versions/1.9.1/api/python/docs/_modules/mxnet/contrib/quantization.html - mxnet-site - Git at Google

 <!DOCTYPE html>

 <html xmlns="http://www.w3.org/1999/xhtml">
   <head>
     <meta charset="utf-8" />
     <meta charset="utf-8">
     <meta name="viewport" content="width=device-width, initial-scale=1, shrink-to-fit=no">
     <meta http-equiv="x-ua-compatible" content="ie=edge">
     <style>
     .dropdown {
         position: relative;
         display: inline-block;
     }

     .dropdown-content {
         display: none;
         position: absolute;
         background-color: #f9f9f9;
         min-width: 160px;
         box-shadow: 0px 8px 16px 0px rgba(0,0,0,0.2);
         padding: 12px 16px;
         z-index: 1;
         text-align: left;
     }

     .dropdown:hover .dropdown-content {
         display: block;
     }

     .dropdown-option:hover {
         color: #FF4500;
     }

     .dropdown-option-active {
         color: #FF4500;
         font-weight: lighter;
     }

     .dropdown-option {
         color: #000000;
         font-weight: lighter;
     }

     .dropdown-header {
         color: #FFFFFF;
         display: inline-flex;
     }

     .dropdown-caret {
         width: 18px;
     }

     .dropdown-caret-path {
         fill: #FFFFFF;
     }
     </style>

     <title>mxnet.contrib.quantization &#8212; Apache MXNet  documentation</title>

     <link rel="stylesheet" href="../../../_static/basic.css" type="text/css" />
     <link rel="stylesheet" href="../../../_static/pygments.css" type="text/css" />
     <link rel="stylesheet" type="text/css" href="../../../_static/mxnet.css" />
     <link rel="stylesheet" href="../../../_static/material-design-lite-1.3.0/material.blue-deep_orange.min.css" type="text/css" />
     <link rel="stylesheet" href="../../../_static/sphinx_materialdesign_theme.css" type="text/css" />
     <link rel="stylesheet" href="../../../_static/fontawesome/all.css" type="text/css" />
     <link rel="stylesheet" href="../../../_static/fonts.css" type="text/css" />
     <link rel="stylesheet" href="../../../_static/feedback.css" type="text/css" />
     <script id="documentation_options" data-url_root="../../../" src="../../../_static/documentation_options.js"></script>
     <script src="../../../_static/jquery.js"></script>
     <script src="../../../_static/underscore.js"></script>
     <script src="../../../_static/doctools.js"></script>
     <script src="../../../_static/language_data.js"></script>
     <script src="../../../_static/matomo_analytics.js"></script>
     <script src="../../../_static/autodoc.js"></script>
     <script crossorigin="anonymous" integrity="sha256-Ae2Vz/4ePdIu6ZyI/5ZGsYnb+m0JlOmKPjt6XZ9JJkA=" src="https://cdnjs.cloudflare.com/ajax/libs/require.js/2.3.4/require.min.js"></script>
     <script async="async" src="https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.5/latest.js?config=TeX-AMS-MML_HTMLorMML"></script>
     <script type="text/x-mathjax-config">MathJax.Hub.Config({"tex2jax": {"inlineMath": [["$", "$"], ["\\(", "\\)"]], "processEscapes": true, "ignoreClass": "document", "processClass": "math|output_area"}})</script>
     <link rel="shortcut icon" href="../../../_static/mxnet-icon.png"/>
     <link rel="index" title="Index" href="../../../genindex.html" />
     <link rel="search" title="Search" href="../../../search.html" />
   </head>
 <body><header class="site-header" role="banner">
   <div class="wrapper">
       <a class="site-title" rel="author" href="/versions/1.9.1/"><img
             src="../../../_static/mxnet_logo.png" class="site-header-logo"></a>
     <nav class="site-nav">
       <input type="checkbox" id="nav-trigger" class="nav-trigger"/>
       <label for="nav-trigger">
           <span class="menu-icon">
             <svg viewBox="0 0 18 15" width="18px" height="15px">
               <path d="M18,1.484c0,0.82-0.665,1.484-1.484,1.484H1.484C0.665,2.969,0,2.304,0,1.484l0,0C0,0.665,0.665,0,1.484,0 h15.032C17.335,0,18,0.665,18,1.484L18,1.484z M18,7.516C18,8.335,17.335,9,16.516,9H1.484C0.665,9,0,8.335,0,7.516l0,0 c0-0.82,0.665-1.484,1.484-1.484h15.032C17.335,6.031,18,6.696,18,7.516L18,7.516z M18,13.516C18,14.335,17.335,15,16.516,15H1.484 C0.665,15,0,14.335,0,13.516l0,0c0-0.82,0.665-1.483,1.484-1.483h15.032C17.335,12.031,18,12.695,18,13.516L18,13.516z"/>
             </svg>
           </span>
       </label>

       <div class="trigger">
         <a class="page-link" href="/versions/1.9.1/get_started">Get Started</a>
         <a class="page-link" href="/versions/1.9.1/features">Features</a>
         <a class="page-link" href="/versions/1.9.1/ecosystem">Ecosystem</a>
         <a class="page-link page-current" href="/versions/1.9.1/api">Docs & Tutorials</a>
         <a class="page-link" href="/versions/1.9.1/trusted_by">Trusted By</a>
         <a class="page-link" href="https://github.com/apache/incubator-mxnet">GitHub</a>
         <div class="dropdown" style="min-width:100px">
           <span class="dropdown-header">Apache
             <svg class="dropdown-caret" viewBox="0 0 32 32" class="icon icon-caret-bottom" aria-hidden="true"><path class="dropdown-caret-path" d="M24 11.305l-7.997 11.39L8 11.305z"></path></svg>
           </span>
           <div class="dropdown-content" style="min-width:250px">
             <a href="https://www.apache.org/foundation/">Apache Software Foundation</a>
             <a href="https://incubator.apache.org/">Apache Incubator</a>
             <a href="https://www.apache.org/licenses/">License</a>
             <a href="/versions/1.9.1/api/faq/security.html">Security</a>
             <a href="https://privacy.apache.org/policies/privacy-policy-public.html">Privacy</a>
             <a href="https://www.apache.org/events/current-event">Events</a>
             <a href="https://www.apache.org/foundation/sponsorship.html">Sponsorship</a>
             <a href="https://www.apache.org/foundation/thanks.html">Thanks</a>
           </div>
         </div>
         <div class="dropdown">
           <span class="dropdown-header">1.9.1
             <svg class="dropdown-caret" viewBox="0 0 32 32" class="icon icon-caret-bottom" aria-hidden="true"><path class="dropdown-caret-path" d="M24 11.305l-7.997 11.39L8 11.305z"></path></svg>
           </span>
           <div class="dropdown-content">
             <a class="dropdown-option" href="/">master</a><br>
             <a class="dropdown-option-active" href="/versions/1.9.1/">1.9.1</a><br>
             <a class="dropdown-option" href="/versions/1.8.0/">1.8.0</a><br>
             <a class="dropdown-option" href="/versions/1.7.0/">1.7.0</a><br>
             <a class="dropdown-option" href="/versions/1.6.0/">1.6.0</a><br>
             <a class="dropdown-option" href="/versions/1.5.0/">1.5.0</a><br>
             <a class="dropdown-option" href="/versions/1.4.1/">1.4.1</a><br>
             <a class="dropdown-option" href="/versions/1.3.1/">1.3.1</a><br>
             <a class="dropdown-option" href="/versions/1.2.1/">1.2.1</a><br>
             <a class="dropdown-option" href="/versions/1.1.0/">1.1.0</a><br>
             <a class="dropdown-option" href="/versions/1.0.0/">1.0.0</a><br>
             <a class="dropdown-option" href="/versions/0.12.1/">0.12.1</a><br>
             <a class="dropdown-option" href="/versions/0.11.0/">0.11.0</a>
           </div>
         </div>
       </div>
     </nav>
   </div>
 </header>
     <div class="mdl-layout mdl-js-layout mdl-layout--fixed-header mdl-layout--fixed-drawer"><header class="mdl-layout__header mdl-layout__header--waterfall ">
     <div class="mdl-layout__header-row">

         <nav class="mdl-navigation breadcrumb">
             <a class="mdl-navigation__link" href="../../index.html">Module code</a><i class="material-icons">navigate_next</i>
             <a class="mdl-navigation__link is-active">mxnet.contrib.quantization</a>
         </nav>
         <div class="mdl-layout-spacer"></div>
         <nav class="mdl-navigation">

 <form class="form-inline pull-sm-right" action="../../../search.html" method="get">
       <div class="mdl-textfield mdl-js-textfield mdl-textfield--expandable mdl-textfield--floating-label mdl-textfield--align-right">
         <label id="quick-search-icon" class="mdl-button mdl-js-button mdl-button--icon"  for="waterfall-exp">
           <i class="material-icons">search</i>
         </label>
         <div class="mdl-textfield__expandable-holder">
           <input class="mdl-textfield__input" type="text" name="q"  id="waterfall-exp" placeholder="Search" />
           <input type="hidden" name="check_keywords" value="yes" />
           <input type="hidden" name="area" value="default" />
         </div>
       </div>
       <div class="mdl-tooltip" data-mdl-for="quick-search-icon">
       Quick search
       </div>
 </form>

         </nav>
     </div>
     <div class="mdl-layout__header-row header-links">
       <div class="mdl-layout-spacer"></div>
       <nav class="mdl-navigation">
       </nav>
     </div>
 </header><header class="mdl-layout__drawer">

       <div class="globaltoc">
         <span class="mdl-layout-title toc">Table Of Contents</span>


             <nav class="mdl-navigation">
                 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../../../tutorials/index.html">Python Tutorials</a><ul>
 <li class="toctree-l2"><a class="reference internal" href="../../../tutorials/getting-started/index.html">Getting Started</a><ul>
 <li class="toctree-l3"><a class="reference internal" href="../../../tutorials/getting-started/crash-course/index.html">Crash Course</a><ul>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/getting-started/crash-course/1-ndarray.html">Manipulate data with <code class="docutils literal notranslate"><span class="pre">ndarray</span></code></a></li>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/getting-started/crash-course/2-nn.html">Create a neural network</a></li>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/getting-started/crash-course/3-autograd.html">Automatic differentiation with <code class="docutils literal notranslate"><span class="pre">autograd</span></code></a></li>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/getting-started/crash-course/4-train.html">Train the neural network</a></li>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/getting-started/crash-course/5-predict.html">Predict with a pre-trained model</a></li>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/getting-started/crash-course/6-use_gpus.html">Use GPUs</a></li>
 </ul>
 </li>
 <li class="toctree-l3"><a class="reference internal" href="../../../tutorials/getting-started/to-mxnet/index.html">Moving to MXNet from Other Frameworks</a><ul>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/getting-started/to-mxnet/pytorch.html">PyTorch vs Apache MXNet</a></li>
 </ul>
 </li>
 <li class="toctree-l3"><a class="reference internal" href="../../../tutorials/getting-started/gluon_from_experiment_to_deployment.html">Gluon: from experiment to deployment</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../tutorials/getting-started/logistic_regression_explained.html">Logistic regression explained</a></li>
 <li class="toctree-l3"><a class="reference external" href="https://mxnet.apache.org/api/python/docs/tutorials/packages/gluon/image/mnist.html">MNIST</a></li>
 </ul>
 </li>
 <li class="toctree-l2"><a class="reference internal" href="../../../tutorials/packages/index.html">Packages</a><ul>
 <li class="toctree-l3"><a class="reference internal" href="../../../tutorials/packages/autograd/index.html">Automatic Differentiation</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../tutorials/packages/gluon/index.html">Gluon</a><ul>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/packages/gluon/blocks/index.html">Blocks</a><ul>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/gluon/blocks/custom-layer.html">Custom Layers</a></li>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/gluon/blocks/custom_layer_beginners.html">Customer Layers (Beginners)</a></li>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/gluon/blocks/hybridize.html">Hybridize</a></li>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/gluon/blocks/init.html">Initialization</a></li>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/gluon/blocks/naming.html">Parameter and Block Naming</a></li>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/gluon/blocks/nn.html">Layers and Blocks</a></li>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/gluon/blocks/parameters.html">Parameter Management</a></li>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/gluon/blocks/save_load_params.html">Saving and Loading Gluon Models</a></li>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/gluon/blocks/activations/activations.html">Activation Blocks</a></li>
 </ul>
 </li>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/packages/gluon/data/index.html">Data Tutorials</a><ul>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/gluon/data/data_augmentation.html">Image Augmentation</a></li>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/gluon/data/data_augmentation.html#Spatial-Augmentation">Spatial Augmentation</a></li>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/gluon/data/data_augmentation.html#Color-Augmentation">Color Augmentation</a></li>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/gluon/data/data_augmentation.html#Composed-Augmentations">Composed Augmentations</a></li>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/gluon/data/datasets.html">Gluon <code class="docutils literal notranslate"><span class="pre">Dataset</span></code>s and <code class="docutils literal notranslate"><span class="pre">DataLoader</span></code></a></li>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/gluon/data/datasets.html#Using-own-data-with-included-Datasets">Using own data with included <code class="docutils literal notranslate"><span class="pre">Dataset</span></code>s</a></li>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/gluon/data/datasets.html#Using-own-data-with-custom-Datasets">Using own data with custom <code class="docutils literal notranslate"><span class="pre">Dataset</span></code>s</a></li>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/gluon/data/datasets.html#Appendix:-Upgrading-from-Module-DataIter-to-Gluon-DataLoader">Appendix: Upgrading from Module <code class="docutils literal notranslate"><span class="pre">DataIter</span></code> to Gluon <code class="docutils literal notranslate"><span class="pre">DataLoader</span></code></a></li>
 </ul>
 </li>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/packages/gluon/image/index.html">Image Tutorials</a><ul>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/gluon/image/image-augmentation.html">Image Augmentation</a></li>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/gluon/image/info_gan.html">Image similarity search with InfoGAN</a></li>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/gluon/image/mnist.html">Handwritten Digit Recognition</a></li>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/gluon/image/pretrained_models.html">Using pre-trained models in MXNet</a></li>
 </ul>
 </li>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/packages/gluon/loss/index.html">Losses</a><ul>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/gluon/loss/custom-loss.html">Custom Loss Blocks</a></li>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/gluon/loss/kl_divergence.html">Kullback-Leibler (KL) Divergence</a></li>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/gluon/loss/loss.html">Loss functions</a></li>
 </ul>
 </li>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/packages/gluon/text/index.html">Text Tutorials</a><ul>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/gluon/text/gnmt.html">Google Neural Machine Translation</a></li>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/gluon/text/transformer.html">Machine Translation with Transformer</a></li>
 </ul>
 </li>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/packages/gluon/training/index.html">Training</a><ul>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/gluon/training/fit_api_tutorial.html">MXNet Gluon Fit API</a></li>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/gluon/training/trainer.html">Trainer</a></li>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/gluon/training/learning_rates/index.html">Learning Rates</a><ul>
 <li class="toctree-l6"><a class="reference internal" href="../../../tutorials/packages/gluon/training/learning_rates/learning_rate_finder.html">Learning Rate Finder</a></li>
 <li class="toctree-l6"><a class="reference internal" href="../../../tutorials/packages/gluon/training/learning_rates/learning_rate_schedules.html">Learning Rate Schedules</a></li>
 <li class="toctree-l6"><a class="reference internal" href="../../../tutorials/packages/gluon/training/learning_rates/learning_rate_schedules_advanced.html">Advanced Learning Rate Schedules</a></li>
 </ul>
 </li>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/gluon/training/normalization/index.html">Normalization Blocks</a></li>
 </ul>
 </li>
 </ul>
 </li>
 <li class="toctree-l3"><a class="reference internal" href="../../../tutorials/packages/kvstore/index.html">KVStore</a><ul>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/packages/kvstore/kvstore.html">Distributed Key-Value Store</a></li>
 </ul>
 </li>
 <li class="toctree-l3"><a class="reference internal" href="../../../tutorials/packages/ndarray/index.html">NDArray</a><ul>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/packages/ndarray/01-ndarray-intro.html">An Intro: Manipulate Data the MXNet Way with NDArray</a></li>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/packages/ndarray/02-ndarray-operations.html">NDArray Operations</a></li>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/packages/ndarray/03-ndarray-contexts.html">NDArray Contexts</a></li>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/packages/ndarray/gotchas_numpy_in_mxnet.html">Gotchas using NumPy in Apache MXNet</a></li>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/packages/ndarray/sparse/index.html">Tutorials</a><ul>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/ndarray/sparse/csr.html">CSRNDArray - NDArray in Compressed Sparse Row Storage Format</a></li>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/ndarray/sparse/row_sparse.html">RowSparseNDArray - NDArray for Sparse Gradient Updates</a></li>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/ndarray/sparse/train.html">Train a Linear Regression Model with Sparse Symbols</a></li>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/ndarray/sparse/train_gluon.html">Sparse NDArrays with Gluon</a></li>
 </ul>
 </li>
 </ul>
 </li>
 <li class="toctree-l3"><a class="reference internal" href="../../../tutorials/packages/onnx/index.html">ONNX</a><ul>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/packages/onnx/fine_tuning_gluon.html">Fine-tuning an ONNX model</a></li>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/packages/onnx/inference_on_onnx_model.html">Running inference on MXNet/Gluon from an ONNX model</a></li>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/packages/onnx/super_resolution.html">Importing an ONNX model into MXNet</a></li>
 <li class="toctree-l4"><a class="reference external" href="https://mxnet.apache.org/api/python/docs/tutorials/deploy/export/onnx.html">Export ONNX Models</a></li>
 </ul>
 </li>
 <li class="toctree-l3"><a class="reference internal" href="../../../tutorials/packages/optimizer/index.html">Optimizers</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../tutorials/packages/viz/index.html">Visualization</a><ul>
 <li class="toctree-l4"><a class="reference external" href="https://mxnet.apache.org/api/faq/visualize_graph">Visualize networks</a></li>
 </ul>
 </li>
 </ul>
 </li>
 <li class="toctree-l2"><a class="reference internal" href="../../../tutorials/performance/index.html">Performance</a><ul>
 <li class="toctree-l3"><a class="reference internal" href="../../../tutorials/performance/compression/index.html">Compression</a><ul>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/performance/compression/int8.html">Deploy with int-8</a></li>
 <li class="toctree-l4"><a class="reference external" href="https://mxnet.apache.org/api/faq/float16">Float16</a></li>
 <li class="toctree-l4"><a class="reference external" href="https://mxnet.apache.org/api/faq/gradient_compression">Gradient Compression</a></li>
 <li class="toctree-l4"><a class="reference external" href="https://gluon-cv.mxnet.io/build/examples_deployment/int8_inference.html">GluonCV with Quantized Models</a></li>
 </ul>
 </li>
 <li class="toctree-l3"><a class="reference internal" href="../../../tutorials/performance/backend/index.html">Accelerated Backend Tools</a><ul>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/performance/backend/mkldnn/index.html">Intel MKL-DNN</a><ul>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/performance/backend/mkldnn/mkldnn_quantization.html">Quantize with MKL-DNN backend</a></li>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/performance/backend/mkldnn/mkldnn_quantization.html#Improving-accuracy-with-Intel®-Neural-Compressor">Improving accuracy with Intel® Neural Compressor</a></li>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/performance/backend/mkldnn/mkldnn_readme.html">Install MXNet with MKL-DNN</a></li>
 </ul>
 </li>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/performance/backend/tensorrt/index.html">TensorRT</a><ul>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/performance/backend/tensorrt/tensorrt.html">Optimizing Deep Learning Computation Graphs with TensorRT</a></li>
 </ul>
 </li>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/performance/backend/tvm.html">Use TVM</a></li>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/performance/backend/profiler.html">Profiling MXNet Models</a></li>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/performance/backend/amp.html">Using AMP: Automatic Mixed Precision</a></li>
 </ul>
 </li>
 </ul>
 </li>
 <li class="toctree-l2"><a class="reference internal" href="../../../tutorials/deploy/index.html">Deployment</a><ul>
 <li class="toctree-l3"><a class="reference internal" href="../../../tutorials/deploy/export/index.html">Export</a><ul>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/deploy/export/onnx.html">Exporting to ONNX format</a></li>
 <li class="toctree-l4"><a class="reference external" href="https://gluon-cv.mxnet.io/build/examples_deployment/export_network.html">Export Gluon CV Models</a></li>
 <li class="toctree-l4"><a class="reference external" href="https://mxnet.apache.org/api/python/docs/tutorials/packages/gluon/blocks/save_load_params.html">Save / Load Parameters</a></li>
 </ul>
 </li>
 <li class="toctree-l3"><a class="reference internal" href="../../../tutorials/deploy/inference/index.html">Inference</a><ul>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/deploy/inference/cpp.html">Deploy into C++</a></li>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/deploy/inference/image_classification_jetson.html">Image Classication using pretrained ResNet-50 model on Jetson module</a></li>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/deploy/inference/scala.html">Deploy into a Java or Scala Environment</a></li>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/deploy/inference/wine_detector.html">Real-time Object Detection with MXNet On The Raspberry Pi</a></li>
 </ul>
 </li>
 <li class="toctree-l3"><a class="reference internal" href="../../../tutorials/deploy/run-on-aws/index.html">Run on AWS</a><ul>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/deploy/run-on-aws/use_ec2.html">Run on an EC2 Instance</a></li>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/deploy/run-on-aws/use_sagemaker.html">Run on Amazon SageMaker</a></li>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/deploy/run-on-aws/cloud.html">MXNet on the Cloud</a></li>
 </ul>
 </li>
 </ul>
 </li>
 <li class="toctree-l2"><a class="reference internal" href="../../../tutorials/extend/index.html">Extend</a><ul>
 <li class="toctree-l3"><a class="reference internal" href="../../../tutorials/extend/custom_layer.html">Custom Layers</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../tutorials/extend/customop.html">Custom Numpy Operators</a></li>
 <li class="toctree-l3"><a class="reference external" href="https://mxnet.apache.org/api/faq/new_op">New Operator Creation</a></li>
 <li class="toctree-l3"><a class="reference external" href="https://mxnet.apache.org/api/faq/add_op_in_backend">New Operator in MXNet Backend</a></li>
 </ul>
 </li>
 </ul>
 </li>
 <li class="toctree-l1"><a class="reference internal" href="../../../api/index.html">Python API</a><ul>
 <li class="toctree-l2"><a class="reference internal" href="../../../api/ndarray/index.html">mxnet.ndarray</a><ul>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/ndarray/ndarray.html">ndarray</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/ndarray/contrib/index.html">ndarray.contrib</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/ndarray/image/index.html">ndarray.image</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/ndarray/linalg/index.html">ndarray.linalg</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/ndarray/op/index.html">ndarray.op</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/ndarray/random/index.html">ndarray.random</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/ndarray/register/index.html">ndarray.register</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/ndarray/sparse/index.html">ndarray.sparse</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/ndarray/utils/index.html">ndarray.utils</a></li>
 </ul>
 </li>
 <li class="toctree-l2"><a class="reference internal" href="../../../api/gluon/index.html">mxnet.gluon</a><ul>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/gluon/block.html">gluon.Block</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/gluon/hybrid_block.html">gluon.HybridBlock</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/gluon/symbol_block.html">gluon.SymbolBlock</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/gluon/constant.html">gluon.Constant</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/gluon/parameter.html">gluon.Parameter</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/gluon/parameter_dict.html">gluon.ParameterDict</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/gluon/trainer.html">gluon.Trainer</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/gluon/contrib/index.html">gluon.contrib</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/gluon/data/index.html">gluon.data</a><ul>
 <li class="toctree-l4"><a class="reference internal" href="../../../api/gluon/data/vision/index.html">data.vision</a><ul>
 <li class="toctree-l5"><a class="reference internal" href="../../../api/gluon/data/vision/datasets/index.html">vision.datasets</a></li>
 <li class="toctree-l5"><a class="reference internal" href="../../../api/gluon/data/vision/transforms/index.html">vision.transforms</a></li>
 </ul>
 </li>
 </ul>
 </li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/gluon/loss/index.html">gluon.loss</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/gluon/model_zoo/index.html">gluon.model_zoo.vision</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/gluon/nn/index.html">gluon.nn</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/gluon/rnn/index.html">gluon.rnn</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/gluon/utils/index.html">gluon.utils</a></li>
 </ul>
 </li>
 <li class="toctree-l2"><a class="reference internal" href="../../../api/autograd/index.html">mxnet.autograd</a></li>
 <li class="toctree-l2"><a class="reference internal" href="../../../api/initializer/index.html">mxnet.initializer</a></li>
 <li class="toctree-l2"><a class="reference internal" href="../../../api/optimizer/index.html">mxnet.optimizer</a></li>
 <li class="toctree-l2"><a class="reference internal" href="../../../api/lr_scheduler/index.html">mxnet.lr_scheduler</a></li>
 <li class="toctree-l2"><a class="reference internal" href="../../../api/metric/index.html">mxnet.metric</a></li>
 <li class="toctree-l2"><a class="reference internal" href="../../../api/kvstore/index.html">mxnet.kvstore</a></li>
 <li class="toctree-l2"><a class="reference internal" href="../../../api/symbol/index.html">mxnet.symbol</a><ul>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/symbol/symbol.html">symbol</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/symbol/contrib/index.html">symbol.contrib</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/symbol/image/index.html">symbol.image</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/symbol/linalg/index.html">symbol.linalg</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/symbol/op/index.html">symbol.op</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/symbol/random/index.html">symbol.random</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/symbol/register/index.html">symbol.register</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/symbol/sparse/index.html">symbol.sparse</a></li>
 </ul>
 </li>
 <li class="toctree-l2"><a class="reference internal" href="../../../api/module/index.html">mxnet.module</a></li>
 <li class="toctree-l2"><a class="reference internal" href="../../../api/contrib/index.html">mxnet.contrib</a><ul>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/contrib/autograd/index.html">contrib.autograd</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/contrib/io/index.html">contrib.io</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/contrib/ndarray/index.html">contrib.ndarray</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/contrib/onnx/index.html">contrib.onnx</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/contrib/quantization/index.html">contrib.quantization</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/contrib/symbol/index.html">contrib.symbol</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/contrib/tensorboard/index.html">contrib.tensorboard</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/contrib/tensorrt/index.html">contrib.tensorrt</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/contrib/text/index.html">contrib.text</a></li>
 </ul>
 </li>
 <li class="toctree-l2"><a class="reference internal" href="../../../api/mxnet/index.html">mxnet</a><ul>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/attribute/index.html">mxnet.attribute</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/base/index.html">mxnet.base</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/callback/index.html">mxnet.callback</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/context/index.html">mxnet.context</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/engine/index.html">mxnet.engine</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/executor/index.html">mxnet.executor</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/executor_manager/index.html">mxnet.executor_manager</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/image/index.html">mxnet.image</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/io/index.html">mxnet.io</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/kvstore_server/index.html">mxnet.kvstore_server</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/libinfo/index.html">mxnet.libinfo</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/log/index.html">mxnet.log</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/model/index.html">mxnet.model</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/monitor/index.html">mxnet.monitor</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/name/index.html">mxnet.name</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/notebook/index.html">mxnet.notebook</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/operator/index.html">mxnet.operator</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/profiler/index.html">mxnet.profiler</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/random/index.html">mxnet.random</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/recordio/index.html">mxnet.recordio</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/registry/index.html">mxnet.registry</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/rtc/index.html">mxnet.rtc</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/runtime/index.html">mxnet.runtime</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/test_utils/index.html">mxnet.test_utils</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/torch/index.html">mxnet.torch</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/util/index.html">mxnet.util</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/visualization/index.html">mxnet.visualization</a></li>
 </ul>
 </li>
 </ul>
 </li>
 </ul>

             </nav>

         </div>

 </header>
         <main class="mdl-layout__content" tabIndex="0">

         <script type="text/javascript" src="../../../_static/sphinx_materialdesign_theme.js "></script>
         <script type="text/javascript" src="../../../_static/feedback.js"></script>
     <header class="mdl-layout__drawer">

       <div class="globaltoc">
         <span class="mdl-layout-title toc">Table Of Contents</span>


             <nav class="mdl-navigation">
                 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../../../tutorials/index.html">Python Tutorials</a><ul>
 <li class="toctree-l2"><a class="reference internal" href="../../../tutorials/getting-started/index.html">Getting Started</a><ul>
 <li class="toctree-l3"><a class="reference internal" href="../../../tutorials/getting-started/crash-course/index.html">Crash Course</a><ul>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/getting-started/crash-course/1-ndarray.html">Manipulate data with <code class="docutils literal notranslate"><span class="pre">ndarray</span></code></a></li>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/getting-started/crash-course/2-nn.html">Create a neural network</a></li>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/getting-started/crash-course/3-autograd.html">Automatic differentiation with <code class="docutils literal notranslate"><span class="pre">autograd</span></code></a></li>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/getting-started/crash-course/4-train.html">Train the neural network</a></li>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/getting-started/crash-course/5-predict.html">Predict with a pre-trained model</a></li>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/getting-started/crash-course/6-use_gpus.html">Use GPUs</a></li>
 </ul>
 </li>
 <li class="toctree-l3"><a class="reference internal" href="../../../tutorials/getting-started/to-mxnet/index.html">Moving to MXNet from Other Frameworks</a><ul>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/getting-started/to-mxnet/pytorch.html">PyTorch vs Apache MXNet</a></li>
 </ul>
 </li>
 <li class="toctree-l3"><a class="reference internal" href="../../../tutorials/getting-started/gluon_from_experiment_to_deployment.html">Gluon: from experiment to deployment</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../tutorials/getting-started/logistic_regression_explained.html">Logistic regression explained</a></li>
 <li class="toctree-l3"><a class="reference external" href="https://mxnet.apache.org/api/python/docs/tutorials/packages/gluon/image/mnist.html">MNIST</a></li>
 </ul>
 </li>
 <li class="toctree-l2"><a class="reference internal" href="../../../tutorials/packages/index.html">Packages</a><ul>
 <li class="toctree-l3"><a class="reference internal" href="../../../tutorials/packages/autograd/index.html">Automatic Differentiation</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../tutorials/packages/gluon/index.html">Gluon</a><ul>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/packages/gluon/blocks/index.html">Blocks</a><ul>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/gluon/blocks/custom-layer.html">Custom Layers</a></li>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/gluon/blocks/custom_layer_beginners.html">Customer Layers (Beginners)</a></li>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/gluon/blocks/hybridize.html">Hybridize</a></li>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/gluon/blocks/init.html">Initialization</a></li>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/gluon/blocks/naming.html">Parameter and Block Naming</a></li>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/gluon/blocks/nn.html">Layers and Blocks</a></li>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/gluon/blocks/parameters.html">Parameter Management</a></li>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/gluon/blocks/save_load_params.html">Saving and Loading Gluon Models</a></li>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/gluon/blocks/activations/activations.html">Activation Blocks</a></li>
 </ul>
 </li>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/packages/gluon/data/index.html">Data Tutorials</a><ul>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/gluon/data/data_augmentation.html">Image Augmentation</a></li>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/gluon/data/data_augmentation.html#Spatial-Augmentation">Spatial Augmentation</a></li>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/gluon/data/data_augmentation.html#Color-Augmentation">Color Augmentation</a></li>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/gluon/data/data_augmentation.html#Composed-Augmentations">Composed Augmentations</a></li>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/gluon/data/datasets.html">Gluon <code class="docutils literal notranslate"><span class="pre">Dataset</span></code>s and <code class="docutils literal notranslate"><span class="pre">DataLoader</span></code></a></li>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/gluon/data/datasets.html#Using-own-data-with-included-Datasets">Using own data with included <code class="docutils literal notranslate"><span class="pre">Dataset</span></code>s</a></li>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/gluon/data/datasets.html#Using-own-data-with-custom-Datasets">Using own data with custom <code class="docutils literal notranslate"><span class="pre">Dataset</span></code>s</a></li>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/gluon/data/datasets.html#Appendix:-Upgrading-from-Module-DataIter-to-Gluon-DataLoader">Appendix: Upgrading from Module <code class="docutils literal notranslate"><span class="pre">DataIter</span></code> to Gluon <code class="docutils literal notranslate"><span class="pre">DataLoader</span></code></a></li>
 </ul>
 </li>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/packages/gluon/image/index.html">Image Tutorials</a><ul>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/gluon/image/image-augmentation.html">Image Augmentation</a></li>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/gluon/image/info_gan.html">Image similarity search with InfoGAN</a></li>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/gluon/image/mnist.html">Handwritten Digit Recognition</a></li>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/gluon/image/pretrained_models.html">Using pre-trained models in MXNet</a></li>
 </ul>
 </li>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/packages/gluon/loss/index.html">Losses</a><ul>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/gluon/loss/custom-loss.html">Custom Loss Blocks</a></li>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/gluon/loss/kl_divergence.html">Kullback-Leibler (KL) Divergence</a></li>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/gluon/loss/loss.html">Loss functions</a></li>
 </ul>
 </li>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/packages/gluon/text/index.html">Text Tutorials</a><ul>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/gluon/text/gnmt.html">Google Neural Machine Translation</a></li>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/gluon/text/transformer.html">Machine Translation with Transformer</a></li>
 </ul>
 </li>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/packages/gluon/training/index.html">Training</a><ul>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/gluon/training/fit_api_tutorial.html">MXNet Gluon Fit API</a></li>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/gluon/training/trainer.html">Trainer</a></li>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/gluon/training/learning_rates/index.html">Learning Rates</a><ul>
 <li class="toctree-l6"><a class="reference internal" href="../../../tutorials/packages/gluon/training/learning_rates/learning_rate_finder.html">Learning Rate Finder</a></li>
 <li class="toctree-l6"><a class="reference internal" href="../../../tutorials/packages/gluon/training/learning_rates/learning_rate_schedules.html">Learning Rate Schedules</a></li>
 <li class="toctree-l6"><a class="reference internal" href="../../../tutorials/packages/gluon/training/learning_rates/learning_rate_schedules_advanced.html">Advanced Learning Rate Schedules</a></li>
 </ul>
 </li>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/gluon/training/normalization/index.html">Normalization Blocks</a></li>
 </ul>
 </li>
 </ul>
 </li>
 <li class="toctree-l3"><a class="reference internal" href="../../../tutorials/packages/kvstore/index.html">KVStore</a><ul>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/packages/kvstore/kvstore.html">Distributed Key-Value Store</a></li>
 </ul>
 </li>
 <li class="toctree-l3"><a class="reference internal" href="../../../tutorials/packages/ndarray/index.html">NDArray</a><ul>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/packages/ndarray/01-ndarray-intro.html">An Intro: Manipulate Data the MXNet Way with NDArray</a></li>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/packages/ndarray/02-ndarray-operations.html">NDArray Operations</a></li>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/packages/ndarray/03-ndarray-contexts.html">NDArray Contexts</a></li>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/packages/ndarray/gotchas_numpy_in_mxnet.html">Gotchas using NumPy in Apache MXNet</a></li>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/packages/ndarray/sparse/index.html">Tutorials</a><ul>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/ndarray/sparse/csr.html">CSRNDArray - NDArray in Compressed Sparse Row Storage Format</a></li>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/ndarray/sparse/row_sparse.html">RowSparseNDArray - NDArray for Sparse Gradient Updates</a></li>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/ndarray/sparse/train.html">Train a Linear Regression Model with Sparse Symbols</a></li>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/packages/ndarray/sparse/train_gluon.html">Sparse NDArrays with Gluon</a></li>
 </ul>
 </li>
 </ul>
 </li>
 <li class="toctree-l3"><a class="reference internal" href="../../../tutorials/packages/onnx/index.html">ONNX</a><ul>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/packages/onnx/fine_tuning_gluon.html">Fine-tuning an ONNX model</a></li>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/packages/onnx/inference_on_onnx_model.html">Running inference on MXNet/Gluon from an ONNX model</a></li>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/packages/onnx/super_resolution.html">Importing an ONNX model into MXNet</a></li>
 <li class="toctree-l4"><a class="reference external" href="https://mxnet.apache.org/api/python/docs/tutorials/deploy/export/onnx.html">Export ONNX Models</a></li>
 </ul>
 </li>
 <li class="toctree-l3"><a class="reference internal" href="../../../tutorials/packages/optimizer/index.html">Optimizers</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../tutorials/packages/viz/index.html">Visualization</a><ul>
 <li class="toctree-l4"><a class="reference external" href="https://mxnet.apache.org/api/faq/visualize_graph">Visualize networks</a></li>
 </ul>
 </li>
 </ul>
 </li>
 <li class="toctree-l2"><a class="reference internal" href="../../../tutorials/performance/index.html">Performance</a><ul>
 <li class="toctree-l3"><a class="reference internal" href="../../../tutorials/performance/compression/index.html">Compression</a><ul>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/performance/compression/int8.html">Deploy with int-8</a></li>
 <li class="toctree-l4"><a class="reference external" href="https://mxnet.apache.org/api/faq/float16">Float16</a></li>
 <li class="toctree-l4"><a class="reference external" href="https://mxnet.apache.org/api/faq/gradient_compression">Gradient Compression</a></li>
 <li class="toctree-l4"><a class="reference external" href="https://gluon-cv.mxnet.io/build/examples_deployment/int8_inference.html">GluonCV with Quantized Models</a></li>
 </ul>
 </li>
 <li class="toctree-l3"><a class="reference internal" href="../../../tutorials/performance/backend/index.html">Accelerated Backend Tools</a><ul>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/performance/backend/mkldnn/index.html">Intel MKL-DNN</a><ul>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/performance/backend/mkldnn/mkldnn_quantization.html">Quantize with MKL-DNN backend</a></li>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/performance/backend/mkldnn/mkldnn_quantization.html#Improving-accuracy-with-Intel®-Neural-Compressor">Improving accuracy with Intel® Neural Compressor</a></li>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/performance/backend/mkldnn/mkldnn_readme.html">Install MXNet with MKL-DNN</a></li>
 </ul>
 </li>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/performance/backend/tensorrt/index.html">TensorRT</a><ul>
 <li class="toctree-l5"><a class="reference internal" href="../../../tutorials/performance/backend/tensorrt/tensorrt.html">Optimizing Deep Learning Computation Graphs with TensorRT</a></li>
 </ul>
 </li>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/performance/backend/tvm.html">Use TVM</a></li>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/performance/backend/profiler.html">Profiling MXNet Models</a></li>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/performance/backend/amp.html">Using AMP: Automatic Mixed Precision</a></li>
 </ul>
 </li>
 </ul>
 </li>
 <li class="toctree-l2"><a class="reference internal" href="../../../tutorials/deploy/index.html">Deployment</a><ul>
 <li class="toctree-l3"><a class="reference internal" href="../../../tutorials/deploy/export/index.html">Export</a><ul>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/deploy/export/onnx.html">Exporting to ONNX format</a></li>
 <li class="toctree-l4"><a class="reference external" href="https://gluon-cv.mxnet.io/build/examples_deployment/export_network.html">Export Gluon CV Models</a></li>
 <li class="toctree-l4"><a class="reference external" href="https://mxnet.apache.org/api/python/docs/tutorials/packages/gluon/blocks/save_load_params.html">Save / Load Parameters</a></li>
 </ul>
 </li>
 <li class="toctree-l3"><a class="reference internal" href="../../../tutorials/deploy/inference/index.html">Inference</a><ul>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/deploy/inference/cpp.html">Deploy into C++</a></li>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/deploy/inference/image_classification_jetson.html">Image Classication using pretrained ResNet-50 model on Jetson module</a></li>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/deploy/inference/scala.html">Deploy into a Java or Scala Environment</a></li>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/deploy/inference/wine_detector.html">Real-time Object Detection with MXNet On The Raspberry Pi</a></li>
 </ul>
 </li>
 <li class="toctree-l3"><a class="reference internal" href="../../../tutorials/deploy/run-on-aws/index.html">Run on AWS</a><ul>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/deploy/run-on-aws/use_ec2.html">Run on an EC2 Instance</a></li>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/deploy/run-on-aws/use_sagemaker.html">Run on Amazon SageMaker</a></li>
 <li class="toctree-l4"><a class="reference internal" href="../../../tutorials/deploy/run-on-aws/cloud.html">MXNet on the Cloud</a></li>
 </ul>
 </li>
 </ul>
 </li>
 <li class="toctree-l2"><a class="reference internal" href="../../../tutorials/extend/index.html">Extend</a><ul>
 <li class="toctree-l3"><a class="reference internal" href="../../../tutorials/extend/custom_layer.html">Custom Layers</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../tutorials/extend/customop.html">Custom Numpy Operators</a></li>
 <li class="toctree-l3"><a class="reference external" href="https://mxnet.apache.org/api/faq/new_op">New Operator Creation</a></li>
 <li class="toctree-l3"><a class="reference external" href="https://mxnet.apache.org/api/faq/add_op_in_backend">New Operator in MXNet Backend</a></li>
 </ul>
 </li>
 </ul>
 </li>
 <li class="toctree-l1"><a class="reference internal" href="../../../api/index.html">Python API</a><ul>
 <li class="toctree-l2"><a class="reference internal" href="../../../api/ndarray/index.html">mxnet.ndarray</a><ul>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/ndarray/ndarray.html">ndarray</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/ndarray/contrib/index.html">ndarray.contrib</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/ndarray/image/index.html">ndarray.image</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/ndarray/linalg/index.html">ndarray.linalg</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/ndarray/op/index.html">ndarray.op</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/ndarray/random/index.html">ndarray.random</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/ndarray/register/index.html">ndarray.register</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/ndarray/sparse/index.html">ndarray.sparse</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/ndarray/utils/index.html">ndarray.utils</a></li>
 </ul>
 </li>
 <li class="toctree-l2"><a class="reference internal" href="../../../api/gluon/index.html">mxnet.gluon</a><ul>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/gluon/block.html">gluon.Block</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/gluon/hybrid_block.html">gluon.HybridBlock</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/gluon/symbol_block.html">gluon.SymbolBlock</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/gluon/constant.html">gluon.Constant</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/gluon/parameter.html">gluon.Parameter</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/gluon/parameter_dict.html">gluon.ParameterDict</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/gluon/trainer.html">gluon.Trainer</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/gluon/contrib/index.html">gluon.contrib</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/gluon/data/index.html">gluon.data</a><ul>
 <li class="toctree-l4"><a class="reference internal" href="../../../api/gluon/data/vision/index.html">data.vision</a><ul>
 <li class="toctree-l5"><a class="reference internal" href="../../../api/gluon/data/vision/datasets/index.html">vision.datasets</a></li>
 <li class="toctree-l5"><a class="reference internal" href="../../../api/gluon/data/vision/transforms/index.html">vision.transforms</a></li>
 </ul>
 </li>
 </ul>
 </li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/gluon/loss/index.html">gluon.loss</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/gluon/model_zoo/index.html">gluon.model_zoo.vision</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/gluon/nn/index.html">gluon.nn</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/gluon/rnn/index.html">gluon.rnn</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/gluon/utils/index.html">gluon.utils</a></li>
 </ul>
 </li>
 <li class="toctree-l2"><a class="reference internal" href="../../../api/autograd/index.html">mxnet.autograd</a></li>
 <li class="toctree-l2"><a class="reference internal" href="../../../api/initializer/index.html">mxnet.initializer</a></li>
 <li class="toctree-l2"><a class="reference internal" href="../../../api/optimizer/index.html">mxnet.optimizer</a></li>
 <li class="toctree-l2"><a class="reference internal" href="../../../api/lr_scheduler/index.html">mxnet.lr_scheduler</a></li>
 <li class="toctree-l2"><a class="reference internal" href="../../../api/metric/index.html">mxnet.metric</a></li>
 <li class="toctree-l2"><a class="reference internal" href="../../../api/kvstore/index.html">mxnet.kvstore</a></li>
 <li class="toctree-l2"><a class="reference internal" href="../../../api/symbol/index.html">mxnet.symbol</a><ul>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/symbol/symbol.html">symbol</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/symbol/contrib/index.html">symbol.contrib</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/symbol/image/index.html">symbol.image</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/symbol/linalg/index.html">symbol.linalg</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/symbol/op/index.html">symbol.op</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/symbol/random/index.html">symbol.random</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/symbol/register/index.html">symbol.register</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/symbol/sparse/index.html">symbol.sparse</a></li>
 </ul>
 </li>
 <li class="toctree-l2"><a class="reference internal" href="../../../api/module/index.html">mxnet.module</a></li>
 <li class="toctree-l2"><a class="reference internal" href="../../../api/contrib/index.html">mxnet.contrib</a><ul>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/contrib/autograd/index.html">contrib.autograd</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/contrib/io/index.html">contrib.io</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/contrib/ndarray/index.html">contrib.ndarray</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/contrib/onnx/index.html">contrib.onnx</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/contrib/quantization/index.html">contrib.quantization</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/contrib/symbol/index.html">contrib.symbol</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/contrib/tensorboard/index.html">contrib.tensorboard</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/contrib/tensorrt/index.html">contrib.tensorrt</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/contrib/text/index.html">contrib.text</a></li>
 </ul>
 </li>
 <li class="toctree-l2"><a class="reference internal" href="../../../api/mxnet/index.html">mxnet</a><ul>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/attribute/index.html">mxnet.attribute</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/base/index.html">mxnet.base</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/callback/index.html">mxnet.callback</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/context/index.html">mxnet.context</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/engine/index.html">mxnet.engine</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/executor/index.html">mxnet.executor</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/executor_manager/index.html">mxnet.executor_manager</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/image/index.html">mxnet.image</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/io/index.html">mxnet.io</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/kvstore_server/index.html">mxnet.kvstore_server</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/libinfo/index.html">mxnet.libinfo</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/log/index.html">mxnet.log</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/model/index.html">mxnet.model</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/monitor/index.html">mxnet.monitor</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/name/index.html">mxnet.name</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/notebook/index.html">mxnet.notebook</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/operator/index.html">mxnet.operator</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/profiler/index.html">mxnet.profiler</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/random/index.html">mxnet.random</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/recordio/index.html">mxnet.recordio</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/registry/index.html">mxnet.registry</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/rtc/index.html">mxnet.rtc</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/runtime/index.html">mxnet.runtime</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/test_utils/index.html">mxnet.test_utils</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/torch/index.html">mxnet.torch</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/util/index.html">mxnet.util</a></li>
 <li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/visualization/index.html">mxnet.visualization</a></li>
 </ul>
 </li>
 </ul>
 </li>
 </ul>

             </nav>

         </div>

 </header>

     <div class="document">
         <div class="page-content" role="main">

   <h1>Source code for mxnet.contrib.quantization</h1><div class="highlight"><pre>
 <span></span><span class="c1"># Licensed to the Apache Software Foundation (ASF) under one</span>
 <span class="c1"># or more contributor license agreements.  See the NOTICE file</span>
 <span class="c1"># distributed with this work for additional information</span>
 <span class="c1"># regarding copyright ownership.  The ASF licenses this file</span>
 <span class="c1"># to you under the Apache License, Version 2.0 (the</span>
 <span class="c1"># &quot;License&quot;); you may not use this file except in compliance</span>
 <span class="c1"># with the License.  You may obtain a copy of the License at</span>
 <span class="c1">#</span>
 <span class="c1">#   http://www.apache.org/licenses/LICENSE-2.0</span>
 <span class="c1">#</span>
 <span class="c1"># Unless required by applicable law or agreed to in writing,</span>
 <span class="c1"># software distributed under the License is distributed on an</span>
 <span class="c1"># &quot;AS IS&quot; BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY</span>
 <span class="c1"># KIND, either express or implied.  See the License for the</span>
 <span class="c1"># specific language governing permissions and limitations</span>
 <span class="c1"># under the License.</span>
 <span class="sd">&quot;&quot;&quot;Quantization module for generating quantized (INT8) models from FP32 models.&quot;&quot;&quot;</span>


 <span class="k">try</span><span class="p">:</span>
     <span class="kn">from</span> <span class="nn">scipy</span> <span class="kn">import</span> <span class="n">stats</span>
 <span class="k">except</span> <span class="ne">ImportError</span><span class="p">:</span>
     <span class="n">stats</span> <span class="o">=</span> <span class="kc">None</span>

 <span class="kn">import</span> <span class="nn">ctypes</span>
 <span class="kn">import</span> <span class="nn">logging</span>
 <span class="kn">import</span> <span class="nn">os</span>
 <span class="kn">import</span> <span class="nn">shutil</span>
 <span class="kn">import</span> <span class="nn">warnings</span>
 <span class="kn">import</span> <span class="nn">numpy</span> <span class="k">as</span> <span class="nn">np</span>
 <span class="kn">from</span> <span class="nn">..base</span> <span class="kn">import</span> <span class="n">_LIB</span><span class="p">,</span> <span class="n">check_call</span><span class="p">,</span> <span class="n">py_str</span>
 <span class="kn">from</span> <span class="nn">..base</span> <span class="kn">import</span> <span class="n">c_array</span><span class="p">,</span> <span class="n">c_str</span><span class="p">,</span> <span class="n">mx_uint</span><span class="p">,</span> <span class="n">c_str_array</span>
 <span class="kn">from</span> <span class="nn">..base</span> <span class="kn">import</span> <span class="n">NDArrayHandle</span><span class="p">,</span> <span class="n">SymbolHandle</span>
 <span class="kn">from</span> <span class="nn">..symbol</span> <span class="kn">import</span> <span class="n">Symbol</span>
 <span class="kn">from</span> <span class="nn">..symbol</span> <span class="kn">import</span> <span class="n">load</span> <span class="k">as</span> <span class="n">sym_load</span>
 <span class="kn">from</span> <span class="nn">..</span> <span class="kn">import</span> <span class="n">ndarray</span>
 <span class="kn">from</span> <span class="nn">..ndarray</span> <span class="kn">import</span> <span class="n">load</span> <span class="k">as</span> <span class="n">nd_load</span>
 <span class="kn">from</span> <span class="nn">..ndarray</span> <span class="kn">import</span> <span class="n">save</span> <span class="k">as</span> <span class="n">nd_save</span>
 <span class="kn">from</span> <span class="nn">..ndarray</span> <span class="kn">import</span> <span class="n">NDArray</span>
 <span class="kn">from</span> <span class="nn">..io</span> <span class="kn">import</span> <span class="n">DataIter</span><span class="p">,</span> <span class="n">DataDesc</span><span class="p">,</span> <span class="n">DataBatch</span>
 <span class="kn">from</span> <span class="nn">..context</span> <span class="kn">import</span> <span class="n">cpu</span><span class="p">,</span> <span class="n">Context</span>
 <span class="kn">from</span> <span class="nn">..module</span> <span class="kn">import</span> <span class="n">Module</span>


 <span class="k">def</span> <span class="nf">_quantize_params</span><span class="p">(</span><span class="n">qsym</span><span class="p">,</span> <span class="n">params</span><span class="p">,</span> <span class="n">th_dict</span><span class="p">):</span>
     <span class="sd">&quot;&quot;&quot;Given a quantized symbol and a dict of params that have not been quantized,</span>
 <span class="sd">    generate quantized params. Currently only supports quantizing the arg_params</span>
 <span class="sd">    with names of `weight` or `bias`, not aux_params. If `qsym` contains symbols</span>
 <span class="sd">    that are excluded from being quantized, their corresponding params will</span>
 <span class="sd">    not be quantized, but saved together with quantized params of the symbols that</span>
 <span class="sd">    have been quantized.</span>

 <span class="sd">    Parameters</span>
 <span class="sd">    ----------</span>
 <span class="sd">    qsym : Symbol</span>
 <span class="sd">        Quantized symbol from FP32 symbol.</span>
 <span class="sd">    params : dict of str-&gt;NDArray</span>
 <span class="sd">    th_dict: dict of min/max pairs of layers&#39; output</span>
 <span class="sd">    &quot;&quot;&quot;</span>
     <span class="n">inputs_name</span> <span class="o">=</span> <span class="n">qsym</span><span class="o">.</span><span class="n">list_arguments</span><span class="p">()</span>
     <span class="n">quantized_params</span> <span class="o">=</span> <span class="p">{}</span>
     <span class="k">for</span> <span class="n">name</span> <span class="ow">in</span> <span class="n">inputs_name</span><span class="p">:</span>
         <span class="k">if</span> <span class="n">name</span><span class="o">.</span><span class="n">endswith</span><span class="p">((</span><span class="s1">&#39;weight_quantize&#39;</span><span class="p">,</span> <span class="s1">&#39;bias_quantize&#39;</span><span class="p">)):</span>
             <span class="n">original_name</span> <span class="o">=</span> <span class="n">name</span><span class="p">[:</span><span class="o">-</span><span class="nb">len</span><span class="p">(</span><span class="s1">&#39;_quantize&#39;</span><span class="p">)]</span>
             <span class="n">param</span> <span class="o">=</span> <span class="n">params</span><span class="p">[</span><span class="n">original_name</span><span class="p">]</span>
             <span class="c1"># pylint: disable=unbalanced-tuple-unpacking</span>
             <span class="n">val</span><span class="p">,</span> <span class="n">vmin</span><span class="p">,</span> <span class="n">vmax</span> <span class="o">=</span> <span class="n">ndarray</span><span class="o">.</span><span class="n">contrib</span><span class="o">.</span><span class="n">quantize</span><span class="p">(</span><span class="n">data</span><span class="o">=</span><span class="n">param</span><span class="p">,</span>
                                                        <span class="n">min_range</span><span class="o">=</span><span class="n">ndarray</span><span class="o">.</span><span class="n">min</span><span class="p">(</span><span class="n">param</span><span class="p">),</span>
                                                        <span class="n">max_range</span><span class="o">=</span><span class="n">ndarray</span><span class="o">.</span><span class="n">max</span><span class="p">(</span><span class="n">param</span><span class="p">),</span>
                                                        <span class="n">out_type</span><span class="o">=</span><span class="s1">&#39;int8&#39;</span><span class="p">)</span>
             <span class="n">quantized_params</span><span class="p">[</span><span class="n">name</span><span class="p">]</span> <span class="o">=</span> <span class="n">val</span>
             <span class="n">quantized_params</span><span class="p">[</span><span class="n">name</span><span class="o">+</span><span class="s1">&#39;_min&#39;</span><span class="p">]</span> <span class="o">=</span> <span class="n">vmin</span>
             <span class="n">quantized_params</span><span class="p">[</span><span class="n">name</span><span class="o">+</span><span class="s1">&#39;_max&#39;</span><span class="p">]</span> <span class="o">=</span> <span class="n">vmax</span>
         <span class="k">elif</span> <span class="n">name</span> <span class="ow">in</span> <span class="n">params</span><span class="p">:</span>
             <span class="n">quantized_params</span><span class="p">[</span><span class="n">name</span><span class="p">]</span> <span class="o">=</span> <span class="n">params</span><span class="p">[</span><span class="n">name</span><span class="p">]</span>
         <span class="k">elif</span> <span class="n">name</span><span class="o">.</span><span class="n">endswith</span><span class="p">((</span><span class="s1">&#39;_min&#39;</span><span class="p">)):</span>
             <span class="n">output</span> <span class="o">=</span> <span class="n">name</span><span class="p">[:</span> <span class="o">-</span> <span class="nb">len</span><span class="p">(</span><span class="s1">&#39;_min&#39;</span><span class="p">)]</span>
             <span class="k">if</span> <span class="n">output</span> <span class="ow">in</span> <span class="n">th_dict</span><span class="p">:</span>
                 <span class="n">quantized_params</span><span class="p">[</span><span class="n">name</span><span class="p">]</span> <span class="o">=</span> <span class="n">ndarray</span><span class="o">.</span><span class="n">array</span><span class="p">([</span><span class="n">th_dict</span><span class="p">[</span><span class="n">output</span><span class="p">][</span><span class="mi">0</span><span class="p">]])</span>
         <span class="k">elif</span> <span class="n">name</span><span class="o">.</span><span class="n">endswith</span><span class="p">((</span><span class="s1">&#39;_max&#39;</span><span class="p">)):</span>
             <span class="n">output</span> <span class="o">=</span> <span class="n">name</span><span class="p">[:</span> <span class="o">-</span> <span class="nb">len</span><span class="p">(</span><span class="s1">&#39;_min&#39;</span><span class="p">)]</span>
             <span class="k">if</span> <span class="n">output</span> <span class="ow">in</span> <span class="n">th_dict</span><span class="p">:</span>
                 <span class="n">quantized_params</span><span class="p">[</span><span class="n">name</span><span class="p">]</span> <span class="o">=</span> <span class="n">ndarray</span><span class="o">.</span><span class="n">array</span><span class="p">([</span><span class="n">th_dict</span><span class="p">[</span><span class="n">output</span><span class="p">][</span><span class="mi">1</span><span class="p">]])</span>
     <span class="k">return</span> <span class="n">quantized_params</span>

 <span class="k">def</span> <span class="nf">_quantize_symbol</span><span class="p">(</span><span class="n">sym</span><span class="p">,</span> <span class="n">ctx</span><span class="p">,</span> <span class="n">excluded_symbols</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span> <span class="n">excluded_operators</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
                      <span class="n">offline_params</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span> <span class="n">quantized_dtype</span><span class="o">=</span><span class="s1">&#39;int8&#39;</span><span class="p">,</span> <span class="n">quantize_mode</span><span class="o">=</span><span class="s1">&#39;smart&#39;</span><span class="p">,</span>
                      <span class="n">quantize_granularity</span><span class="o">=</span><span class="s1">&#39;tensor-wise&#39;</span><span class="p">):</span>
     <span class="sd">&quot;&quot;&quot;Given a symbol object representing a neural network of data type FP32,</span>
 <span class="sd">    quantize it into a INT8 network.</span>

 <span class="sd">    Parameters</span>
 <span class="sd">    ----------</span>
 <span class="sd">    sym : Symbol</span>
 <span class="sd">        FP32 neural network symbol.</span>
 <span class="sd">    ctx : Context</span>
 <span class="sd">        Defines the device that users want to run quantized symbol.</span>
 <span class="sd">    excluded_symbols : list of strings</span>
 <span class="sd">        A list of strings representing the names of the symbols that users want to excluding</span>
 <span class="sd">        from being quantized.</span>
 <span class="sd">    excluded_operators : list of strings</span>
 <span class="sd">        A list of strings representing the names of the operators that users want to excluding</span>
 <span class="sd">        from being quantized.</span>
 <span class="sd">    offline_params : list of strs</span>
 <span class="sd">        Names of the parameters that users want to quantize offline. It&#39;s always recommended to</span>
 <span class="sd">        quantize parameters offline so that quantizing parameters during the inference can be</span>
 <span class="sd">        avoided.</span>
 <span class="sd">    quantized_dtype: str</span>
 <span class="sd">        The quantized destination type for input data.</span>
 <span class="sd">    quantize_mode: str</span>
 <span class="sd">        The mode that quantization pass to apply.</span>
 <span class="sd">    quantize_granularity: str</span>
 <span class="sd">        The granularity of quantization, currently supports &#39;tensor-wise&#39; and &#39;channel-wise&#39;</span>
 <span class="sd">        quantization. The default value is &#39;tensor-wise&#39;.</span>

 <span class="sd">    &quot;&quot;&quot;</span>
     <span class="n">num_excluded_symbols</span> <span class="o">=</span> <span class="mi">0</span>
     <span class="k">if</span> <span class="n">excluded_symbols</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
         <span class="k">assert</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">excluded_symbols</span><span class="p">,</span> <span class="nb">list</span><span class="p">)</span>
         <span class="n">num_excluded_symbols</span> <span class="o">=</span> <span class="nb">len</span><span class="p">(</span><span class="n">excluded_symbols</span><span class="p">)</span>
     <span class="k">else</span><span class="p">:</span>
         <span class="n">excluded_symbols</span> <span class="o">=</span> <span class="p">[]</span>

     <span class="n">num_excluded_ops</span> <span class="o">=</span> <span class="mi">0</span>
     <span class="k">if</span> <span class="n">excluded_operators</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
         <span class="k">assert</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">excluded_operators</span><span class="p">,</span> <span class="nb">list</span><span class="p">)</span>
         <span class="n">num_excluded_ops</span> <span class="o">=</span> <span class="nb">len</span><span class="p">(</span><span class="n">excluded_operators</span><span class="p">)</span>
     <span class="k">else</span><span class="p">:</span>
         <span class="n">excluded_operators</span> <span class="o">=</span> <span class="p">[]</span>

     <span class="n">num_offline</span> <span class="o">=</span> <span class="mi">0</span>
     <span class="n">offline</span> <span class="o">=</span> <span class="p">[]</span>
     <span class="k">if</span> <span class="n">offline_params</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
         <span class="n">num_offline</span> <span class="o">=</span> <span class="nb">len</span><span class="p">(</span><span class="n">offline_params</span><span class="p">)</span>
         <span class="k">for</span> <span class="n">k</span> <span class="ow">in</span> <span class="n">offline_params</span><span class="p">:</span>
             <span class="n">offline</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="n">c_str</span><span class="p">(</span><span class="n">k</span><span class="p">))</span>

     <span class="n">out</span> <span class="o">=</span> <span class="n">SymbolHandle</span><span class="p">()</span>
     <span class="n">size</span> <span class="o">=</span> <span class="n">mx_uint</span><span class="p">()</span>
     <span class="n">calib_str</span> <span class="o">=</span> <span class="n">ctypes</span><span class="o">.</span><span class="n">POINTER</span><span class="p">(</span><span class="n">ctypes</span><span class="o">.</span><span class="n">c_char_p</span><span class="p">)()</span>
     <span class="n">check_call</span><span class="p">(</span><span class="n">_LIB</span><span class="o">.</span><span class="n">MXQuantizeSymbol</span><span class="p">(</span><span class="n">sym</span><span class="o">.</span><span class="n">handle</span><span class="p">,</span>
                                      <span class="n">ctypes</span><span class="o">.</span><span class="n">byref</span><span class="p">(</span><span class="n">out</span><span class="p">),</span>
                                      <span class="n">ctypes</span><span class="o">.</span><span class="n">byref</span><span class="p">(</span><span class="n">ctypes</span><span class="o">.</span><span class="n">c_int</span><span class="p">(</span><span class="n">ctx</span><span class="o">.</span><span class="n">device_typeid</span><span class="p">)),</span>
                                      <span class="n">mx_uint</span><span class="p">(</span><span class="n">num_excluded_symbols</span><span class="p">),</span>
                                      <span class="n">c_str_array</span><span class="p">(</span><span class="n">excluded_symbols</span><span class="p">),</span>
                                      <span class="n">mx_uint</span><span class="p">(</span><span class="n">num_excluded_ops</span><span class="p">),</span>
                                      <span class="n">c_str_array</span><span class="p">(</span><span class="n">excluded_operators</span><span class="p">),</span>
                                      <span class="n">mx_uint</span><span class="p">(</span><span class="n">num_offline</span><span class="p">),</span>
                                      <span class="n">c_array</span><span class="p">(</span><span class="n">ctypes</span><span class="o">.</span><span class="n">c_char_p</span><span class="p">,</span> <span class="n">offline</span><span class="p">),</span>
                                      <span class="n">c_str</span><span class="p">(</span><span class="n">quantized_dtype</span><span class="p">),</span>
                                      <span class="n">ctypes</span><span class="o">.</span><span class="n">c_bool</span><span class="p">(</span><span class="kc">True</span><span class="p">),</span>
                                      <span class="n">c_str</span><span class="p">(</span><span class="n">quantize_mode</span><span class="p">),</span>
                                      <span class="n">c_str</span><span class="p">(</span><span class="n">quantize_granularity</span><span class="p">),</span>
                                      <span class="n">ctypes</span><span class="o">.</span><span class="n">byref</span><span class="p">(</span><span class="n">size</span><span class="p">),</span>
                                      <span class="n">ctypes</span><span class="o">.</span><span class="n">byref</span><span class="p">(</span><span class="n">calib_str</span><span class="p">)))</span>
     <span class="n">calib_layer</span> <span class="o">=</span> <span class="p">[]</span>
     <span class="n">calib_layer</span> <span class="o">=</span> <span class="p">[</span><span class="n">py_str</span><span class="p">(</span><span class="n">calib_str</span><span class="p">[</span><span class="n">i</span><span class="p">])</span> <span class="k">for</span> <span class="n">i</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="n">size</span><span class="o">.</span><span class="n">value</span><span class="p">)]</span>
     <span class="k">return</span> <span class="n">Symbol</span><span class="p">(</span><span class="n">out</span><span class="p">),</span> <span class="n">calib_layer</span>

 <div class="viewcode-block" id="combine_histogram"><a class="viewcode-back" href="../../../api/contrib/quantization/index.html#mxnet.contrib.quantization.combine_histogram">[docs]</a><span class="k">def</span> <span class="nf">combine_histogram</span><span class="p">(</span><span class="n">old_hist</span><span class="p">,</span> <span class="n">arr</span><span class="p">,</span> <span class="n">new_min</span><span class="p">,</span> <span class="n">new_max</span><span class="p">,</span> <span class="n">new_th</span><span class="p">):</span>
     <span class="sd">&quot;&quot;&quot; Collect layer histogram for arr and combine it with old histogram.</span>
 <span class="sd">    &quot;&quot;&quot;</span>
     <span class="p">(</span><span class="n">old_hist</span><span class="p">,</span> <span class="n">old_hist_edges</span><span class="p">,</span> <span class="n">old_min</span><span class="p">,</span> <span class="n">old_max</span><span class="p">,</span> <span class="n">old_th</span><span class="p">)</span> <span class="o">=</span> <span class="n">old_hist</span>
     <span class="k">if</span> <span class="n">new_th</span> <span class="o">&lt;=</span> <span class="n">old_th</span><span class="p">:</span>
         <span class="n">hist</span><span class="p">,</span> <span class="n">_</span> <span class="o">=</span> <span class="n">np</span><span class="o">.</span><span class="n">histogram</span><span class="p">(</span><span class="n">arr</span><span class="p">,</span> <span class="n">bins</span><span class="o">=</span><span class="nb">len</span><span class="p">(</span><span class="n">old_hist</span><span class="p">),</span> <span class="nb">range</span><span class="o">=</span><span class="p">(</span><span class="o">-</span><span class="n">old_th</span><span class="p">,</span> <span class="n">old_th</span><span class="p">))</span>
         <span class="k">return</span> <span class="p">(</span><span class="n">old_hist</span> <span class="o">+</span> <span class="n">hist</span><span class="p">,</span> <span class="n">old_hist_edges</span><span class="p">,</span> <span class="nb">min</span><span class="p">(</span><span class="n">old_min</span><span class="p">,</span> <span class="n">new_min</span><span class="p">),</span> <span class="nb">max</span><span class="p">(</span><span class="n">old_max</span><span class="p">,</span> <span class="n">new_max</span><span class="p">),</span> <span class="n">old_th</span><span class="p">)</span>
     <span class="k">else</span><span class="p">:</span>
         <span class="c1"># Need to generate new histogram with new_th</span>
         <span class="n">old_num_bins</span> <span class="o">=</span> <span class="nb">len</span><span class="p">(</span><span class="n">old_hist</span><span class="p">)</span>
         <span class="n">old_step</span> <span class="o">=</span> <span class="mi">2</span> <span class="o">*</span> <span class="n">old_th</span> <span class="o">/</span> <span class="n">old_num_bins</span>
         <span class="n">half_increased_bins</span> <span class="o">=</span> <span class="nb">int</span><span class="p">((</span><span class="n">new_th</span> <span class="o">-</span> <span class="n">old_th</span><span class="p">)</span> <span class="o">//</span> <span class="n">old_step</span> <span class="o">+</span> <span class="mi">1</span><span class="p">)</span>
         <span class="n">new_num_bins</span> <span class="o">=</span> <span class="n">half_increased_bins</span> <span class="o">*</span> <span class="mi">2</span> <span class="o">+</span> <span class="n">old_num_bins</span>
         <span class="n">new_th</span> <span class="o">=</span> <span class="n">half_increased_bins</span> <span class="o">*</span> <span class="n">old_step</span> <span class="o">+</span> <span class="n">old_th</span>
         <span class="n">hist</span><span class="p">,</span> <span class="n">hist_edges</span> <span class="o">=</span> <span class="n">np</span><span class="o">.</span><span class="n">histogram</span><span class="p">(</span><span class="n">arr</span><span class="p">,</span> <span class="n">bins</span><span class="o">=</span><span class="n">new_num_bins</span><span class="p">,</span> <span class="nb">range</span><span class="o">=</span><span class="p">(</span><span class="o">-</span><span class="n">new_th</span><span class="p">,</span> <span class="n">new_th</span><span class="p">))</span>
         <span class="n">hist</span><span class="p">[</span><span class="n">half_increased_bins</span><span class="p">:</span><span class="n">new_num_bins</span> <span class="o">-</span> <span class="n">half_increased_bins</span><span class="p">]</span> <span class="o">+=</span> <span class="n">old_hist</span>
         <span class="k">return</span> <span class="p">(</span><span class="n">hist</span><span class="p">,</span> <span class="n">hist_edges</span><span class="p">,</span> <span class="nb">min</span><span class="p">(</span><span class="n">old_min</span><span class="p">,</span> <span class="n">new_min</span><span class="p">),</span> <span class="nb">max</span><span class="p">(</span><span class="n">old_max</span><span class="p">,</span> <span class="n">new_max</span><span class="p">),</span> <span class="n">new_th</span><span class="p">)</span></div>

 <span class="k">class</span> <span class="nc">_LayerHistogramCollector</span><span class="p">(</span><span class="nb">object</span><span class="p">):</span>
     <span class="sd">&quot;&quot;&quot;Saves layer histogram in a dict with layer names as keys and lists of NDArrays as</span>
 <span class="sd">    values. The collected histogram will be used for calculating the optimal thresholds for</span>
 <span class="sd">    quantization using KL divergence.</span>
 <span class="sd">    &quot;&quot;&quot;</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">num_bins</span><span class="o">=</span><span class="mi">8001</span><span class="p">,</span> <span class="n">include_layer</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span> <span class="n">logger</span><span class="o">=</span><span class="kc">None</span><span class="p">):</span>
         <span class="bp">self</span><span class="o">.</span><span class="n">hist_dict</span> <span class="o">=</span> <span class="p">{}</span>
         <span class="bp">self</span><span class="o">.</span><span class="n">num_bins</span> <span class="o">=</span> <span class="n">num_bins</span>
         <span class="bp">self</span><span class="o">.</span><span class="n">include_layer</span> <span class="o">=</span> <span class="n">include_layer</span>
         <span class="bp">self</span><span class="o">.</span><span class="n">logger</span> <span class="o">=</span> <span class="n">logger</span>

     <span class="k">def</span> <span class="nf">collect</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">name</span><span class="p">,</span> <span class="n">arr</span><span class="p">):</span>
         <span class="sd">&quot;&quot;&quot;Callback function for collecting layer output NDArrays.&quot;&quot;&quot;</span>
         <span class="n">name</span> <span class="o">=</span> <span class="n">py_str</span><span class="p">(</span><span class="n">name</span><span class="p">)</span>
         <span class="k">if</span> <span class="n">name</span> <span class="ow">not</span> <span class="ow">in</span> <span class="bp">self</span><span class="o">.</span><span class="n">include_layer</span><span class="p">:</span>
             <span class="k">return</span>
         <span class="n">handle</span> <span class="o">=</span> <span class="n">ctypes</span><span class="o">.</span><span class="n">cast</span><span class="p">(</span><span class="n">arr</span><span class="p">,</span> <span class="n">NDArrayHandle</span><span class="p">)</span>
         <span class="n">arr</span> <span class="o">=</span> <span class="n">NDArray</span><span class="p">(</span><span class="n">handle</span><span class="p">,</span> <span class="n">writable</span><span class="o">=</span><span class="kc">False</span><span class="p">)</span><span class="o">.</span><span class="n">copyto</span><span class="p">(</span><span class="n">cpu</span><span class="p">())</span><span class="o">.</span><span class="n">asnumpy</span><span class="p">()</span>
         <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">logger</span><span class="p">:</span>
             <span class="bp">self</span><span class="o">.</span><span class="n">logger</span><span class="o">.</span><span class="n">debug</span><span class="p">(</span><span class="s2">&quot;Collecting layer </span><span class="si">%s</span><span class="s2"> histogram of shape </span><span class="si">%s</span><span class="s2">&quot;</span> <span class="o">%</span> <span class="p">(</span><span class="n">name</span><span class="p">,</span> <span class="n">arr</span><span class="o">.</span><span class="n">shape</span><span class="p">))</span>
         <span class="n">min_range</span> <span class="o">=</span> <span class="n">np</span><span class="o">.</span><span class="n">min</span><span class="p">(</span><span class="n">arr</span><span class="p">)</span>
         <span class="n">max_range</span> <span class="o">=</span> <span class="n">np</span><span class="o">.</span><span class="n">max</span><span class="p">(</span><span class="n">arr</span><span class="p">)</span>
         <span class="n">th</span> <span class="o">=</span> <span class="nb">max</span><span class="p">(</span><span class="nb">abs</span><span class="p">(</span><span class="n">min_range</span><span class="p">),</span> <span class="nb">abs</span><span class="p">(</span><span class="n">max_range</span><span class="p">))</span>
         <span class="k">if</span> <span class="n">name</span> <span class="ow">in</span> <span class="bp">self</span><span class="o">.</span><span class="n">hist_dict</span><span class="p">:</span>
             <span class="bp">self</span><span class="o">.</span><span class="n">hist_dict</span><span class="p">[</span><span class="n">name</span><span class="p">]</span> <span class="o">=</span> <span class="n">combine_histogram</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">hist_dict</span><span class="p">[</span><span class="n">name</span><span class="p">],</span> <span class="n">arr</span><span class="p">,</span> <span class="n">min_range</span><span class="p">,</span> <span class="n">max_range</span><span class="p">,</span> <span class="n">th</span><span class="p">)</span>
         <span class="k">else</span><span class="p">:</span>
             <span class="n">hist</span><span class="p">,</span> <span class="n">hist_edges</span> <span class="o">=</span> <span class="n">np</span><span class="o">.</span><span class="n">histogram</span><span class="p">(</span><span class="n">arr</span><span class="p">,</span> <span class="n">bins</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">num_bins</span><span class="p">,</span> <span class="nb">range</span><span class="o">=</span><span class="p">(</span><span class="o">-</span><span class="n">th</span><span class="p">,</span> <span class="n">th</span><span class="p">))</span>
             <span class="bp">self</span><span class="o">.</span><span class="n">hist_dict</span><span class="p">[</span><span class="n">name</span><span class="p">]</span> <span class="o">=</span> <span class="p">(</span><span class="n">hist</span><span class="p">,</span> <span class="n">hist_edges</span><span class="p">,</span> <span class="n">min_range</span><span class="p">,</span> <span class="n">max_range</span><span class="p">,</span> <span class="n">th</span><span class="p">)</span>

 <span class="k">class</span> <span class="nc">_LayerOutputMinMaxCollector</span><span class="p">(</span><span class="nb">object</span><span class="p">):</span>
     <span class="sd">&quot;&quot;&quot;Saves layer output min and max values in a dict with layer names as keys.</span>
 <span class="sd">    The collected min and max values will be directly used as thresholds for quantization.</span>
 <span class="sd">    &quot;&quot;&quot;</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">quantized_dtype</span><span class="p">,</span> <span class="n">include_layer</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span> <span class="n">logger</span><span class="o">=</span><span class="kc">None</span><span class="p">):</span>
         <span class="bp">self</span><span class="o">.</span><span class="n">min_max_dict</span> <span class="o">=</span> <span class="p">{}</span>
         <span class="bp">self</span><span class="o">.</span><span class="n">quantized_dtype</span> <span class="o">=</span> <span class="n">quantized_dtype</span>
         <span class="bp">self</span><span class="o">.</span><span class="n">include_layer</span> <span class="o">=</span> <span class="n">include_layer</span>
         <span class="bp">self</span><span class="o">.</span><span class="n">logger</span> <span class="o">=</span> <span class="n">logger</span>

     <span class="k">def</span> <span class="nf">collect</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">name</span><span class="p">,</span> <span class="n">arr</span><span class="p">):</span>
         <span class="sd">&quot;&quot;&quot;Callback function for collecting min and max values from an NDArray.&quot;&quot;&quot;</span>
         <span class="n">name</span> <span class="o">=</span> <span class="n">py_str</span><span class="p">(</span><span class="n">name</span><span class="p">)</span>
         <span class="k">if</span> <span class="n">name</span> <span class="ow">not</span> <span class="ow">in</span> <span class="bp">self</span><span class="o">.</span><span class="n">include_layer</span><span class="p">:</span>
             <span class="k">return</span>
         <span class="n">handle</span> <span class="o">=</span> <span class="n">ctypes</span><span class="o">.</span><span class="n">cast</span><span class="p">(</span><span class="n">arr</span><span class="p">,</span> <span class="n">NDArrayHandle</span><span class="p">)</span>
         <span class="n">arr</span> <span class="o">=</span> <span class="n">NDArray</span><span class="p">(</span><span class="n">handle</span><span class="p">,</span> <span class="n">writable</span><span class="o">=</span><span class="kc">False</span><span class="p">)</span>
         <span class="n">min_range</span> <span class="o">=</span> <span class="n">ndarray</span><span class="o">.</span><span class="n">min</span><span class="p">(</span><span class="n">arr</span><span class="p">)</span><span class="o">.</span><span class="n">asscalar</span><span class="p">()</span>
         <span class="n">max_range</span> <span class="o">=</span> <span class="n">ndarray</span><span class="o">.</span><span class="n">max</span><span class="p">(</span><span class="n">arr</span><span class="p">)</span><span class="o">.</span><span class="n">asscalar</span><span class="p">()</span>
         <span class="k">if</span> <span class="n">name</span> <span class="ow">in</span> <span class="bp">self</span><span class="o">.</span><span class="n">min_max_dict</span><span class="p">:</span>
             <span class="n">cur_min_max</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">min_max_dict</span><span class="p">[</span><span class="n">name</span><span class="p">]</span>
             <span class="bp">self</span><span class="o">.</span><span class="n">min_max_dict</span><span class="p">[</span><span class="n">name</span><span class="p">]</span> <span class="o">=</span> <span class="p">(</span><span class="nb">min</span><span class="p">(</span><span class="n">cur_min_max</span><span class="p">[</span><span class="mi">0</span><span class="p">],</span> <span class="n">min_range</span><span class="p">),</span>
                                        <span class="nb">max</span><span class="p">(</span><span class="n">cur_min_max</span><span class="p">[</span><span class="mi">1</span><span class="p">],</span> <span class="n">max_range</span><span class="p">))</span>
         <span class="k">else</span><span class="p">:</span>
             <span class="bp">self</span><span class="o">.</span><span class="n">min_max_dict</span><span class="p">[</span><span class="n">name</span><span class="p">]</span> <span class="o">=</span> <span class="p">(</span><span class="n">min_range</span><span class="p">,</span> <span class="n">max_range</span><span class="p">)</span>
         <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">logger</span><span class="p">:</span>
             <span class="bp">self</span><span class="o">.</span><span class="n">logger</span><span class="o">.</span><span class="n">debug</span><span class="p">(</span><span class="s2">&quot;Collecting layer </span><span class="si">%s</span><span class="s2"> min_range=</span><span class="si">%f</span><span class="s2">, max_range=</span><span class="si">%f</span><span class="s2">&quot;</span>
                               <span class="o">%</span> <span class="p">(</span><span class="n">name</span><span class="p">,</span> <span class="n">min_range</span><span class="p">,</span> <span class="n">max_range</span><span class="p">))</span>

 <span class="k">def</span> <span class="nf">_calibrate_quantized_sym</span><span class="p">(</span><span class="n">qsym</span><span class="p">,</span> <span class="n">th_dict</span><span class="p">):</span>
     <span class="sd">&quot;&quot;&quot;Given a dictionary containing the thresholds for quantizing the layers,</span>
 <span class="sd">    set the thresholds into the quantized symbol as the params of requantize operators.</span>
 <span class="sd">    &quot;&quot;&quot;</span>
     <span class="k">if</span> <span class="n">th_dict</span> <span class="ow">is</span> <span class="kc">None</span> <span class="ow">or</span> <span class="nb">len</span><span class="p">(</span><span class="n">th_dict</span><span class="p">)</span> <span class="o">==</span> <span class="mi">0</span><span class="p">:</span>
         <span class="k">return</span> <span class="n">qsym</span>
     <span class="n">num_layer_outputs</span> <span class="o">=</span> <span class="nb">len</span><span class="p">(</span><span class="n">th_dict</span><span class="p">)</span>
     <span class="n">layer_output_names</span> <span class="o">=</span> <span class="p">[]</span>
     <span class="n">min_vals</span> <span class="o">=</span> <span class="p">[]</span>
     <span class="n">max_vals</span> <span class="o">=</span> <span class="p">[]</span>
     <span class="k">for</span> <span class="n">k</span><span class="p">,</span> <span class="n">v</span> <span class="ow">in</span> <span class="n">th_dict</span><span class="o">.</span><span class="n">items</span><span class="p">():</span>
         <span class="n">layer_output_names</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="n">k</span><span class="p">)</span>
         <span class="n">min_vals</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="n">v</span><span class="p">[</span><span class="mi">0</span><span class="p">])</span>
         <span class="n">max_vals</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="n">v</span><span class="p">[</span><span class="mi">1</span><span class="p">])</span>

     <span class="n">calibrated_sym</span> <span class="o">=</span> <span class="n">SymbolHandle</span><span class="p">()</span>
     <span class="n">check_call</span><span class="p">(</span><span class="n">_LIB</span><span class="o">.</span><span class="n">MXSetCalibTableToQuantizedSymbol</span><span class="p">(</span><span class="n">qsym</span><span class="o">.</span><span class="n">handle</span><span class="p">,</span>
                                                      <span class="n">mx_uint</span><span class="p">(</span><span class="n">num_layer_outputs</span><span class="p">),</span>
                                                      <span class="n">c_str_array</span><span class="p">(</span><span class="n">layer_output_names</span><span class="p">),</span>
                                                      <span class="n">c_array</span><span class="p">(</span><span class="n">ctypes</span><span class="o">.</span><span class="n">c_float</span><span class="p">,</span> <span class="n">min_vals</span><span class="p">),</span>
                                                      <span class="n">c_array</span><span class="p">(</span><span class="n">ctypes</span><span class="o">.</span><span class="n">c_float</span><span class="p">,</span> <span class="n">max_vals</span><span class="p">),</span>
                                                      <span class="n">ctypes</span><span class="o">.</span><span class="n">byref</span><span class="p">(</span><span class="n">calibrated_sym</span><span class="p">)))</span>
     <span class="k">return</span> <span class="n">Symbol</span><span class="p">(</span><span class="n">calibrated_sym</span><span class="p">)</span>


 <span class="k">def</span> <span class="nf">_collect_layer_statistics</span><span class="p">(</span><span class="n">mod</span><span class="p">,</span> <span class="n">data</span><span class="p">,</span> <span class="n">collector</span><span class="p">,</span> <span class="n">max_num_examples</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span> <span class="n">logger</span><span class="o">=</span><span class="kc">None</span><span class="p">):</span>
     <span class="k">if</span> <span class="ow">not</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">data</span><span class="p">,</span> <span class="n">DataIter</span><span class="p">):</span>
         <span class="k">raise</span> <span class="ne">ValueError</span><span class="p">(</span><span class="s1">&#39;Only supports data as a type of DataIter, while received type </span><span class="si">%s</span><span class="s1">&#39;</span>
                          <span class="o">%</span> <span class="nb">str</span><span class="p">(</span><span class="nb">type</span><span class="p">(</span><span class="n">data</span><span class="p">)))</span>
     <span class="n">mod</span><span class="o">.</span><span class="n">_exec_group</span><span class="o">.</span><span class="n">execs</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span><span class="o">.</span><span class="n">set_monitor_callback</span><span class="p">(</span><span class="n">collector</span><span class="o">.</span><span class="n">collect</span><span class="p">,</span> <span class="n">monitor_all</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span>
     <span class="n">num_batches</span> <span class="o">=</span> <span class="mi">0</span>
     <span class="n">num_examples</span> <span class="o">=</span> <span class="mi">0</span>
     <span class="k">for</span> <span class="n">batch</span> <span class="ow">in</span> <span class="n">data</span><span class="p">:</span>
         <span class="n">mod</span><span class="o">.</span><span class="n">forward</span><span class="p">(</span><span class="n">data_batch</span><span class="o">=</span><span class="n">batch</span><span class="p">,</span> <span class="n">is_train</span><span class="o">=</span><span class="kc">False</span><span class="p">)</span>
         <span class="n">num_batches</span> <span class="o">+=</span> <span class="mi">1</span>
         <span class="n">num_examples</span> <span class="o">+=</span> <span class="n">data</span><span class="o">.</span><span class="n">batch_size</span>
         <span class="k">if</span> <span class="n">max_num_examples</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span> <span class="ow">and</span> <span class="n">num_examples</span> <span class="o">&gt;=</span> <span class="n">max_num_examples</span><span class="p">:</span>
             <span class="k">break</span>
     <span class="k">if</span> <span class="n">logger</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
         <span class="n">logger</span><span class="o">.</span><span class="n">info</span><span class="p">(</span><span class="s2">&quot;Collected statistics from </span><span class="si">%d</span><span class="s2"> batches with batch_size=</span><span class="si">%d</span><span class="s2">&quot;</span>
                     <span class="o">%</span> <span class="p">(</span><span class="n">num_batches</span><span class="p">,</span> <span class="n">data</span><span class="o">.</span><span class="n">batch_size</span><span class="p">))</span>
     <span class="k">return</span> <span class="n">num_examples</span>


 <span class="k">def</span> <span class="nf">_collect_layer_output_min_max</span><span class="p">(</span><span class="n">mod</span><span class="p">,</span> <span class="n">data</span><span class="p">,</span> <span class="n">quantized_dtype</span><span class="p">,</span> <span class="n">include_layer</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
                                   <span class="n">max_num_examples</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span> <span class="n">logger</span><span class="o">=</span><span class="kc">None</span><span class="p">):</span>
     <span class="sd">&quot;&quot;&quot;Collect min and max values from layer outputs and save them in</span>
 <span class="sd">    a dictionary mapped by layer names.</span>
 <span class="sd">    &quot;&quot;&quot;</span>
     <span class="n">collector</span> <span class="o">=</span> <span class="n">_LayerOutputMinMaxCollector</span><span class="p">(</span><span class="n">quantized_dtype</span><span class="o">=</span><span class="n">quantized_dtype</span><span class="p">,</span>
                                             <span class="n">include_layer</span><span class="o">=</span><span class="n">include_layer</span><span class="p">,</span> <span class="n">logger</span><span class="o">=</span><span class="n">logger</span><span class="p">)</span>
     <span class="n">num_examples</span> <span class="o">=</span> <span class="n">_collect_layer_statistics</span><span class="p">(</span><span class="n">mod</span><span class="p">,</span> <span class="n">data</span><span class="p">,</span> <span class="n">collector</span><span class="p">,</span> <span class="n">max_num_examples</span><span class="p">,</span> <span class="n">logger</span><span class="p">)</span>
     <span class="k">return</span> <span class="n">collector</span><span class="o">.</span><span class="n">min_max_dict</span><span class="p">,</span> <span class="n">num_examples</span>


 <span class="k">def</span> <span class="nf">_collect_layer_histogram</span><span class="p">(</span><span class="n">mod</span><span class="p">,</span> <span class="n">data</span><span class="p">,</span> <span class="n">include_layer</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
                              <span class="n">max_num_examples</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span> <span class="n">logger</span><span class="o">=</span><span class="kc">None</span><span class="p">):</span>
     <span class="sd">&quot;&quot;&quot;Collect layer outputs and save them in a dictionary mapped by layer names.&quot;&quot;&quot;</span>
     <span class="n">collector</span> <span class="o">=</span> <span class="n">_LayerHistogramCollector</span><span class="p">(</span><span class="n">include_layer</span><span class="o">=</span><span class="n">include_layer</span><span class="p">,</span> <span class="n">logger</span><span class="o">=</span><span class="n">logger</span><span class="p">)</span>
     <span class="n">num_examples</span> <span class="o">=</span> <span class="n">_collect_layer_statistics</span><span class="p">(</span><span class="n">mod</span><span class="p">,</span> <span class="n">data</span><span class="p">,</span> <span class="n">collector</span><span class="p">,</span> <span class="n">max_num_examples</span><span class="p">,</span> <span class="n">logger</span><span class="p">)</span>
     <span class="k">return</span> <span class="n">collector</span><span class="o">.</span><span class="n">hist_dict</span><span class="p">,</span> <span class="n">num_examples</span>


 <span class="k">def</span> <span class="nf">_smooth_distribution</span><span class="p">(</span><span class="n">p</span><span class="p">,</span> <span class="n">eps</span><span class="o">=</span><span class="mf">0.0001</span><span class="p">):</span>
     <span class="sd">&quot;&quot;&quot;Given a discrete distribution (may have not been normalized to 1),</span>
 <span class="sd">    smooth it by replacing zeros with eps multiplied by a scaling factor and taking the</span>
 <span class="sd">    corresponding amount off the non-zero values.</span>
 <span class="sd">    Ref: http://web.engr.illinois.edu/~hanj/cs412/bk3/KL-divergence.pdf</span>
 <span class="sd">    &quot;&quot;&quot;</span>
     <span class="n">is_zeros</span> <span class="o">=</span> <span class="p">(</span><span class="n">p</span> <span class="o">==</span> <span class="mi">0</span><span class="p">)</span><span class="o">.</span><span class="n">astype</span><span class="p">(</span><span class="n">np</span><span class="o">.</span><span class="n">float32</span><span class="p">)</span>
     <span class="n">is_nonzeros</span> <span class="o">=</span> <span class="p">(</span><span class="n">p</span> <span class="o">!=</span> <span class="mi">0</span><span class="p">)</span><span class="o">.</span><span class="n">astype</span><span class="p">(</span><span class="n">np</span><span class="o">.</span><span class="n">float32</span><span class="p">)</span>
     <span class="n">n_zeros</span> <span class="o">=</span> <span class="n">is_zeros</span><span class="o">.</span><span class="n">sum</span><span class="p">()</span>
     <span class="n">n_nonzeros</span> <span class="o">=</span> <span class="n">p</span><span class="o">.</span><span class="n">size</span> <span class="o">-</span> <span class="n">n_zeros</span>
     <span class="k">if</span> <span class="ow">not</span> <span class="n">n_nonzeros</span><span class="p">:</span>
         <span class="k">raise</span> <span class="ne">ValueError</span><span class="p">(</span><span class="s1">&#39;The discrete probability distribution is malformed. All entries are 0.&#39;</span><span class="p">)</span>
     <span class="n">eps1</span> <span class="o">=</span> <span class="n">eps</span> <span class="o">*</span> <span class="nb">float</span><span class="p">(</span><span class="n">n_zeros</span><span class="p">)</span> <span class="o">/</span> <span class="nb">float</span><span class="p">(</span><span class="n">n_nonzeros</span><span class="p">)</span>
     <span class="k">assert</span> <span class="n">eps1</span> <span class="o">&lt;</span> <span class="mf">1.0</span><span class="p">,</span> <span class="s1">&#39;n_zeros=</span><span class="si">%d</span><span class="s1">, n_nonzeros=</span><span class="si">%d</span><span class="s1">, eps1=</span><span class="si">%f</span><span class="s1">&#39;</span> <span class="o">%</span> <span class="p">(</span><span class="n">n_zeros</span><span class="p">,</span> <span class="n">n_nonzeros</span><span class="p">,</span> <span class="n">eps1</span><span class="p">)</span>
     <span class="n">hist</span> <span class="o">=</span> <span class="n">p</span><span class="o">.</span><span class="n">astype</span><span class="p">(</span><span class="n">np</span><span class="o">.</span><span class="n">float32</span><span class="p">)</span>
     <span class="n">hist</span> <span class="o">+=</span> <span class="n">eps</span> <span class="o">*</span> <span class="n">is_zeros</span> <span class="o">+</span> <span class="p">(</span><span class="o">-</span><span class="n">eps1</span><span class="p">)</span> <span class="o">*</span> <span class="n">is_nonzeros</span>
     <span class="k">assert</span> <span class="p">(</span><span class="n">hist</span> <span class="o">&lt;=</span> <span class="mi">0</span><span class="p">)</span><span class="o">.</span><span class="n">sum</span><span class="p">()</span> <span class="o">==</span> <span class="mi">0</span>
     <span class="k">return</span> <span class="n">hist</span>


 <span class="c1"># pylint: disable=line-too-long</span>
 <span class="k">def</span> <span class="nf">_get_optimal_threshold</span><span class="p">(</span><span class="n">hist_data</span><span class="p">,</span> <span class="n">quantized_dtype</span><span class="p">,</span> <span class="n">num_quantized_bins</span><span class="o">=</span><span class="mi">255</span><span class="p">):</span>
     <span class="sd">&quot;&quot;&quot;Given a dataset, find the optimal threshold for quantizing it.</span>
 <span class="sd">    The reference distribution is `q`, and the candidate distribution is `p`.</span>
 <span class="sd">    `q` is a truncated version of the original distribution.</span>

 <span class="sd">    Ref: http://on-demand.gputechconf.com/gtc/2017/presentation/s7310-8-bit-inference-with-tensorrt.pdf</span>
 <span class="sd">    &quot;&quot;&quot;</span>
     <span class="p">(</span><span class="n">hist</span><span class="p">,</span> <span class="n">hist_edges</span><span class="p">,</span> <span class="n">min_val</span><span class="p">,</span> <span class="n">max_val</span><span class="p">,</span> <span class="n">_</span><span class="p">)</span> <span class="o">=</span> <span class="n">hist_data</span>
     <span class="n">num_bins</span> <span class="o">=</span> <span class="nb">len</span><span class="p">(</span><span class="n">hist</span><span class="p">)</span>
     <span class="k">assert</span> <span class="p">(</span><span class="n">num_bins</span> <span class="o">%</span> <span class="mi">2</span> <span class="o">==</span> <span class="mi">1</span><span class="p">)</span>
     <span class="k">if</span> <span class="n">min_val</span> <span class="o">&gt;=</span> <span class="mi">0</span> <span class="ow">and</span> <span class="n">quantized_dtype</span> <span class="ow">in</span> <span class="p">[</span><span class="s1">&#39;auto&#39;</span><span class="p">,</span> <span class="s1">&#39;uint8&#39;</span><span class="p">]:</span>
         <span class="c1"># We need to move negative bins to positive bins to fit uint8 range.</span>
         <span class="n">num_quantized_bins</span> <span class="o">=</span> <span class="n">num_quantized_bins</span> <span class="o">*</span> <span class="mi">2</span> <span class="o">+</span> <span class="mi">1</span>
     <span class="n">hist</span> <span class="o">=</span> <span class="n">ndarray</span><span class="o">.</span><span class="n">array</span><span class="p">(</span><span class="n">hist</span><span class="p">,</span> <span class="n">ctx</span><span class="o">=</span><span class="n">cpu</span><span class="p">())</span>
     <span class="n">hist_edges</span> <span class="o">=</span> <span class="n">ndarray</span><span class="o">.</span><span class="n">array</span><span class="p">(</span><span class="n">hist_edges</span><span class="p">,</span> <span class="n">ctx</span><span class="o">=</span><span class="n">cpu</span><span class="p">())</span>
     <span class="n">threshold</span><span class="p">,</span> <span class="n">divergence</span> <span class="o">=</span> <span class="n">ndarray</span><span class="o">.</span><span class="n">contrib</span><span class="o">.</span><span class="n">calibrate_entropy</span><span class="p">(</span><span class="n">hist</span><span class="o">=</span><span class="n">hist</span><span class="p">,</span>
                                                               <span class="n">hist_edges</span><span class="o">=</span><span class="n">hist_edges</span><span class="p">,</span>
                                                               <span class="n">num_quantized_bins</span><span class="o">=</span><span class="n">num_quantized_bins</span><span class="p">)</span>
     <span class="n">threshold</span> <span class="o">=</span> <span class="n">threshold</span><span class="o">.</span><span class="n">asnumpy</span><span class="p">()</span>
     <span class="n">divergence</span> <span class="o">=</span> <span class="n">divergence</span><span class="o">.</span><span class="n">asnumpy</span><span class="p">()</span>
     <span class="k">return</span> <span class="n">min_val</span><span class="p">,</span> <span class="n">max_val</span><span class="p">,</span> <span class="n">threshold</span><span class="p">,</span> <span class="n">divergence</span>
 <span class="c1"># pylint: enable=line-too-long</span>

 <span class="k">def</span> <span class="nf">_get_optimal_thresholds</span><span class="p">(</span><span class="n">hist_dict</span><span class="p">,</span> <span class="n">quantized_dtype</span><span class="p">,</span> <span class="n">num_quantized_bins</span><span class="o">=</span><span class="mi">255</span><span class="p">,</span> <span class="n">logger</span><span class="o">=</span><span class="kc">None</span><span class="p">):</span>
     <span class="sd">&quot;&quot;&quot;Given a ndarray dict, find the optimal threshold for quantizing each value of the key.&quot;&quot;&quot;</span>
     <span class="k">if</span> <span class="n">stats</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
         <span class="k">raise</span> <span class="ne">ImportError</span><span class="p">(</span><span class="s1">&#39;scipy.stats is required for running entropy mode of calculating&#39;</span>
                           <span class="s1">&#39; the optimal thresholds for quantizing FP32 ndarrays into int8.&#39;</span>
                           <span class="s1">&#39; Please check if the scipy python bindings are installed.&#39;</span><span class="p">)</span>
     <span class="k">assert</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">hist_dict</span><span class="p">,</span> <span class="nb">dict</span><span class="p">)</span>
     <span class="k">if</span> <span class="n">logger</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
         <span class="n">logger</span><span class="o">.</span><span class="n">info</span><span class="p">(</span><span class="s1">&#39;Calculating optimal thresholds for quantization using KL divergence&#39;</span>
                     <span class="s1">&#39; with num_quantized_bins=</span><span class="si">%d</span><span class="s1">&#39;</span> <span class="o">%</span> <span class="n">num_quantized_bins</span><span class="p">)</span>
     <span class="n">th_dict</span> <span class="o">=</span> <span class="p">{}</span>
     <span class="c1"># copy hist_dict keys since the keys() only returns a view in python3</span>
     <span class="n">layer_names</span> <span class="o">=</span> <span class="nb">list</span><span class="p">(</span><span class="n">hist_dict</span><span class="o">.</span><span class="n">keys</span><span class="p">())</span>
     <span class="k">for</span> <span class="n">name</span> <span class="ow">in</span> <span class="n">layer_names</span><span class="p">:</span>
         <span class="k">assert</span> <span class="n">name</span> <span class="ow">in</span> <span class="n">hist_dict</span>
         <span class="n">min_val</span><span class="p">,</span> <span class="n">max_val</span><span class="p">,</span> <span class="n">th</span><span class="p">,</span> <span class="n">divergence</span> <span class="o">=</span> \
             <span class="n">_get_optimal_threshold</span><span class="p">(</span><span class="n">hist_dict</span><span class="p">[</span><span class="n">name</span><span class="p">],</span> <span class="n">quantized_dtype</span><span class="p">,</span>
                                    <span class="n">num_quantized_bins</span><span class="o">=</span><span class="n">num_quantized_bins</span><span class="p">)</span>
         <span class="k">if</span> <span class="n">min_val</span> <span class="o">&gt;=</span> <span class="mi">0</span> <span class="ow">and</span> <span class="n">quantized_dtype</span> <span class="ow">in</span> <span class="p">[</span><span class="s1">&#39;auto&#39;</span><span class="p">,</span> <span class="s1">&#39;uint8&#39;</span><span class="p">]:</span>
             <span class="n">th_dict</span><span class="p">[</span><span class="n">name</span><span class="p">]</span> <span class="o">=</span> <span class="p">(</span><span class="mi">0</span><span class="p">,</span> <span class="n">th</span><span class="p">)</span>
         <span class="k">else</span><span class="p">:</span>
             <span class="n">th_dict</span><span class="p">[</span><span class="n">name</span><span class="p">]</span> <span class="o">=</span> <span class="p">(</span><span class="o">-</span><span class="n">th</span><span class="p">,</span> <span class="n">th</span><span class="p">)</span>
         <span class="k">del</span> <span class="n">hist_dict</span><span class="p">[</span><span class="n">name</span><span class="p">]</span>  <span class="c1"># release the memory</span>
         <span class="k">if</span> <span class="n">logger</span><span class="p">:</span>
             <span class="n">logger</span><span class="o">.</span><span class="n">debug</span><span class="p">(</span><span class="s1">&#39;layer=</span><span class="si">%s</span><span class="s1">, min_val=</span><span class="si">%f</span><span class="s1">, max_val=</span><span class="si">%f</span><span class="s1">, th=</span><span class="si">%f</span><span class="s1">, divergence=</span><span class="si">%f</span><span class="s1">&#39;</span>
                          <span class="o">%</span> <span class="p">(</span><span class="n">name</span><span class="p">,</span> <span class="n">min_val</span><span class="p">,</span> <span class="n">max_val</span><span class="p">,</span> <span class="n">th</span><span class="p">,</span> <span class="n">divergence</span><span class="p">))</span>
     <span class="k">return</span> <span class="n">th_dict</span>


 <span class="k">def</span> <span class="nf">_load_sym</span><span class="p">(</span><span class="n">sym</span><span class="p">,</span> <span class="n">logger</span><span class="o">=</span><span class="kc">None</span><span class="p">):</span>
     <span class="sd">&quot;&quot;&quot;Given a str as a path the symbol .json file or a symbol, returns a Symbol object.&quot;&quot;&quot;</span>
     <span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">sym</span><span class="p">,</span> <span class="nb">str</span><span class="p">):</span>  <span class="c1"># sym is a symbol file path</span>
         <span class="n">cur_path</span> <span class="o">=</span> <span class="n">os</span><span class="o">.</span><span class="n">path</span><span class="o">.</span><span class="n">dirname</span><span class="p">(</span><span class="n">os</span><span class="o">.</span><span class="n">path</span><span class="o">.</span><span class="n">realpath</span><span class="p">(</span><span class="vm">__file__</span><span class="p">))</span>
         <span class="n">symbol_file_path</span> <span class="o">=</span> <span class="n">os</span><span class="o">.</span><span class="n">path</span><span class="o">.</span><span class="n">join</span><span class="p">(</span><span class="n">cur_path</span><span class="p">,</span> <span class="n">sym</span><span class="p">)</span>
         <span class="k">if</span> <span class="n">logger</span><span class="p">:</span>
             <span class="n">logger</span><span class="o">.</span><span class="n">info</span><span class="p">(</span><span class="s1">&#39;Loading symbol from file </span><span class="si">%s</span><span class="s1">&#39;</span> <span class="o">%</span> <span class="n">symbol_file_path</span><span class="p">)</span>
         <span class="k">return</span> <span class="n">sym_load</span><span class="p">(</span><span class="n">symbol_file_path</span><span class="p">)</span>
     <span class="k">elif</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">sym</span><span class="p">,</span> <span class="n">Symbol</span><span class="p">):</span>
         <span class="k">return</span> <span class="n">sym</span>
     <span class="k">else</span><span class="p">:</span>
         <span class="k">raise</span> <span class="ne">ValueError</span><span class="p">(</span><span class="s1">&#39;_load_sym only accepts Symbol or path to the symbol file,&#39;</span>
                          <span class="s1">&#39; while received type </span><span class="si">%s</span><span class="s1">&#39;</span> <span class="o">%</span> <span class="nb">str</span><span class="p">(</span><span class="nb">type</span><span class="p">(</span><span class="n">sym</span><span class="p">)))</span>


 <span class="k">def</span> <span class="nf">_load_params</span><span class="p">(</span><span class="n">params</span><span class="p">,</span> <span class="n">logger</span><span class="o">=</span><span class="kc">None</span><span class="p">):</span>
     <span class="sd">&quot;&quot;&quot;Given a str as a path to the .params file or a pair of params,</span>
 <span class="sd">    returns two dictionaries representing arg_params and aux_params.</span>
 <span class="sd">    &quot;&quot;&quot;</span>
     <span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">params</span><span class="p">,</span> <span class="nb">str</span><span class="p">):</span>
         <span class="n">cur_path</span> <span class="o">=</span> <span class="n">os</span><span class="o">.</span><span class="n">path</span><span class="o">.</span><span class="n">dirname</span><span class="p">(</span><span class="n">os</span><span class="o">.</span><span class="n">path</span><span class="o">.</span><span class="n">realpath</span><span class="p">(</span><span class="vm">__file__</span><span class="p">))</span>
         <span class="n">param_file_path</span> <span class="o">=</span> <span class="n">os</span><span class="o">.</span><span class="n">path</span><span class="o">.</span><span class="n">join</span><span class="p">(</span><span class="n">cur_path</span><span class="p">,</span> <span class="n">params</span><span class="p">)</span>
         <span class="k">if</span> <span class="n">logger</span><span class="p">:</span>
             <span class="n">logger</span><span class="o">.</span><span class="n">info</span><span class="p">(</span><span class="s1">&#39;Loading params from file </span><span class="si">%s</span><span class="s1">&#39;</span> <span class="o">%</span> <span class="n">param_file_path</span><span class="p">)</span>
         <span class="n">save_dict</span> <span class="o">=</span> <span class="n">nd_load</span><span class="p">(</span><span class="n">param_file_path</span><span class="p">)</span>
         <span class="n">arg_params</span> <span class="o">=</span> <span class="p">{}</span>
         <span class="n">aux_params</span> <span class="o">=</span> <span class="p">{}</span>
         <span class="k">for</span> <span class="n">k</span><span class="p">,</span> <span class="n">v</span> <span class="ow">in</span> <span class="n">save_dict</span><span class="o">.</span><span class="n">items</span><span class="p">():</span>
             <span class="n">tp</span><span class="p">,</span> <span class="n">name</span> <span class="o">=</span> <span class="n">k</span><span class="o">.</span><span class="n">split</span><span class="p">(</span><span class="s1">&#39;:&#39;</span><span class="p">,</span> <span class="mi">1</span><span class="p">)</span>
             <span class="k">if</span> <span class="n">tp</span> <span class="o">==</span> <span class="s1">&#39;arg&#39;</span><span class="p">:</span>
                 <span class="n">arg_params</span><span class="p">[</span><span class="n">name</span><span class="p">]</span> <span class="o">=</span> <span class="n">v</span>
             <span class="k">if</span> <span class="n">tp</span> <span class="o">==</span> <span class="s1">&#39;aux&#39;</span><span class="p">:</span>
                 <span class="n">aux_params</span><span class="p">[</span><span class="n">name</span><span class="p">]</span> <span class="o">=</span> <span class="n">v</span>
         <span class="k">return</span> <span class="n">arg_params</span><span class="p">,</span> <span class="n">aux_params</span>
     <span class="k">elif</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">params</span><span class="p">,</span> <span class="p">(</span><span class="nb">tuple</span><span class="p">,</span> <span class="nb">list</span><span class="p">))</span> <span class="ow">and</span> <span class="nb">len</span><span class="p">(</span><span class="n">params</span><span class="p">)</span> <span class="o">==</span> <span class="mi">2</span><span class="p">:</span>
         <span class="k">return</span> <span class="n">params</span><span class="p">[</span><span class="mi">0</span><span class="p">],</span> <span class="n">params</span><span class="p">[</span><span class="mi">1</span><span class="p">]</span>
     <span class="k">else</span><span class="p">:</span>
         <span class="k">raise</span> <span class="ne">ValueError</span><span class="p">(</span><span class="s1">&#39;Unsupported params provided. Must be either a path to the param file or&#39;</span>
                          <span class="s1">&#39; a pair of dictionaries representing arg_params and aux_params&#39;</span><span class="p">)</span>

 <span class="c1"># pylint: disable=super-init-not-called</span>
 <span class="k">class</span> <span class="nc">_DataIterWrapper</span><span class="p">(</span><span class="n">DataIter</span><span class="p">):</span>
     <span class="sd">&quot;&quot;&quot;DataIter wrapper for general iterator, e.g., gluon dataloader&quot;&quot;&quot;</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">calib_data</span><span class="p">):</span>
         <span class="bp">self</span><span class="o">.</span><span class="n">_data</span> <span class="o">=</span> <span class="n">calib_data</span>
         <span class="k">try</span><span class="p">:</span>
             <span class="n">calib_iter</span> <span class="o">=</span> <span class="nb">iter</span><span class="p">(</span><span class="n">calib_data</span><span class="p">)</span>
         <span class="k">except</span> <span class="ne">TypeError</span> <span class="k">as</span> <span class="n">e</span><span class="p">:</span>
             <span class="k">raise</span> <span class="ne">TypeError</span><span class="p">(</span><span class="s1">&#39;calib_data is not a valid iterator. </span><span class="si">{}</span><span class="s1">&#39;</span><span class="o">.</span><span class="n">format</span><span class="p">(</span><span class="nb">str</span><span class="p">(</span><span class="n">e</span><span class="p">)))</span>
         <span class="n">data_example</span> <span class="o">=</span> <span class="nb">next</span><span class="p">(</span><span class="n">calib_iter</span><span class="p">)</span>
         <span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">data_example</span><span class="p">,</span> <span class="p">(</span><span class="nb">list</span><span class="p">,</span> <span class="nb">tuple</span><span class="p">)):</span>
             <span class="n">data_example</span> <span class="o">=</span> <span class="nb">list</span><span class="p">(</span><span class="n">data_example</span><span class="p">)</span>
         <span class="k">else</span><span class="p">:</span>
             <span class="n">data_example</span> <span class="o">=</span> <span class="p">[</span><span class="n">data_example</span><span class="p">]</span>
         <span class="c1"># suppose there must be one label in data_example</span>
         <span class="c1"># TODO(xinyu-intel): little tricky here, need to refactor.</span>
         <span class="n">num_data</span> <span class="o">=</span> <span class="nb">len</span><span class="p">(</span><span class="n">data_example</span><span class="p">)</span>
         <span class="k">assert</span> <span class="n">num_data</span> <span class="o">&gt;</span> <span class="mi">0</span>
         <span class="c1"># here reshape is to handle the 5D/6D input data</span>
         <span class="k">if</span> <span class="nb">len</span><span class="p">(</span><span class="n">data_example</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span><span class="o">.</span><span class="n">shape</span><span class="p">)</span> <span class="o">&gt;</span> <span class="mi">4</span><span class="p">:</span>
             <span class="n">data_example</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span> <span class="o">=</span> <span class="n">data_example</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span><span class="o">.</span><span class="n">reshape</span><span class="p">((</span><span class="o">-</span><span class="mi">1</span><span class="p">,)</span> <span class="o">+</span> <span class="n">data_example</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span><span class="o">.</span><span class="n">shape</span><span class="p">[</span><span class="mi">2</span><span class="p">:])</span>
         <span class="bp">self</span><span class="o">.</span><span class="n">provide_data</span> <span class="o">=</span> <span class="p">[</span><span class="n">DataDesc</span><span class="p">(</span><span class="n">name</span><span class="o">=</span><span class="s1">&#39;data&#39;</span><span class="p">,</span> <span class="n">shape</span><span class="o">=</span><span class="p">(</span><span class="n">data_example</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span><span class="o">.</span><span class="n">shape</span><span class="p">))]</span>
         <span class="bp">self</span><span class="o">.</span><span class="n">provide_data</span> <span class="o">+=</span> <span class="p">[</span><span class="n">DataDesc</span><span class="p">(</span><span class="n">name</span><span class="o">=</span><span class="s1">&#39;data</span><span class="si">{}</span><span class="s1">&#39;</span><span class="o">.</span><span class="n">format</span><span class="p">(</span><span class="n">i</span><span class="p">),</span> <span class="n">shape</span><span class="o">=</span><span class="n">x</span><span class="o">.</span><span class="n">shape</span><span class="p">)</span> <span class="k">for</span> <span class="n">i</span><span class="p">,</span> <span class="n">x</span> <span class="ow">in</span> <span class="nb">enumerate</span><span class="p">(</span><span class="n">data_example</span><span class="p">[</span><span class="mi">1</span><span class="p">:])]</span>
         <span class="c1"># data0, data1, ..., label</span>
         <span class="k">if</span> <span class="n">num_data</span> <span class="o">&gt;=</span> <span class="mi">3</span><span class="p">:</span>
             <span class="bp">self</span><span class="o">.</span><span class="n">provide_data</span> <span class="o">=</span> <span class="p">[</span><span class="n">DataDesc</span><span class="p">(</span><span class="n">name</span><span class="o">=</span><span class="s1">&#39;data</span><span class="si">{}</span><span class="s1">&#39;</span><span class="o">.</span><span class="n">format</span><span class="p">(</span><span class="n">i</span><span class="p">),</span> <span class="n">shape</span><span class="o">=</span><span class="n">x</span><span class="o">.</span><span class="n">shape</span><span class="p">)</span>
                                  <span class="k">for</span> <span class="n">i</span><span class="p">,</span> <span class="n">x</span> <span class="ow">in</span> <span class="nb">enumerate</span><span class="p">(</span><span class="n">data_example</span><span class="p">[</span><span class="mi">0</span><span class="p">:])]</span>
         <span class="bp">self</span><span class="o">.</span><span class="n">batch_size</span> <span class="o">=</span> <span class="n">data_example</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span><span class="o">.</span><span class="n">shape</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span>
         <span class="bp">self</span><span class="o">.</span><span class="n">reset</span><span class="p">()</span>

     <span class="k">def</span> <span class="nf">reset</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
         <span class="bp">self</span><span class="o">.</span><span class="n">_iter</span> <span class="o">=</span> <span class="nb">iter</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">_data</span><span class="p">)</span>

     <span class="k">def</span> <span class="nf">next</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
         <span class="n">next_data</span> <span class="o">=</span> <span class="nb">next</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">_iter</span><span class="p">)</span>
         <span class="c1"># here reshape is to handle the 5D/6D input data</span>
         <span class="k">if</span> <span class="nb">len</span><span class="p">(</span><span class="n">next_data</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span><span class="o">.</span><span class="n">shape</span><span class="p">)</span> <span class="o">&gt;</span> <span class="mi">4</span><span class="p">:</span>
             <span class="n">next_data</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span> <span class="o">=</span> <span class="n">next_data</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span><span class="o">.</span><span class="n">reshape</span><span class="p">((</span><span class="o">-</span><span class="mi">1</span><span class="p">,)</span> <span class="o">+</span> <span class="n">next_data</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span><span class="o">.</span><span class="n">shape</span><span class="p">[</span><span class="mi">2</span><span class="p">:])</span>
         <span class="k">return</span> <span class="n">DataBatch</span><span class="p">(</span><span class="n">data</span><span class="o">=</span><span class="n">next_data</span><span class="p">)</span>
 <span class="c1"># pylint: enable=super-init-not-called</span>

 <span class="k">def</span> <span class="nf">_as_data_iter</span><span class="p">(</span><span class="n">calib_data</span><span class="p">):</span>
     <span class="sd">&quot;&quot;&quot;Convert normal iterator to mx.io.DataIter while parsing the data_shapes&quot;&quot;&quot;</span>
     <span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">calib_data</span><span class="p">,</span> <span class="n">DataIter</span><span class="p">):</span>
         <span class="c1"># already validated DataIter, just return</span>
         <span class="k">return</span> <span class="n">calib_data</span><span class="p">,</span> <span class="n">calib_data</span><span class="o">.</span><span class="n">provide_data</span>

     <span class="n">calib_data</span> <span class="o">=</span> <span class="n">_DataIterWrapper</span><span class="p">(</span><span class="n">calib_data</span><span class="p">)</span>
     <span class="k">return</span> <span class="n">calib_data</span><span class="p">,</span> <span class="n">calib_data</span><span class="o">.</span><span class="n">provide_data</span>

 <div class="viewcode-block" id="quantize_model"><a class="viewcode-back" href="../../../api/contrib/quantization/index.html#mxnet.contrib.quantization.quantize_model">[docs]</a><span class="k">def</span> <span class="nf">quantize_model</span><span class="p">(</span><span class="n">sym</span><span class="p">,</span> <span class="n">arg_params</span><span class="p">,</span> <span class="n">aux_params</span><span class="p">,</span>
                    <span class="n">data_names</span><span class="o">=</span><span class="p">(</span><span class="s1">&#39;data&#39;</span><span class="p">,),</span> <span class="n">label_names</span><span class="o">=</span><span class="p">(</span><span class="s1">&#39;softmax_label&#39;</span><span class="p">,),</span>
                    <span class="n">ctx</span><span class="o">=</span><span class="n">cpu</span><span class="p">(),</span> <span class="n">excluded_sym_names</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span> <span class="n">excluded_op_names</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span> <span class="n">calib_mode</span><span class="o">=</span><span class="s1">&#39;entropy&#39;</span><span class="p">,</span>
                    <span class="n">calib_data</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span> <span class="n">num_calib_examples</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
                    <span class="n">quantized_dtype</span><span class="o">=</span><span class="s1">&#39;int8&#39;</span><span class="p">,</span> <span class="n">quantize_mode</span><span class="o">=</span><span class="s1">&#39;smart&#39;</span><span class="p">,</span>
                    <span class="n">quantize_granularity</span><span class="o">=</span><span class="s1">&#39;tensor-wise&#39;</span><span class="p">,</span> <span class="n">logger</span><span class="o">=</span><span class="kc">None</span><span class="p">):</span>
     <span class="sd">&quot;&quot;&quot;User-level API for generating a quantized model from a FP32 model w/ or w/o calibration.</span>
 <span class="sd">    The backend quantized operators are only enabled for Linux systems. Please do not run</span>
 <span class="sd">    inference using the quantized models on Windows for now.</span>
 <span class="sd">    The quantization implementation adopts the TensorFlow&#39;s approach:</span>
 <span class="sd">    https://www.tensorflow.org/performance/quantization.</span>
 <span class="sd">    The calibration implementation borrows the idea of Nvidia&#39;s 8-bit Inference with TensorRT:</span>
 <span class="sd">    http://on-demand.gputechconf.com/gtc/2017/presentation/s7310-8-bit-inference-with-tensorrt.pdf</span>
 <span class="sd">    and adapts the method to MXNet.</span>

 <span class="sd">    Parameters</span>
 <span class="sd">    ----------</span>
 <span class="sd">    sym : str or Symbol</span>
 <span class="sd">        Defines the structure of a neural network for FP32 data types.</span>
 <span class="sd">    arg_params : dict</span>
 <span class="sd">        Dictionary of name to `NDArray`.</span>
 <span class="sd">    aux_params : dict</span>
 <span class="sd">        Dictionary of name to `NDArray`.</span>
 <span class="sd">    data_names : a list of strs</span>
 <span class="sd">        Data names required for creating a Module object to run forward propagation on the</span>
 <span class="sd">        calibration dataset.</span>
 <span class="sd">    label_names : a list of strs</span>
 <span class="sd">        Label names required for creating a Module object to run forward propagation on the</span>
 <span class="sd">        calibration dataset.</span>
 <span class="sd">    ctx : Context</span>
 <span class="sd">        Defines the device that users want to run forward propagation on the calibration</span>
 <span class="sd">        dataset for collecting layer output statistics. Currently, only supports single context.</span>
 <span class="sd">    excluded_sym_names : list of strings</span>
 <span class="sd">        A list of strings representing the names of the symbols that users want to excluding</span>
 <span class="sd">        from being quantized.</span>
 <span class="sd">    excluded_op_names : list of strings</span>
 <span class="sd">        A list of strings representing the names of the operators that users want to excluding</span>
 <span class="sd">        from being quantized.</span>
 <span class="sd">    calib_mode : str</span>
 <span class="sd">        If calib_mode=&#39;none&#39;, no calibration will be used and the thresholds for</span>
 <span class="sd">        requantization after the corresponding layers will be calculated at runtime by</span>
 <span class="sd">        calling min and max operators. The quantized models generated in this</span>
 <span class="sd">        mode are normally 10-20% slower than those with calibrations during inference.</span>
 <span class="sd">        If calib_mode=&#39;naive&#39;, the min and max values of the layer outputs from a calibration</span>
 <span class="sd">        dataset will be directly taken as the thresholds for quantization.</span>
 <span class="sd">        If calib_mode=&#39;entropy&#39; (default mode), the thresholds for quantization will be</span>
 <span class="sd">        derived such that the KL divergence between the distributions of FP32 layer outputs and</span>
 <span class="sd">        quantized layer outputs is minimized based upon the calibration dataset.</span>
 <span class="sd">    calib_data : DataIter</span>
 <span class="sd">        A data iterator initialized by the calibration dataset.</span>
 <span class="sd">    num_calib_examples : int or None</span>
 <span class="sd">        The maximum number of examples that user would like to use for calibration. If not provided,</span>
 <span class="sd">        the whole calibration dataset will be used.</span>
 <span class="sd">    quantized_dtype : str</span>
 <span class="sd">        The quantized destination type for input data. Currently support &#39;int8&#39;, &#39;uint8&#39; and &#39;auto&#39;.</span>
 <span class="sd">        &#39;auto&#39; means automatically select output type according to calibration result.</span>
 <span class="sd">        Default value is &#39;int8&#39;.</span>
 <span class="sd">    quantize_mode : str</span>
 <span class="sd">        The mode that quantization pass to apply. Support &#39;full&#39; and &#39;smart&#39;.</span>
 <span class="sd">        &#39;full&#39; means quantize all operator if possible.</span>
 <span class="sd">        &#39;smart&#39; means quantization pass will smartly choice which operator should be quantized.</span>
 <span class="sd">    quantize_granularity: str</span>
 <span class="sd">        The granularity of quantization, currently supports &#39;tensor-wise&#39; and &#39;channel-wise&#39;</span>
 <span class="sd">        quantization. The default value is &#39;tensor-wise&#39;.</span>
 <span class="sd">    logger : Object</span>
 <span class="sd">        A logging object for printing information during the process of quantization.</span>

 <span class="sd">    Returns</span>
 <span class="sd">    -------</span>
 <span class="sd">    tuple</span>
 <span class="sd">        A tuple of quantized symbol, quantized arg_params, and aux_params.</span>
 <span class="sd">    -------</span>
 <span class="sd">    &quot;&quot;&quot;</span>
     <span class="k">if</span> <span class="n">excluded_sym_names</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
         <span class="n">excluded_sym_names</span> <span class="o">=</span> <span class="p">[]</span>
     <span class="k">if</span> <span class="ow">not</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">excluded_sym_names</span><span class="p">,</span> <span class="nb">list</span><span class="p">):</span>
         <span class="k">raise</span> <span class="ne">ValueError</span><span class="p">(</span><span class="s1">&#39;excluded_sym_names must be a list of strings representing&#39;</span>
                          <span class="s1">&#39; the names of the symbols that will not be quantized,&#39;</span>
                          <span class="s1">&#39; while received type </span><span class="si">%s</span><span class="s1">&#39;</span> <span class="o">%</span> <span class="nb">str</span><span class="p">(</span><span class="nb">type</span><span class="p">(</span><span class="n">excluded_sym_names</span><span class="p">)))</span>

     <span class="k">if</span> <span class="n">excluded_op_names</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
         <span class="n">excluded_op_names</span> <span class="o">=</span> <span class="p">[]</span>
     <span class="k">if</span> <span class="ow">not</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">excluded_op_names</span><span class="p">,</span> <span class="nb">list</span><span class="p">):</span>
         <span class="k">raise</span> <span class="ne">ValueError</span><span class="p">(</span><span class="s1">&#39;excluded_op_names must be a list of strings representing&#39;</span>
                          <span class="s1">&#39; the names of the operators that will not be quantized,&#39;</span>
                          <span class="s1">&#39; while received type </span><span class="si">%s</span><span class="s1">&#39;</span> <span class="o">%</span> <span class="nb">str</span><span class="p">(</span><span class="nb">type</span><span class="p">(</span><span class="n">excluded_op_names</span><span class="p">)))</span>

     <span class="k">if</span> <span class="n">logger</span><span class="p">:</span>
         <span class="n">os</span><span class="o">.</span><span class="n">environ</span><span class="p">[</span><span class="s1">&#39;MXNET_QUANTIZATION_VERBOSE&#39;</span><span class="p">]</span> <span class="o">=</span> <span class="s1">&#39;1&#39;</span>
         <span class="n">logger</span><span class="o">.</span><span class="n">info</span><span class="p">(</span><span class="s1">&#39;Quantizing symbol&#39;</span><span class="p">)</span>
     <span class="k">if</span> <span class="n">quantized_dtype</span> <span class="ow">not</span> <span class="ow">in</span> <span class="p">(</span><span class="s1">&#39;int8&#39;</span><span class="p">,</span> <span class="s1">&#39;uint8&#39;</span><span class="p">,</span> <span class="s1">&#39;auto&#39;</span><span class="p">):</span>
         <span class="k">raise</span> <span class="ne">ValueError</span><span class="p">(</span><span class="s1">&#39;unknown quantized_dtype </span><span class="si">%s</span><span class="s1"> received,&#39;</span>
                          <span class="s1">&#39; expected `int8`, `uint8` or `auto`&#39;</span> <span class="o">%</span> <span class="n">quantized_dtype</span><span class="p">)</span>
     <span class="k">if</span> <span class="n">quantize_granularity</span> <span class="ow">not</span> <span class="ow">in</span> <span class="p">(</span><span class="s1">&#39;tensor-wise&#39;</span><span class="p">,</span> <span class="s1">&#39;channel-wise&#39;</span><span class="p">):</span>
         <span class="k">raise</span> <span class="ne">ValueError</span><span class="p">(</span><span class="s1">&#39;unkonwn quantize_granularity </span><span class="si">%s</span><span class="s1"> received,&#39;</span>
                          <span class="s1">&#39; expected `tensor-wise` or `channel-wise`.&#39;</span> <span class="o">%</span> <span class="n">quantize_granularity</span><span class="p">)</span>
     <span class="n">qsym</span><span class="p">,</span> <span class="n">calib_layer</span> <span class="o">=</span> <span class="n">_quantize_symbol</span><span class="p">(</span><span class="n">sym</span><span class="p">,</span> <span class="n">ctx</span><span class="p">,</span> <span class="n">excluded_symbols</span><span class="o">=</span><span class="n">excluded_sym_names</span><span class="p">,</span>
                                          <span class="n">excluded_operators</span><span class="o">=</span><span class="n">excluded_op_names</span><span class="p">,</span>
                                          <span class="n">offline_params</span><span class="o">=</span><span class="nb">list</span><span class="p">(</span><span class="n">arg_params</span><span class="o">.</span><span class="n">keys</span><span class="p">()),</span>
                                          <span class="n">quantized_dtype</span><span class="o">=</span><span class="n">quantized_dtype</span><span class="p">,</span>
                                          <span class="n">quantize_mode</span><span class="o">=</span><span class="n">quantize_mode</span><span class="p">,</span>
                                          <span class="n">quantize_granularity</span><span class="o">=</span><span class="n">quantize_granularity</span><span class="p">)</span>
     <span class="n">th_dict</span> <span class="o">=</span> <span class="p">{}</span>
     <span class="k">if</span> <span class="n">calib_mode</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span> <span class="ow">and</span> <span class="n">calib_mode</span> <span class="o">!=</span> <span class="s1">&#39;none&#39;</span><span class="p">:</span>
         <span class="k">if</span> <span class="ow">not</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">ctx</span><span class="p">,</span> <span class="n">Context</span><span class="p">):</span>
             <span class="k">raise</span> <span class="ne">ValueError</span><span class="p">(</span><span class="s1">&#39;currently only supports single ctx, while received </span><span class="si">%s</span><span class="s1">&#39;</span> <span class="o">%</span> <span class="nb">str</span><span class="p">(</span><span class="n">ctx</span><span class="p">))</span>
         <span class="k">if</span> <span class="n">calib_data</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
             <span class="k">raise</span> <span class="ne">ValueError</span><span class="p">(</span><span class="s1">&#39;calib_data must be provided when calib_mode=</span><span class="si">%s</span><span class="s1">&#39;</span> <span class="o">%</span> <span class="n">calib_mode</span><span class="p">)</span>
         <span class="k">if</span> <span class="ow">not</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">calib_data</span><span class="p">,</span> <span class="n">DataIter</span><span class="p">):</span>
             <span class="k">raise</span> <span class="ne">ValueError</span><span class="p">(</span><span class="s1">&#39;calib_data must be of DataIter type when calib_mode=</span><span class="si">%s</span><span class="s1">,&#39;</span>
                              <span class="s1">&#39; while received type </span><span class="si">%s</span><span class="s1">&#39;</span> <span class="o">%</span> <span class="p">(</span><span class="n">calib_mode</span><span class="p">,</span> <span class="nb">str</span><span class="p">(</span><span class="nb">type</span><span class="p">(</span><span class="n">calib_data</span><span class="p">))))</span>

         <span class="n">mod</span> <span class="o">=</span> <span class="n">Module</span><span class="p">(</span><span class="n">symbol</span><span class="o">=</span><span class="n">sym</span><span class="p">,</span> <span class="n">data_names</span><span class="o">=</span><span class="n">data_names</span><span class="p">,</span> <span class="n">label_names</span><span class="o">=</span><span class="n">label_names</span><span class="p">,</span> <span class="n">context</span><span class="o">=</span><span class="n">ctx</span><span class="p">)</span>
         <span class="k">if</span> <span class="nb">len</span><span class="p">(</span><span class="n">calib_data</span><span class="o">.</span><span class="n">provide_label</span><span class="p">)</span> <span class="o">&gt;</span> <span class="mi">0</span><span class="p">:</span>
             <span class="n">mod</span><span class="o">.</span><span class="n">bind</span><span class="p">(</span><span class="n">for_training</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span> <span class="n">data_shapes</span><span class="o">=</span><span class="n">calib_data</span><span class="o">.</span><span class="n">provide_data</span><span class="p">,</span>
                      <span class="n">label_shapes</span><span class="o">=</span><span class="n">calib_data</span><span class="o">.</span><span class="n">provide_label</span><span class="p">)</span>
         <span class="k">else</span><span class="p">:</span>
             <span class="n">mod</span><span class="o">.</span><span class="n">bind</span><span class="p">(</span><span class="n">for_training</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span> <span class="n">data_shapes</span><span class="o">=</span><span class="n">calib_data</span><span class="o">.</span><span class="n">provide_data</span><span class="p">)</span>
         <span class="n">mod</span><span class="o">.</span><span class="n">set_params</span><span class="p">(</span><span class="n">arg_params</span><span class="p">,</span> <span class="n">aux_params</span><span class="p">)</span>
         <span class="k">if</span> <span class="n">calib_mode</span> <span class="o">==</span> <span class="s1">&#39;entropy&#39;</span><span class="p">:</span>
             <span class="n">hist_dict</span><span class="p">,</span> <span class="n">num_examples</span> <span class="o">=</span> <span class="n">_collect_layer_histogram</span><span class="p">(</span><span class="n">mod</span><span class="p">,</span> <span class="n">calib_data</span><span class="p">,</span>
                                                                <span class="n">include_layer</span><span class="o">=</span><span class="n">calib_layer</span><span class="p">,</span>
                                                                <span class="n">max_num_examples</span><span class="o">=</span><span class="n">num_calib_examples</span><span class="p">,</span>
                                                                <span class="n">logger</span><span class="o">=</span><span class="n">logger</span><span class="p">)</span>
             <span class="k">if</span> <span class="n">logger</span><span class="p">:</span>
                 <span class="n">logger</span><span class="o">.</span><span class="n">info</span><span class="p">(</span><span class="s1">&#39;Collected layer outputs from FP32 model using </span><span class="si">%d</span><span class="s1"> examples&#39;</span> <span class="o">%</span> <span class="n">num_examples</span><span class="p">)</span>
                 <span class="n">logger</span><span class="o">.</span><span class="n">info</span><span class="p">(</span><span class="s1">&#39;Calculating optimal thresholds for quantization&#39;</span><span class="p">)</span>
             <span class="n">th_dict</span> <span class="o">=</span> <span class="n">_get_optimal_thresholds</span><span class="p">(</span><span class="n">hist_dict</span><span class="p">,</span> <span class="n">quantized_dtype</span><span class="p">,</span> <span class="n">logger</span><span class="o">=</span><span class="n">logger</span><span class="p">)</span>
         <span class="k">elif</span> <span class="n">calib_mode</span> <span class="o">==</span> <span class="s1">&#39;naive&#39;</span><span class="p">:</span>
             <span class="n">th_dict</span><span class="p">,</span> <span class="n">num_examples</span> <span class="o">=</span> <span class="n">_collect_layer_output_min_max</span><span class="p">(</span>
                 <span class="n">mod</span><span class="p">,</span> <span class="n">calib_data</span><span class="p">,</span> <span class="n">quantized_dtype</span><span class="p">,</span> <span class="n">include_layer</span><span class="o">=</span><span class="n">calib_layer</span><span class="p">,</span> <span class="n">max_num_examples</span><span class="o">=</span><span class="n">num_calib_examples</span><span class="p">,</span>
                 <span class="n">logger</span><span class="o">=</span><span class="n">logger</span><span class="p">)</span>
             <span class="k">if</span> <span class="n">logger</span><span class="p">:</span>
                 <span class="n">logger</span><span class="o">.</span><span class="n">info</span><span class="p">(</span><span class="s1">&#39;Collected layer output min/max values from FP32 model using </span><span class="si">%d</span><span class="s1"> examples&#39;</span>
                             <span class="o">%</span> <span class="n">num_examples</span><span class="p">)</span>
         <span class="k">else</span><span class="p">:</span>
             <span class="k">raise</span> <span class="ne">ValueError</span><span class="p">(</span><span class="s1">&#39;unknown calibration mode </span><span class="si">%s</span><span class="s1"> received,&#39;</span>
                              <span class="s1">&#39; expected `none`, `naive`, or `entropy`&#39;</span> <span class="o">%</span> <span class="n">calib_mode</span><span class="p">)</span>
         <span class="n">qsym</span> <span class="o">=</span> <span class="n">_calibrate_quantized_sym</span><span class="p">(</span><span class="n">qsym</span><span class="p">,</span> <span class="n">th_dict</span><span class="p">)</span>

     <span class="k">if</span> <span class="n">logger</span><span class="p">:</span>
         <span class="n">logger</span><span class="o">.</span><span class="n">info</span><span class="p">(</span><span class="s1">&#39;Quantizing parameters&#39;</span><span class="p">)</span>
     <span class="n">qarg_params</span> <span class="o">=</span> <span class="n">_quantize_params</span><span class="p">(</span><span class="n">qsym</span><span class="p">,</span> <span class="n">arg_params</span><span class="p">,</span> <span class="n">th_dict</span><span class="p">)</span>

     <span class="k">return</span> <span class="n">qsym</span><span class="p">,</span> <span class="n">qarg_params</span><span class="p">,</span> <span class="n">aux_params</span></div>

 <div class="viewcode-block" id="quantize_model_mkldnn"><a class="viewcode-back" href="../../../api/contrib/quantization/index.html#mxnet.contrib.quantization.quantize_model_mkldnn">[docs]</a><span class="k">def</span> <span class="nf">quantize_model_mkldnn</span><span class="p">(</span><span class="n">sym</span><span class="p">,</span> <span class="n">arg_params</span><span class="p">,</span> <span class="n">aux_params</span><span class="p">,</span>
                           <span class="n">data_names</span><span class="o">=</span><span class="p">(</span><span class="s1">&#39;data&#39;</span><span class="p">,),</span> <span class="n">label_names</span><span class="o">=</span><span class="p">(</span><span class="s1">&#39;softmax_label&#39;</span><span class="p">,),</span>
                           <span class="n">ctx</span><span class="o">=</span><span class="n">cpu</span><span class="p">(),</span> <span class="n">excluded_sym_names</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span> <span class="n">excluded_op_names</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
                           <span class="n">calib_mode</span><span class="o">=</span><span class="s1">&#39;entropy&#39;</span><span class="p">,</span> <span class="n">calib_data</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span> <span class="n">num_calib_examples</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
                           <span class="n">quantized_dtype</span><span class="o">=</span><span class="s1">&#39;int8&#39;</span><span class="p">,</span> <span class="n">quantize_mode</span><span class="o">=</span><span class="s1">&#39;smart&#39;</span><span class="p">,</span>
                           <span class="n">quantize_granularity</span><span class="o">=</span><span class="s1">&#39;tensor-wise&#39;</span><span class="p">,</span> <span class="n">logger</span><span class="o">=</span><span class="kc">None</span><span class="p">):</span>
     <span class="sd">&quot;&quot;&quot;User-level API for generating a fusion + quantized model from a FP32 model</span>
 <span class="sd">    w/ or w/o calibration with Intel MKL-DNN.</span>
 <span class="sd">    The backend quantized operators are only enabled for Linux systems. Please do not run</span>
 <span class="sd">    inference using the quantized models on Windows for now.</span>

 <span class="sd">    Parameters</span>
 <span class="sd">    ----------</span>
 <span class="sd">    same with quantize_model</span>

 <span class="sd">    Returns</span>
 <span class="sd">    -------</span>
 <span class="sd">    tuple</span>
 <span class="sd">        A tuple of quantized symbol, quantized arg_params, and aux_params.</span>
 <span class="sd">    -------</span>
 <span class="sd">    &quot;&quot;&quot;</span>
     <span class="k">if</span> <span class="ow">not</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">ctx</span><span class="p">,</span> <span class="n">Context</span><span class="p">):</span>
         <span class="k">raise</span> <span class="ne">ValueError</span><span class="p">(</span><span class="s1">&#39;currently only supports single ctx, while received </span><span class="si">%s</span><span class="s1">&#39;</span> <span class="o">%</span> <span class="nb">str</span><span class="p">(</span><span class="n">ctx</span><span class="p">))</span>
     <span class="k">if</span> <span class="n">ctx</span><span class="o">.</span><span class="n">device_type</span> <span class="o">!=</span> <span class="s1">&#39;cpu&#39;</span><span class="p">:</span>
         <span class="k">raise</span> <span class="ne">ValueError</span><span class="p">(</span>
             <span class="s1">&#39;quantize_model_mkldnn only support Intel cpu platform with MKL-DNN Backend&#39;</span><span class="p">)</span>

     <span class="n">sym</span> <span class="o">=</span> <span class="n">sym</span><span class="o">.</span><span class="n">get_backend_symbol</span><span class="p">(</span><span class="s1">&#39;MKLDNN_QUANTIZE&#39;</span><span class="p">)</span>

     <span class="n">qsym</span><span class="p">,</span> <span class="n">qarg_params</span><span class="p">,</span> <span class="n">aux_params</span> <span class="o">=</span> <span class="n">quantize_model</span><span class="p">(</span><span class="n">sym</span><span class="o">=</span><span class="n">sym</span><span class="p">,</span> <span class="n">arg_params</span><span class="o">=</span><span class="n">arg_params</span><span class="p">,</span> <span class="n">aux_params</span><span class="o">=</span><span class="n">aux_params</span><span class="p">,</span>
                                                    <span class="n">data_names</span><span class="o">=</span><span class="n">data_names</span><span class="p">,</span> <span class="n">label_names</span><span class="o">=</span><span class="n">label_names</span><span class="p">,</span>
                                                    <span class="n">ctx</span><span class="o">=</span><span class="n">ctx</span><span class="p">,</span> <span class="n">excluded_sym_names</span><span class="o">=</span><span class="n">excluded_sym_names</span><span class="p">,</span>
                                                    <span class="n">excluded_op_names</span><span class="o">=</span><span class="n">excluded_op_names</span><span class="p">,</span>
                                                    <span class="n">calib_mode</span><span class="o">=</span><span class="n">calib_mode</span><span class="p">,</span> <span class="n">calib_data</span><span class="o">=</span><span class="n">calib_data</span><span class="p">,</span>
                                                    <span class="n">num_calib_examples</span><span class="o">=</span><span class="n">num_calib_examples</span><span class="p">,</span>
                                                    <span class="n">quantized_dtype</span><span class="o">=</span><span class="n">quantized_dtype</span><span class="p">,</span> <span class="n">quantize_mode</span><span class="o">=</span><span class="n">quantize_mode</span><span class="p">,</span>
                                                    <span class="n">quantize_granularity</span><span class="o">=</span><span class="n">quantize_granularity</span><span class="p">,</span> <span class="n">logger</span><span class="o">=</span><span class="n">logger</span><span class="p">)</span>

     <span class="n">qsym</span> <span class="o">=</span> <span class="n">qsym</span><span class="o">.</span><span class="n">get_backend_symbol</span><span class="p">(</span><span class="s1">&#39;MKLDNN_QUANTIZE&#39;</span><span class="p">)</span>

     <span class="k">return</span> <span class="n">qsym</span><span class="p">,</span> <span class="n">qarg_params</span><span class="p">,</span> <span class="n">aux_params</span></div>

 <div class="viewcode-block" id="quantize_graph"><a class="viewcode-back" href="../../../api/contrib/quantization/index.html#mxnet.contrib.quantization.quantize_graph">[docs]</a><span class="k">def</span> <span class="nf">quantize_graph</span><span class="p">(</span><span class="n">sym</span><span class="p">,</span> <span class="n">arg_params</span><span class="p">,</span> <span class="n">aux_params</span><span class="p">,</span> <span class="n">ctx</span><span class="o">=</span><span class="n">cpu</span><span class="p">(),</span>
                    <span class="n">excluded_sym_names</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span> <span class="n">excluded_op_names</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
                    <span class="n">calib_mode</span><span class="o">=</span><span class="s1">&#39;entropy&#39;</span><span class="p">,</span> <span class="n">quantized_dtype</span><span class="o">=</span><span class="s1">&#39;int8&#39;</span><span class="p">,</span>
                    <span class="n">quantize_mode</span><span class="o">=</span><span class="s1">&#39;full&#39;</span><span class="p">,</span> <span class="n">quantize_granularity</span><span class="o">=</span><span class="s1">&#39;tensor-wise&#39;</span><span class="p">,</span>
                    <span class="n">LayerOutputCollector</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span> <span class="n">logger</span><span class="o">=</span><span class="kc">None</span><span class="p">):</span>
     <span class="sd">&quot;&quot;&quot;User-level API for generating a quantized model from a FP32 model w/o calibration</span>
 <span class="sd">    and a collector for naive or entropy calibration.</span>
 <span class="sd">    The backend quantized operators are only enabled for Linux systems. Please do not run</span>
 <span class="sd">    inference using the quantized models on Windows for now.</span>
 <span class="sd">    Parameters</span>
 <span class="sd">    ----------</span>
 <span class="sd">    sym : str or Symbol</span>
 <span class="sd">        Defines the structure of a neural network for FP32 data types.</span>
 <span class="sd">    ctx : Context</span>
 <span class="sd">        Defines the device that users want to run forward propagation on the calibration</span>
 <span class="sd">        dataset for collecting layer output statistics. Currently, only supports single context.</span>
 <span class="sd">    arg_params : dict</span>
 <span class="sd">        Dictionary of name to `NDArray`.</span>
 <span class="sd">    aux_params : dict</span>
 <span class="sd">        Dictionary of name to `NDArray`.</span>
 <span class="sd">    excluded_sym_names : list of strings</span>
 <span class="sd">        A list of strings representing the names of the symbols that users want to excluding</span>
 <span class="sd">        from being quantized.</span>
 <span class="sd">    excluded_op_names : list of strings</span>
 <span class="sd">        A list of strings representing the names of the operators that users want to excluding</span>
 <span class="sd">    calib_mode : str</span>
 <span class="sd">        If calib_mode=&#39;none&#39;, no calibration will be used and the thresholds for</span>
 <span class="sd">        requantization after the corresponding layers will be calculated at runtime by</span>
 <span class="sd">        calling min and max operators. The quantized models generated in this</span>
 <span class="sd">        mode are normally 10-20% slower than those with calibrations during inference.</span>
 <span class="sd">        If calib_mode=&#39;naive&#39;, the min and max values of the layer outputs from a calibration</span>
 <span class="sd">        dataset will be directly taken as the thresholds for quantization.</span>
 <span class="sd">        If calib_mode=&#39;entropy&#39; (default mode), the thresholds for quantization will be</span>
 <span class="sd">        derived such that the KL divergence between the distributions of FP32 layer outputs and</span>
 <span class="sd">        quantized layer outputs is minimized based upon the calibration dataset.</span>
 <span class="sd">    quantized_dtype : str</span>
 <span class="sd">        The quantized destination type for input data. Currently support &#39;int8&#39;</span>
 <span class="sd">        , &#39;uint8&#39; and &#39;auto&#39;. &#39;auto&#39; means automatically select output type according to calibration result.</span>
 <span class="sd">        Default value is &#39;int8&#39;.</span>
 <span class="sd">    quantize_mode : str</span>
 <span class="sd">        The mode that quantization pass to apply. Support &#39;full&#39; and &#39;smart&#39;.</span>
 <span class="sd">        &#39;full&#39; means quantize all operator if possible.</span>
 <span class="sd">        &#39;smart&#39; means quantization pass will smartly choice which operator should be quantized.</span>
 <span class="sd">    quantize_granularity: str</span>
 <span class="sd">        The granularity of quantization, currently supports &#39;tensor-wise&#39; and &#39;channel-wise&#39;</span>
 <span class="sd">        quantization. The default value is &#39;tensor-wise&#39;.</span>
 <span class="sd">    LayerOutputCollector : class</span>
 <span class="sd">        For customize calibration method usage.</span>
 <span class="sd">    logger : Object</span>
 <span class="sd">        A logging object for printing information during the process of quantization.</span>
 <span class="sd">    Returns</span>
 <span class="sd">    -------</span>
 <span class="sd">    tuple</span>
 <span class="sd">        A tuple of quantized symbol, quantized arg_params, aux_params and collector.</span>
 <span class="sd">    -------</span>
 <span class="sd">    &quot;&quot;&quot;</span>
     <span class="k">if</span> <span class="n">excluded_sym_names</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
         <span class="n">excluded_sym_names</span> <span class="o">=</span> <span class="p">[]</span>
     <span class="k">if</span> <span class="ow">not</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">excluded_sym_names</span><span class="p">,</span> <span class="nb">list</span><span class="p">):</span>
         <span class="k">raise</span> <span class="ne">ValueError</span><span class="p">(</span><span class="s1">&#39;excluded_sym_names must be a list of strings representing&#39;</span>
                          <span class="s1">&#39; the names of the symbols that will not be quantized,&#39;</span>
                          <span class="s1">&#39; while received type </span><span class="si">%s</span><span class="s1">&#39;</span> <span class="o">%</span> <span class="nb">str</span><span class="p">(</span><span class="nb">type</span><span class="p">(</span><span class="n">excluded_sym_names</span><span class="p">)))</span>
     <span class="k">if</span> <span class="ow">not</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">ctx</span><span class="p">,</span> <span class="n">Context</span><span class="p">):</span>
         <span class="k">raise</span> <span class="ne">ValueError</span><span class="p">(</span><span class="s1">&#39;currently only supports single ctx, while received </span><span class="si">%s</span><span class="s1">&#39;</span> <span class="o">%</span> <span class="nb">str</span><span class="p">(</span><span class="n">ctx</span><span class="p">))</span>
     <span class="k">if</span> <span class="n">logger</span><span class="p">:</span>
         <span class="n">os</span><span class="o">.</span><span class="n">environ</span><span class="p">[</span><span class="s1">&#39;MXNET_QUANTIZATION_VERBOSE&#39;</span><span class="p">]</span> <span class="o">=</span> <span class="s1">&#39;1&#39;</span>
         <span class="n">logger</span><span class="o">.</span><span class="n">info</span><span class="p">(</span><span class="s1">&#39;Quantizing graph&#39;</span><span class="p">)</span>
     <span class="k">if</span> <span class="n">quantized_dtype</span> <span class="ow">not</span> <span class="ow">in</span> <span class="p">(</span><span class="s1">&#39;int8&#39;</span><span class="p">,</span> <span class="s1">&#39;uint8&#39;</span><span class="p">,</span> <span class="s1">&#39;auto&#39;</span><span class="p">):</span>
         <span class="k">raise</span> <span class="ne">ValueError</span><span class="p">(</span><span class="s1">&#39;unknown quantized_dtype </span><span class="si">%s</span><span class="s1"> received,&#39;</span>
                          <span class="s1">&#39; expected `int8`, `uint8` or `auto`&#39;</span> <span class="o">%</span> <span class="n">quantized_dtype</span><span class="p">)</span>
     <span class="k">if</span> <span class="n">quantize_granularity</span> <span class="ow">not</span> <span class="ow">in</span> <span class="p">(</span><span class="s1">&#39;tensor-wise&#39;</span><span class="p">,</span> <span class="s1">&#39;channel-wise&#39;</span><span class="p">):</span>
         <span class="k">raise</span> <span class="ne">ValueError</span><span class="p">(</span><span class="s1">&#39;unkonwn quantize_granularity </span><span class="si">%s</span><span class="s1"> received,&#39;</span>
                          <span class="s1">&#39; expected `tensor-wise` or `channel-wise`.&#39;</span> <span class="o">%</span> <span class="n">quantize_granularity</span><span class="p">)</span>
     <span class="n">qsym</span><span class="p">,</span> <span class="n">calib_layer</span> <span class="o">=</span> <span class="n">_quantize_symbol</span><span class="p">(</span><span class="n">sym</span><span class="p">,</span> <span class="n">ctx</span><span class="p">,</span> <span class="n">excluded_symbols</span><span class="o">=</span><span class="n">excluded_sym_names</span><span class="p">,</span>
                                          <span class="n">excluded_operators</span><span class="o">=</span><span class="n">excluded_op_names</span><span class="p">,</span>
                                          <span class="n">offline_params</span><span class="o">=</span><span class="nb">list</span><span class="p">(</span>
                                              <span class="n">arg_params</span><span class="o">.</span><span class="n">keys</span><span class="p">()),</span>
                                          <span class="n">quantized_dtype</span><span class="o">=</span><span class="n">quantized_dtype</span><span class="p">,</span>
                                          <span class="n">quantize_mode</span><span class="o">=</span><span class="n">quantize_mode</span><span class="p">,</span>
                                          <span class="n">quantize_granularity</span><span class="o">=</span><span class="n">quantize_granularity</span><span class="p">)</span>

     <span class="n">th_dict</span> <span class="o">=</span> <span class="p">{}</span>
     <span class="n">collector</span> <span class="o">=</span> <span class="kc">None</span>
     <span class="k">if</span> <span class="n">calib_mode</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span> <span class="ow">and</span> <span class="n">calib_mode</span> <span class="o">!=</span> <span class="s1">&#39;none&#39;</span><span class="p">:</span>
         <span class="k">if</span> <span class="n">calib_mode</span> <span class="o">==</span> <span class="s1">&#39;entropy&#39;</span><span class="p">:</span>
             <span class="n">collector</span> <span class="o">=</span> <span class="n">_LayerHistogramCollector</span><span class="p">(</span>
                 <span class="n">include_layer</span><span class="o">=</span><span class="n">calib_layer</span><span class="p">,</span> <span class="n">logger</span><span class="o">=</span><span class="n">logger</span><span class="p">)</span>
             <span class="k">if</span> <span class="n">logger</span><span class="p">:</span>
                 <span class="n">logger</span><span class="o">.</span><span class="n">info</span><span class="p">(</span>
                     <span class="s1">&#39;Create a layer output collector for entropy calibration.&#39;</span><span class="p">)</span>
         <span class="k">elif</span> <span class="n">calib_mode</span> <span class="o">==</span> <span class="s1">&#39;naive&#39;</span><span class="p">:</span>
             <span class="n">collector</span> <span class="o">=</span> <span class="n">_LayerOutputMinMaxCollector</span><span class="p">(</span><span class="n">quantized_dtype</span><span class="o">=</span><span class="n">quantized_dtype</span><span class="p">,</span>
                                                     <span class="n">include_layer</span><span class="o">=</span><span class="n">calib_layer</span><span class="p">,</span> <span class="n">logger</span><span class="o">=</span><span class="n">logger</span><span class="p">)</span>
             <span class="k">if</span> <span class="n">logger</span><span class="p">:</span>
                 <span class="n">logger</span><span class="o">.</span><span class="n">info</span><span class="p">(</span>
                     <span class="s1">&#39;Create a layer output minmax collector for naive calibration&#39;</span><span class="p">)</span>
         <span class="k">elif</span> <span class="n">calib_mode</span> <span class="o">==</span> <span class="s1">&#39;customize&#39;</span> <span class="ow">and</span> <span class="n">LayerOutputCollector</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
             <span class="n">collector</span> <span class="o">=</span> <span class="n">LayerOutputCollector</span>
             <span class="k">if</span> <span class="n">logger</span><span class="p">:</span>
                 <span class="n">logger</span><span class="o">.</span><span class="n">info</span><span class="p">(</span>
                     <span class="s1">&#39;Create a customize layer output minmax collector for calibration&#39;</span><span class="p">)</span>
         <span class="k">else</span><span class="p">:</span>
             <span class="k">raise</span> <span class="ne">ValueError</span><span class="p">(</span><span class="s1">&#39;unknown calibration mode </span><span class="si">%s</span><span class="s1"> received,&#39;</span>
                              <span class="s1">&#39; expected `none`, `naive`, `entropy` or `customize`&#39;</span> <span class="o">%</span> <span class="n">calib_mode</span><span class="p">)</span>
         <span class="k">if</span> <span class="n">logger</span><span class="p">:</span>
             <span class="n">logger</span><span class="o">.</span><span class="n">info</span><span class="p">(</span><span class="s1">&#39;Collector created, please use set_monitor_callback&#39;</span>
                         <span class="s1">&#39; to collect calibration information.&#39;</span><span class="p">)</span>

     <span class="k">if</span> <span class="n">logger</span><span class="p">:</span>
         <span class="n">logger</span><span class="o">.</span><span class="n">info</span><span class="p">(</span><span class="s1">&#39;Quantizing parameters&#39;</span><span class="p">)</span>
     <span class="n">qarg_params</span> <span class="o">=</span> <span class="n">_quantize_params</span><span class="p">(</span><span class="n">qsym</span><span class="p">,</span> <span class="n">arg_params</span><span class="p">,</span> <span class="n">th_dict</span><span class="p">)</span>

     <span class="k">return</span> <span class="n">qsym</span><span class="p">,</span> <span class="n">qarg_params</span><span class="p">,</span> <span class="n">aux_params</span><span class="p">,</span> <span class="n">collector</span></div>

 <div class="viewcode-block" id="calib_graph"><a class="viewcode-back" href="../../../api/contrib/quantization/index.html#mxnet.contrib.quantization.calib_graph">[docs]</a><span class="k">def</span> <span class="nf">calib_graph</span><span class="p">(</span><span class="n">qsym</span><span class="p">,</span> <span class="n">arg_params</span><span class="p">,</span> <span class="n">aux_params</span><span class="p">,</span> <span class="n">collector</span><span class="p">,</span>
                 <span class="n">calib_mode</span><span class="o">=</span><span class="s1">&#39;entropy&#39;</span><span class="p">,</span> <span class="n">quantized_dtype</span><span class="o">=</span><span class="s1">&#39;int8&#39;</span><span class="p">,</span> <span class="n">logger</span><span class="o">=</span><span class="n">logging</span><span class="p">):</span>
     <span class="sd">&quot;&quot;&quot;User-level API for calibrating a quantized model using a filled collector.</span>
 <span class="sd">    The backend quantized operators are only enabled for Linux systems. Please do not run</span>
 <span class="sd">    inference using the quantized models on Windows for now.</span>
 <span class="sd">    Parameters</span>
 <span class="sd">    ----------</span>
 <span class="sd">    qsym : str or Symbol</span>
 <span class="sd">        Defines the structure of a neural network for INT8 data types.</span>
 <span class="sd">    arg_params : dict</span>
 <span class="sd">        Dictionary of name to `NDArray`.</span>
 <span class="sd">    aux_params : dict</span>
 <span class="sd">        Dictionary of name to `NDArray`.</span>
 <span class="sd">    collector : function</span>
 <span class="sd">        layer collector for naive or entropy calibration.</span>
 <span class="sd">    calib_mode : str</span>
 <span class="sd">        If calib_mode=&#39;none&#39;, no calibration will be used and the thresholds for</span>
 <span class="sd">        requantization after the corresponding layers will be calculated at runtime by</span>
 <span class="sd">        calling min and max operators. The quantized models generated in this</span>
 <span class="sd">        mode are normally 10-20% slower than those with calibrations during inference.</span>
 <span class="sd">        If calib_mode=&#39;naive&#39;, the min and max values of the layer outputs from a calibration</span>
 <span class="sd">        dataset will be directly taken as the thresholds for quantization.</span>
 <span class="sd">        If calib_mode=&#39;entropy&#39; (default mode), the thresholds for quantization will be</span>
 <span class="sd">        derived such that the KL divergence between the distributions of FP32 layer outputs and</span>
 <span class="sd">        quantized layer outputs is minimized based upon the calibration dataset.</span>
 <span class="sd">    quantized_dtype : str</span>
 <span class="sd">        The quantized destination type for input data. Currently support &#39;int8&#39;</span>
 <span class="sd">        , &#39;uint8&#39; and &#39;auto&#39;. &#39;auto&#39; means automatically select output type according to calibration result.</span>
 <span class="sd">        Default value is &#39;int8&#39;.</span>
 <span class="sd">    logger : Object</span>
 <span class="sd">        A logging object for printing information during the process of quantization.</span>
 <span class="sd">    Returns</span>
 <span class="sd">    -------</span>
 <span class="sd">    tuple</span>
 <span class="sd">        A tuple of calibrated symbol, quantized arg_params, aux_params.</span>
 <span class="sd">    -------</span>
 <span class="sd">    &quot;&quot;&quot;</span>
     <span class="n">th_dict</span> <span class="o">=</span> <span class="p">{}</span>
     <span class="k">if</span> <span class="n">calib_mode</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span> <span class="ow">and</span> <span class="n">calib_mode</span> <span class="o">!=</span> <span class="s1">&#39;none&#39;</span><span class="p">:</span>
         <span class="k">if</span> <span class="n">calib_mode</span> <span class="o">==</span> <span class="s1">&#39;entropy&#39;</span><span class="p">:</span>
             <span class="k">if</span> <span class="n">logger</span><span class="p">:</span>
                 <span class="n">logger</span><span class="o">.</span><span class="n">info</span><span class="p">(</span><span class="s1">&#39;Calculating optimal thresholds for quantization&#39;</span><span class="p">)</span>
             <span class="n">th_dict</span> <span class="o">=</span> <span class="n">_get_optimal_thresholds</span><span class="p">(</span>
                 <span class="n">collector</span><span class="o">.</span><span class="n">hist_dict</span><span class="p">,</span> <span class="n">quantized_dtype</span><span class="p">,</span> <span class="n">logger</span><span class="o">=</span><span class="n">logger</span><span class="p">)</span>
         <span class="k">elif</span> <span class="n">calib_mode</span> <span class="o">==</span> <span class="s1">&#39;naive&#39;</span><span class="p">:</span>
             <span class="n">th_dict</span> <span class="o">=</span> <span class="n">collector</span><span class="o">.</span><span class="n">min_max_dict</span>
         <span class="k">elif</span> <span class="n">calib_mode</span> <span class="o">==</span> <span class="s1">&#39;customize&#39;</span><span class="p">:</span>
             <span class="n">th_dict</span> <span class="o">=</span> <span class="n">collector</span><span class="o">.</span><span class="n">min_max_dict</span>
         <span class="k">else</span><span class="p">:</span>
             <span class="k">raise</span> <span class="ne">ValueError</span><span class="p">(</span><span class="s1">&#39;unknown calibration mode </span><span class="si">%s</span><span class="s1"> received,&#39;</span>
                              <span class="s1">&#39; expected `none`, `naive`, `entropy` or `customize`&#39;</span> <span class="o">%</span> <span class="n">calib_mode</span><span class="p">)</span>
         <span class="n">qsym</span> <span class="o">=</span> <span class="n">_calibrate_quantized_sym</span><span class="p">(</span><span class="n">qsym</span><span class="p">,</span> <span class="n">th_dict</span><span class="p">)</span>
     <span class="k">else</span><span class="p">:</span>
         <span class="k">raise</span> <span class="ne">ValueError</span><span class="p">(</span><span class="s1">&#39;please set calibration mode to naive or entropy.&#39;</span><span class="p">)</span>

     <span class="k">if</span> <span class="n">logger</span><span class="p">:</span>
         <span class="n">logger</span><span class="o">.</span><span class="n">info</span><span class="p">(</span><span class="s1">&#39;Quantizing parameters&#39;</span><span class="p">)</span>
     <span class="n">qarg_params</span> <span class="o">=</span> <span class="n">_quantize_params</span><span class="p">(</span><span class="n">qsym</span><span class="p">,</span> <span class="n">arg_params</span><span class="p">,</span> <span class="n">th_dict</span><span class="p">)</span>

     <span class="k">return</span> <span class="n">qsym</span><span class="p">,</span> <span class="n">qarg_params</span><span class="p">,</span> <span class="n">aux_params</span></div>

 <div class="viewcode-block" id="quantize_net_v2"><a class="viewcode-back" href="../../../api/contrib/quantization/index.html#mxnet.contrib.quantization.quantize_net_v2">[docs]</a><span class="k">def</span> <span class="nf">quantize_net_v2</span><span class="p">(</span><span class="n">network</span><span class="p">,</span> <span class="n">quantized_dtype</span><span class="o">=</span><span class="s1">&#39;auto&#39;</span><span class="p">,</span> <span class="n">quantize_mode</span><span class="o">=</span><span class="s1">&#39;full&#39;</span><span class="p">,</span> <span class="n">quantize_granularity</span><span class="o">=</span><span class="s1">&#39;tensor-wise&#39;</span><span class="p">,</span>
                     <span class="n">exclude_layers</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span> <span class="n">exclude_layers_match</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span> <span class="n">exclude_operators</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
                     <span class="n">calib_data</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span> <span class="n">data_shapes</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span> <span class="n">calib_mode</span><span class="o">=</span><span class="s1">&#39;none&#39;</span><span class="p">,</span>
                     <span class="n">num_calib_examples</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span> <span class="n">ctx</span><span class="o">=</span><span class="n">cpu</span><span class="p">(),</span> <span class="n">LayerOutputCollector</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span> <span class="n">logger</span><span class="o">=</span><span class="kc">None</span><span class="p">):</span>
     <span class="sd">&quot;&quot;&quot;User-level API for Gluon users to generate a quantized SymbolBlock from a FP32 HybridBlock w/ or w/o calibration.</span>
 <span class="sd">    The backend quantized operators are only enabled for Linux systems. Please do not run</span>
 <span class="sd">    inference using the quantized models on Windows for now.</span>

 <span class="sd">    Parameters</span>
 <span class="sd">    ----------</span>
 <span class="sd">    network : Gluon HybridBlock</span>
 <span class="sd">        Defines the structure of a neural network for FP32 data types.</span>
 <span class="sd">    quantized_dtype : str</span>
 <span class="sd">        The quantized destination type for input data. Currently support &#39;int8&#39;</span>
 <span class="sd">        , &#39;uint8&#39; and &#39;auto&#39;. &#39;auto&#39; means automatically select output type according to calibration result.</span>
 <span class="sd">        Default value is &#39;int8&#39;.</span>
 <span class="sd">    quantize_mode : str</span>
 <span class="sd">        The mode that quantization pass to apply. Support &#39;full&#39; and &#39;smart&#39;.</span>
 <span class="sd">        &#39;full&#39; means quantize all operator if possible.</span>
 <span class="sd">        &#39;smart&#39; means quantization pass will smartly choice which operator should be quantized.</span>
 <span class="sd">    quantize_granularity: str</span>
 <span class="sd">        The granularity of quantization, currently supports &#39;tensor-wise&#39; and &#39;channel-wise&#39;</span>
 <span class="sd">        quantization. The default value is &#39;tensor-wise&#39;.</span>
 <span class="sd">    exclude_layers : list of strings</span>
 <span class="sd">        A list of strings representing the names of the symbols that users want to excluding</span>
 <span class="sd">    exclude_layers_match : list of strings</span>
 <span class="sd">        A list of strings wildcard matching the names of the symbols that users want to excluding</span>
 <span class="sd">        from being quantized.</span>
 <span class="sd">    exclude_operators : list of strings</span>
 <span class="sd">        A list of strings representing the names of the operators that users want to excluding</span>
 <span class="sd">    calib_data : mx.io.DataIter or gluon.DataLoader</span>
 <span class="sd">        A iterable data loading object.</span>
 <span class="sd">    data_shapes : list</span>
 <span class="sd">        List of DataDesc, required if calib_data is not provided</span>
 <span class="sd">    calib_mode : str</span>
 <span class="sd">        If calib_mode=&#39;none&#39;, no calibration will be used and the thresholds for</span>
 <span class="sd">        requantization after the corresponding layers will be calculated at runtime by</span>
 <span class="sd">        calling min and max operators. The quantized models generated in this</span>
 <span class="sd">        mode are normally 10-20% slower than those with calibrations during inference.</span>
 <span class="sd">        If calib_mode=&#39;naive&#39;, the min and max values of the layer outputs from a calibration</span>
 <span class="sd">        dataset will be directly taken as the thresholds for quantization.</span>
 <span class="sd">        If calib_mode=&#39;entropy&#39; (default mode), the thresholds for quantization will be</span>
 <span class="sd">        derived such that the KL divergence between the distributions of FP32 layer outputs and</span>
 <span class="sd">        quantized layer outputs is minimized based upon the calibration dataset.</span>
 <span class="sd">    num_calib_examples : int or None</span>
 <span class="sd">        The maximum number of examples that user would like to use for calibration. If not provided,</span>
 <span class="sd">        the whole calibration dataset will be used.</span>
 <span class="sd">    ctx : Context</span>
 <span class="sd">        Defines the device that users want to run forward propagation on the calibration</span>
 <span class="sd">        dataset for collecting layer output statistics. Currently, only supports single context.</span>
 <span class="sd">    LayerOutputCollector : class</span>
 <span class="sd">        For customize calibration method usage.</span>
 <span class="sd">    logger : Object</span>
 <span class="sd">        A logging object for printing information during the process of quantization.</span>

 <span class="sd">    Returns</span>
 <span class="sd">    -------</span>
 <span class="sd">    network : Gluon SymbolBlock</span>
 <span class="sd">        Defines the structure of a neural network for INT8 data types.</span>
 <span class="sd">    -------</span>
 <span class="sd">    &quot;&quot;&quot;</span>

     <span class="k">if</span> <span class="n">logger</span><span class="p">:</span>
         <span class="n">logger</span><span class="o">.</span><span class="n">info</span><span class="p">(</span><span class="s1">&#39;Export HybridBlock&#39;</span><span class="p">)</span>
     <span class="n">network</span><span class="o">.</span><span class="n">hybridize</span><span class="p">()</span>
     <span class="kn">import</span> <span class="nn">mxnet</span> <span class="k">as</span> <span class="nn">mx</span>
     <span class="k">if</span> <span class="n">calib_data</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
         <span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">calib_data</span><span class="p">,</span> <span class="n">DataIter</span><span class="p">):</span>
             <span class="n">dshapes</span> <span class="o">=</span> <span class="n">calib_data</span><span class="o">.</span><span class="n">provide_data</span>
         <span class="k">else</span><span class="p">:</span>
             <span class="n">calib_data</span><span class="p">,</span> <span class="n">dshapes</span> <span class="o">=</span> <span class="n">_as_data_iter</span><span class="p">(</span><span class="n">calib_data</span><span class="p">)</span>
         <span class="k">if</span> <span class="ow">not</span> <span class="n">data_shapes</span><span class="p">:</span>
             <span class="n">data_shapes</span> <span class="o">=</span> <span class="n">dshapes</span>
     <span class="k">if</span> <span class="ow">not</span> <span class="n">data_shapes</span><span class="p">:</span>
         <span class="k">raise</span> <span class="ne">ValueError</span><span class="p">(</span><span class="s1">&#39;data_shapes required&#39;</span><span class="p">)</span>
     <span class="n">data_nd</span> <span class="o">=</span> <span class="p">[]</span>
     <span class="k">for</span> <span class="n">shape</span> <span class="ow">in</span> <span class="n">data_shapes</span><span class="p">:</span>
         <span class="n">data_nd</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="n">mx</span><span class="o">.</span><span class="n">nd</span><span class="o">.</span><span class="n">zeros</span><span class="p">(</span><span class="n">shape</span><span class="o">.</span><span class="n">shape</span><span class="p">))</span>
     <span class="k">while</span> <span class="kc">True</span><span class="p">:</span>
         <span class="k">try</span><span class="p">:</span>
             <span class="n">network</span><span class="p">(</span><span class="o">*</span><span class="n">data_nd</span><span class="p">)</span>
         <span class="k">except</span> <span class="ne">TypeError</span><span class="p">:</span>
             <span class="k">del</span> <span class="n">data_nd</span><span class="p">[</span><span class="o">-</span><span class="mi">1</span><span class="p">]</span>
             <span class="k">del</span> <span class="n">calib_data</span><span class="o">.</span><span class="n">provide_data</span><span class="p">[</span><span class="o">-</span><span class="mi">1</span><span class="p">]</span>
             <span class="k">continue</span>
         <span class="k">else</span><span class="p">:</span>
             <span class="k">break</span>

     <span class="kn">import</span> <span class="nn">tempfile</span>
     <span class="k">try</span><span class="p">:</span>
         <span class="kn">from</span> <span class="nn">tempfile</span> <span class="kn">import</span> <span class="n">TemporaryDirectory</span>
     <span class="k">except</span> <span class="ne">ImportError</span><span class="p">:</span>
         <span class="c1"># really simple implementation of TemporaryDirectory</span>
         <span class="k">class</span> <span class="nc">TemporaryDirectory</span><span class="p">(</span><span class="nb">object</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">suffix</span><span class="o">=</span><span class="s1">&#39;&#39;</span><span class="p">,</span> <span class="n">prefix</span><span class="o">=</span><span class="s1">&#39;&#39;</span><span class="p">,</span> <span class="nb">dir</span><span class="o">=</span><span class="s1">&#39;&#39;</span><span class="p">):</span>
                 <span class="bp">self</span><span class="o">.</span><span class="n">_dirname</span> <span class="o">=</span> <span class="n">tempfile</span><span class="o">.</span><span class="n">mkdtemp</span><span class="p">(</span><span class="n">suffix</span><span class="p">,</span> <span class="n">prefix</span><span class="p">,</span> <span class="nb">dir</span><span class="p">)</span>

             <span class="k">def</span> <span class="fm">__enter__</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
                 <span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">_dirname</span>

             <span class="k">def</span> <span class="fm">__exit__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">exc_type</span><span class="p">,</span> <span class="n">exc_value</span><span class="p">,</span> <span class="n">traceback</span><span class="p">):</span>
                 <span class="n">shutil</span><span class="o">.</span><span class="n">rmtree</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">_dirname</span><span class="p">)</span>
     <span class="c1"># TODO(xinyu-intel): tmp solution to save and reload for mxnet.mod.Module.</span>
     <span class="c1"># will enhance `export` function to return `sym, args, auxs` directly.</span>
     <span class="k">with</span> <span class="n">TemporaryDirectory</span><span class="p">()</span> <span class="k">as</span> <span class="n">tmpdirname</span><span class="p">:</span>
         <span class="n">prefix</span> <span class="o">=</span> <span class="n">os</span><span class="o">.</span><span class="n">path</span><span class="o">.</span><span class="n">join</span><span class="p">(</span><span class="n">tmpdirname</span><span class="p">,</span> <span class="s1">&#39;tmp&#39;</span><span class="p">)</span>
         <span class="n">network</span><span class="o">.</span><span class="n">export</span><span class="p">(</span><span class="n">prefix</span><span class="p">,</span> <span class="n">epoch</span><span class="o">=</span><span class="mi">0</span><span class="p">)</span>
         <span class="n">symnet</span><span class="p">,</span> <span class="n">args</span><span class="p">,</span> <span class="n">auxs</span> <span class="o">=</span> <span class="n">mx</span><span class="o">.</span><span class="n">model</span><span class="o">.</span><span class="n">load_checkpoint</span><span class="p">(</span><span class="n">prefix</span><span class="p">,</span> <span class="mi">0</span><span class="p">)</span>

     <span class="k">if</span> <span class="n">exclude_layers</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
         <span class="n">exclude_layers</span> <span class="o">=</span> <span class="p">[]</span>
     <span class="k">if</span> <span class="n">exclude_layers_match</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
         <span class="n">exclude_layers_match</span> <span class="o">=</span> <span class="p">[]</span>
     <span class="k">if</span> <span class="n">exclude_operators</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
         <span class="n">exclude_operators</span> <span class="o">=</span> <span class="p">[]</span>
     <span class="k">for</span> <span class="n">name_match</span> <span class="ow">in</span> <span class="n">exclude_layers_match</span><span class="p">:</span>
         <span class="k">for</span> <span class="n">layers</span> <span class="ow">in</span> <span class="nb">list</span><span class="p">(</span><span class="n">symnet</span><span class="o">.</span><span class="n">get_internals</span><span class="p">()):</span>
             <span class="k">if</span> <span class="n">layers</span><span class="o">.</span><span class="n">name</span><span class="o">.</span><span class="n">find</span><span class="p">(</span><span class="n">name_match</span><span class="p">)</span> <span class="o">!=</span> <span class="o">-</span><span class="mi">1</span><span class="p">:</span>
                 <span class="n">exclude_layers</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="n">layers</span><span class="o">.</span><span class="n">name</span><span class="p">)</span>
     <span class="k">if</span> <span class="n">logger</span><span class="p">:</span>
         <span class="n">logger</span><span class="o">.</span><span class="n">info</span><span class="p">(</span><span class="s1">&#39;These layers have been excluded </span><span class="si">%s</span><span class="s1">&#39;</span> <span class="o">%</span> <span class="n">exclude_layers</span><span class="p">)</span>

     <span class="k">if</span> <span class="n">ctx</span> <span class="o">==</span> <span class="n">mx</span><span class="o">.</span><span class="n">cpu</span><span class="p">():</span>
         <span class="n">symnet</span> <span class="o">=</span> <span class="n">symnet</span><span class="o">.</span><span class="n">get_backend_symbol</span><span class="p">(</span><span class="s1">&#39;MKLDNN_QUANTIZE&#39;</span><span class="p">)</span>

     <span class="n">qsym</span><span class="p">,</span> <span class="n">qarg_params</span><span class="p">,</span> <span class="n">aux_params</span><span class="p">,</span> <span class="n">collector</span> <span class="o">=</span> <span class="n">quantize_graph</span><span class="p">(</span>
         <span class="n">sym</span><span class="o">=</span><span class="n">symnet</span><span class="p">,</span> <span class="n">arg_params</span><span class="o">=</span><span class="n">args</span><span class="p">,</span> <span class="n">aux_params</span><span class="o">=</span><span class="n">auxs</span><span class="p">,</span> <span class="n">ctx</span><span class="o">=</span><span class="n">ctx</span><span class="p">,</span>
         <span class="n">excluded_sym_names</span><span class="o">=</span><span class="n">exclude_layers</span><span class="p">,</span> <span class="n">excluded_op_names</span><span class="o">=</span><span class="n">exclude_operators</span><span class="p">,</span>
         <span class="n">calib_mode</span><span class="o">=</span><span class="n">calib_mode</span><span class="p">,</span> <span class="n">quantized_dtype</span><span class="o">=</span><span class="n">quantized_dtype</span><span class="p">,</span> <span class="n">quantize_mode</span><span class="o">=</span><span class="n">quantize_mode</span><span class="p">,</span>
         <span class="n">quantize_granularity</span><span class="o">=</span><span class="n">quantize_granularity</span><span class="p">,</span> <span class="n">LayerOutputCollector</span><span class="o">=</span><span class="n">LayerOutputCollector</span><span class="p">,</span>
         <span class="n">logger</span><span class="o">=</span><span class="n">logger</span><span class="p">)</span>

     <span class="k">if</span> <span class="n">calib_mode</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span> <span class="ow">and</span> <span class="n">calib_mode</span> <span class="o">!=</span> <span class="s1">&#39;none&#39;</span><span class="p">:</span>
         <span class="k">if</span> <span class="ow">not</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">ctx</span><span class="p">,</span> <span class="n">Context</span><span class="p">):</span>
             <span class="k">raise</span> <span class="ne">ValueError</span><span class="p">(</span>
                 <span class="s1">&#39;currently only supports single ctx, while received </span><span class="si">%s</span><span class="s1">&#39;</span> <span class="o">%</span> <span class="nb">str</span><span class="p">(</span><span class="n">ctx</span><span class="p">))</span>
         <span class="k">if</span> <span class="n">calib_data</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
             <span class="k">raise</span> <span class="ne">ValueError</span><span class="p">(</span>
                 <span class="s1">&#39;calib_data must be provided when calib_mode=</span><span class="si">%s</span><span class="s1">&#39;</span> <span class="o">%</span> <span class="n">calib_mode</span><span class="p">)</span>
         <span class="k">if</span> <span class="n">calib_mode</span> <span class="ow">in</span> <span class="p">[</span><span class="s1">&#39;naive&#39;</span><span class="p">,</span> <span class="s1">&#39;entropy&#39;</span><span class="p">,</span> <span class="s1">&#39;customize&#39;</span><span class="p">]:</span>
             <span class="n">data_names</span> <span class="o">=</span> <span class="p">[</span><span class="n">pair</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span> <span class="k">for</span> <span class="n">pair</span> <span class="ow">in</span> <span class="n">calib_data</span><span class="o">.</span><span class="n">provide_data</span><span class="p">]</span>
             <span class="n">mod</span> <span class="o">=</span> <span class="n">Module</span><span class="p">(</span><span class="n">symbol</span><span class="o">=</span><span class="n">symnet</span><span class="p">,</span> <span class="n">context</span><span class="o">=</span><span class="n">ctx</span><span class="p">,</span>
                          <span class="n">data_names</span><span class="o">=</span><span class="n">data_names</span><span class="p">,</span> <span class="n">label_names</span><span class="o">=</span><span class="kc">None</span><span class="p">)</span>
             <span class="n">mod</span><span class="o">.</span><span class="n">bind</span><span class="p">(</span><span class="n">for_training</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span> <span class="n">data_shapes</span><span class="o">=</span><span class="n">data_shapes</span><span class="p">)</span>
             <span class="n">mod</span><span class="o">.</span><span class="n">set_params</span><span class="p">(</span><span class="n">args</span><span class="p">,</span> <span class="n">auxs</span><span class="p">,</span> <span class="n">allow_missing</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span> <span class="n">force_init</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span>
             <span class="n">num_examples</span> <span class="o">=</span> <span class="n">_collect_layer_statistics</span><span class="p">(</span><span class="n">mod</span><span class="p">,</span> <span class="n">calib_data</span><span class="p">,</span> <span class="n">collector</span><span class="p">,</span>
                                                      <span class="n">num_calib_examples</span><span class="p">,</span> <span class="n">logger</span><span class="p">)</span>
             <span class="k">if</span> <span class="n">logger</span><span class="p">:</span>
                 <span class="n">logger</span><span class="o">.</span><span class="n">info</span><span class="p">(</span><span class="s1">&#39;Collected layer output values from FP32 model using </span><span class="si">%d</span><span class="s1"> examples&#39;</span>
                             <span class="o">%</span> <span class="n">num_examples</span><span class="p">)</span>
             <span class="n">qsym</span><span class="p">,</span> <span class="n">qarg_params</span><span class="p">,</span> <span class="n">aux_params</span> <span class="o">=</span> <span class="n">calib_graph</span><span class="p">(</span>
                 <span class="n">qsym</span><span class="o">=</span><span class="n">qsym</span><span class="p">,</span> <span class="n">arg_params</span><span class="o">=</span><span class="n">args</span><span class="p">,</span> <span class="n">aux_params</span><span class="o">=</span><span class="n">auxs</span><span class="p">,</span> <span class="n">collector</span><span class="o">=</span><span class="n">collector</span><span class="p">,</span>
                 <span class="n">calib_mode</span><span class="o">=</span><span class="n">calib_mode</span><span class="p">,</span> <span class="n">quantized_dtype</span><span class="o">=</span><span class="n">quantized_dtype</span><span class="p">,</span> <span class="n">logger</span><span class="o">=</span><span class="n">logger</span><span class="p">)</span>
         <span class="k">else</span><span class="p">:</span>
             <span class="k">raise</span> <span class="ne">ValueError</span><span class="p">(</span>
                 <span class="s1">&#39;please set calibration mode to naive or entropy.&#39;</span><span class="p">)</span>
     <span class="k">elif</span> <span class="n">calib_mode</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span> <span class="ow">and</span> <span class="n">calib_mode</span> <span class="o">==</span> <span class="s1">&#39;none&#39;</span><span class="p">:</span>
         <span class="n">data_names</span> <span class="o">=</span> <span class="p">[</span><span class="n">pair</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span> <span class="k">for</span> <span class="n">pair</span> <span class="ow">in</span> <span class="n">data_shapes</span><span class="p">]</span>

     <span class="k">if</span> <span class="n">ctx</span> <span class="o">==</span> <span class="n">mx</span><span class="o">.</span><span class="n">cpu</span><span class="p">():</span>
         <span class="n">qsym</span> <span class="o">=</span> <span class="n">qsym</span><span class="o">.</span><span class="n">get_backend_symbol</span><span class="p">(</span><span class="s1">&#39;MKLDNN_QUANTIZE&#39;</span><span class="p">)</span>

     <span class="kn">from</span> <span class="nn">..gluon</span> <span class="kn">import</span> <span class="n">SymbolBlock</span>
     <span class="n">data_sym</span> <span class="o">=</span> <span class="p">[]</span>
     <span class="k">for</span> <span class="n">name</span> <span class="ow">in</span> <span class="n">data_names</span><span class="p">:</span>
         <span class="n">data_sym</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="n">mx</span><span class="o">.</span><span class="n">sym</span><span class="o">.</span><span class="n">var</span><span class="p">(</span><span class="n">name</span><span class="p">))</span>
     <span class="n">net</span> <span class="o">=</span> <span class="n">SymbolBlock</span><span class="p">(</span><span class="n">qsym</span><span class="p">,</span> <span class="n">data_sym</span><span class="p">)</span>
     <span class="c1"># TODO(xinyu-intel): tmp solution to save param_dict and reload for SymbolBlock</span>
     <span class="c1"># will enhance SymbolBlock to load args, auxs directly.</span>
     <span class="k">with</span> <span class="n">TemporaryDirectory</span><span class="p">()</span> <span class="k">as</span> <span class="n">tmpdirname</span><span class="p">:</span>
         <span class="n">prefix</span> <span class="o">=</span> <span class="n">os</span><span class="o">.</span><span class="n">path</span><span class="o">.</span><span class="n">join</span><span class="p">(</span><span class="n">tmpdirname</span><span class="p">,</span> <span class="s1">&#39;tmp&#39;</span><span class="p">)</span>
         <span class="n">param_name</span> <span class="o">=</span> <span class="s1">&#39;</span><span class="si">%s</span><span class="s1">-</span><span class="si">%04d</span><span class="s1">.params&#39;</span> <span class="o">%</span> <span class="p">(</span><span class="n">prefix</span> <span class="o">+</span> <span class="s1">&#39;net-quantized&#39;</span><span class="p">,</span> <span class="mi">0</span><span class="p">)</span>
         <span class="n">save_dict</span> <span class="o">=</span> <span class="p">{(</span><span class="s1">&#39;arg:</span><span class="si">%s</span><span class="s1">&#39;</span> <span class="o">%</span> <span class="n">k</span><span class="p">):</span> <span class="n">v</span><span class="o">.</span><span class="n">as_in_context</span><span class="p">(</span><span class="n">cpu</span><span class="p">())</span>
                      <span class="k">for</span> <span class="n">k</span><span class="p">,</span> <span class="n">v</span> <span class="ow">in</span> <span class="n">qarg_params</span><span class="o">.</span><span class="n">items</span><span class="p">()}</span>
         <span class="n">save_dict</span><span class="o">.</span><span class="n">update</span><span class="p">({(</span><span class="s1">&#39;aux:</span><span class="si">%s</span><span class="s1">&#39;</span> <span class="o">%</span> <span class="n">k</span><span class="p">):</span> <span class="n">v</span><span class="o">.</span><span class="n">as_in_context</span><span class="p">(</span><span class="n">cpu</span><span class="p">())</span>
                           <span class="k">for</span> <span class="n">k</span><span class="p">,</span> <span class="n">v</span> <span class="ow">in</span> <span class="n">aux_params</span><span class="o">.</span><span class="n">items</span><span class="p">()})</span>
         <span class="n">nd_save</span><span class="p">(</span><span class="n">param_name</span><span class="p">,</span> <span class="n">save_dict</span><span class="p">)</span>
         <span class="n">net</span><span class="o">.</span><span class="n">collect_params</span><span class="p">()</span><span class="o">.</span><span class="n">load</span><span class="p">(</span><span class="n">param_name</span><span class="p">,</span> <span class="n">cast_dtype</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span> <span class="n">dtype_source</span><span class="o">=</span><span class="s1">&#39;saved&#39;</span><span class="p">)</span>
         <span class="n">net</span><span class="o">.</span><span class="n">collect_params</span><span class="p">()</span><span class="o">.</span><span class="n">reset_ctx</span><span class="p">(</span><span class="n">ctx</span><span class="p">)</span>
     <span class="k">return</span> <span class="n">net</span></div>

 <div class="viewcode-block" id="quantize_net"><a class="viewcode-back" href="../../../api/contrib/quantization/index.html#mxnet.contrib.quantization.quantize_net">[docs]</a><span class="k">def</span> <span class="nf">quantize_net</span><span class="p">(</span><span class="n">network</span><span class="p">,</span> <span class="n">quantized_dtype</span><span class="o">=</span><span class="s1">&#39;auto&#39;</span><span class="p">,</span> <span class="n">quantize_mode</span><span class="o">=</span><span class="s1">&#39;full&#39;</span><span class="p">,</span>
                  <span class="n">exclude_layers</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span> <span class="n">exclude_layers_match</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span> <span class="n">exclude_operators</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
                  <span class="n">calib_data</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span> <span class="n">data_shapes</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span> <span class="n">calib_mode</span><span class="o">=</span><span class="s1">&#39;none&#39;</span><span class="p">,</span>
                  <span class="n">num_calib_examples</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span> <span class="n">ctx</span><span class="o">=</span><span class="n">cpu</span><span class="p">(),</span> <span class="n">logger</span><span class="o">=</span><span class="kc">None</span><span class="p">):</span>
     <span class="sd">&quot;&quot;&quot;User-level API for Gluon users to generate a quantized SymbolBlock from a FP32 HybridBlock w/ or w/o calibration.</span>
 <span class="sd">       Will be deprecated after MXNet 2.0, please use quantize_net_v2.</span>
 <span class="sd">    &quot;&quot;&quot;</span>
     <span class="n">warnings</span><span class="o">.</span><span class="n">warn</span><span class="p">(</span><span class="s1">&#39;WARNING: This will be deprecated after MXNet 2.0, please use quantize_net_v2.&#39;</span><span class="p">)</span>
     <span class="k">return</span> <span class="n">quantize_net_v2</span><span class="p">(</span><span class="n">network</span><span class="o">=</span><span class="n">network</span><span class="p">,</span> <span class="n">quantized_dtype</span><span class="o">=</span><span class="n">quantized_dtype</span><span class="p">,</span>
                            <span class="n">quantize_mode</span><span class="o">=</span><span class="n">quantize_mode</span><span class="p">,</span>
                            <span class="n">quantize_granularity</span><span class="o">=</span><span class="s1">&#39;tensor-wise&#39;</span><span class="p">,</span>
                            <span class="n">exclude_layers</span><span class="o">=</span><span class="n">exclude_layers</span><span class="p">,</span>
                            <span class="n">exclude_layers_match</span><span class="o">=</span><span class="n">exclude_layers_match</span><span class="p">,</span>
                            <span class="n">exclude_operators</span><span class="o">=</span><span class="n">exclude_operators</span><span class="p">,</span>
                            <span class="n">calib_data</span><span class="o">=</span><span class="n">calib_data</span><span class="p">,</span> <span class="n">data_shapes</span><span class="o">=</span><span class="n">data_shapes</span><span class="p">,</span>
                            <span class="n">calib_mode</span><span class="o">=</span><span class="n">calib_mode</span><span class="p">,</span> <span class="n">num_calib_examples</span><span class="o">=</span><span class="n">num_calib_examples</span><span class="p">,</span>
                            <span class="n">ctx</span><span class="o">=</span><span class="n">ctx</span><span class="p">,</span> <span class="n">LayerOutputCollector</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span> <span class="n">logger</span><span class="o">=</span><span class="n">logger</span><span class="p">)</span></div>
 </pre></div>

         <hr class="feedback-hr-top" />
 <div class="feedback-container">
     <div class="feedback-question">Did this page help you?</div>
     <div class="feedback-answer-container">
         <div class="feedback-answer yes-link" data-response="yes">Yes</div>
         <div class="feedback-answer no-link" data-response="no">No</div>
     </div>
     <div class="feedback-thank-you">Thanks for your feedback!</div>
 </div>
 <hr class="feedback-hr-bottom" />
         </div>
         <div class="side-doc-outline">
             <div class="side-doc-outline--content">
             </div>
         </div>

       <div class="clearer"></div>
     </div><div class="pagenation">
   </div>
             <footer class="site-footer h-card">
     <div class="wrapper">
         <div class="row">
             <div class="col-4">
                 <h4 class="footer-category-title">Resources</h4>
                 <ul class="contact-list">
                     <li><a class="u-email" href="mailto:dev@mxnet.apache.org">Dev list</a></li>
                     <li><a class="u-email" href="mailto:user@mxnet.apache.org">User mailing list</a></li>
                     <li><a href="https://cwiki.apache.org/confluence/display/MXNET/Apache+MXNet+Home">Developer Wiki</a></li>
                     <li><a href="https://issues.apache.org/jira/projects/MXNET/issues">Jira Tracker</a></li>
                     <li><a href="https://github.com/apache/incubator-mxnet/labels/Roadmap">Github Roadmap</a></li>
                     <li><a href="https://medium.com/apache-mxnet">Blog</a></li>
                     <li><a href="https://discuss.mxnet.io">Forum</a></li>
                     <li><a href="/community/contribute">Contribute</a></li>

                 </ul>
             </div>

             <div class="col-4"><ul class="social-media-list"><li><a href="https://github.com/apache/incubator-mxnet"><svg class="svg-icon"><use xlink:href="../../../_static/minima-social-icons.svg#github"></use></svg> <span class="username">apache/incubator-mxnet</span></a></li><li><a href="https://www.twitter.com/apachemxnet"><svg class="svg-icon"><use xlink:href="../../../_static/minima-social-icons.svg#twitter"></use></svg> <span class="username">apachemxnet</span></a></li><li><a href="https://youtube.com/apachemxnet"><svg class="svg-icon"><use xlink:href="../../../_static/minima-social-icons.svg#youtube"></use></svg> <span class="username">apachemxnet</span></a></li></ul>
 </div>

             <div class="col-4 footer-text">
                 <p>A flexible and efficient library for deep learning.</p>
             </div>
         </div>
     </div>
 </footer>

 <footer class="site-footer2">
     <div class="wrapper">
         <div class="row">
             <div class="col-3">
                 <img src="../../../_static/apache_incubator_logo.png" class="footer-logo col-2">
             </div>
             <div class="footer-bottom-warning col-9">
                 <p>Apache MXNet is an effort undergoing incubation at <a href="http://www.apache.org/">The Apache Software Foundation</a> (ASF), <span style="font-weight:bold">sponsored by the <i>Apache Incubator</i></span>. Incubation is required
                     of all newly accepted projects until a further review indicates that the infrastructure,
                     communications, and decision making process have stabilized in a manner consistent with other
                     successful ASF projects. While incubation status is not necessarily a reflection of the completeness
                     or stability of the code, it does indicate that the project has yet to be fully endorsed by the ASF.
                 </p><p>"Copyright © 2017-2018, The Apache Software Foundation Apache MXNet, MXNet, Apache, the Apache
                     feather, and the Apache MXNet project logo are either registered trademarks or trademarks of the
                     Apache Software Foundation."</p>
             </div>
         </div>
     </div>
 </footer>

   </body>
 </html>