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
use_case.html 18.0 KB
Newer Older
Y
Update doc  
Yu Yang 已提交
1 2 3 4 5 6 7 8 9 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 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 
<!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" />
    
    <title>Use Case &mdash; PaddlePaddle  documentation</title>
    
    <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="Recurrent Neural Network Configuration" href="../api/rnn/index.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/rnn/index.html" title="Recurrent Neural Network Configuration"
             accesskey="P">previous</a> |</li>
        <li class="nav-item nav-item-0"><a href="../../index.html">PaddlePaddle  documentation</a> &raquo;</li>
          <li class="nav-item nav-item-1"><a href="../index.html" accesskey="U">User Interface</a> &raquo;</li> 
      </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>
<div class="highlight-python"><div class="highlight"><pre><span></span>paddle train \
  --use_gpu=1/0 \                        #1:GPU,0:CPU(default:true)
  --config=network_config \
  --save_dir=output \
  --trainer_count=COUNT \                #(default:1)
  --test_period=M \                      #(default:1000）
  --test_all_data_in_one_period=true \   #(default:false) 
  --num_passes=N \                       #(defalut:100）
  --log_period=K \                       #(default:100)
  --dot_period=1000 \                    #(default:1)
  #[--show_parameter_stats_period=100] \ #(default:0)
  #[--saving_period_by_batches=200] \    #(default:0)
</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>
<div class="highlight-python"><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> \
</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>
<div class="highlight-python"><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="bp">False</span><span class="p">)</span>
<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>
<div class="highlight-python"><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>
</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>
<div class="highlight-python"><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>
<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>
<div class="highlight-python"><div class="highlight"><pre><span></span>paddle train --job=test \
             --use_gpu=1/0 \ 
             --config=network_config \
             --trainer_count=COUNT \ 
             --init_model_path=model_path \
</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>
<div class="highlight-python"><div class="highlight"><pre><span></span>paddle train --job=test \
             --use_gpu=1/0 \ 
             --config=network_config \
             --trainer_count=COUNT \ 
             --model_list=model.list \
</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>
<div class="highlight-python"><div class="highlight"><pre><span></span>./alexnet_pass1
./alexnet_pass2
</pre></div>
</div>
<p>Method 3:</p>
<div class="highlight-python"><div class="highlight"><pre><span></span>paddle train --job=test \
             --use_gpu=1/0 \
             --config=network_config \
             --trainer_count=COUNT \
             --save_dir=model \
             --test_pass=M \
             --num_passes=N \
</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>
<div class="highlight-python"><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>
</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>
<div class="highlight-python"><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>
</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>
<div class="highlight-python"><div class="highlight"><pre><span></span>#command line:
paddle train --use_gpu=true --parallel_nn=true trainer_count=COUNT

default_device(0)

fc1=fc_layer(...)
fc2=fc_layer(...)
fc3=fc_layer(...,layer_attr=ExtraAttr(device=-1))
</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>
<div class="highlight-python"><div class="highlight"><pre><span></span>#command line:
paddle train --use_gpu=true --parallel_nn=true --trainer_count=COUNT

#network:
fc2=fc_layer(input=l1, layer_attr=ExtraAttr(device=0), ...)
fc3=fc_layer(input=l1, layer_attr=ExtraAttr(device=1), ...)
fc4=fc_layer(input=fc2, layer_attr=ExtraAttr(device=-1), ...)
</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>
<div class="highlight-python"><div class="highlight"><pre><span></span>(deviceId + gpu_id + threadId * numLogicalDevices_) % numDevices_

deviceId:             specified in layer.
gpu_id:               0 by default.
threadId:             thread ID, range: 0,1,..., trainer_count-1
numDevices_:          device (GPU) count in machine.
numLogicalDevices_:   min(max(deviceId + 1), numDevices_)
</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/rnn/index.html"
                        title="previous chapter">Recurrent Neural Network Configuration</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">
      <input type="text" name="q" />
      <input type="submit" value="Go" />
      <input type="hidden" name="check_keywords" value="yes" />
      <input type="hidden" name="area" value="default" />
    </form>
    <p class="searchtip" style="font-size: 90%">
    Enter search terms or a module, class or function name.
    </p>
</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/rnn/index.html" title="Recurrent Neural Network Configuration"
             >previous</a> |</li>
        <li class="nav-item nav-item-0"><a href="../../index.html">PaddlePaddle  documentation</a> &raquo;</li>
          <li class="nav-item nav-item-1"><a href="../index.html" >User Interface</a> &raquo;</li> 
      </ul>
    </div>
    <div class="footer" role="contentinfo">
        &copy; Copyright 2016, PaddlePaddle developers.
      Created using <a href="http://sphinx-doc.org/">Sphinx</a> 1.3.5.
    </div>
  </body>
</html>