tr>
<tr class="row-even"><td><p>Release
Status</p></td>
<td><p>beta</p></td>
<td><p>prototype</p></td>
</tr>
<tr class="row-odd"><td><p>Operator
Fusion</p></td>
<td><p>Manual</p></td>
<td><p>Automatic</p></td>
</tr>
<tr class="row-even"><td><p>Quant/DeQuant
Placement</p></td>
<td><p>Manual</p></td>
<td><p>Automatic</p></td>
</tr>
<tr class="row-odd"><td><p>Quantizing
Modules</p></td>
<td><p>Supported</p></td>
<td><p>Supported</p></td>
</tr>
<tr class="row-even"><td><p>Quantizing
Functionals/Torch
Ops</p></td>
<td><p>Manual</p></td>
<td><p>Automatic</p></td>
</tr>
<tr class="row-odd"><td><p>Support for
Customization</p></td>
<td><p>Limited Support</p></td>
<td><p>Fully
Supported</p></td>
</tr>
<tr class="row-even"><td><p>Quantization Mode
Support</p></td>
<td><p>Post Training
Quantization:
Static, Dynamic,
Weight Only</p>
<p>Quantization Aware
Training:
Static</p>
</td>
<td><p>Post Training
Quantization:
Static, Dynamic,
Weight Only</p>
<p>Quantization Aware
Training:
Static</p>
</td>
</tr>
<tr class="row-odd"><td><p>Input/Output
Model Type</p></td>
<td><p><code class="docutils literal notranslate"><span class="pre">torch.nn.Module</span></code></p></td>
<td><p><code class="docutils literal notranslate"><span class="pre">torch.nn.Module</span></code>
(May need some
refactors to make
the model
compatible with FX
Graph Mode
Quantization)</p></td>
</tr>
</tbody>
</table>
</div>
<div class="section" id="backend-hardware-support">
<h3>Backend/Hardware Support<a class="headerlink" href="#backend-hardware-support" title="Permalink to this heading">¶</a></h3>
<table class="docutils colwidths-auto align-default">
<tbody>
<tr class="row-odd"><td><p>Hardware</p></td>
<td><p>Kernel Library</p></td>
<td><p>Eager Mode
Quantization</p></td>
<td><p>FX Graph
Mode
Quantization</p></td>
<td><p>Quantization
Mode Support</p></td>
</tr>
<tr class="row-even"><td><p>server CPU</p></td>
<td><p>fbgemm/onednn</p></td>
<td colspan="2" rowspan="2"><p>Supported</p></td>
<td rowspan="2"><p>All
Supported</p></td>
</tr>
<tr class="row-odd"><td><p>mobile CPU</p></td>
<td><p>qnnpack/xnnpack</p></td>
</tr>
<tr class="row-even"><td><p>server GPU</p></td>
<td><p>TensorRT (early
prototype)</p></td>
<td><p>Not support
this it
requires a
graph</p></td>
<td><p>Supported</p></td>
<td><p>Static
Quantization</p></td>
</tr>
</tbody>
</table>
<p>Today, PyTorch supports the following backends for running quantized operators efficiently:</p>
<ul class="simple">
<li><p>x86 CPUs with AVX2 support or higher (without AVX2 some operations have inefficient implementations), via <cite>x86</cite> optimized by <a class="reference external" href="https://github.com/pytorch/FBGEMM">fbgemm</a> and <a class="reference external" href="https://github.com/oneapi-src/oneDNN">onednn</a> (see the details at <a class="reference external" href="https://github.com/pytorch/pytorch/issues/83888">RFC</a>)</p></li>
<li><p>ARM CPUs (typically found in mobile/embedded devices), via <a class="reference external" href="https://github.com/pytorch/pytorch/tree/main/aten/src/ATen/native/quantized/cpu/qnnpack">qnnpack</a></p></li>
<li><p>(early prototype) support for NVidia GPU via <a class="reference external" href="https://developer.nvidia.com/tensorrt">TensorRT</a> through <cite>fx2trt</cite> (to be open sourced)</p></li>
</ul>
<div class="section" id="note-for-native-cpu-backends">
<h4>Note for native CPU backends<a class="headerlink" href="#note-for-native-cpu-backends" title="Permalink to this heading">¶</a></h4>
<p>We expose both <cite>x86</cite> and <cite>qnnpack</cite> with the same native pytorch quantized operators, so we need additional flag to distinguish between them. The corresponding implementation of  <cite>x86</cite> and <cite>qnnpack</cite> is chosen automatically based on the PyTorch build mode, though users have the option to override this by setting <cite>torch.backends.quantization.engine</cite> to <cite>x86</cite> or <cite>qnnpack</cite>.</p>
<p>When preparing a quantized model, it is necessary to ensure that qconfig
and the engine used for quantized computations match the backend on which
the model will be executed. The qconfig controls the type of observers used
during the quantization passes. The qengine controls whether <cite>x86</cite> or <cite>qnnpack</cite>
specific packing function is used when packing weights for
linear and convolution functions and modules. For example:</p>
<p>Default settings for x86:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="c1"># set the qconfig for PTQ</span>
<span class="c1"># Note: the old &#39;fbgemm&#39; is still available but &#39;x86&#39; is the recommended default on x86 CPUs</span>
<span class="n">qconfig</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">ao</span><span class="o">.</span><span class="n">quantization</span><span class="o">.</span><span class="n">get_default_qconfig</span><span class="p">(</span><span class="s1">&#39;x86&#39;</span><span class="p">)</span>
<span class="c1"># or, set the qconfig for QAT</span>
<span class="n">qconfig</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">ao</span><span class="o">.</span><span class="n">quantization</span><span class="o">.</span><span class="n">get_default_qat_qconfig</span><span class="p">(</span><span class="s1">&#39;x86&#39;</span><span class="p">)</span>
<span class="c1"># set the qengine to control weight packing</span>
<span class="n">torch</span><span class="o">.</span><span class="n">backends</span><span class="o">.</span><span class="n">quantized</span><span class="o">.</span><span class="n">engine</span> <span class="o">=</span> <span class="s1">&#39;x86&#39;</span>
</pre></div>
</div>
<p>Default settings for qnnpack:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="c1"># set the qconfig for PTQ</span>
<span class="n">qconfig</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">ao</span><span class="o">.</span><span class="n">quantization</span><span class="o">.</span><span class="n">get_default_qconfig</span><span class="p">(</span><span class="s1">&#39;qnnpack&#39;</span><span class="p">)</span>
<span class="c1"># or, set the qconfig for QAT</span>
<span class="n">qconfig</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">ao</span><span class="o">.</span><span class="n">quantization</span><span class="o">.</span><span class="n">get_default_qat_qconfig</span><span class="p">(</span><span class="s1">&#39;qnnpack&#39;</span><span class="p">)</span>
<span class="c1"># set the qengine to control weight packing</span>
<span class="n">torch</span><span class="o">.</span><span class="n">backends</span><span class="o">.</span><span class="n">quantized</span><span class="o">.</span><span class="n">engine</span> <span class="o">=</span> <span class="s1">&#39;qnnpack&#39;</span>
</pre></div>
</div>
</div>
</div>
<div class="section" id="operator-support">
<h3>Operator Support<a class="headerlink" href="#operator-support" title="Permalink to this heading">¶</a></h3>
<p>Operator coverage varies between dynamic and static quantization and is captured in the table below.
Note that for FX Graph Mode Quantization, the corresponding functionals are also supported.</p>
<table class="docutils colwidths-auto align-default">
<tbody>
<tr class="row-odd"><td></td>
<td><p>Static
Quantization</p></td>
<td><p>Dynamic
Quantization</p></td>
</tr>
<tr class="row-even"><td><div class="line-block">
<div class="line">nn.Linear</div>
<div class="line">nn.Conv1d/2d/3d</div>
</div>
</td>
<td><div class="line-block">
<div class="line">Y</div>
<div class="line">Y</div>
</div>
</td>
<td><div class="line-block">
<div class="line">Y</div>
<div class="line">N</div>
</div>
</td>
</tr>
<tr class="row-odd"><td><div class="line-block">
<div class="line">nn.LSTM</div>
<div class="line">nn.GRU</div>
</div>
</td>
<td><div class="line-block">
<div class="line">N</div>
<div class="line">N</div>
</div>
</td>
<td><div class="line-block">
<div class="line">Y</div>
<div class="line">Y</div>
</div>
</td>
</tr>
<tr class="row-even"><td><div class="line-block">
<div class="line">nn.RNNCell</div>
<div class="line">nn.GRUCell</div>
<div class="line">nn.LSTMCell</div>
</div>
</td>
<td><div class="line-block">
<div class="line">N</div>
<div class="line">N</div>
<div class="line">N</div>
</div>
</td>
<td><div class="line-block">
<div class="line">Y</div>
<div class="line">Y</div>
<div class="line">Y</div>
</div>
</td>
</tr>
<tr class="row-odd"><td><p>nn.EmbeddingBag</p></td>
<td><p>Y (activations
are in fp32)</p></td>
<td><p>Y</p></td>
</tr>
<tr class="row-even"><td><p>nn.Embedding</p></td>
<td><p>Y</p></td>
<td><p>Y</p></td>
</tr>
<tr class="row-odd"><td><p>nn.MultiheadAttention</p></td>
<td><p>Not Supported</p></td>
<td><p>Not supported</p></td>
</tr>
<tr class="row-even"><td><p>Activations</p></td>
<td><p>Broadly supported</p></td>
<td><p>Un-changed,
computations
stay in fp32</p></td>
</tr>
</tbody>
</table>
<p>Note: this will be updated with some information generated from native backend_config_dict soon.</p>
</div>
</div>
<div class="section" id="quantization-api-reference">
<h2>Quantization API Reference<a class="headerlink" href="#quantization-api-reference" title="Permalink to this heading">¶</a></h2>
<p>The <a class="reference internal" href="quantization-support.html"><span class="doc">Quantization API Reference</span></a> contains documentation
of quantization APIs, such as quantization passes, quantized tensor operations,
and supported quantized modules and functions.</p>
<div class="toctree-wrapper compound">
</div>
</div>
<div class="section" id="quantization-backend-configuration">
<h2>Quantization Backend Configuration<a class="headerlink" href="#quantization-backend-configuration" title="Permalink to this heading">¶</a></h2>
<p>The <a class="reference internal" href="quantization-backend-configuration.html"><span class="doc">Quantization Backend Configuration</span></a> contains documentation
on how to configure the quantization workflows for various backends.</p>
<div class="toctree-wrapper compound">
</div>
</div>
<div class="section" id="quantization-accuracy-debugging">
<h2>Quantization Accuracy Debugging<a class="headerlink" href="#quantization-accuracy-debugging" title="Permalink to this heading">¶</a></h2>
<p>The <a class="reference internal" href="quantization-accuracy-debugging.html"><span class="doc">Quantization Accuracy Debugging</span></a> contains documentation
on how to debug quantization accuracy.</p>
<div class="toctree-wrapper compound">
</div>
</div>
<div class="section" id="quantization-customizations">
<h2>Quantization Customizations<a class="headerlink" href="#quantization-customizations" title="Permalink to this heading">¶</a></h2>
<p>While default implementations of observers to select the scale factor and bias
based on observed tensor data are provided, developers can provide their own
quantization functions. Quantization can be applied selectively to different
parts of the model or configured differently for different parts of the model.</p>
<p>We also provide support for per channel quantization for <strong>conv1d()</strong>, <strong>conv2d()</strong>,
<strong>conv3d()</strong> and <strong>linear()</strong>.</p>
<p>Quantization workflows work by adding (e.g. adding observers as
<code class="docutils literal notranslate"><span class="pre">.observer</span></code> submodule) or replacing (e.g. converting <code class="docutils literal notranslate"><span class="pre">nn.Conv2d</span></code> to
<code class="docutils literal notranslate"><span class="pre">nn.quantized.Conv2d</span></code>) submodules in the model’s module hierarchy. It
means that the model stays a regular <code class="docutils literal notranslate"><span class="pre">nn.Module</span></code>-based instance throughout the
process and thus can work with the rest of PyTorch APIs.</p>
<div class="section" id="quantization-custom-module-api">
<h3>Quantization Custom Module API<a class="headerlink" href="#quantization-custom-module-api" title="Permalink to this heading">¶</a></h3>
<p>Both Eager mode and FX graph mode quantization APIs provide a hook for the user
to specify module quantized in a custom way, with user defined logic for
observation and quantization. The user needs to specify:</p>
<ol class="arabic simple">
<li><p>The Python type of the source fp32 module (existing in the model)</p></li>
<li><p>The Python type of the observed module (provided by user). This module needs
to define a <cite>from_float</cite> function which defines how the observed module is
created from the original fp32 module.</p></li>
<li><p>The Python type of the quantized module (provided by user). This module needs
to define a <cite>from_observed</cite> function which defines how the quantized module is
created from the observed module.</p></li>
<li><p>A configuration describing (1), (2), (3) above, passed to the quantization APIs.</p></li>
</ol>
<p>The framework will then do the following:</p>
<ol class="arabic simple">
<li><p>during the <cite>prepare</cite> module swaps, it will convert every module of type
specified in (1) to the type specified in (2), using the <cite>from_float</cite> function of
the class in (2).</p></li>
<li><p>during the <cite>convert</cite> module swaps, it will convert every module of type
specified in (2) to the type specified in (3), using the <cite>from_observed</cite> function
of the class in (3).</p></li>
</ol>
<p>Currently, there is a requirement that <cite>ObservedCustomModule</cite> will have a single
Tensor output, and an observer will be added by the framework (not by the user)
on that output. The observer will be stored under the <cite>activation_post_process</cite> key
as an attribute of the custom module instance. Relaxing these restrictions may
be done at a future time.</p>
<p>Custom API Example:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="kn">import</span> <span class="nn">torch</span>
<span class="kn">import</span> <span class="nn">torch.ao.nn.quantized</span> <span class="k">as</span> <span class="nn">nnq</span>
<span class="kn">from</span> <span class="nn">torch.ao.quantization</span> <span class="kn">import</span> <span class="n">QConfigMapping</span>
<span class="kn">import</span> <span class="nn">torch.ao.quantization.quantize_fx</span>

<span class="c1"># original fp32 module to replace</span>
<span class="k">class</span> <span class="nc">CustomModule</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">nn</span><span class="o">.</span><span class="n">Module</span><span class="p">):</span>
    <span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
        <span class="nb">super</span><span class="p">()</span><span class="o">.</span><span class="fm">__init__</span><span class="p">()</span>
        <span class="bp">self</span><span class="o">.</span><span class="n">linear</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">nn</span><span class="o">.</span><span class="n">Linear</span><span class="p">(</span><span class="mi">3</span><span class="p">,</span> <span class="mi">3</span><span class="p">)</span>

    <span class="k">def</span> <span class="nf">forward</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">x</span><span class="p">):</span>
        <span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">linear</span><span class="p">(</span><span class="n">x</span><span class="p">)</span>

<span class="c1"># custom observed module, provided by user</span>
<span class="k">class</span> <span class="nc">ObservedCustomModule</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">nn</span><span class="o">.</span><span class="n">Module</span><span class="p">):</span>
    <span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">linear</span><span class="p">):</span>
        <span class="nb">super</span><span class="p">()</span><span class="o">.</span><span class="fm">__init__</span><span class="p">()</span>
        <span class="bp">self</span><span class="o">.</span><span class="n">linear</span> <span class="o">=</span> <span class="n">linear</span>

    <span class="k">def</span> <span class="nf">forward</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">x</span><span class="p">):</span>
        <span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">linear</span><span class="p">(</span><span class="n">x</span><span class="p">)</span>

    <span class="nd">@classmethod</span>
    <span class="k">def</span> <span class="nf">from_float</span><span class="p">(</span><span class="bp">cls</span><span class="p">,</span> <span class="n">float_module</span><span class="p">):</span>
        <span class="k">assert</span> <span class="nb">hasattr</span><span class="p">(</span><span class="n">float_module</span><span class="p">,</span> <span class="s1">&#39;qconfig&#39;</span><span class="p">)</span>
        <span class="n">observed</span> <span class="o">=</span> <span class="bp">cls</span><span class="p">(</span><span class="n">float_module</span><span class="o">.</span><span class="n">linear</span><span class="p">)</span>
        <span class="n">observed</span><span class="o">.</span><span class="n">qconfig</span> <span class="o">=</span> <span class="n">float_module</span><span class="o">.</span><span class="n">qconfig</span>
        <span class="k">return</span> <span class="n">observed</span>

<span class="c1"># custom quantized module, provided by user</span>
<span class="k">class</span> <span class="nc">StaticQuantCustomModule</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">nn</span><span class="o">.</span><span class="n">Module</span><span class="p">):</span>
    <span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">linear</span><span class="p">):</span>
        <span class="nb">super</span><span class="p">()</span><span class="o">.</span><span class="fm">__init__</span><span class="p">()</span>
        <span class="bp">self</span><span class="o">.</span><span class="n">linear</span> <span class="o">=</span> <span class="n">linear</span>

    <span class="k">def</span> <span class="nf">forward</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">x</span><span class="p">):</span>
        <span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">linear</span><span class="p">(</span><span class="n">x</span><span class="p">)</span>

    <span class="nd">@classmethod</span>
    <span class="k">def</span> <span class="nf">from_observed</span><span class="p">(</span><span class="bp">cls</span><span class="p">,</span> <span class="n">observed_module</span><span class="p">):</span>
        <span class="k">assert</span> <span class="nb">hasattr</span><span class="p">(</span><span class="n">observed_module</span><span class="p">,</span> <span class="s1">&#39;qconfig&#39;</span><span class="p">)</span>
        <span class="k">assert</span> <span class="nb">hasattr</span><span class="p">(</span><span class="n">observed_module</span><span class="p">,</span> <span class="s1">&#39;activation_post_process&#39;</span><span class="p">)</span>
        <span class="n">observed_module</span><span class="o">.</span><span class="n">linear</span><span class="o">.</span><span class="n">activation_post_process</span> <span class="o">=</span> \
            <span class="n">observed_module</span><span class="o">.</span><span class="n">activation_post_process</span>
        <span class="n">quantized</span> <span class="o">=</span> <span class="bp">cls</span><span class="p">(</span><span class="n">nnq</span><span class="o">.</span><span class="n">Linear</span><span class="o">.</span><span class="n">from_float</span><span class="p">(</span><span class="n">observed_module</span><span class="o">.</span><span class="n">linear</span><span class="p">))</span>
        <span class="k">return</span> <span class="n">quantized</span>

<span class="c1">#</span>
<span class="c1"># example API call (Eager mode quantization)</span>
<span class="c1">#</span>

<span class="n">m</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">nn</span><span class="o">.</span><span class="n">Sequential</span><span class="p">(</span><span class="n">CustomModule</span><span class="p">())</span><span class="o">.</span><span class="n">eval</span><span class="p">()</span>
<span class="n">prepare_custom_config_dict</span> <span class="o">=</span> <span class="p">{</span>
    <span class="s2">&quot;float_to_observed_custom_module_class&quot;</span><span class="p">:</span> <span class="p">{</span>
        <span class="n">CustomModule</span><span class="p">:</span> <span class="n">ObservedCustomModule</span>
    <span class="p">}</span>
<span class="p">}</span>
<span class="n">convert_custom_config_dict</span> <span class="o">=</span> <span class="p">{</span>
    <span class="s2">&quot;observed_to_quantized_custom_module_class&quot;</span><span class="p">:</span> <span class="p">{</span>
        <span class="n">ObservedCustomModule</span><span class="p">:</span> <span class="n">StaticQuantCustomModule</span>
    <span class="p">}</span>
<span class="p">}</span>
<span class="n">m</span><span class="o">.</span><span class="n">qconfig</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">ao</span><span class="o">.</span><span class="n">quantization</span><span class="o">.</span><span class="n">default_qconfig</span>
<span class="n">mp</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">ao</span><span class="o">.</span><span class="n">quantization</span><span class="o">.</span><span class="n">prepare</span><span class="p">(</span>
    <span class="n">m</span><span class="p">,</span> <span class="n">prepare_custom_config_dict</span><span class="o">=</span><span class="n">prepare_custom_config_dict</span><span class="p">)</span>
<span class="c1"># calibration (not shown)</span>
<span class="n">mq</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">ao</span><span class="o">.</span><span class="n">quantization</span><span class="o">.</span><span class="n">convert</span><span class="p">(</span>
    <span class="n">mp</span><span class="p">,</span> <span class="n">convert_custom_config_dict</span><span class="o">=</span><span class="n">convert_custom_config_dict</span><span class="p">)</span>
<span class="c1">#</span>
<span class="c1"># example API call (FX graph mode quantization)</span>
<span class="c1">#</span>
<span class="n">m</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">nn</span><span class="o">.</span><span class="n">Sequential</span><span class="p">(</span><span class="n">CustomModule</span><span class="p">())</span><span class="o">.</span><span class="n">eval</span><span class="p">()</span>
<span class="n">qconfig_mapping</span> <span class="o">=</span> <span class="n">QConfigMapping</span><span class="p">()</span><span class="o">.</span><span class="n">set_global</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">ao</span><span class="o">.</span><span class="n">quantization</span><span class="o">.</span><span class="n">default_qconfig</span><span class="p">)</span>
<span class="n">prepare_custom_config_dict</span> <span class="o">=</span> <span class="p">{</span>
    <span class="s2">&quot;float_to_observed_custom_module_class&quot;</span><span class="p">:</span> <span class="p">{</span>
        <span class="s2">&quot;static&quot;</span><span class="p">:</span> <span class="p">{</span>
            <span class="n">CustomModule</span><span class="p">:</span> <span class="n">ObservedCustomModule</span><span class="p">,</span>
        <span class="p">}</span>
    <span class="p">}</span>
<span class="p">}</span>
<span class="n">convert_custom_config_dict</span> <span class="o">=</span> <span class="p">{</span>
    <span class="s2">&quot;observed_to_quantized_custom_module_class&quot;</span><span class="p">:</span> <span class="p">{</span>
        <span class="s2">&quot;static&quot;</span><span class="p">:</span> <span class="p">{</span>
            <span class="n">ObservedCustomModule</span><span class="p">:</span> <span class="n">StaticQuantCustomModule</span><span class="p">,</span>
        <span class="p">}</span>
    <span class="p">}</span>
<span class="p">}</span>
<span class="n">mp</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">ao</span><span class="o">.</span><span class="n">quantization</span><span class="o">.</span><span class="n">quantize_fx</span><span class="o">.</span><span class="n">prepare_fx</span><span class="p">(</span>
    <span class="n">m</span><span class="p">,</span> <span class="n">qconfig_mapping</span><span class="p">,</span> <span class="n">torch</span><span class="o">.</span><span class="n">randn</span><span class="p">(</span><span class="mi">3</span><span class="p">,</span><span class="mi">3</span><span class="p">),</span> <span class="n">prepare_custom_config</span><span class="o">=</span><span class="n">prepare_custom_config_dict</span><span class="p">)</span>
<span class="c1"># calibration (not shown)</span>
<span class="n">mq</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">ao</span><span class="o">.</span><span class="n">quantization</span><span class="o">.</span><span class="n">quantize_fx</span><span class="o">.</span><span class="n">convert_fx</span><span class="p">(</span>
    <span class="n">mp</span><span class="p">,</span> <span class="n">convert_custom_config</span><span class="o">=</span><span class="n">convert_custom_config_dict</span><span class="p">)</span>
</pre></div>
</div>
</div>
</div>
<div class="section" id="best-practices">
<h2>Best Practices<a class="headerlink" href="#best-practices" title="Permalink to this heading">¶</a></h2>
<p>1. If you are using the <code class="docutils literal notranslate"><span class="pre">x86</span></code> backend, we need to use 7 bits instead of 8 bits. Make sure you reduce the range for the <code class="docutils literal notranslate"><span class="pre">quant\_min</span></code>, <code class="docutils literal notranslate"><span class="pre">quant\_max</span></code>, e.g.
if <code class="docutils literal notranslate"><span class="pre">dtype</span></code> is <code class="docutils literal notranslate"><span class="pre">torch.quint8</span></code>, make sure to set a custom <code class="docutils literal notranslate"><span class="pre">quant_min</span></code> to be <code class="docutils literal notranslate"><span class="pre">0</span></code> and <code class="docutils literal notranslate"><span class="pre">quant_max</span></code> to be <code class="docutils literal notranslate"><span class="pre">127</span></code> (<code class="docutils literal notranslate"><span class="pre">255</span></code> / <code class="docutils literal notranslate"><span class="pre">2</span></code>)
if <code class="docutils literal notranslate"><span class="pre">dtype</span></code> is <code class="docutils literal notranslate"><span class="pre">torch.qint8</span></code>, make sure to set a custom <code class="docutils literal notranslate"><span class="pre">quant_min</span></code> to be <code class="docutils literal notranslate"><span class="pre">-64</span></code> (<code class="docutils literal notranslate"><span class="pre">-128</span></code> / <code class="docutils literal notranslate"><span class="pre">2</span></code>) and <code class="docutils literal notranslate"><span class="pre">quant_max</span></code> to be <code class="docutils literal notranslate"><span class="pre">63</span></code> (<code class="docutils literal notranslate"><span class="pre">127</span></code> / <code class="docutils literal notranslate"><span class="pre">2</span></code>), we already set this correctly if
you call the <cite>torch.ao.quantization.get_default_qconfig(backend)</cite> or <cite>torch.ao.quantization.get_default_qat_qconfig(backend)</cite> function to get the default <code class="docutils literal notranslate"><span class="pre">qconfig</span></code> for
<code class="docutils literal notranslate"><span class="pre">x86</span></code> or <code class="docutils literal notranslate"><span class="pre">qnnpack</span></code> backend</p>
<p>2. If <code class="docutils literal notranslate"><span class="pre">onednn</span></code> backend is selected, 8 bits for activation will be used in the default qconfig mapping <code class="docutils literal notranslate"><span class="pre">torch.ao.quantization.get_default_qconfig_mapping('onednn')</span></code>
and default qconfig <code class="docutils literal notranslate"><span class="pre">torch.ao.quantization.get_default_qconfig('onednn')</span></code>. It is recommended to be used on CPUs with Vector Neural Network Instruction (VNNI)
support. Otherwise, setting <code class="docutils literal notranslate"><span class="pre">reduce_range</span></code> to True of the activation’s observer to get better accuracy on CPUs without VNNI support.</p>
</div>
<div class="section" id="frequently-asked-questions">
<h2>Frequently Asked Questions<a class="headerlink" href="#frequently-asked-questions" title="Permalink to this heading">¶</a></h2>
<ol class="arabic">
<li><p>How can I do quantized inference on GPU?:</p>
<p>We don’t have official GPU support yet, but this is an area of active development, you can find more information
<a class="reference external" href="https://github.com/pytorch/pytorch/issues/87395">here</a></p>
</li>
<li><p>Where can I get ONNX support for my quantized model?</p>
<p>If you get errors exporting the model (using APIs under <code class="docutils literal notranslate"><span class="pre">torch.onnx</span></code>), you may open an issue in the PyTorch repository. Prefix the issue title with <code class="docutils literal notranslate"><span class="pre">[ONNX]</span></code> and tag the issue as <code class="docutils literal notranslate"><span class="pre">module:</span> <span class="pre">onnx</span></code>.</p>
<p>If you encounter issues with ONNX Runtime, open an issue at <a class="reference external" href="https://github.com/microsoft/onnxruntime/issues/">GitHub - microsoft/onnxruntime</a>.</p>
</li>
<li><p>How can I use quantization with LSTM’s?:</p>
<p>LSTM is supported through our custom module api in both eager mode and fx graph mode quantization. Examples can be found at
Eager Mode: <a class="reference external" href="https://github.com/pytorch/pytorch/blob/9b88dcf248e717ca6c3f8c5e11f600825547a561/test/quantization/core/test_quantized_op.py#L2782">pytorch/test_quantized_op.py TestQuantizedOps.test_custom_module_lstm</a>
FX Graph Mode: <a class="reference external" href="https://github.com/pytorch/pytorch/blob/9b88dcf248e717ca6c3f8c5e11f600825547a561/test/quantization/fx/test_quantize_fx.py#L4116">pytorch/test_quantize_fx.py TestQuantizeFx.test_static_lstm</a></p>
</li>
</ol>
</div>
<div class="section" id="common-errors">
<h2>Common Errors<a class="headerlink" href="#common-errors" title="Permalink to this heading">¶</a></h2>
<div class="section" id="passing-a-non-quantized-tensor-into-a-quantized-kernel">
<h3>Passing a non-quantized Tensor into a quantized kernel<a class="headerlink" href="#passing-a-non-quantized-tensor-into-a-quantized-kernel" title="Permalink to this heading">¶</a></h3>
<p>If you see an error similar to:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="ne">RuntimeError</span><span class="p">:</span> <span class="n">Could</span> <span class="ow">not</span> <span class="n">run</span> <span class="s1">&#39;quantized::some_operator&#39;</span> <span class="k">with</span> <span class="n">arguments</span> <span class="kn">from</span> <span class="nn">the</span> <span class="s1">&#39;CPU&#39;</span> <span class="n">backend</span><span class="o">...</span>
</pre></div>
</div>
<p>This means that you are trying to pass a non-quantized Tensor to a quantized
kernel. A common workaround is to use <code class="docutils literal notranslate"><span class="pre">torch.ao.quantization.QuantStub</span></code> to
quantize the tensor.  This needs to be done manually in Eager mode quantization.
An e2e example:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="k">class</span> <span class="nc">M</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">nn</span><span class="o">.</span><span class="n">Module</span><span class="p">):</span>
    <span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
        <span class="nb">super</span><span class="p">()</span><span class="o">.</span><span class="fm">__init__</span><span class="p">()</span>
        <span class="bp">self</span><span class="o">.</span><span class="n">quant</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">ao</span><span class="o">.</span><span class="n">quantization</span><span class="o">.</span><span class="n">QuantStub</span><span class="p">()</span>
        <span class="bp">self</span><span class="o">.</span><span class="n">conv</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">nn</span><span class="o">.</span><span class="n">Conv2d</span><span class="p">(</span><span class="mi">1</span><span class="p">,</span> <span class="mi">1</span><span class="p">,</span> <span class="mi">1</span><span class="p">)</span>

    <span class="k">def</span> <span class="nf">forward</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">x</span><span class="p">):</span>
        <span class="c1"># during the convert step, this will be replaced with a</span>
        <span class="c1"># `quantize_per_tensor` call</span>
        <span class="n">x</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">quant</span><span class="p">(</span><span class="n">x</span><span class="p">)</span>
        <span class="n">x</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">conv</span><span class="p">(</span><span class="n">x</span><span class="p">)</span>
        <span class="k">return</span> <span class="n">x</span>
</pre></div>
</div>
</div>
<div class="section" id="passing-a-quantized-tensor-into-a-non-quantized-kernel">
<h3>Passing a quantized Tensor into a non-quantized kernel<a class="headerlink" href="#passing-a-quantized-tensor-into-a-non-quantized-kernel" title="Permalink to this heading">¶</a></h3>
<p>If you see an error similar to:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="ne">RuntimeError</span><span class="p">:</span> <span class="n">Could</span> <span class="ow">not</span> <span class="n">run</span> <span class="s1">&#39;aten::thnn_conv2d_forward&#39;</span> <span class="k">with</span> <span class="n">arguments</span> <span class="kn">from</span> <span class="nn">the</span> <span class="s1">&#39;QuantizedCPU&#39;</span> <span class="n">backend</span><span class="o">.</span>
</pre></div>
</div>
<p>This means that you are trying to pass a quantized Tensor to a non-quantized
kernel. A common workaround is to use <code class="docutils literal notranslate"><span class="pre">torch.ao.quantization.DeQuantStub</span></code> to
dequantize the tensor.  This needs to be done manually in Eager mode quantization.
An e2e example:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="k">class</span> <span class="nc">M</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">nn</span><span class="o">.</span><span class="n">Module</span><span class="p">):</span>
    <span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
        <span class="nb">super</span><span class="p">()</span><span class="o">.</span><span class="fm">__init__</span><span class="p">()</span>
        <span class="bp">self</span><span class="o">.</span><span class="n">quant</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">ao</span><span class="o">.</span><span class="n">quantization</span><span class="o">.</span><span class="n">QuantStub</span><span class="p">()</span>
        <span class="bp">self</span><span class="o">.</span><span class="n">conv1</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">nn</span><span class="o">.</span><span class="n">Conv2d</span><span class="p">(</span><span class="mi">1</span><span class="p">,</span> <span class="mi">1</span><span class="p">,</span> <span class="mi">1</span><span class="p">)</span>
        <span class="c1"># this module will not be quantized (see `qconfig = None` logic below)</span>
        <span class="bp">self</span><span class="o">.</span><span class="n">conv2</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">nn</span><span class="o">.</span><span class="n">Conv2d</span><span class="p">(</span><span class="mi">1</span><span class="p">,</span> <span class="mi">1</span><span class="p">,</span> <span class="mi">1</span><span class="p">)</span>
        <span class="bp">self</span><span class="o">.</span><span class="n">dequant</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">ao</span><span class="o">.</span><span class="n">quantization</span><span class="o">.</span><span class="n">DeQuantStub</span><span class="p">()</span>

    <span class="k">def</span> <span class="nf">forward</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">x</span><span class="p">):</span>
        <span class="c1"># during the convert step, this will be replaced with a</span>
        <span class="c1"># `quantize_per_tensor` call</span>
        <span class="n">x</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">quant</span><span class="p">(</span><span class="n">x</span><span class="p">)</span>
        <span class="n">x</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">conv1</span><span class="p">(</span><span class="n">x</span><span class="p">)</span>
        <span class="c1"># during the convert step, this will be replaced with a</span>
        <span class="c1"># `dequantize` call</span>
        <span class="n">x</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">dequant</span><span class="p">(</span><span class="n">x</span><span class="p">)</span>
        <span class="n">x</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">conv2</span><span class="p">(</span><span class="n">x</span><span class="p">)</span>
        <span class="k">return</span> <span class="n">x</span>

<span class="n">m</span> <span class="o">=</span> <span class="n">M</span><span class="p">()</span>
<span class="n">m</span><span class="o">.</span><span class="n">qconfig</span> <span class="o">=</span> <span class="n">some_qconfig</span>
<span class="c1"># turn off quantization for conv2</span>
<span class="n">m</span><span class="o">.</span><span class="n">conv2</span><span class="o">.</span><span class="n">qconfig</span> <span class="o">=</span> <span class="kc">None</span>
</pre></div>
</div>
</div>
<div class="section" id="saving-and-loading-quantized-models">
<h3>Saving and Loading Quantized models<a class="headerlink" href="#saving-and-loading-quantized-models" title="Permalink to this heading">¶</a></h3>
<p>When calling <code class="docutils literal notranslate"><span class="pre">torch.load</span></code> on a quantized model, if you see an error like:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="ne">AttributeError</span><span class="p">:</span> <span class="s1">&#39;LinearPackedParams&#39;</span> <span class="nb">object</span> <span class="n">has</span> <span class="n">no</span> <span class="n">attribute</span> <span class="s1">&#39;_modules&#39;</span>
</pre></div>
</div>
<p>This is because directly saving and loading a quantized model using <code class="docutils literal notranslate"><span class="pre">torch.save</span></code> and <code class="docutils literal notranslate"><span class="pre">torch.load</span></code>
is not supported. To save/load quantized models, the following ways can be used:</p>
<ol class="arabic simple">
<li><p>Saving/Loading the quantized model state_dict</p></li>
</ol>
<p>An example:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="k">class</span> <span class="nc">M</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">nn</span><span class="o">.</span><span class="n">Module</span><span class="p">):</span>
    <span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
        <span class="nb">super</span><span class="p">()</span><span class="o">.</span><span class="fm">__init__</span><span class="p">()</span>
        <span class="bp">self</span><span class="o">.</span><span class="n">linear</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">Linear</span><span class="p">(</span><span class="mi">5</span><span class="p">,</span> <span class="mi">5</span><span class="p">)</span>
        <span class="bp">self</span><span class="o">.</span><span class="n">relu</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">ReLU</span><span class="p">()</span>

    <span class="k">def</span> <span class="nf">forward</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">x</span><span class="p">):</span>
        <span class="n">x</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">linear</span><span class="p">(</span><span class="n">x</span><span class="p">)</span>
        <span class="n">x</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">relu</span><span class="p">(</span><span class="n">x</span><span class="p">)</span>
        <span class="k">return</span> <span class="n">x</span>

<span class="n">m</span> <span class="o">=</span> <span class="n">M</span><span class="p">()</span><span class="o">.</span><span class="n">eval</span><span class="p">()</span>
<span class="n">prepare_orig</span> <span class="o">=</span> <span class="n">prepare_fx</span><span class="p">(</span><span class="n">m</span><span class="p">,</span> <span class="p">{</span><span class="s1">&#39;&#39;</span> <span class="p">:</span> <span class="n">default_qconfig</span><span class="p">})</span>
<span class="n">prepare_orig</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">rand</span><span class="p">(</span><span class="mi">5</span><span class="p">,</span> <span class="mi">5</span><span class="p">))</span>
<span class="n">quantized_orig</span> <span class="o">=</span> <span class="n">convert_fx</span><span class="p">(</span><span class="n">prepare_orig</span><span class="p">)</span>

<span class="c1"># Save/load using state_dict</span>
<span class="n">b</span> <span class="o">=</span> <span class="n">io</span><span class="o">.</span><span class="n">BytesIO</span><span class="p">()</span>
<span class="n">torch</span><span class="o">.</span><span class="n">save</span><span class="p">(</span><span class="n">quantized_orig</span><span class="o">.</span><span class="n">state_dict</span><span class="p">(),</span> <span class="n">b</span><span class="p">)</span>

<span class="n">m2</span> <span class="o">=</span> <span class="n">M</span><span class="p">()</span><span class="o">.</span><span class="n">eval</span><span class="p">()</span>
<span class="n">prepared</span> <span class="o">=</span> <span class="n">prepare_fx</span><span class="p">(</span><span class="n">m2</span><span class="p">,</span> <span class="p">{</span><span class="s1">&#39;&#39;</span> <span class="p">:</span> <span class="n">default_qconfig</span><span class="p">})</span>
<span class="n">quantized</span> <span class="o">=</span> <span class="n">convert_fx</span><span class="p">(</span><span class="n">prepared</span><span class="p">)</span>
<span class="n">b</span><span class="o">.</span><span class="n">seek</span><span class="p">(</span><span class="mi">0</span><span class="p">)</span>
<span class="n">quantized</span><span class="o">.</span><span class="n">load_state_dict</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">load</span><span class="p">(</span><span class="n">b</span><span class="p">))</span>
</pre></div>
</div>
<ol class="arabic simple" start="2">
<li><p>Saving/Loading scripted quantized models using <code class="docutils literal notranslate"><span class="pre">torch.jit.save</span></code> and <code class="docutils literal notranslate"><span class="pre">torch.jit.load</span></code></p></li>
</ol>
<p>An example:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="c1"># Note: using the same model M from previous example</span>
<span class="n">m</span> <span class="o">=</span> <span class="n">M</span><span class="p">()</span><span class="o">.</span><span class="n">eval</span><span class="p">()</span>
<span class="n">prepare_orig</span> <span class="o">=</span> <span class="n">prepare_fx</span><span class="p">(</span><span class="n">m</span><span class="p">,</span> <span class="p">{</span><span class="s1">&#39;&#39;</span> <span class="p">:</span> <span class="n">default_qconfig</span><span class="p">})</span>
<span class="n">prepare_orig</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">rand</span><span class="p">(</span><span class="mi">5</span><span class="p">,</span> <span class="mi">5</span><span class="p">))</span>
<span class="n">quantized_orig</span> <span class="o">=</span> <span class="n">convert_fx</span><span class="p">(</span><span class="n">prepare_orig</span><span class="p">)</span>

<span class="c1"># save/load using scripted model</span>
<span class="n">scripted</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">jit</span><span class="o">.</span><span class="n">script</span><span class="p">(</span><span class="n">quantized_orig</span><span class="p">)</span>
<span class="n">b</span> <span class="o">=</span> <span class="n">io</span><span class="o">.</span><span class="n">BytesIO</span><span class="p">()</span>
<span class="n">torch</span><span class="o">.</span><span class="n">jit</span><span class="o">.</span><span class="n">save</span><span class="p">(</span><span class="n">scripted</span><span class="p">,</span> <span class="n">b</span><span class="p">)</span>
<span class="n">b</span><span class="o">.</span><span class="n">seek</span><span class="p">(</span><span class="mi">0</span><span class="p">)</span>
<span class="n">scripted_quantized</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">jit</span><span class="o">.</span><span class="n">load</span><span class="p">(</span><span class="n">b</span><span class="p">)</span>
</pre></div>
</div>
</div>
<div class="section" id="symbolic-trace-error-when-using-fx-graph-mode-quantization">
<h3>Symbolic Trace Error when using FX Graph Mode Quantization<a class="headerlink" href="#symbolic-trace-error-when-using-fx-graph-mode-quantization" title="Permalink to this heading">¶</a></h3>
<p>Symbolic traceability is a requirement for <a class="reference internal" href="#prototype-maintaince-mode-fx-graph-mode-quantization">(Prototype - maintaince mode) FX Graph Mode Quantization</a>, so if you pass a PyTorch Model that is not symbolically traceable to <cite>torch.ao.quantization.prepare_fx</cite> or <cite>torch.ao.quantization.prepare_qat_fx</cite>, we might see an error like the following:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">torch</span><span class="o">.</span><span class="n">fx</span><span class="o">.</span><span class="n">proxy</span><span class="o">.</span><span class="n">TraceError</span><span class="p">:</span> <span class="n">symbolically</span> <span class="n">traced</span> <span class="n">variables</span> <span class="n">cannot</span> <span class="n">be</span> <span class="n">used</span> <span class="k">as</span> <span class="n">inputs</span> <span class="n">to</span> <span class="n">control</span> <span class="n">flow</span>
</pre></div>
</div>
<p>Please take a look at <a class="reference external" href="https://pytorch.org/docs/2.0/fx.html#limitations-of-symbolic-tracing">Limitations of Symbolic Tracing</a> and use - <a class="reference external" href="https://pytorch.org/tutorials/prototype/fx_graph_mode_quant_guide.html">User Guide on Using FX Graph Mode Quantization</a> to workaround the problem.</p>
<span class="target" id="module-torch.ao"></span><span class="target" id="module-torch.ao.nn"></span><span class="target" id="module-torch.ao.nn.quantizable"></span><span class="target" id="module-torch.ao.nn.quantizable.modules"></span><span class="target" id="module-torch.ao.nn.quantized"></span><span class="target" id="module-torch.ao.nn.quantized.reference"></span><span class="target" id="module-torch.ao.nn.quantized.reference.modules"></span><span class="target" id="module-torch.ao.nn.sparse"></span><span class="target" id="module-torch.ao.nn.sparse.quantized"></span><span class="target" id="module-torch.ao.nn.sparse.quantized.dynamic"></span><span class="target" id="module-torch.ao.ns"></span><span class="target" id="module-torch.ao.ns.fx"></span><span class="target" id="module-torch.ao.quantization.backend_config"></span><span class="target" id="module-torch.ao.pruning"></span><span class="target" id="module-torch.ao.pruning.scheduler"></span><span class="target" id="module-torch.ao.pruning.sparsifier"></span><span class="target" id="module-torch.ao.nn.intrinsic.modules.fused"></span><span class="target" id="module-torch.ao.nn.intrinsic.qat.modules.conv_fused"></span><span class="target" id="module-torch.ao.nn.intrinsic.qat.modules.linear_fused"></span><span class="target" id="module-torch.ao.nn.intrinsic.qat.modules.linear_relu"></span><span class="target" id="module-torch.ao.nn.intrinsic.quantized.dynamic.modules.linear_relu"></span><span class="target" id="module-torch.ao.nn.intrinsic.quantized.modules.bn_relu"></span><span class="target" id="module-torch.ao.nn.intrinsic.quantized.modules.conv_add"></span><span class="target" id="module-torch.ao.nn.intrinsic.quantized.modules.conv_relu"></span><span class="target" id="module-torch.ao.nn.intrinsic.quantized.modules.linear_relu"></span><span class="target" id="module-torch.ao.nn.qat.dynamic.modules.linear"></span><span class="target" id="module-torch.ao.nn.qat.modules.conv"></span><span class="target" id="module-torch.ao.nn.qat.modules.embedding_ops"></span><span class="target" id="module-torch.ao.nn.qat.modules.linear"></span><span class="target" id="module-torch.ao.nn.quantizable.modules.activation"></span><span class="target" id="module-torch.ao.nn.quantizable.modules.rnn"></span><span class="target" id="module-torch.ao.nn.quantized.dynamic.modules.conv"></span><span class="target" id="module-torch.ao.nn.quantized.dynamic.modules.linear"></span><span class="target" id="module-torch.ao.nn.quantized.dynamic.modules.rnn"></span><span class="target" id="module-torch.ao.nn.quantized.modules.activation"></span><span class="target" id="module-torch.ao.nn.quantized.modules.batchnorm"></span><span class="target" id="module-torch.ao.nn.quantized.modules.conv"></span><span class="target" id="module-torch.ao.nn.quantized.modules.dropout"></span><span class="target" id="module-torch.ao.nn.quantized.modules.embedding_ops"></span><span class="target" id="module-torch.ao.nn.quantized.modules.functional_modules"></span><span class="target" id="module-torch.ao.nn.quantized.modules.linear"></span><span class="target" id="module-torch.ao.nn.quantized.modules.normalization"></span><span class="target" id="module-torch.ao.nn.quantized.modules.rnn"></span><span class="target" id="module-torch.ao.nn.quantized.modules.utils"></span><span class="target" id="module-torch.ao.nn.quantized.reference.modules.conv"></span><span class="target" id="module-torch.ao.nn.quantized.reference.modules.linear"></span><span class="target" id="module-torch.ao.nn.quantized.reference.modules.rnn"></span><span class="target" id="module-torch.ao.nn.quantized.reference.modules.sparse"></span><span class="target" id="module-torch.ao.nn.quantized.reference.modules.utils"></span><span class="target" id="module-torch.ao.nn.sparse.quantized.dynamic.linear"></span><span class="target" id="module-torch.ao.nn.sparse.quantized.linear"></span><span class="target" id="module-torch.ao.nn.sparse.quantized.utils"></span><span class="target" id="module-torch.ao.ns.fx.graph_matcher"></span><span class="target" id="module-torch.ao.ns.fx.graph_passes"></span><span class="target" id="module-torch.ao.ns.fx.mappings"></span><span class="target" id="module-torch.ao.ns.fx.n_shadows_utils"></span><span class="target" id="module-torch.ao.ns.fx.ns_types"></span><span class="target" id="module-torch.ao.ns.fx.pattern_utils"></span><span class="target" id="module-torch.ao.ns.fx.qconfig_multi_mapping"></span><span class="target" id="module-torch.ao.ns.fx.utils"></span><span class="target" id="module-torch.ao.ns.fx.weight_utils"></span><span class="target" id="module-torch.ao.pruning.scheduler.base_scheduler"></span><span class="target" id="module-torch.ao.pruning.scheduler.cubic_scheduler"></span><span class="target" id="module-torch.ao.pruning.scheduler.lambda_scheduler"></span><span class="target" id="module-torch.ao.pruning.sparsifier.base_sparsifier"></span><span class="target" id="module-torch.ao.pruning.sparsifier.nearly_diagonal_sparsifier"></span><span class="target" id="module-torch.ao.pruning.sparsifier.utils"></span><span class="target" id="module-torch.ao.pruning.sparsifier.weight_norm_sparsifier"></span><span class="target" id="module-torch.ao.quantization.backend_config.backend_config"></span><span class="target" id="module-torch.ao.quantization.backend_config.executorch"></span><span class="target" id="module-torch.ao.quantization.backend_config.fbgemm"></span><span class="target" id="module-torch.ao.quantization.backend_config.native"></span><span class="target" id="module-torch.ao.quantization.backend_config.observation_type"></span><span class="target" id="module-torch.ao.quantization.backend_config.onednn"></span><span class="target" id="module-torch.ao.quantization.backend_config.qnnpack"></span><span class="target" id="module-torch.ao.quantization.backend_config.tensorrt"></span><span class="target" id="module-torch.ao.quantization.backend_config.utils"></span><span class="target" id="module-torch.ao.quantization.backend_config.x86"></span><span class="target" id="module-torch.ao.quantization.fake_quantize"></span><span class="target" id="module-torch.ao.quantization.fuser_method_mappings"></span><span class="target" id="module-torch.ao.quantization.fuse_modules"></span><span class="target" id="module-torch.ao.quantization.fx.convert"></span><span class="target" id="module-torch.ao.quantization.fx.custom_config"></span><span class="target" id="module-torch.ao.quantization.fx.fuse"></span><span class="target" id="module-torch.ao.quantization.fx.fuse_handler"></span><span class="target" id="module-torch.ao.quantization.fx.graph_module"></span><span class="target" id="module-torch.ao.quantization.fx.lower_to_fbgemm"></span><span class="target" id="module-torch.ao.quantization.fx.lower_to_qnnpack"></span><span class="target" id="module-torch.ao.quantization.fx.lstm_utils"></span><span class="target" id="module-torch.ao.quantization.fx.match_utils"></span><span class="target" id="module-torch.ao.quantization.fx.pattern_utils"></span><span class="target" id="module-torch.ao.quantization.fx.prepare"></span><span class="target" id="module-torch.ao.quantization.fx.qconfig_mapping_utils"></span><span class="target" id="module-torch.ao.quantization.fx.quantize_handler"></span><span class="target" id="module-torch.ao.quantization.fx.tracer"></span><span class="target" id="module-torch.ao.quantization.fx.utils"></span><span class="target" id="module-torch.ao.quantization.observer"></span><span class="target" id="module-torch.ao.quantization.pt2e.duplicate_dq_pass"></span><span class="target" id="module-torch.ao.quantization.pt2e.eval_utils"></span><span class="target" id="module-torch.ao.quantization.pt2e.graph_utils"></span><span class="target" id="module-torch.ao.quantization.pt2e.port_metadata_pass"></span><span class="target" id="module-torch.ao.quantization.pt2e.prepare"></span><span class="target" id="module-torch.ao.quantization.pt2e.qat_utils"></span><span class="target" id="module-torch.ao.quantization.pt2e.representation.rewrite"></span><span class="target" id="module-torch.ao.quantization.pt2e.utils"></span><span class="target" id="module-torch.ao.quantization.qconfig"></span><span class="target" id="module-torch.ao.quantization.qconfig_mapping"></span><span class="target" id="module-torch.ao.quantization.quant_type"></span><span class="target" id="module-torch.ao.quantization.quantization_mappings"></span><span class="target" id="module-torch.ao.quantization.quantize"></span><span class="target" id="module-torch.ao.quantization.quantize_fx"></span><span class="target" id="module-torch.ao.quantization.quantize_jit"></span><span class="target" id="module-torch.ao.quantization.quantize_pt2e"></span><span class="target" id="module-torch.ao.quantization.quantizer.composable_quantizer"></span><span class="target" id="module-torch.ao.quantization.quantizer.embedding_quantizer"></span><span class="target" id="module-torch.ao.quantization.quantizer.quantizer"></span><span class="target" id="module-torch.ao.quantization.quantizer.utils"></span><span class="target" id="module-torch.ao.quantization.quantizer.x86_inductor_quantizer"></span><span class="target" id="module-torch.ao.quantization.quantizer.xnnpack_quantizer"></span><span class="target" id="module-torch.ao.quantization.quantizer.xnnpack_quantizer_utils"></span><span class="target" id="module-torch.ao.quantization.stubs"></span><span class="target" id="module-torch.ao.quantization.utils"></span><span class="target" id="module-torch.nn.intrinsic.modules.fused"></span><span class="target" id="module-torch.nn.intrinsic.qat.modules.conv_fused"></span><span class="target" id="module-torch.nn.intrinsic.qat.modules.linear_fused"></span><span class="target" id="module-torch.nn.intrinsic.qat.modules.linear_relu"></span><span class="target" id="module-torch.nn.intrinsic.quantized.dynamic.modules.linear_relu"></span><span class="target" id="module-torch.nn.intrinsic.quantized.modules.bn_relu"></span><span class="target" id="module-torch.nn.intrinsic.quantized.modules.conv_relu"></span><span class="target" id="module-torch.nn.intrinsic.quantized.modules.linear_relu"></span><span class="target" id="module-torch.nn.qat.dynamic.modules.linear"></span><span class="target" id="module-torch.nn.qat.modules.conv"></span><span class="target" id="module-torch.nn.qat.modules.embedding_ops"></span><span class="target" id="module-torch.nn.qat.modules.linear"></span><span class="target" id="module-torch.nn.quantizable.modules.activation"></span><span class="target" id="module-torch.nn.quantizable.modules.rnn"></span><span class="target" id="module-torch.nn.quantized.dynamic.modules.conv"></span><span class="target" id="module-torch.nn.quantized.dynamic.modules.linear"></span><span class="target" id="module-torch.nn.quantized.dynamic.modules.rnn"></span><span class="target" id="module-torch.nn.quantized.functional"></span><span class="target" id="module-torch.nn.quantized.modules.activation"></span><span class="target" id="module-torch.nn.quantized.modules.batchnorm"></span><span class="target" id="module-torch.nn.quantized.modules.conv"></span><span class="target" id="module-torch.nn.quantized.modules.dropout"></span><span class="target" id="module-torch.nn.quantized.modules.embedding_ops"></span><span class="target" id="module-torch.nn.quantized.modules.functional_modules"></span><span class="target" id="module-torch.nn.quantized.modules.linear"></span><span class="target" id="module-torch.nn.quantized.modules.normalization"></span><span class="target" id="module-torch.nn.quantized.modules.rnn"></span><span class="target" id="module-torch.nn.quantized.modules.utils"></span><span class="target" id="module-torch.quantization.fake_quantize"></span><span class="target" id="module-torch.quantization.fuse_modules"></span><span class="target" id="module-torch.quantization.fuser_method_mappings"></span><span class="target" id="module-torch.quantization.fx.convert"></span><span class="target" id="module-torch.quantization.fx.fuse"></span><span class="target" id="module-torch.quantization.fx.fusion_patterns"></span><span class="target" id="module-torch.quantization.fx.graph_module"></span><span class="target" id="module-torch.quantization.fx.match_utils"></span><span class="target" id="module-torch.quantization.fx.pattern_utils"></span><span class="target" id="module-torch.quantization.fx.prepare"></span><span class="target" id="module-torch.quantization.fx.quantization_patterns"></span><span class="target" id="module-torch.quantization.fx.quantization_types"></span><span class="target" id="module-torch.quantization.fx.utils"></span><span class="target" id="module-torch.quantization.observer"></span><span class="target" id="module-torch.quantization.qconfig"></span><span class="target" id="module-torch.quantization.quant_type"></span><span class="target" id="module-torch.quantization.quantization_mappings"></span><span class="target" id="module-torch.quantization.quantize"></span><span class="target" id="module-torch.quantization.quantize_fx"></span><span class="target" id="module-torch.quantization.quantize_jit"></span><span class="target" id="module-torch.quantization.stubs"></span><span class="target" id="module-torch.quantization.utils"></span></div>
</div>
</div>


             </article>
             
            </div>
            <footer>
  
    <div class="rst-footer-buttons" role="navigation" aria-label="footer navigation">
      
        <a href="quantization-support.html" class="btn btn-neutral float-right" title="Quantization API Reference" accesskey="n" rel="next">Next <img src="_static/images/chevron-right-orange.svg" class="next-page"></a>
      
      
        <a href="pipeline.html" class="btn btn-neutral" title="Pipeline Parallelism" 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="#">Quantization</a><ul>
<li><a class="reference internal" href="#introduction-to-quantization">Introduction to Quantization</a></li>
<li><a class="reference internal" href="#quantization-api-summary">Quantization API Summary</a><ul>
<li><a class="reference internal" href="#eager-mode-quantization">Eager Mode Quantization</a><ul>
<li><a class="reference internal" href="#post-training-dynamic-quantization">Post Training Dynamic Quantization</a></li>
<li><a class="reference internal" href="#post-training-static-quantization">Post Training Static Quantization</a></li>
<li><a class="reference internal" href="#quantization-aware-training-for-static-quantization">Quantization Aware Training for Static Quantization</a></li>
<li><a class="reference internal" href="#model-preparation-for-eager-mode-static-quantization">Model Preparation for Eager Mode Static Quantization</a></li>
</ul>
</li>
<li><a class="reference internal" href="#prototype-maintaince-mode-fx-graph-mode-quantization">(Prototype - maintaince mode) FX Graph Mode Quantization</a></li>
<li><a class="reference internal" href="#prototype-pytorch-2-export-quantization">(Prototype) PyTorch 2 Export Quantization</a></li>
</ul>
</li>
<li><a class="reference internal" href="#quantization-stack">Quantization Stack</a><ul>
<li><a class="reference internal" href="#quantized-model">Quantized Model</a><ul>
<li><a class="reference internal" href="#quantized-tensor">Quantized Tensor</a></li>
<li><a class="reference internal" href="#quantize-and-dequantize">Quantize and Dequantize</a></li>
<li><a class="reference internal" href="#quantized-operators-modules">Quantized Operators/Modules</a></li>
<li><a class="reference internal" href="#quantized-engine">Quantized Engine</a></li>
</ul>
</li>
<li><a class="reference internal" href="#quantization-flow">Quantization Flow</a><ul>
<li><a class="reference internal" href="#observer-and-fakequantize">Observer and FakeQuantize</a></li>
<li><a class="reference internal" href="#qconfig">QConfig</a></li>
<li><a class="reference internal" href="#general-quantization-flow">General Quantization Flow</a></li>
</ul>
</li>
</ul>
</li>
<li><a class="reference internal" href="#quantization-support-matrix">Quantization Support Matrix</a><ul>
<li><a class="reference internal" href="#quantization-mode-support">Quantization Mode Support</a></li>
<li><a class="reference internal" href="#quantization-flow-support">Quantization Flow Support</a></li>
<li><a class="reference internal" href="#backend-hardware-support">Backend/Hardware Support</a><ul>
<li><a class="reference internal" href="#note-for-native-cpu-backends">Note for native CPU backends</a></li>
</ul>
</li>
<li><a class="reference internal" href="#operator-support">Operator Support</a></li>
</ul>
</li>
<li><a class="reference internal" href="#quantization-api-reference">Quantization API Reference</a></li>
<li><a class="reference internal" href="#quantization-backend-configuration">Quantization Backend Configuration</a></li>
<li><a class="reference internal" href="#quantization-accuracy-debugging">Quantization Accuracy Debugging</a></li>
<li><a class="reference internal" href="#quantization-customizations">Quantization Customizations</a><ul>
<li><a class="reference internal" href="#quantization-custom-module-api">Quantization Custom Module API</a></li>
</ul>
</li>
<li><a class="reference internal" href="#best-practices">Best Practices</a></li>
<li><a class="reference internal" href="#frequently-asked-questions">Frequently Asked Questions</a></li>
<li><a class="reference internal" href="#common-errors">Common Errors</a><ul>
<li><a class="reference internal" href="#passing-a-non-quantized-tensor-into-a-quantized-kernel">Passing a non-quantized Tensor into a quantized kernel</a></li>
<li><a class="reference internal" href="#passing-a-quantized-tensor-into-a-non-quantized-kernel">Passing a quantized Tensor into a non-quantized kernel</a></li>
<li><a class="reference internal" href="#saving-and-loading-quantized-models">Saving and Loading Quantized models</a></li>
<li><a class="reference internal" href="#symbolic-trace-error-when-using-fx-graph-mode-quantization">Symbolic Trace Error when using FX Graph Mode Quantization</a></li>
</ul>
</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>
            <a href="https://pytorch.org/get-started">Get Started</a>
          </li>

          <li>
            <a href="https://pytorch.org/ecosystem">Ecosystem</a>
          </li>
            
          <li>
            <a href="">Mobile</a>
          </li>

          <li>
            <a href="https://pytorch.org/blog/">Blog</a>
          </li>

          <li>
            <a href="https://pytorch.org/tutorials">Tutorials</a>
          </li>

          <li class="resources-mobile-menu-title" class="active">
            Docs
          </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/audio/stable/index.html">torchaudio</a>
            </li>

            <li>
              <a href="https://pytorch.org/text/stable/index.html">torchtext</a>
            </li>

            <li>
              <a href="https://pytorch.org/vision/stable/index.html">torchvision</a>
            </li>

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

            <li>
              <a href="https://pytorch.org/data">TorchData</a>
            </li>

            <li>
              <a href="https://pytorch.org/torchrec">TorchRec</a>
            </li>

            <li>
              <a href="https://pytorch.org/serve/">TorchServe</a>
            </li>

            <li>
              <a href="https://pytorch.org/torchx/">TorchX</a>
            </li>

            <li>
              <a href="https://pytorch.org/xla">PyTorch on XLA Devices</a>
            </li>
          </ul>

          <li class="resources-mobile-menu-title">
            Resources
          </li>
            
           <ul class="resources-mobile-menu-items">

            <li>
              <a href="https://pytorch.org/features">About</a>
            </li>

            <li>
              <a href="https://pytorch.org/foundation">PyTorch Foundation</a>
            </li>

            <li>
              <a href="https://pytorch.org/#community-module">Community</a>
            </li>

            <li>
              <a href="https://pytorch.org/community-stories">Community Stories</a>
            </li>

            <li>
              <a href="https://pytorch.org/resources">Developer Resources</a>
            </li>

            <li>
              <a href="https://pytorch.org/events">Events</a>
            </li>

            <li>
              <a href="https://discuss.pytorch.org/">Forums</a>
            </li>

            <li>
              <a href="https://pytorch.org/hub">Models (Beta)</a>
            </li>
          </ul>

          <li>
            <a href="https://github.com/pytorch/pytorch">Github</a>
          </li>
        </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>