Skip to content

P
Paddle

PaddlePaddle / Paddle
大约 1 年 前同步成功

通知 2298
Star 20931
Fork 5422
P
Paddle
切换分支/标签
查找文件 普通视图历史永久链接Permalink
cluster_train_en.html 40.8 KB
Newer Older
T
Deploy to GitHub Pages: 2c98becba4fb48311b80c53c1793fae47e7fe975  
Travis CI 已提交
1 2 3 4 5 6 7 8 9 10 


<!DOCTYPE html>
<!--[if IE 8]><html class="no-js lt-ie9" lang="en" > <![endif]-->
<!--[if gt IE 8]><!--> <html class="no-js" lang="en" > <!--<![endif]-->
<head>
  <meta charset="utf-8">
  
  <meta name="viewport" content="width=device-width, initial-scale=1.0">
T
Deploy to GitHub Pages: 1ff47b284c565d030b198705d5f18b4bd4ce53e5  
Travis CI 已提交
11 
  <title>PaddlePaddle Distributed Training &mdash; PaddlePaddle  documentation</title>
T
Deploy to GitHub Pages: 2c98becba4fb48311b80c53c1793fae47e7fe975  
Travis CI 已提交
12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 
  

  
  

  

  
  
    

  

  
  
    <link rel="stylesheet" href="../../../_static/css/theme.css" type="text/css" />
  

  
  
        <link rel="index" title="Index"
              href="../../../genindex.html"/>
        <link rel="search" title="Search" href="../../../search.html"/>
    <link rel="top" title="PaddlePaddle  documentation" href="../../../index.html"/>
        <link rel="up" title="HOW TO" href="../../index_en.html"/>
        <link rel="next" title="Paddle On Kubernetes" href="../k8s/k8s_en.html"/>
        <link rel="prev" title="Detail Description" href="../cmd_parameter/detail_introduction_en.html"/> 

  <link rel="stylesheet" href="https://cdn.jsdelivr.net/perfect-scrollbar/0.6.14/css/perfect-scrollbar.min.css" type="text/css" />
  <link rel="stylesheet" href="../../../_static/css/override.css" type="text/css" />
  <script>
  var _hmt = _hmt || [];
  (function() {
    var hm = document.createElement("script");
    hm.src = "//hm.baidu.com/hm.js?b9a314ab40d04d805655aab1deee08ba";
    var s = document.getElementsByTagName("script")[0]; 
    s.parentNode.insertBefore(hm, s);
  })();
  </script>

  

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

</head>

<body class="wy-body-for-nav" role="document">

  
  <header class="site-header">
    <div class="site-logo">
      <a href="/"><img src="../../../_static/images/PP_w.png"></a>
    </div>
    <div class="site-nav-links">
      <div class="site-menu">
T
Deploy to GitHub Pages: 1ff47b284c565d030b198705d5f18b4bd4ce53e5  
Travis CI 已提交
68 
        <a class="fork-on-github" href="https://github.com/PaddlePaddle/Paddle" target="_blank"><i class="fa fa-github"></i>Fork me on Github</a>
T
Deploy to GitHub Pages: 2c98becba4fb48311b80c53c1793fae47e7fe975  
Travis CI 已提交
69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 
        <div class="language-switcher dropdown">
          <a type="button" data-toggle="dropdown">
            <span>English</span>
            <i class="fa fa-angle-up"></i>
            <i class="fa fa-angle-down"></i>
          </a>
          <ul class="dropdown-menu">
            <li><a href="/doc_cn">中文</a></li>
            <li><a href="/doc">English</a></li>
          </ul>
        </div>
        <ul class="site-page-links">
          <li><a href="/">Home</a></li>
        </ul>
      </div>
      <div class="doc-module">
        
        <ul class="current">
<li class="toctree-l1"><a class="reference internal" href="../../../getstarted/index_en.html">GET STARTED</a></li>
<li class="toctree-l1 current"><a class="reference internal" href="../../index_en.html">HOW TO</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../../api/index_en.html">API</a></li>
T
Deploy to GitHub Pages: 1ff47b284c565d030b198705d5f18b4bd4ce53e5  
Travis CI 已提交
90 
<li class="toctree-l1"><a class="reference internal" href="../../../mobile/index_en.html">MOBILE</a></li>
T
Deploy to GitHub Pages: 2c98becba4fb48311b80c53c1793fae47e7fe975  
Travis CI 已提交
91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 
</ul>

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

    
    <nav class="doc-menu-vertical" role="navigation">
        
          
          <ul class="current">
<li class="toctree-l1"><a class="reference internal" href="../../../getstarted/index_en.html">GET STARTED</a><ul>
<li class="toctree-l2"><a class="reference internal" href="../../../getstarted/build_and_install/index_en.html">Install and Build</a><ul>
T
Deploy to GitHub Pages: 1ff47b284c565d030b198705d5f18b4bd4ce53e5  
Travis CI 已提交
114 115 116 117 
<li class="toctree-l3"><a class="reference internal" href="../../../getstarted/build_and_install/pip_install_en.html">Install Using pip</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../getstarted/build_and_install/docker_install_en.html">Run in Docker Containers</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../dev/build_en.html">Build using Docker</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../getstarted/build_and_install/build_from_source_en.html">Build from Sources</a></li>
T
Deploy to GitHub Pages: 2c98becba4fb48311b80c53c1793fae47e7fe975  
Travis CI 已提交
118 119 120 121 122 123 124 125 126 127 128 
</ul>
</li>
</ul>
</li>
<li class="toctree-l1 current"><a class="reference internal" href="../../index_en.html">HOW TO</a><ul class="current">
<li class="toctree-l2"><a class="reference internal" href="../cmd_parameter/index_en.html">Set Command-line Parameters</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../cmd_parameter/use_case_en.html">Use Case</a></li>
<li class="toctree-l3"><a class="reference internal" href="../cmd_parameter/arguments_en.html">Argument Outline</a></li>
<li class="toctree-l3"><a class="reference internal" href="../cmd_parameter/detail_introduction_en.html">Detail Description</a></li>
</ul>
</li>
T
Deploy to GitHub Pages: 1ff47b284c565d030b198705d5f18b4bd4ce53e5  
Travis CI 已提交
129 
<li class="toctree-l2 current"><a class="current reference internal" href="#">PaddlePaddle Distributed Training</a></li>
T
Deploy to GitHub Pages: 2c98becba4fb48311b80c53c1793fae47e7fe975  
Travis CI 已提交
130 131 132 133 
<li class="toctree-l2"><a class="reference internal" href="../k8s/k8s_en.html">Paddle On Kubernetes</a></li>
<li class="toctree-l2"><a class="reference internal" href="../k8s/k8s_aws_en.html">Distributed PaddlePaddle Training on AWS with Kubernetes</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../dev/new_layer_en.html">Write New Layers</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../dev/contribute_to_paddle_en.html">Contribute Code</a></li>
T
Deploy to GitHub Pages: 1ff47b284c565d030b198705d5f18b4bd4ce53e5  
Travis CI 已提交
134 135 136 137 138 
<li class="toctree-l2"><a class="reference internal" href="../../dev/write_docs_en.html">Contribute Documentation</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../deep_model/rnn/index_en.html">RNN Models</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../../deep_model/rnn/rnn_config_en.html">RNN Configuration</a></li>
</ul>
</li>
T
Deploy to GitHub Pages: 2c98becba4fb48311b80c53c1793fae47e7fe975  
Travis CI 已提交
139 140 141 142 143 144 145 
<li class="toctree-l2"><a class="reference internal" href="../../optimization/gpu_profiling_en.html">Tune GPU Performance</a></li>
</ul>
</li>
<li class="toctree-l1"><a class="reference internal" href="../../../api/index_en.html">API</a><ul>
<li class="toctree-l2"><a class="reference internal" href="../../../api/v2/model_configs.html">Model Configuration</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../../../api/v2/config/activation.html">Activation</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/v2/config/layer.html">Layers</a></li>
T
Deploy to GitHub Pages: 1ff47b284c565d030b198705d5f18b4bd4ce53e5  
Travis CI 已提交
146 
<li class="toctree-l3"><a class="reference internal" href="../../../api/v2/config/evaluators.html">Evaluators</a></li>
T
Deploy to GitHub Pages: 2c98becba4fb48311b80c53c1793fae47e7fe975  
Travis CI 已提交
147 148 149 150 151 152 
<li class="toctree-l3"><a class="reference internal" href="../../../api/v2/config/optimizer.html">Optimizer</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/v2/config/pooling.html">Pooling</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/v2/config/networks.html">Networks</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/v2/config/attr.html">Parameter Attribute</a></li>
</ul>
</li>
T
Deploy to GitHub Pages: 1ff47b284c565d030b198705d5f18b4bd4ce53e5  
Travis CI 已提交
153 154 155 156 157 158 
<li class="toctree-l2"><a class="reference internal" href="../../../api/v2/data.html">Data Reader Interface and DataSets</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../../../api/v2/data/data_reader.html">Data Reader Interface</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/v2/data/image.html">Image Interface</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/v2/data/dataset.html">Dataset</a></li>
</ul>
</li>
T
Deploy to GitHub Pages: 2c98becba4fb48311b80c53c1793fae47e7fe975  
Travis CI 已提交
159 
<li class="toctree-l2"><a class="reference internal" href="../../../api/v2/run_logic.html">Training and Inference</a></li>
T
Deploy to GitHub Pages: 1ff47b284c565d030b198705d5f18b4bd4ce53e5  
Travis CI 已提交
160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 
<li class="toctree-l2"><a class="reference internal" href="../../../api/v2/fluid.html">Fluid</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../../../api/v2/fluid/layers.html">Layers</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/v2/fluid/data_feeder.html">DataFeeder</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/v2/fluid/executor.html">Executor</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/v2/fluid/initializer.html">Initializer</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/v2/fluid/evaluator.html">Evaluator</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/v2/fluid/nets.html">Nets</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/v2/fluid/optimizer.html">Optimizer</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/v2/fluid/param_attr.html">ParamAttr</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/v2/fluid/profiler.html">Profiler</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/v2/fluid/regularizer.html">Regularizer</a></li>
</ul>
</li>
</ul>
</li>
<li class="toctree-l1"><a class="reference internal" href="../../../mobile/index_en.html">MOBILE</a><ul>
<li class="toctree-l2"><a class="reference internal" href="../../../mobile/cross_compiling_for_android_en.html">Build PaddlePaddle for Android</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../../mobile/cross_compiling_for_raspberry_en.html">Build PaddlePaddle for Raspberry Pi</a></li>
T
Deploy to GitHub Pages: 2c98becba4fb48311b80c53c1793fae47e7fe975  
Travis CI 已提交
178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 
</ul>
</li>
</ul>

        
    </nav>
    
    <section class="doc-content-wrap">

      

 







<div role="navigation" aria-label="breadcrumbs navigation">
  <ul class="wy-breadcrumbs">
      
        <li><a href="../../index_en.html">HOW TO</a> > </li>
T
Deploy to GitHub Pages: 1ff47b284c565d030b198705d5f18b4bd4ce53e5  
Travis CI 已提交
202 
    <li>PaddlePaddle Distributed Training</li>
T
Deploy to GitHub Pages: 2c98becba4fb48311b80c53c1793fae47e7fe975  
Travis CI 已提交
203 204 205 206 207 208 209 210 
  </ul>
</div>
      
      <div class="wy-nav-content" id="doc-content">
        <div class="rst-content">
          <div role="main" class="document" itemscope="itemscope" itemtype="http://schema.org/Article">
           <div itemprop="articleBody">
T
Deploy to GitHub Pages: 1ff47b284c565d030b198705d5f18b4bd4ce53e5  
Travis CI 已提交
211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 
  <div class="section" id="paddlepaddle-distributed-training">
<span id="paddlepaddle-distributed-training"></span><h1>PaddlePaddle Distributed Training<a class="headerlink" href="#paddlepaddle-distributed-training" title="Permalink to this headline"></a></h1>
<ul class="simple">
<li><a class="reference external" href="#introduction">Introduction</a></li>
<li><a class="reference external" href="#preparations">Preparations</a></li>
<li><a class="reference external" href="#command-line-arguments">Command-line arguments</a><ul>
<li><a class="reference external" href="#starting-parameter-server">Starting parameter server</a></li>
<li><a class="reference external" href="#starting-trainer">Starting trainer</a></li>
<li><a class="reference external" href="#prepare-training-dataset">Prepare Training Dataset</a></li>
<li><a class="reference external" href="#prepare-training-program">Prepare Training program</a></li>
</ul>
</li>
<li><a class="reference external" href="#use-cluster-platforms-or-cluster-management-tools">Use cluster platforms or cluster management tools</a><ul>
<li><a class="reference external" href="#cluster-training-using-fabric">Cluster Training Using Fabric</a><ul>
<li><a class="reference external" href="#prepare-a-linux-cluster">Prepare a Linux cluster</a></li>
<li><a class="reference external" href="#launching-cluster-job">Launching Cluster Job</a></li>
<li><a class="reference external" href="#kill-cluster-job">Kill Cluster Job</a></li>
<li><a class="reference external" href="#check-cluster-training-result">Check Cluster Training Result</a></li>
<li><a class="reference external" href="#check-model-output">Check Model Output</a></li>
</ul>
T
Deploy to GitHub Pages: 2c98becba4fb48311b80c53c1793fae47e7fe975  
Travis CI 已提交
231 
</li>
T
Deploy to GitHub Pages: 1ff47b284c565d030b198705d5f18b4bd4ce53e5  
Travis CI 已提交
232 233 234 235 
<li><a class="reference external" href="#cluster-training-using-openmpi">Cluster Training Using OpenMPI</a><ul>
<li><a class="reference external" href="#prepare-an-openmpi-cluster">Prepare an OpenMPI cluster</a></li>
<li><a class="reference external" href="#launching-cluster-job-1">Launching Cluster Job</a></li>
</ul>
T
Deploy to GitHub Pages: 2c98becba4fb48311b80c53c1793fae47e7fe975  
Travis CI 已提交
236 
</li>
T
Deploy to GitHub Pages: 1ff47b284c565d030b198705d5f18b4bd4ce53e5  
Travis CI 已提交
237 238 
<li><a class="reference external" href="#cluster-training-using-kubernetes">Cluster Training Using Kubernetes</a></li>
</ul>
T
Deploy to GitHub Pages: 2c98becba4fb48311b80c53c1793fae47e7fe975  
Travis CI 已提交
239 
</li>
T
Deploy to GitHub Pages: 1ff47b284c565d030b198705d5f18b4bd4ce53e5  
Travis CI 已提交
240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 
</ul>
<div class="section" id="introduction">
<span id="introduction"></span><h2>Introduction<a class="headerlink" href="#introduction" title="Permalink to this headline"></a></h2>
<p>In this article, we&#8217;ll explain how to run distributed training jobs with PaddlePaddle on different types of clusters. The diagram below shows the main architecture of a distributed trainning job:</p>
<p><img src="https://user-images.githubusercontent.com/13348433/31772146-41523d84-b511-11e7-8a12-a69fd136c283.png" width="500"></p>
<ul class="simple">
<li>Data shard: training data will be split into multiple partitions, trainers use the partitions of the whole dataset to do the training job.</li>
<li>Trainer: each trainer reads the data shard, and train the neural network. Then the trainer will upload calculated &#8220;gradients&#8221; to parameter servers, and wait for parameters to be optimized on the parameter server side. When that finishes, the trainer download optimized parameters and continues its training.</li>
<li>Parameter server: every parameter server stores part of the whole neural network model data. They will do optimization calculations when gradients are uploaded from trainers, and then send updated parameters to trainers.</li>
</ul>
<p>PaddlePaddle can support both synchronize stochastic gradient descent (SGD) and asynchronous SGD.</p>
<p>When training with synchronize SGD, PaddlePaddle uses an internal &#8220;synchronize barrier&#8221; which makes gradients update and parameter download in strict order. On the other hand, asynchronous SGD won&#8217;t wait for all trainers to finish upload at a single step, this will increase the parallelism of distributed training: parameter servers do not depend on each other, they&#8217;ll do parameter optimization concurrently. Parameter servers will not wait for trainers, so trainers will also do their work concurrently. But asynchronous SGD will introduce more randomness and noises in the gradient.</p>
</div>
<div class="section" id="preparations">
<span id="preparations"></span><h2>Preparations<a class="headerlink" href="#preparations" title="Permalink to this headline"></a></h2>
<ol class="simple">
<li>Prepare your computer cluster. It&#8217;s normally a bunch of Linux servers connected by LAN. Each server will be assigned a unique IP address. The computers in the cluster can be called &#8220;nodes&#8221;.</li>
<li>Install PaddlePaddle on every node. If you are going to take advantage of GPU cards, you&#8217;ll also need to install proper driver and CUDA libraries. To install PaddlePaddle please read <a class="reference external" href="https://github.com/PaddlePaddle/Paddle/tree/develop/doc/getstarted/build_and_install">this build and install</a> document. We strongly recommend using <a class="reference external" href="https://github.com/PaddlePaddle/Paddle/blob/develop/doc/getstarted/build_and_install/docker_install_en.rst">Docker installation</a>.</li>
T
Deploy to GitHub Pages: 2c98becba4fb48311b80c53c1793fae47e7fe975  
Travis CI 已提交
258 
</ol>
T
Deploy to GitHub Pages: 1ff47b284c565d030b198705d5f18b4bd4ce53e5  
Travis CI 已提交
259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 
<p>After installation, you can check the version by typing the below command (run a docker container  if using docker: <code class="docutils literal"><span class="pre">docker</span> <span class="pre">run</span> <span class="pre">-it</span> <span class="pre">paddlepaddle/paddle:[tag]</span> <span class="pre">/bin/bash</span></code>):</p>
<div class="highlight-bash"><div class="highlight"><pre><span></span>$ paddle version
PaddlePaddle <span class="m">0</span>.10.0rc, compiled with
    with_avx: ON
    with_gpu: OFF
    with_double: OFF
    with_python: ON
    with_rdma: OFF
    with_timer: OFF
</pre></div>
</div>
<p>We&#8217;ll take <code class="docutils literal"><span class="pre">doc/howto/usage/cluster/src/word2vec</span></code> as an example to introduce distributed training using PaddlePaddle v2 API.</p>
</div>
<div class="section" id="command-line-arguments">
<span id="command-line-arguments"></span><h2>Command-line arguments<a class="headerlink" href="#command-line-arguments" title="Permalink to this headline"></a></h2>
<div class="section" id="starting-parameter-server">
<span id="starting-parameter-server"></span><h3>Starting parameter server<a class="headerlink" href="#starting-parameter-server" title="Permalink to this headline"></a></h3>
<p>Type the below command to start a parameter server which will wait for trainers to connect:</p>
<div class="highlight-bash"><div class="highlight"><pre><span></span>$ paddle pserver --port<span class="o">=</span><span class="m">7164</span> --ports_num<span class="o">=</span><span class="m">1</span> --ports_num_for_sparse<span class="o">=</span><span class="m">1</span> --num_gradient_servers<span class="o">=</span><span class="m">1</span>
</pre></div>
</div>
<p>If you wish to run parameter servers in background, and save a log file, you can type:</p>
<div class="highlight-bash"><div class="highlight"><pre><span></span>$ stdbuf -oL /usr/bin/nohup paddle pserver --port<span class="o">=</span><span class="m">7164</span> --ports_num<span class="o">=</span><span class="m">1</span> --ports_num_for_sparse<span class="o">=</span><span class="m">1</span> --num_gradient_servers<span class="o">=</span><span class="m">1</span> <span class="p">&amp;</span>&gt; pserver.log
</pre></div>
</div>
<p>| param  | required | default | description |
| &#8212;&#8212;&#8212;&#8212;- | &#8212;&#8212;&#8212;&#8212;- | &#8212;&#8212;&#8212;&#8212;- | &#8212;&#8212;&#8212;&#8212;- |
| port  | required | 7164 | port which parameter server will listen on. If ports_num greater than 1, parameter server will listen on multiple ports for more network throughput |
| ports_num  | required | 1 | total number of ports will listen on  |
| ports_num_for_sparse  | required | 1 | number of ports which serves sparse parameter update  |
| num_gradient_servers  | required | 1 | total number of gradient servers |</p>
</div>
<div class="section" id="starting-trainer">
<span id="starting-trainer"></span><h3>Starting trainer<a class="headerlink" href="#starting-trainer" title="Permalink to this headline"></a></h3>
<p>Type the command below to start the trainer(name the file whatever you want, like &#8220;train.py&#8221;)</p>
<div class="highlight-bash"><div class="highlight"><pre><span></span>$ python train.py
</pre></div>
</div>
<p>Trainers&#8217; network need to be connected with parameter servers&#8217; network to finish the job. Trainers need to know port and IPs to locate parameter servers. You can pass arguments to trainers through <a class="reference external" href="https://en.wikipedia.org/wiki/Environment_variable">environment variables</a> or pass to <code class="docutils literal"><span class="pre">paddle.init()</span></code> function. Arguments passed to the <code class="docutils literal"><span class="pre">paddle.init()</span></code> function will overwrite environment variables.</p>
<p>Use environment viriables:</p>
<div class="highlight-bash"><div class="highlight"><pre><span></span><span class="nb">export</span> <span class="nv">PADDLE_INIT_USE_GPU</span><span class="o">=</span>False
<span class="nb">export</span> <span class="nv">PADDLE_INIT_TRAINER_COUNT</span><span class="o">=</span><span class="m">1</span>
<span class="nb">export</span> <span class="nv">PADDLE_INIT_PORT</span><span class="o">=</span><span class="m">7164</span>
<span class="nb">export</span> <span class="nv">PADDLE_INIT_PORTS_NUM</span><span class="o">=</span><span class="m">1</span>
<span class="nb">export</span> <span class="nv">PADDLE_INIT_PORTS_NUM_FOR_SPARSE</span><span class="o">=</span><span class="m">1</span>
<span class="nb">export</span> <span class="nv">PADDLE_INIT_NUM_GRADIENT_SERVERS</span><span class="o">=</span><span class="m">1</span>
<span class="nb">export</span> <span class="nv">PADDLE_INIT_TRAINER_ID</span><span class="o">=</span><span class="m">0</span>
<span class="nb">export</span> <span class="nv">PADDLE_INIT_PSERVERS</span><span class="o">=</span><span class="m">127</span>.0.0.1
python train.py
</pre></div>
</div>
<p>Pass arguments:</p>
<div class="highlight-python"><div class="highlight"><pre><span></span><span class="n">paddle</span><span class="o">.</span><span class="n">init</span><span class="p">(</span>
        <span class="n">use_gpu</span><span class="o">=</span><span class="bp">False</span><span class="p">,</span>
        <span class="n">trainer_count</span><span class="o">=</span><span class="mi">1</span><span class="p">,</span>
        <span class="n">port</span><span class="o">=</span><span class="mi">7164</span><span class="p">,</span>
        <span class="n">ports_num</span><span class="o">=</span><span class="mi">1</span><span class="p">,</span>
        <span class="n">ports_num_for_sparse</span><span class="o">=</span><span class="mi">1</span><span class="p">,</span>
        <span class="n">num_gradient_servers</span><span class="o">=</span><span class="mi">1</span><span class="p">,</span>
        <span class="n">trainer_id</span><span class="o">=</span><span class="mi">0</span><span class="p">,</span>
        <span class="n">pservers</span><span class="o">=</span><span class="s2">&quot;127.0.0.1&quot;</span><span class="p">)</span>
</pre></div>
T
Deploy to GitHub Pages: 2c98becba4fb48311b80c53c1793fae47e7fe975  
Travis CI 已提交
321 
</div>
T
Deploy to GitHub Pages: 1ff47b284c565d030b198705d5f18b4bd4ce53e5  
Travis CI 已提交
322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 
<p>| param  | required | default | description |
| &#8212;&#8212;&#8212;&#8212;- | &#8212;&#8212;&#8212;&#8212;- | &#8212;&#8212;&#8212;&#8212;- | &#8212;&#8212;&#8212;&#8212;- |
| use_gpu  | optional | False | set to &#8220;True&#8221; to enable GPU training |
| trainer_count  | required | 1 | total count of trainers in the training job |
| port  | required | 7164 | port to connect to parameter server  |
| ports_num  | required | 1 | number of ports for communication |
| ports_num_for_sparse  | required | 1 | number of ports for sparse type caculation |
| num_gradient_servers  | required | 1 | total number of gradient server |
| trainer_id  | required | 0 | ID for every trainer, start from 0 |
| pservers  | required | 127.0.0.1 | list of IPs of parameter servers, separated by &#8221;,&#8221; |</p>
</div>
<div class="section" id="prepare-training-dataset">
<span id="prepare-training-dataset"></span><h3>Prepare Training Dataset<a class="headerlink" href="#prepare-training-dataset" title="Permalink to this headline"></a></h3>
<p>Here&#8217;s some example code <a class="reference external" href="https://github.com/PaddlePaddle/Paddle/tree/develop/doc/howto/usage/cluster/src/word2vec/prepare.py">prepare.py</a>, it will download public <code class="docutils literal"><span class="pre">imikolov</span></code> dataset and split it into multiple files according to job parallelism(trainers count). Modify <code class="docutils literal"><span class="pre">SPLIT_COUNT</span></code> at the begining of <code class="docutils literal"><span class="pre">prepare.py</span></code> to change the count of output files.</p>
<p>In the real world, we often use <code class="docutils literal"><span class="pre">MapReduce</span></code> job&#8217;s output as training data, so there will be lots of files. You can use <code class="docutils literal"><span class="pre">mod</span></code> to assign training file to trainers:</p>
<div class="highlight-python"><div class="highlight"><pre><span></span><span class="kn">import</span> <span class="nn">os</span>
<span class="n">train_list</span> <span class="o">=</span> <span class="p">[]</span>
<span class="n">flist</span> <span class="o">=</span> <span class="n">os</span><span class="o">.</span><span class="n">listdir</span><span class="p">(</span><span class="s2">&quot;/train_data/&quot;</span><span class="p">)</span>
<span class="k">for</span> <span class="n">f</span> <span class="ow">in</span> <span class="n">flist</span><span class="p">:</span>
  <span class="n">suffix</span> <span class="o">=</span> <span class="nb">int</span><span class="p">(</span><span class="n">f</span><span class="o">.</span><span class="n">split</span><span class="p">(</span><span class="s2">&quot;-&quot;</span><span class="p">)[</span><span class="mi">1</span><span class="p">])</span>
  <span class="k">if</span> <span class="n">suffix</span> <span class="o">%</span> <span class="n">TRAINER_COUNT</span> <span class="o">==</span> <span class="n">TRAINER_ID</span><span class="p">:</span>
    <span class="n">train_list</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="n">f</span><span class="p">)</span>
</pre></div>
</div>
<p>Example code <code class="docutils literal"><span class="pre">prepare.py</span></code> will split training data and testing data into 3 files with digital suffix like <code class="docutils literal"><span class="pre">-00000</span></code>, <code class="docutils literal"><span class="pre">-00001</span></code> and<code class="docutils literal"><span class="pre">-00002</span></code>:</p>
<div class="highlight-default"><div class="highlight"><pre><span></span><span class="n">train</span><span class="o">.</span><span class="n">txt</span>
<span class="n">train</span><span class="o">.</span><span class="n">txt</span><span class="o">-</span><span class="mi">00000</span>
<span class="n">train</span><span class="o">.</span><span class="n">txt</span><span class="o">-</span><span class="mi">00001</span>
<span class="n">train</span><span class="o">.</span><span class="n">txt</span><span class="o">-</span><span class="mi">00002</span>
<span class="n">test</span><span class="o">.</span><span class="n">txt</span>
<span class="n">test</span><span class="o">.</span><span class="n">txt</span><span class="o">-</span><span class="mi">00000</span>
<span class="n">test</span><span class="o">.</span><span class="n">txt</span><span class="o">-</span><span class="mi">00001</span>
<span class="n">test</span><span class="o">.</span><span class="n">txt</span><span class="o">-</span><span class="mi">00002</span>
</pre></div>
</div>
<p>When job started, every trainer needs to get it&#8217;s own part of data. In some distributed systems a storage service will be provided, so the date under that path can be accessed by all the trainer nodes. Without the storage service, you must copy the training data to each trainer node.</p>
<p>Different training jobs may have different data format and <code class="docutils literal"><span class="pre">reader()</span></code> function, developers may need to write different data prepare scripts and <code class="docutils literal"><span class="pre">reader()</span></code> functions for their job.</p>
</div>
<div class="section" id="prepare-training-program">
<span id="prepare-training-program"></span><h3>Prepare Training program<a class="headerlink" href="#prepare-training-program" title="Permalink to this headline"></a></h3>
<p>We&#8217;ll create a <em>workspace</em> directory on each node, storing your training program, dependencies, mounted or downloaded dataset directory.</p>
<p>Your workspace may looks like:</p>
T
Deploy to GitHub Pages: 2c98becba4fb48311b80c53c1793fae47e7fe975  
Travis CI 已提交
364 
<div class="highlight-default"><div class="highlight"><pre><span></span>.
T
Deploy to GitHub Pages: 1ff47b284c565d030b198705d5f18b4bd4ce53e5  
Travis CI 已提交
365 366 367 368 369 370 371 372 373 374 375 
|-- my_lib.py
|-- word_dict.pickle
|-- train.py
|-- train_data_dir/
|   |-- train.txt-00000
|   |-- train.txt-00001
|   |-- train.txt-00002
`-- test_data_dir/
    |-- test.txt-00000
    |-- test.txt-00001
    `-- test.txt-00002
T
Deploy to GitHub Pages: 2c98becba4fb48311b80c53c1793fae47e7fe975  
Travis CI 已提交
376 377 
</pre></div>
</div>
T
Deploy to GitHub Pages: 1ff47b284c565d030b198705d5f18b4bd4ce53e5  
Travis CI 已提交
378 379 380 381 382 383 384 385 386 387 388 
<ul>
<li><p class="first"><code class="docutils literal"><span class="pre">my_lib.py</span></code>: user defined libraries, like PIL libs. This is optional.</p>
</li>
<li><p class="first"><code class="docutils literal"><span class="pre">word_dict.pickle</span></code>: dict file for training word embeding.</p>
</li>
<li><p class="first"><code class="docutils literal"><span class="pre">train.py</span></code>: training program. Sample code: <a class="reference external" href="https://github.com/PaddlePaddle/Paddle/tree/develop/doc/howto/usage/cluster/src/word2vec/prepare.py">api_train_v2_cluster.py</a>. <strong><em>NOTE:</em></strong> You may need to modify the head part of <code class="docutils literal"><span class="pre">train.py</span></code> when using different cluster platform to retrive configuration environment variables:</p>
<div class="highlight-python"><div class="highlight"><pre><span></span><span class="n">cluster_train_file</span> <span class="o">=</span> <span class="s2">&quot;./train_data_dir/train/train.txt&quot;</span>
<span class="n">cluster_test_file</span> <span class="o">=</span> <span class="s2">&quot;./test_data_dir/test/test.txt&quot;</span>
<span class="n">node_id</span> <span class="o">=</span> <span class="n">os</span><span class="o">.</span><span class="n">getenv</span><span class="p">(</span><span class="s2">&quot;OMPI_COMM_WORLD_RANK&quot;</span><span class="p">)</span>
<span class="k">if</span> <span class="ow">not</span> <span class="n">node_id</span><span class="p">:</span>
    <span class="k">raise</span> <span class="ne">EnvironmentError</span><span class="p">(</span><span class="s2">&quot;must provied OMPI_COMM_WORLD_RANK&quot;</span><span class="p">)</span>
T
Deploy to GitHub Pages: 2c98becba4fb48311b80c53c1793fae47e7fe975  
Travis CI 已提交
389 390 
</pre></div>
</div>
T
Deploy to GitHub Pages: 1ff47b284c565d030b198705d5f18b4bd4ce53e5  
Travis CI 已提交
391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 
</li>
<li><p class="first"><code class="docutils literal"><span class="pre">train_data_dir</span></code>: containing training data. Mount from storage service or copy trainning data to here.</p>
</li>
<li><p class="first"><code class="docutils literal"><span class="pre">test_data_dir</span></code>: containing testing data.</p>
</li>
</ul>
</div>
</div>
<div class="section" id="use-cluster-platforms-or-cluster-management-tools">
<span id="use-cluster-platforms-or-cluster-management-tools"></span><h2>Use cluster platforms or cluster management tools<a class="headerlink" href="#use-cluster-platforms-or-cluster-management-tools" title="Permalink to this headline"></a></h2>
<p>PaddlePaddle supports running jobs on several platforms including:</p>
<ul class="simple">
<li><a class="reference external" href="http://kubernetes.io">Kubernetes</a> open-source system for automating deployment, scaling, and management of containerized applications from Google.</li>
<li><a class="reference external" href="https://www.open-mpi.org">OpenMPI</a> Mature high performance parallel computing framework.</li>
<li><a class="reference external" href="http://www.fabfile.org">Fabric</a> A cluster management tool. Write scripts to submit jobs or manage the cluster.</li>
</ul>
<p>We&#8217;ll introduce cluster job management on these platforms. The examples can be found under <a class="reference external" href="https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/scripts/cluster_train_v2">cluster_train_v2</a>.</p>
<p>These cluster platforms provide API or environment variables for training processes, when the job is dispatched to different nodes. Like node ID, IP or total number of nodes etc.</p>
<div class="section" id="cluster-training-using-fabric">
<span id="cluster-training-using-fabric"></span><h3>Cluster Training Using Fabric<a class="headerlink" href="#cluster-training-using-fabric" title="Permalink to this headline"></a></h3>
<div class="section" id="prepare-a-linux-cluster">
<span id="prepare-a-linux-cluster"></span><h4>Prepare a Linux cluster<a class="headerlink" href="#prepare-a-linux-cluster" title="Permalink to this headline"></a></h4>
<p>Run <code class="docutils literal"><span class="pre">kubectl</span> <span class="pre">-f</span> <span class="pre">ssh_servers.yaml</span></code> under the directory:  <code class="docutils literal"><span class="pre">paddle/scripts/cluster_train_v2/fabric/docker_cluster</span></code> will launch a demo cluster. Run <code class="docutils literal"><span class="pre">kubectl</span> <span class="pre">get</span> <span class="pre">po</span> <span class="pre">-o</span> <span class="pre">wide</span></code> to get IP addresses of these nodes.</p>
</div>
T
Deploy to GitHub Pages: 2c98becba4fb48311b80c53c1793fae47e7fe975  
Travis CI 已提交
415 
<div class="section" id="launching-cluster-job">
T
Deploy to GitHub Pages: 1ff47b284c565d030b198705d5f18b4bd4ce53e5  
Travis CI 已提交
416 417 
<span id="launching-cluster-job"></span><h4>Launching Cluster Job<a class="headerlink" href="#launching-cluster-job" title="Permalink to this headline"></a></h4>
<p><code class="docutils literal"><span class="pre">paddle.py</span></code> provides automatical scripts to start all PaddlePaddle cluster processes in different nodes. By default, all command line options can be set as <code class="docutils literal"><span class="pre">paddle.py</span></code> command options and <code class="docutils literal"><span class="pre">paddle.py</span></code> will transparently and automatically set these options to PaddlePaddle lower level processes.</p>
T
Deploy to GitHub Pages: 2c98becba4fb48311b80c53c1793fae47e7fe975  
Travis CI 已提交
418 
<p><code class="docutils literal"><span class="pre">paddle.py</span></code>provides two distinguished command option for easy job launching.</p>
T
Deploy to GitHub Pages: 1ff47b284c565d030b198705d5f18b4bd4ce53e5  
Travis CI 已提交
419 420 421 422 423 
<ul class="simple">
<li><code class="docutils literal"><span class="pre">job_dispatch_package</span></code> set it with local <code class="docutils literal"><span class="pre">workspace</span></code> directory, it will be dispatched to all nodes which is set in <code class="docutils literal"><span class="pre">conf.py</span></code>. It could be helpful for frequently manipulating workspace files. otherwise, frequent multi-nodes workspace deployment is very annoying.</li>
<li><code class="docutils literal"><span class="pre">job_workspace</span></code>  set it with already deployed workspace directory, <code class="docutils literal"><span class="pre">paddle.py</span></code> will skip dispatch stage to directly launch cluster job with all nodes. It could help to reduce heavy
dispatch latency.</li>
</ul>
T
Deploy to GitHub Pages: 2c98becba4fb48311b80c53c1793fae47e7fe975  
Travis CI 已提交
424 425 426 427 428 429 430 
<p><code class="docutils literal"><span class="pre">cluster_train/run.sh</span></code> provides command line sample to run <code class="docutils literal"><span class="pre">demo/recommendation</span></code> cluster job, just modify <code class="docutils literal"><span class="pre">job_dispatch_package</span></code> and <code class="docutils literal"><span class="pre">job_workspace</span></code> with your defined directory, then:</p>
<div class="highlight-default"><div class="highlight"><pre><span></span><span class="n">sh</span> <span class="n">run</span><span class="o">.</span><span class="n">sh</span>
</pre></div>
</div>
<p>The cluster Job will start in several seconds.</p>
</div>
<div class="section" id="kill-cluster-job">
T
Deploy to GitHub Pages: 1ff47b284c565d030b198705d5f18b4bd4ce53e5  
Travis CI 已提交
431 432 
<span id="kill-cluster-job"></span><h4>Kill Cluster Job<a class="headerlink" href="#kill-cluster-job" title="Permalink to this headline"></a></h4>
<p><code class="docutils literal"><span class="pre">paddle.py</span></code> can capture <code class="docutils literal"><span class="pre">Ctrl</span> <span class="pre">+</span> <span class="pre">C</span></code> SIGINT signal to automatically kill all processes launched by it. So just stop <code class="docutils literal"><span class="pre">paddle.py</span></code> to kill cluster job. You should manually kill the job if the program crashed.</p>
T
Deploy to GitHub Pages: 2c98becba4fb48311b80c53c1793fae47e7fe975  
Travis CI 已提交
433 434 
</div>
<div class="section" id="check-cluster-training-result">
T
Deploy to GitHub Pages: 1ff47b284c565d030b198705d5f18b4bd4ce53e5  
Travis CI 已提交
435 
<span id="check-cluster-training-result"></span><h4>Check Cluster Training Result<a class="headerlink" href="#check-cluster-training-result" title="Permalink to this headline"></a></h4>
T
Deploy to GitHub Pages: 2c98becba4fb48311b80c53c1793fae47e7fe975  
Travis CI 已提交
436 437 
<p>Check log in $workspace/log for details, each node owns same log structure.</p>
<p><code class="docutils literal"><span class="pre">paddle_trainer.INFO</span></code>
T
Deploy to GitHub Pages: 1ff47b284c565d030b198705d5f18b4bd4ce53e5  
Travis CI 已提交
438 
It provides almost all internal output log for training,  same as local training. Check runtime model convergence here.</p>
T
Deploy to GitHub Pages: 2c98becba4fb48311b80c53c1793fae47e7fe975  
Travis CI 已提交
439 
<p><code class="docutils literal"><span class="pre">paddle_pserver2.INFO</span></code>
T
Deploy to GitHub Pages: 1ff47b284c565d030b198705d5f18b4bd4ce53e5  
Travis CI 已提交
440 
It provides parameter server running log, which could help to diagnose distributed error.</p>
T
Deploy to GitHub Pages: 2c98becba4fb48311b80c53c1793fae47e7fe975  
Travis CI 已提交
441 
<p><code class="docutils literal"><span class="pre">server.log</span></code>
T
Deploy to GitHub Pages: 1ff47b284c565d030b198705d5f18b4bd4ce53e5  
Travis CI 已提交
442 
It provides stderr and stdout of parameter server process. Check error log if training crashes.</p>
T
Deploy to GitHub Pages: 2c98becba4fb48311b80c53c1793fae47e7fe975  
Travis CI 已提交
443 
<p><code class="docutils literal"><span class="pre">train.log</span></code>
T
Deploy to GitHub Pages: 1ff47b284c565d030b198705d5f18b4bd4ce53e5  
Travis CI 已提交
444 
It provides stderr and stdout of trainer process. Check error log if training crashes.</p>
T
Deploy to GitHub Pages: 2c98becba4fb48311b80c53c1793fae47e7fe975  
Travis CI 已提交
445 446 
</div>
<div class="section" id="check-model-output">
T
Deploy to GitHub Pages: 1ff47b284c565d030b198705d5f18b4bd4ce53e5  
Travis CI 已提交
447 448 
<span id="check-model-output"></span><h4>Check Model Output<a class="headerlink" href="#check-model-output" title="Permalink to this headline"></a></h4>
<p>After one pass finished, model files will be written in <code class="docutils literal"><span class="pre">output</span></code> directory in node 0.
T
Deploy to GitHub Pages: 2c98becba4fb48311b80c53c1793fae47e7fe975  
Travis CI 已提交
449 450 451 
<code class="docutils literal"><span class="pre">nodefile</span></code> in workspace indicates the node id of current cluster job.</p>
</div>
</div>
T
Deploy to GitHub Pages: 1ff47b284c565d030b198705d5f18b4bd4ce53e5  
Travis CI 已提交
452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 
<div class="section" id="cluster-training-using-openmpi">
<span id="cluster-training-using-openmpi"></span><h3>Cluster Training Using OpenMPI<a class="headerlink" href="#cluster-training-using-openmpi" title="Permalink to this headline"></a></h3>
<div class="section" id="prepare-an-openmpi-cluster">
<span id="prepare-an-openmpi-cluster"></span><h4>Prepare an OpenMPI cluster<a class="headerlink" href="#prepare-an-openmpi-cluster" title="Permalink to this headline"></a></h4>
<p>Run the following command to start a 3-node MPI cluster and one &#8220;head&#8221; node.</p>
<div class="highlight-bash"><div class="highlight"><pre><span></span><span class="nb">cd</span> paddle/scripts/cluster_train_v2/openmpi/docker_cluster
kubectl create -f head.yaml
kubectl create -f mpi-nodes.yaml
</pre></div>
</div>
<p>Then you can log in to every OpenMPI node using ssh without input any passwords.</p>
</div>
<div class="section" id="launching-cluster-job">
<span id="id1"></span><h4>Launching Cluster Job<a class="headerlink" href="#launching-cluster-job" title="Permalink to this headline"></a></h4>
<p>Follow the steps to launch a PaddlePaddle training job in OpenMPI cluster:\</p>
<div class="highlight-bash"><div class="highlight"><pre><span></span><span class="c1"># find out node IP addresses</span>
kubectl get po -o wide
<span class="c1"># generate a &quot;machines&quot; file containing node IP addresses</span>
kubectl get po -o wide <span class="p">|</span> grep nodes <span class="p">|</span> awk <span class="s1">&#39;{print $6}&#39;</span> &gt; machines
<span class="c1"># copy necessary files onto &quot;head&quot; node</span>
scp -i ssh/id_rsa.mpi.pub machines prepare.py train.py start_mpi_train.sh tutorial@<span class="o">[</span>headIP<span class="o">]</span>:~
<span class="c1"># login to head node using ssh</span>
ssh -i ssh/id_rsa.mpi.pub tutorial@<span class="o">[</span>headIP<span class="o">]</span>
<span class="c1"># --------------- in head node ---------------</span>
<span class="c1"># prepare training data</span>
python prepare.py
<span class="c1"># copy training data and dict file to MPI nodes</span>
cat machines <span class="p">|</span> xargs -i scp word_dict.pickle train.py start_mpi_train.sh machines <span class="o">{}</span>:/home/tutorial
<span class="c1"># creat a directory for storing log files</span>
mpirun -hostfile machines -n <span class="m">3</span> mkdir /home/tutorial/logs
<span class="c1"># copy training data to every node</span>
scp train.txt-00000 test.txt-00000 <span class="o">[</span>node1IP<span class="o">]</span>:/home/tutorial
scp train.txt-00001 test.txt-00001 <span class="o">[</span>node2IP<span class="o">]</span>:/home/tutorial
scp train.txt-00002 test.txt-00002 <span class="o">[</span>node3IP<span class="o">]</span>:/home/tutorial
<span class="c1"># start the job</span>
mpirun -hostfile machines -n <span class="m">3</span>  /home/tutorial/start_mpi_train.sh
</pre></div>
</div>
</div>
</div>
<div class="section" id="cluster-training-using-kubernetes">
<span id="cluster-training-using-kubernetes"></span><h3>Cluster Training Using Kubernetes<a class="headerlink" href="#cluster-training-using-kubernetes" title="Permalink to this headline"></a></h3>
<p>The details can be found <span class="xref doc">here</span></p>
</div>
</div>
T
Deploy to GitHub Pages: 2c98becba4fb48311b80c53c1793fae47e7fe975  
Travis CI 已提交
497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 
</div>


           </div>
          </div>
          <footer>
  
    <div class="rst-footer-buttons" role="navigation" aria-label="footer navigation">
      
        <a href="../k8s/k8s_en.html" class="btn btn-neutral float-right" title="Paddle On Kubernetes" accesskey="n">Next <span class="fa fa-arrow-circle-right"></span></a>
      
      
        <a href="../cmd_parameter/detail_introduction_en.html" class="btn btn-neutral" title="Detail Description" accesskey="p"><span class="fa fa-arrow-circle-left"></span> Previous</a>
      
    </div>
  

  <hr/>

  <div role="contentinfo">
    <p>
        &copy; Copyright 2016, PaddlePaddle developers.

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

</footer>

        </div>
      </div>

    </section>

  </div>
  


  

    <script type="text/javascript">
        var DOCUMENTATION_OPTIONS = {
            URL_ROOT:'../../../',
            VERSION:'',
            COLLAPSE_INDEX:false,
            FILE_SUFFIX:'.html',
            HAS_SOURCE:  true,
            SOURCELINK_SUFFIX: ".txt",
        };
    </script>
      <script type="text/javascript" src="../../../_static/jquery.js"></script>
      <script type="text/javascript" src="../../../_static/underscore.js"></script>
      <script type="text/javascript" src="../../../_static/doctools.js"></script>
      <script type="text/javascript" src="https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.0/MathJax.js?config=TeX-AMS-MML_HTMLorMML"></script>
       
  

  
  
    <script type="text/javascript" src="../../../_static/js/theme.js"></script>
  
  
  <script src="https://maxcdn.bootstrapcdn.com/bootstrap/3.3.7/js/bootstrap.min.js" integrity="sha384-Tc5IQib027qvyjSMfHjOMaLkfuWVxZxUPnCJA7l2mCWNIpG9mGCD8wGNIcPD7Txa" crossorigin="anonymous"></script>
  <script src="https://cdn.jsdelivr.net/perfect-scrollbar/0.6.14/js/perfect-scrollbar.jquery.min.js"></script>
  <script src="../../../_static/js/paddle_doc_init.js"></script> 

</body>
</html>