Skip to content

P
Paddle

BaiXuePrincess / Paddle
与 Fork 源项目一致

Fork自 PaddlePaddle / Paddle

通知 1
Star 1
Fork 0
P
Paddle
切换分支/标签
查找文件 普通视图历史永久链接Permalink
use_case.html 25.0 KB
Newer Older
Y
Update docs  
Yu Yang 已提交
1 2 3 4 5 6 7 8 
<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN"
  "http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd">


<html xmlns="http://www.w3.org/1999/xhtml">
  <head>
    <meta http-equiv="Content-Type" content="text/html; charset=utf-8" />
T
Deploy to GitHub Pages: 703cce35a871c78a5e3f1c33abf7fe788e2bbda5  
Travis CI 已提交
9 
    <title>Use Case &#8212; PaddlePaddle  documentation</title>
Y
Update docs  
Yu Yang 已提交
10 11 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 
    
    <link rel="stylesheet" href="../../_static/classic.css" type="text/css" />
    <link rel="stylesheet" href="../../_static/pygments.css" type="text/css" />
    
    <script type="text/javascript">
      var DOCUMENTATION_OPTIONS = {
        URL_ROOT:    '../../',
        VERSION:     '',
        COLLAPSE_INDEX: false,
        FILE_SUFFIX: '.html',
        HAS_SOURCE:  true
      };
    </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://cdn.mathjax.org/mathjax/latest/MathJax.js?config=TeX-AMS-MML_HTMLorMML"></script>
    <link rel="top" title="PaddlePaddle  documentation" href="../../index.html" />
    <link rel="up" title="User Interface" href="../index.html" />
    <link rel="next" title="Argument Outline" href="argument_outline.html" />
    <link rel="prev" title="Parameter and Extra Layer Attribute" href="../api/trainer_config_helpers/attrs.html" /> 
  </head>
  <body role="document">
    <div class="related" role="navigation" aria-label="related navigation">
      <h3>Navigation</h3>
      <ul>
        <li class="right" style="margin-right: 10px">
          <a href="../../genindex.html" title="General Index"
             accesskey="I">index</a></li>
        <li class="right" >
          <a href="../../py-modindex.html" title="Python Module Index"
             >modules</a> |</li>
        <li class="right" >
          <a href="argument_outline.html" title="Argument Outline"
             accesskey="N">next</a> |</li>
        <li class="right" >
          <a href="../api/trainer_config_helpers/attrs.html" title="Parameter and Extra Layer Attribute"
             accesskey="P">previous</a> |</li>
T
Deploy to GitHub Pages: 703cce35a871c78a5e3f1c33abf7fe788e2bbda5  
Travis CI 已提交
48 49 
        <li class="nav-item nav-item-0"><a href="../../index.html">PaddlePaddle  documentation</a> &#187;</li>
          <li class="nav-item nav-item-1"><a href="../index.html" accesskey="U">User Interface</a> &#187;</li>
Y
Update docs  
Yu Yang 已提交
50 51 52 53 54 55 56 57 58 59 60 61 62 
      </ul>
    </div>  

    <div class="document">
      <div class="documentwrapper">
        <div class="bodywrapper">
          <div class="body" role="main">
            
  <div class="section" id="use-case">
<span id="use-case"></span><h1>Use Case<a class="headerlink" href="#use-case" title="Permalink to this headline"></a></h1>
<div class="section" id="local-training">
<span id="local-training"></span><h2>Local Training<a class="headerlink" href="#local-training" title="Permalink to this headline"></a></h2>
<p>These command line arguments are commonly used by local training experiments, such as image classification, natural language processing, et al.</p>
T
Deploy to GitHub Pages: 703cce35a871c78a5e3f1c33abf7fe788e2bbda5  
Travis CI 已提交
63 64 65 66 67 68 69 70 71 72 73 74 
<div class="highlight-default"><div class="highlight"><pre><span></span><span class="n">paddle</span> <span class="n">train</span> \
  <span class="o">--</span><span class="n">use_gpu</span><span class="o">=</span><span class="mi">1</span><span class="o">/</span><span class="mi">0</span> \                        <span class="c1">#1:GPU,0:CPU(default:true)</span>
  <span class="o">--</span><span class="n">config</span><span class="o">=</span><span class="n">network_config</span> \
  <span class="o">--</span><span class="n">save_dir</span><span class="o">=</span><span class="n">output</span> \
  <span class="o">--</span><span class="n">trainer_count</span><span class="o">=</span><span class="n">COUNT</span> \                <span class="c1">#(default:1)</span>
  <span class="o">--</span><span class="n">test_period</span><span class="o">=</span><span class="n">M</span> \                      <span class="c1">#(default:1000）</span>
  <span class="o">--</span><span class="n">test_all_data_in_one_period</span><span class="o">=</span><span class="n">true</span> \   <span class="c1">#(default:false) </span>
  <span class="o">--</span><span class="n">num_passes</span><span class="o">=</span><span class="n">N</span> \                       <span class="c1">#(defalut:100）</span>
  <span class="o">--</span><span class="n">log_period</span><span class="o">=</span><span class="n">K</span> \                       <span class="c1">#(default:100)</span>
  <span class="o">--</span><span class="n">dot_period</span><span class="o">=</span><span class="mi">1000</span> \                    <span class="c1">#(default:1)</span>
  <span class="c1">#[--show_parameter_stats_period=100] \ #(default:0)</span>
  <span class="c1">#[--saving_period_by_batches=200] \    #(default:0)</span>
Y
Update docs  
Yu Yang 已提交
75 76 77 78 79 80 
</pre></div>
</div>
<p><code class="docutils literal"><span class="pre">show_parameter_stats_period</span></code> and <code class="docutils literal"><span class="pre">saving_period_by_batches</span></code> are optional according to your task.</p>
<div class="section" id="pass-command-argument-to-network-config">
<span id="pass-command-argument-to-network-config"></span><h3>1) Pass Command Argument to Network config<a class="headerlink" href="#pass-command-argument-to-network-config" title="Permalink to this headline"></a></h3>
<p><code class="docutils literal"><span class="pre">config_args</span></code> is a useful parameter to pass arguments to network config.</p>
T
Deploy to GitHub Pages: 703cce35a871c78a5e3f1c33abf7fe788e2bbda5  
Travis CI 已提交
81 
<div class="highlight-default"><div class="highlight"><pre><span></span><span class="o">--</span><span class="n">config_args</span><span class="o">=</span><span class="n">generating</span><span class="o">=</span><span class="mi">1</span><span class="p">,</span><span class="n">beam_size</span><span class="o">=</span><span class="mi">5</span><span class="p">,</span><span class="n">layer_num</span><span class="o">=</span><span class="mi">10</span> \
Y
Update docs  
Yu Yang 已提交
82 83 84 
</pre></div>
</div>
<p>And <code class="docutils literal"><span class="pre">get_config_arg</span></code> can be used to parse these arguments in network config as follows:</p>
T
Deploy to GitHub Pages: 703cce35a871c78a5e3f1c33abf7fe788e2bbda5  
Travis CI 已提交
85 
<div class="highlight-default"><div class="highlight"><pre><span></span><span class="n">generating</span> <span class="o">=</span> <span class="n">get_config_arg</span><span class="p">(</span><span class="s1">&#39;generating&#39;</span><span class="p">,</span> <span class="nb">bool</span><span class="p">,</span> <span class="kc">False</span><span class="p">)</span>
Y
Update docs  
Yu Yang 已提交
86 87 88 89 90 
<span class="n">beam_size</span> <span class="o">=</span> <span class="n">get_config_arg</span><span class="p">(</span><span class="s1">&#39;beam_size&#39;</span><span class="p">,</span> <span class="nb">int</span><span class="p">,</span> <span class="mi">3</span><span class="p">)</span>
<span class="n">layer_num</span> <span class="o">=</span> <span class="n">get_config_arg</span><span class="p">(</span><span class="s1">&#39;layer_num&#39;</span><span class="p">,</span> <span class="nb">int</span><span class="p">,</span> <span class="mi">8</span><span class="p">)</span>
</pre></div>
</div>
<p><code class="docutils literal"><span class="pre">get_config_arg</span></code>:</p>
T
Deploy to GitHub Pages: 703cce35a871c78a5e3f1c33abf7fe788e2bbda5  
Travis CI 已提交
91 
<div class="highlight-default"><div class="highlight"><pre><span></span><span class="n">get_config_arg</span><span class="p">(</span><span class="n">name</span><span class="p">,</span> <span class="nb">type</span><span class="p">,</span> <span class="n">default_value</span><span class="p">)</span>
Y
Update docs  
Yu Yang 已提交
92 93 94 95 96 97 98 99 100 101 102 
</pre></div>
</div>
<ul class="simple">
<li>name: the name specified in the <code class="docutils literal"><span class="pre">--config_args</span></code></li>
<li>type: value type, bool, int, str, float etc.</li>
<li>default_value: default value if not set.</li>
</ul>
</div>
<div class="section" id="use-model-to-initialize-network">
<span id="use-model-to-initialize-network"></span><h3>2) Use Model to Initialize Network<a class="headerlink" href="#use-model-to-initialize-network" title="Permalink to this headline"></a></h3>
<p>add argument:</p>
T
Deploy to GitHub Pages: 703cce35a871c78a5e3f1c33abf7fe788e2bbda5  
Travis CI 已提交
103 
<div class="highlight-default"><div class="highlight"><pre><span></span><span class="o">--</span><span class="n">init_model_path</span><span class="o">=</span><span class="n">model_path</span>
Y
Update docs  
Yu Yang 已提交
104 105 106 107 108 109 110 111 
<span class="o">--</span><span class="n">load_missing_parameter_strategy</span><span class="o">=</span><span class="n">rand</span>
</pre></div>
</div>
</div>
</div>
<div class="section" id="local-testing">
<span id="local-testing"></span><h2>Local Testing<a class="headerlink" href="#local-testing" title="Permalink to this headline"></a></h2>
<p>Method 1:</p>
T
Deploy to GitHub Pages: 703cce35a871c78a5e3f1c33abf7fe788e2bbda5  
Travis CI 已提交
112 113 114 115 116 
<div class="highlight-default"><div class="highlight"><pre><span></span><span class="n">paddle</span> <span class="n">train</span> <span class="o">--</span><span class="n">job</span><span class="o">=</span><span class="n">test</span> \
             <span class="o">--</span><span class="n">use_gpu</span><span class="o">=</span><span class="mi">1</span><span class="o">/</span><span class="mi">0</span> \ 
             <span class="o">--</span><span class="n">config</span><span class="o">=</span><span class="n">network_config</span> \
             <span class="o">--</span><span class="n">trainer_count</span><span class="o">=</span><span class="n">COUNT</span> \ 
             <span class="o">--</span><span class="n">init_model_path</span><span class="o">=</span><span class="n">model_path</span> \
Y
Update docs  
Yu Yang 已提交
117 118 119 120 121 122 123 
</pre></div>
</div>
<ul class="simple">
<li>use init_model_path to specify test model.</li>
<li>only can test one model.</li>
</ul>
<p>Method 2:</p>
T
Deploy to GitHub Pages: 703cce35a871c78a5e3f1c33abf7fe788e2bbda5  
Travis CI 已提交
124 125 126 127 128 
<div class="highlight-default"><div class="highlight"><pre><span></span><span class="n">paddle</span> <span class="n">train</span> <span class="o">--</span><span class="n">job</span><span class="o">=</span><span class="n">test</span> \
             <span class="o">--</span><span class="n">use_gpu</span><span class="o">=</span><span class="mi">1</span><span class="o">/</span><span class="mi">0</span> \ 
             <span class="o">--</span><span class="n">config</span><span class="o">=</span><span class="n">network_config</span> \
             <span class="o">--</span><span class="n">trainer_count</span><span class="o">=</span><span class="n">COUNT</span> \ 
             <span class="o">--</span><span class="n">model_list</span><span class="o">=</span><span class="n">model</span><span class="o">.</span><span class="n">list</span> \
Y
Update docs  
Yu Yang 已提交
129 130 131 132 133 134 
</pre></div>
</div>
<ul class="simple">
<li>use model_list to specify test models</li>
<li>can test several models, where model.list likes:</li>
</ul>
T
Deploy to GitHub Pages: 703cce35a871c78a5e3f1c33abf7fe788e2bbda5  
Travis CI 已提交
135 136 
<div class="highlight-default"><div class="highlight"><pre><span></span><span class="o">./</span><span class="n">alexnet_pass1</span>
<span class="o">./</span><span class="n">alexnet_pass2</span>
Y
Update docs  
Yu Yang 已提交
137 138 139 
</pre></div>
</div>
<p>Method 3:</p>
T
Deploy to GitHub Pages: 703cce35a871c78a5e3f1c33abf7fe788e2bbda5  
Travis CI 已提交
140 141 142 143 144 145 146 
<div class="highlight-default"><div class="highlight"><pre><span></span><span class="n">paddle</span> <span class="n">train</span> <span class="o">--</span><span class="n">job</span><span class="o">=</span><span class="n">test</span> \
             <span class="o">--</span><span class="n">use_gpu</span><span class="o">=</span><span class="mi">1</span><span class="o">/</span><span class="mi">0</span> \
             <span class="o">--</span><span class="n">config</span><span class="o">=</span><span class="n">network_config</span> \
             <span class="o">--</span><span class="n">trainer_count</span><span class="o">=</span><span class="n">COUNT</span> \
             <span class="o">--</span><span class="n">save_dir</span><span class="o">=</span><span class="n">model</span> \
             <span class="o">--</span><span class="n">test_pass</span><span class="o">=</span><span class="n">M</span> \
             <span class="o">--</span><span class="n">num_passes</span><span class="o">=</span><span class="n">N</span> \
Y
Update docs  
Yu Yang 已提交
147 148 149 150 151 152 153 154 155 156 157 158 159 160 
</pre></div>
</div>
<p>This way must use model path saved by Paddle like this: <code class="docutils literal"><span class="pre">model/pass-%5d</span></code>. Testing model is from M-th pass to (N-1)-th pass. For example: M=12 and N=14 will test <code class="docutils literal"><span class="pre">model/pass-00012</span></code> and <code class="docutils literal"><span class="pre">model/pass-00013</span></code>.</p>
</div>
<div class="section" id="sparse-training">
<span id="sparse-training"></span><h2>Sparse Training<a class="headerlink" href="#sparse-training" title="Permalink to this headline"></a></h2>
<p>Sparse training is usually used to accelerate calculation when input is sparse data with highly dimension. For example, dictionary dimension of input data is 1 million, but one sample just have several words. In paddle, sparse matrix multiplication is used in forward propagation and sparse updating is perfomed on weight updating after backward propagation.</p>
<div class="section" id="local-training">
<span id="id1"></span><h3>1) Local training<a class="headerlink" href="#local-training" title="Permalink to this headline"></a></h3>
<p>You need to set <strong>sparse_update=True</strong> in network config.  Check the network config documentation for more details.</p>
</div>
<div class="section" id="cluster-training">
<span id="cluster-training"></span><h3>2) cluster training<a class="headerlink" href="#cluster-training" title="Permalink to this headline"></a></h3>
<p>Add the following argument for cluster training of a sparse model. At the same time you need to set <strong>sparse_remote_update=True</strong> in network config. Check the network config documentation for more details.</p>
T
Deploy to GitHub Pages: 703cce35a871c78a5e3f1c33abf7fe788e2bbda5  
Travis CI 已提交
161 
<div class="highlight-default"><div class="highlight"><pre><span></span><span class="o">--</span><span class="n">ports_num_for_sparse</span><span class="o">=</span><span class="mi">1</span>    <span class="c1">#(default: 0)</span>
Y
Update docs  
Yu Yang 已提交
162 163 164 165 166 167 168 169 
</pre></div>
</div>
</div>
</div>
<div class="section" id="parallel-nn">
<span id="parallel-nn"></span><h2>parallel_nn<a class="headerlink" href="#parallel-nn" title="Permalink to this headline"></a></h2>
<p><code class="docutils literal"><span class="pre">parallel_nn</span></code> can be set to mixed use of GPUs and CPUs to compute layers. That is to say, you can deploy network to use a GPU to compute some layers and use a CPU to compute other layers. The other way is to split layers into different GPUs, which can <strong>reduce GPU memory</strong> or <strong>use parallel computation to accelerate some layers</strong>.</p>
<p>If you want to use these characteristics, you need to specify device ID in network config (denote it as deviceId) and add command line argument:</p>
T
Deploy to GitHub Pages: 703cce35a871c78a5e3f1c33abf7fe788e2bbda5  
Travis CI 已提交
170 
<div class="highlight-default"><div class="highlight"><pre><span></span><span class="o">--</span><span class="n">parallel_nn</span><span class="o">=</span><span class="n">true</span>
Y
Update docs  
Yu Yang 已提交
171 172 173 174 175 
</pre></div>
</div>
<div class="section" id="case-1-mixed-use-of-gpu-and-cpu">
<span id="case-1-mixed-use-of-gpu-and-cpu"></span><h3>case 1: Mixed Use of GPU and CPU<a class="headerlink" href="#case-1-mixed-use-of-gpu-and-cpu" title="Permalink to this headline"></a></h3>
<p>Consider the following example:</p>
T
Deploy to GitHub Pages: 703cce35a871c78a5e3f1c33abf7fe788e2bbda5  
Travis CI 已提交
176 177 
<div class="highlight-default"><div class="highlight"><pre><span></span><span class="c1">#command line:</span>
<span class="n">paddle</span> <span class="n">train</span> <span class="o">--</span><span class="n">use_gpu</span><span class="o">=</span><span class="n">true</span> <span class="o">--</span><span class="n">parallel_nn</span><span class="o">=</span><span class="n">true</span> <span class="n">trainer_count</span><span class="o">=</span><span class="n">COUNT</span>
Y
Update docs  
Yu Yang 已提交
178
T
Deploy to GitHub Pages: 703cce35a871c78a5e3f1c33abf7fe788e2bbda5  
Travis CI 已提交
179 
<span class="n">default_device</span><span class="p">(</span><span class="mi">0</span><span class="p">)</span>
Y
Update docs  
Yu Yang 已提交
180
T
Deploy to GitHub Pages: 703cce35a871c78a5e3f1c33abf7fe788e2bbda5  
Travis CI 已提交
181 182 183 
<span class="n">fc1</span><span class="o">=</span><span class="n">fc_layer</span><span class="p">(</span><span class="o">...</span><span class="p">)</span>
<span class="n">fc2</span><span class="o">=</span><span class="n">fc_layer</span><span class="p">(</span><span class="o">...</span><span class="p">)</span>
<span class="n">fc3</span><span class="o">=</span><span class="n">fc_layer</span><span class="p">(</span><span class="o">...</span><span class="p">,</span><span class="n">layer_attr</span><span class="o">=</span><span class="n">ExtraAttr</span><span class="p">(</span><span class="n">device</span><span class="o">=-</span><span class="mi">1</span><span class="p">))</span>
Y
Update docs  
Yu Yang 已提交
184 185 186 187 188 189 190 191 192 193 194 195 196 197 
</pre></div>
</div>
<ul class="simple">
<li>default_device(0): set default device ID to 0. This means that except the layers with device=-1, all layers will use a GPU, and the specific GPU used for each layer depends on trainer_count and gpu_id (0 by default). Here, layer l1 and l2 are computed on the GPU.</li>
<li>device=-1: use the CPU for layer l3.</li>
<li>trainer_count:<ul>
<li>trainer_count=1: if gpu_id is not set, then use the first GPU to compute layers l1 and l2. Otherwise use the GPU with gpu_id.</li>
<li>trainer_count&gt;1: use trainer_count GPUs to compute one layer using data parallelism. For example, trainer_count=2 means that GPUs 0 and 1 will use data parallelism to compute layer l1 and l2.</li>
</ul>
</li>
</ul>
</div>
<div class="section" id="case-2-specify-layers-in-different-devices">
<span id="case-2-specify-layers-in-different-devices"></span><h3>Case 2: Specify Layers in Different Devices<a class="headerlink" href="#case-2-specify-layers-in-different-devices" title="Permalink to this headline"></a></h3>
T
Deploy to GitHub Pages: 703cce35a871c78a5e3f1c33abf7fe788e2bbda5  
Travis CI 已提交
198 199 
<div class="highlight-default"><div class="highlight"><pre><span></span><span class="c1">#command line:</span>
<span class="n">paddle</span> <span class="n">train</span> <span class="o">--</span><span class="n">use_gpu</span><span class="o">=</span><span class="n">true</span> <span class="o">--</span><span class="n">parallel_nn</span><span class="o">=</span><span class="n">true</span> <span class="o">--</span><span class="n">trainer_count</span><span class="o">=</span><span class="n">COUNT</span>
Y
Update docs  
Yu Yang 已提交
200
T
Deploy to GitHub Pages: 703cce35a871c78a5e3f1c33abf7fe788e2bbda5  
Travis CI 已提交
201 202 203 204 
<span class="c1">#network:</span>
<span class="n">fc2</span><span class="o">=</span><span class="n">fc_layer</span><span class="p">(</span><span class="nb">input</span><span class="o">=</span><span class="n">l1</span><span class="p">,</span> <span class="n">layer_attr</span><span class="o">=</span><span class="n">ExtraAttr</span><span class="p">(</span><span class="n">device</span><span class="o">=</span><span class="mi">0</span><span class="p">),</span> <span class="o">...</span><span class="p">)</span>
<span class="n">fc3</span><span class="o">=</span><span class="n">fc_layer</span><span class="p">(</span><span class="nb">input</span><span class="o">=</span><span class="n">l1</span><span class="p">,</span> <span class="n">layer_attr</span><span class="o">=</span><span class="n">ExtraAttr</span><span class="p">(</span><span class="n">device</span><span class="o">=</span><span class="mi">1</span><span class="p">),</span> <span class="o">...</span><span class="p">)</span>
<span class="n">fc4</span><span class="o">=</span><span class="n">fc_layer</span><span class="p">(</span><span class="nb">input</span><span class="o">=</span><span class="n">fc2</span><span class="p">,</span> <span class="n">layer_attr</span><span class="o">=</span><span class="n">ExtraAttr</span><span class="p">(</span><span class="n">device</span><span class="o">=-</span><span class="mi">1</span><span class="p">),</span> <span class="o">...</span><span class="p">)</span>
Y
Update docs  
Yu Yang 已提交
205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 
</pre></div>
</div>
<p>In this case, we assume that there are 4 GPUs in one machine.</p>
<ul class="simple">
<li>trainer_count=1:<ul>
<li>Use GPU 0 to compute layer l2.</li>
<li>Use GPU 1 to compute layer l3.</li>
<li>Use CPU to compute layer l4.</li>
</ul>
</li>
<li>trainer_count=2:<ul>
<li>Use GPU 0 and 1 to compute layer l2.</li>
<li>Use GPU 2 and 3 to compute layer l3.</li>
<li>Use CPU to compute l4 in two threads.</li>
</ul>
</li>
<li>trainer_count=4:<ul>
<li>It will fail (note, we have assumed that there are 4 GPUs in machine), because argument <code class="docutils literal"><span class="pre">allow_only_one_model_on_one_gpu</span></code> is true by default.</li>
</ul>
</li>
</ul>
<p><strong>Allocation of device ID when <code class="docutils literal"><span class="pre">device!=-1</span></code></strong>:</p>
T
Deploy to GitHub Pages: 703cce35a871c78a5e3f1c33abf7fe788e2bbda5  
Travis CI 已提交
227 
<div class="highlight-default"><div class="highlight"><pre><span></span><span class="p">(</span><span class="n">deviceId</span> <span class="o">+</span> <span class="n">gpu_id</span> <span class="o">+</span> <span class="n">threadId</span> <span class="o">*</span> <span class="n">numLogicalDevices_</span><span class="p">)</span> <span class="o">%</span> <span class="n">numDevices_</span>
Y
Update docs  
Yu Yang 已提交
228
T
Deploy to GitHub Pages: 703cce35a871c78a5e3f1c33abf7fe788e2bbda5  
Travis CI 已提交
229 230 231 232 233 
<span class="n">deviceId</span><span class="p">:</span>             <span class="n">specified</span> <span class="ow">in</span> <span class="n">layer</span><span class="o">.</span>
<span class="n">gpu_id</span><span class="p">:</span>               <span class="mi">0</span> <span class="n">by</span> <span class="n">default</span><span class="o">.</span>
<span class="n">threadId</span><span class="p">:</span>             <span class="n">thread</span> <span class="n">ID</span><span class="p">,</span> <span class="nb">range</span><span class="p">:</span> <span class="mi">0</span><span class="p">,</span><span class="mi">1</span><span class="p">,</span><span class="o">...</span><span class="p">,</span> <span class="n">trainer_count</span><span class="o">-</span><span class="mi">1</span>
<span class="n">numDevices_</span><span class="p">:</span>          <span class="n">device</span> <span class="p">(</span><span class="n">GPU</span><span class="p">)</span> <span class="n">count</span> <span class="ow">in</span> <span class="n">machine</span><span class="o">.</span>
<span class="n">numLogicalDevices_</span><span class="p">:</span>   <span class="nb">min</span><span class="p">(</span><span class="nb">max</span><span class="p">(</span><span class="n">deviceId</span> <span class="o">+</span> <span class="mi">1</span><span class="p">),</span> <span class="n">numDevices_</span><span class="p">)</span>
Y
Update docs  
Yu Yang 已提交
234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 
</pre></div>
</div>
</div>
</div>
</div>


          </div>
        </div>
      </div>
      <div class="sphinxsidebar" role="navigation" aria-label="main navigation">
        <div class="sphinxsidebarwrapper">
  <h3><a href="../../index.html">Table Of Contents</a></h3>
  <ul>
<li><a class="reference internal" href="#">Use Case</a><ul>
<li><a class="reference internal" href="#local-training">Local Training</a><ul>
<li><a class="reference internal" href="#pass-command-argument-to-network-config">1) Pass Command Argument to Network config</a></li>
<li><a class="reference internal" href="#use-model-to-initialize-network">2) Use Model to Initialize Network</a></li>
</ul>
</li>
<li><a class="reference internal" href="#local-testing">Local Testing</a></li>
<li><a class="reference internal" href="#sparse-training">Sparse Training</a><ul>
<li><a class="reference internal" href="#local-training">1) Local training</a></li>
<li><a class="reference internal" href="#cluster-training">2) cluster training</a></li>
</ul>
</li>
<li><a class="reference internal" href="#parallel-nn">parallel_nn</a><ul>
<li><a class="reference internal" href="#case-1-mixed-use-of-gpu-and-cpu">case 1: Mixed Use of GPU and CPU</a></li>
<li><a class="reference internal" href="#case-2-specify-layers-in-different-devices">Case 2: Specify Layers in Different Devices</a></li>
</ul>
</li>
</ul>
</li>
</ul>

  <h4>Previous topic</h4>
  <p class="topless"><a href="../api/trainer_config_helpers/attrs.html"
                        title="previous chapter">Parameter and Extra Layer Attribute</a></p>
  <h4>Next topic</h4>
  <p class="topless"><a href="argument_outline.html"
                        title="next chapter">Argument Outline</a></p>
  <div role="note" aria-label="source link">
    <h3>This Page</h3>
    <ul class="this-page-menu">
      <li><a href="../../_sources/ui/cmd_argument/use_case.txt"
            rel="nofollow">Show Source</a></li>
    </ul>
   </div>
<div id="searchbox" style="display: none" role="search">
  <h3>Quick search</h3>
    <form class="search" action="../../search.html" method="get">
T
Deploy to GitHub Pages: 703cce35a871c78a5e3f1c33abf7fe788e2bbda5  
Travis CI 已提交
285 286 
      <div><input type="text" name="q" /></div>
      <div><input type="submit" value="Go" /></div>
Y
Update docs  
Yu Yang 已提交
287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 
      <input type="hidden" name="check_keywords" value="yes" />
      <input type="hidden" name="area" value="default" />
    </form>
</div>
<script type="text/javascript">$('#searchbox').show(0);</script>
        </div>
      </div>
      <div class="clearer"></div>
    </div>
    <div class="related" role="navigation" aria-label="related navigation">
      <h3>Navigation</h3>
      <ul>
        <li class="right" style="margin-right: 10px">
          <a href="../../genindex.html" title="General Index"
             >index</a></li>
        <li class="right" >
          <a href="../../py-modindex.html" title="Python Module Index"
             >modules</a> |</li>
        <li class="right" >
          <a href="argument_outline.html" title="Argument Outline"
             >next</a> |</li>
        <li class="right" >
          <a href="../api/trainer_config_helpers/attrs.html" title="Parameter and Extra Layer Attribute"
             >previous</a> |</li>
T
Deploy to GitHub Pages: 703cce35a871c78a5e3f1c33abf7fe788e2bbda5  
Travis CI 已提交
311 312 
        <li class="nav-item nav-item-0"><a href="../../index.html">PaddlePaddle  documentation</a> &#187;</li>
          <li class="nav-item nav-item-1"><a href="../index.html" >User Interface</a> &#187;</li>
Y
Update docs  
Yu Yang 已提交
313 314 315 
      </ul>
    </div>
    <div class="footer" role="contentinfo">
T
Deploy to GitHub Pages: 703cce35a871c78a5e3f1c33abf7fe788e2bbda5  
Travis CI 已提交
316 317 
        &#169; Copyright 2016, PaddlePaddle developers.
      Created using <a href="http://sphinx-doc.org/">Sphinx</a> 1.4.6.
Y
Update docs  
Yu Yang 已提交
318 319 320 
    </div>
  </body>
</html>