Skip to content

P
PaddleSlim

PaddlePaddle / PaddleSlim
1 年多 前同步成功

通知 51
Star 1434
Fork 344
P
PaddleSlim
切换分支/标签
查找文件 普通视图历史永久链接Permalink
index.html 49.2 KB
Newer Older
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 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 http-equiv="X-UA-Compatible" content="IE=edge">
  <meta name="viewport" content="width=device-width, initial-scale=1.0">
  
  
  <link rel="shortcut icon" href="../../img/favicon.ico">
B
Deployed 0f9c3ce with MkDocs version: 1.0.4  
baiyfbupt 已提交
11 
  <title>知识蒸馏 - PaddleSlim Docs</title>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
12 13 14 15 16 17 18 19 
  <link href='https://fonts.googleapis.com/css?family=Lato:400,700|Roboto+Slab:400,700|Inconsolata:400,700' rel='stylesheet' type='text/css'>

  <link rel="stylesheet" href="../../css/theme.css" type="text/css" />
  <link rel="stylesheet" href="../../css/theme_extra.css" type="text/css" />
  <link rel="stylesheet" href="//cdnjs.cloudflare.com/ajax/libs/highlight.js/9.12.0/styles/github.min.css">
  
  <script>
    // Current page data
B
Deployed 0f9c3ce with MkDocs version: 1.0.4  
baiyfbupt 已提交
20 
    var mkdocs_page_name = "\u77e5\u8bc6\u84b8\u998f";
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
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 
    var mkdocs_page_input_path = "api/single_distiller_api.md";
    var mkdocs_page_url = null;
  </script>
  
  <script src="../../js/jquery-2.1.1.min.js" defer></script>
  <script src="../../js/modernizr-2.8.3.min.js" defer></script>
  <script src="//cdnjs.cloudflare.com/ajax/libs/highlight.js/9.12.0/highlight.min.js"></script>
  <script>hljs.initHighlightingOnLoad();</script> 
  
</head>

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

  <div class="wy-grid-for-nav">

    
    <nav data-toggle="wy-nav-shift" class="wy-nav-side stickynav">
      <div class="wy-side-nav-search">
        <a href="../.." class="icon icon-home"> PaddleSlim Docs</a>
        <div role="search">
  <form id ="rtd-search-form" class="wy-form" action="../../search.html" method="get">
    <input type="text" name="q" placeholder="Search docs" title="Type search term here" />
  </form>
</div>
      </div>

      <div class="wy-menu wy-menu-vertical" data-spy="affix" role="navigation" aria-label="main navigation">
	<ul class="current">
	  
          
            <li class="toctree-l1">
		
    <a class="" href="../..">Home</a>
	    </li>
          
            <li class="toctree-l1">
		
    <span class="caption-text">API</span>
    <ul class="subnav">
                <li class="">
                    
    <a class="" href="../quantization_api/">量化</a>
                </li>
                <li class="">
B
Deployed 0f9c3ce with MkDocs version: 1.0.4  
baiyfbupt 已提交
66 
    <a class="" href="../prune_api/">剪枝与敏感度</a>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
67 68 69 
                </li>
                <li class="">
B
Deployed 0f9c3ce with MkDocs version: 1.0.4  
baiyfbupt 已提交
70 
    <a class="" href="../analysis_api/">模型分析</a>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
71 72 73 
                </li>
                <li class=" current">
B
Deployed 0f9c3ce with MkDocs version: 1.0.4  
baiyfbupt 已提交
74 
    <a class="current" href="./">知识蒸馏</a>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
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 
    <ul class="subnav">
            
    <li class="toctree-l3"><a href="#merge">merge</a></li>
    

    <li class="toctree-l3"><a href="#fsp_loss">fsp_loss</a></li>
    

    <li class="toctree-l3"><a href="#l2_loss">l2_loss</a></li>
    

    <li class="toctree-l3"><a href="#soft_label_loss">soft_label_loss</a></li>
    

    <li class="toctree-l3"><a href="#loss">loss</a></li>
    

    </ul>
                </li>
                <li class="">
                    
    <a class="" href="../nas_api/">SA搜索</a>
                </li>
                <li class="">
                    
    <a class="" href="../search_space/">搜索空间</a>
                </li>
B
Deployed 0f9c3ce with MkDocs version: 1.0.4  
baiyfbupt 已提交
102 103 104 105 
                <li class="">
                    
    <a class="" href="../../table_latency/">硬件延时评估表</a>
                </li>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
106 107 108 
    </ul>
	    </li>
B
Deployed 0f9c3ce with MkDocs version: 1.0.4  
baiyfbupt 已提交
109 110 
            <li class="toctree-l1">
B
Deployed 9bcff00 with MkDocs version: 1.0.4  
baiyfbupt 已提交
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 
    <span class="caption-text">教程</span>
    <ul class="subnav">
                <li class="">
                    
    <a class="" href="../../tutorials/quant_post_demo/">离线量化</a>
                </li>
                <li class="">
                    
    <a class="" href="../../tutorials/quant_aware_demo/">量化训练</a>
                </li>
                <li class="">
                    
    <a class="" href="../../tutorials/quant_embedding_demo/">Embedding量化</a>
                </li>
                <li class="">
                    
    <a class="" href="../../tutorials/nas_demo/">SA搜索</a>
                </li>
                <li class="">
                    
    <a class="" href="../../tutorials/distillation_demo/">知识蒸馏</a>
                </li>
    </ul>
	    </li>
          
            <li class="toctree-l1">
B
Deployed 0f9c3ce with MkDocs version: 1.0.4  
baiyfbupt 已提交
138 139 140 
    <a class="" href="../../algo/algo/">算法原理</a>
	    </li>
B
Deployed 9bcff00 with MkDocs version: 1.0.4  
baiyfbupt 已提交
141 142 143 144 145 
            <li class="toctree-l1">
		
    <a class="" href="../../model_zoo/">模型库</a>
	    </li>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
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 
        </ul>
      </div>
      &nbsp;
    </nav>

    <section data-toggle="wy-nav-shift" class="wy-nav-content-wrap">

      
      <nav class="wy-nav-top" role="navigation" aria-label="top navigation">
        <i data-toggle="wy-nav-top" class="fa fa-bars"></i>
        <a href="../..">PaddleSlim Docs</a>
      </nav>

      
      <div class="wy-nav-content">
        <div class="rst-content">
          <div role="navigation" aria-label="breadcrumbs navigation">
  <ul class="wy-breadcrumbs">
    <li><a href="../..">Docs</a> &raquo;</li>
    
      
        
          <li>API &raquo;</li>
B
Deployed 0f9c3ce with MkDocs version: 1.0.4  
baiyfbupt 已提交
172 
    <li>知识蒸馏</li>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
173 174 175 176 177 178 179 180 181 182 183 184 185 186 
    <li class="wy-breadcrumbs-aside">
      
        <a href="https://github.com/PaddlePaddle/PaddleSlim/edit/master/docs/api/single_distiller_api.md"
          class="icon icon-github"> Edit on GitHub</a>
      
    </li>
  </ul>
  <hr/>
</div>
          <div role="main">
            <div class="section">
              
                <h2 id="merge">merge<a class="headerlink" href="#merge" title="Permanent link">#</a></h2>
<dl>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
187 
<dt>paddleslim.dist.merge(teacher_program, student_program, data_name_map, place, scope=fluid.global_scope(), name_prefix='teacher_') <a href="https://github.com/PaddlePaddle/PaddleSlim/blob/develop/paddleslim/dist/single_distiller.py#L19">[源代码]</a> </dt>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
188 
<dd>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
189 
<p>merge将两个paddle program（teacher_program, student_program）融合为一个program，并将融合得到的program返回。在融合的program中，可以为其中合适的teacher特征图和student特征图添加蒸馏损失函数，从而达到用teacher模型的暗知识（Dark Knowledge）指导student模型学习的目的。</p>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
190 191 192 193 
</dd>
</dl>
<p><strong>参数：</strong></p>
<ul>
B
Deployed c953571 with MkDocs version: 1.0.4  
baiyfbupt 已提交
194 195 
<li><strong>teacher_program</strong>(Program)-定义了teacher模型的 <a href="https://www.paddlepaddle.org.cn/documentation/docs/zh/api_cn/fluid_cn/Program_cn.html#program"><em>paddle program</em></a></li>
<li><strong>student_program</strong>(Program)-定义了student模型的 <a href="https://www.paddlepaddle.org.cn/documentation/docs/zh/api_cn/fluid_cn/Program_cn.html#program"><em>paddle program</em></a></li>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
196 
<li><strong>data_name_map</strong>(dict)-teacher输入接口名与student输入接口名的映射，其中dict的 <em>key</em> 为teacher的输入名，<em>value</em> 为student的输入名</li>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
197 
<li><strong>place</strong>(fluid.CPUPlace()|fluid.CUDAPlace(N))-该参数表示程序运行在何种设备上，这里的N为GPU对应的ID</li>
B
Deployed c953571 with MkDocs version: 1.0.4  
baiyfbupt 已提交
198 199 
<li><strong>scope</strong>(Scope)-该参数表示程序使用的变量作用域，如果不指定将使用默认的全局作用域。默认值：<a href="https://www.paddlepaddle.org.cn/documentation/docs/zh/api_cn/fluid_cn/global_scope_cn.html#global-scope"><em>fluid.global_scope()</em></a></li>
<li><strong>name_prefix</strong>(str)-merge操作将统一为teacher的<a href="https://www.paddlepaddle.org.cn/documentation/docs/zh/1.3/api_guides/low_level/program.html#variable"><em>Variables</em></a>添加的名称前缀name_prefix。默认值：'teacher_'</li>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
200 201 
</ul>
<p><strong>返回：</strong> 由student_program和teacher_program merge得到的program</p>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
202 203 204 205 
<div class="admonition note">
<p class="admonition-title">Note</p>
<p><em>data_name_map</em> 是 <strong>teacher_var name到student_var name的映射</strong>，如果写反可能无法正确进行merge</p>
</div>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
206 
<p><strong>使用示例：</strong></p>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 
<table class="codehilitetable"><tr><td class="linenos"><div class="linenodiv"><pre><span></span> 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18</pre></div></td><td class="code"><div class="codehilite"><pre><span></span><span class="kn">import</span> <span class="nn">paddle.fluid</span> <span class="kn">as</span> <span class="nn">fluid</span>
<span class="kn">import</span> <span class="nn">paddleslim.dist</span> <span class="kn">as</span> <span class="nn">dist</span>
<span class="n">student_program</span> <span class="o">=</span> <span class="n">fluid</span><span class="o">.</span><span class="n">Program</span><span class="p">()</span>
<span class="k">with</span> <span class="n">fluid</span><span class="o">.</span><span class="n">program_guard</span><span class="p">(</span><span class="n">student_program</span><span class="p">):</span>
    <span class="n">x</span> <span class="o">=</span> <span class="n">fluid</span><span class="o">.</span><span class="n">layers</span><span class="o">.</span><span class="n">data</span><span class="p">(</span><span class="n">name</span><span class="o">=</span><span class="s1">&#39;x&#39;</span><span class="p">,</span> <span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="mi">1</span><span class="p">,</span> <span class="mi">28</span><span class="p">,</span> <span class="mi">28</span><span class="p">])</span>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
229 230 231 232 233 234 235 236 237 238 239 
    <span class="n">conv</span> <span class="o">=</span> <span class="n">fluid</span><span class="o">.</span><span class="n">layers</span><span class="o">.</span><span class="n">conv2d</span><span class="p">(</span><span class="n">x</span><span class="p">,</span> <span class="mi">32</span><span class="p">,</span> <span class="mi">1</span><span class="p">)</span>
    <span class="n">out</span> <span class="o">=</span> <span class="n">fluid</span><span class="o">.</span><span class="n">layers</span><span class="o">.</span><span class="n">conv2d</span><span class="p">(</span><span class="n">conv</span><span class="p">,</span> <span class="mi">64</span><span class="p">,</span> <span class="mi">3</span><span class="p">,</span> <span class="n">padding</span><span class="o">=</span><span class="mi">1</span><span class="p">)</span>
<span class="n">teacher_program</span> <span class="o">=</span> <span class="n">fluid</span><span class="o">.</span><span class="n">Program</span><span class="p">()</span>
<span class="k">with</span> <span class="n">fluid</span><span class="o">.</span><span class="n">program_guard</span><span class="p">(</span><span class="n">teacher_program</span><span class="p">):</span>
    <span class="n">y</span> <span class="o">=</span> <span class="n">fluid</span><span class="o">.</span><span class="n">layers</span><span class="o">.</span><span class="n">data</span><span class="p">(</span><span class="n">name</span><span class="o">=</span><span class="s1">&#39;y&#39;</span><span class="p">,</span> <span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="mi">1</span><span class="p">,</span> <span class="mi">28</span><span class="p">,</span> <span class="mi">28</span><span class="p">])</span>
    <span class="n">conv</span> <span class="o">=</span> <span class="n">fluid</span><span class="o">.</span><span class="n">layers</span><span class="o">.</span><span class="n">conv2d</span><span class="p">(</span><span class="n">y</span><span class="p">,</span> <span class="mi">32</span><span class="p">,</span> <span class="mi">1</span><span class="p">)</span>
    <span class="n">conv</span> <span class="o">=</span> <span class="n">fluid</span><span class="o">.</span><span class="n">layers</span><span class="o">.</span><span class="n">conv2d</span><span class="p">(</span><span class="n">conv</span><span class="p">,</span> <span class="mi">32</span><span class="p">,</span> <span class="mi">3</span><span class="p">,</span> <span class="n">padding</span><span class="o">=</span><span class="mi">1</span><span class="p">)</span>
    <span class="n">out</span> <span class="o">=</span> <span class="n">fluid</span><span class="o">.</span><span class="n">layers</span><span class="o">.</span><span class="n">conv2d</span><span class="p">(</span><span class="n">conv</span><span class="p">,</span> <span class="mi">64</span><span class="p">,</span> <span class="mi">3</span><span class="p">,</span> <span class="n">padding</span><span class="o">=</span><span class="mi">1</span><span class="p">)</span>
<span class="n">data_name_map</span> <span class="o">=</span> <span class="p">{</span><span class="s1">&#39;y&#39;</span><span class="p">:</span><span class="s1">&#39;x&#39;</span><span class="p">}</span>
<span class="n">USE_GPU</span> <span class="o">=</span> <span class="bp">False</span>
<span class="n">place</span> <span class="o">=</span> <span class="n">fluid</span><span class="o">.</span><span class="n">CUDAPlace</span><span class="p">(</span><span class="mi">0</span><span class="p">)</span> <span class="k">if</span> <span class="n">USE_GPU</span> <span class="k">else</span> <span class="n">fluid</span><span class="o">.</span><span class="n">CPUPlace</span><span class="p">()</span>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
240 241 242 243 
<span class="hll"><span class="n">main_program</span> <span class="o">=</span> <span class="n">dist</span><span class="o">.</span><span class="n">merge</span><span class="p">(</span><span class="n">teacher_program</span><span class="p">,</span> <span class="n">student_program</span><span class="p">,</span>
</span><span class="hll">                          <span class="n">data_name_map</span><span class="p">,</span> <span class="n">place</span><span class="p">)</span>
</span></pre></div>
</td></tr></table>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
244 245 246 

<h2 id="fsp_loss">fsp_loss<a class="headerlink" href="#fsp_loss" title="Permanent link">#</a></h2>
<dl>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
247 
<dt>paddleslim.dist.fsp_loss(teacher_var1_name, teacher_var2_name, student_var1_name, student_var2_name, program=fluid.default_main_program()) <a href="https://github.com/PaddlePaddle/PaddleSlim/blob/develop/paddleslim/dist/single_distiller.py#L90">[源代码]</a></dt>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
248 
<dd>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
249 
<p>fsp_loss为program内的teacher var和student var添加fsp loss，出自论文<a href="http://openaccess.thecvf.com/content_cvpr_2017/papers/Yim_A_Gift_From_CVPR_2017_paper.pdf">&lt;&lt;A Gift from Knowledge Distillation: Fast Optimization, Network Minimization and Transfer Learning>></a></p>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
250 251 252 253 254 255 256 257 
</dd>
</dl>
<p><strong>参数：</strong></p>
<ul>
<li><strong>teacher_var1_name</strong>(str): teacher_var1的名称. 对应的variable是一个形为<code>[batch_size, x_channel, height, width]</code>的4-D特征图Tensor，数据类型为float32或float64</li>
<li><strong>teacher_var2_name</strong>(str): teacher_var2的名称. 对应的variable是一个形为<code>[batch_size, y_channel, height, width]</code>的4-D特征图Tensor，数据类型为float32或float64。只有y_channel可以与teacher_var1的x_channel不同，其他维度必须与teacher_var1相同</li>
<li><strong>student_var1_name</strong>(str): student_var1的名称. 对应的variable需与teacher_var1尺寸保持一致，是一个形为<code>[batch_size, x_channel, height, width]</code>的4-D特征图Tensor，数据类型为float32或float64</li>
<li><strong>student_var2_name</strong>(str): student_var2的名称. 对应的variable需与teacher_var2尺寸保持一致，是一个形为<code>[batch_size, y_channel, height, width]</code>的4-D特征图Tensor，数据类型为float32或float64。只有y_channel可以与student_var1的x_channel不同，其他维度必须与student_var1相同</li>
B
Deployed c953571 with MkDocs version: 1.0.4  
baiyfbupt 已提交
258 
<li><strong>program</strong>(Program): 用于蒸馏训练的fluid program。默认值：<a href="https://www.paddlepaddle.org.cn/documentation/docs/zh/1.3/api_cn/fluid_cn.html#default-main-program"><em>fluid.default_main_program()</em></a></li>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
259 
</ul>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
260 
<p><strong>返回：</strong> 由teacher_var1, teacher_var2, student_var1, student_var2组合得到的fsp_loss</p>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
261 
<p><strong>使用示例：</strong></p>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 
<table class="codehilitetable"><tr><td class="linenos"><div class="linenodiv"><pre><span></span> 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20</pre></div></td><td class="code"><div class="codehilite"><pre><span></span><span class="kn">import</span> <span class="nn">paddle.fluid</span> <span class="kn">as</span> <span class="nn">fluid</span>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 
<span class="kn">import</span> <span class="nn">paddleslim.dist</span> <span class="kn">as</span> <span class="nn">dist</span>
<span class="n">student_program</span> <span class="o">=</span> <span class="n">fluid</span><span class="o">.</span><span class="n">Program</span><span class="p">()</span>
<span class="k">with</span> <span class="n">fluid</span><span class="o">.</span><span class="n">program_guard</span><span class="p">(</span><span class="n">student_program</span><span class="p">):</span>
    <span class="n">x</span> <span class="o">=</span> <span class="n">fluid</span><span class="o">.</span><span class="n">layers</span><span class="o">.</span><span class="n">data</span><span class="p">(</span><span class="n">name</span><span class="o">=</span><span class="s1">&#39;x&#39;</span><span class="p">,</span> <span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="mi">1</span><span class="p">,</span> <span class="mi">28</span><span class="p">,</span> <span class="mi">28</span><span class="p">])</span>
    <span class="n">conv</span> <span class="o">=</span> <span class="n">fluid</span><span class="o">.</span><span class="n">layers</span><span class="o">.</span><span class="n">conv2d</span><span class="p">(</span><span class="n">x</span><span class="p">,</span> <span class="mi">32</span><span class="p">,</span> <span class="mi">1</span><span class="p">,</span> <span class="n">name</span><span class="o">=</span><span class="s1">&#39;s1&#39;</span><span class="p">)</span>
    <span class="n">out</span> <span class="o">=</span> <span class="n">fluid</span><span class="o">.</span><span class="n">layers</span><span class="o">.</span><span class="n">conv2d</span><span class="p">(</span><span class="n">conv</span><span class="p">,</span> <span class="mi">64</span><span class="p">,</span> <span class="mi">3</span><span class="p">,</span> <span class="n">padding</span><span class="o">=</span><span class="mi">1</span><span class="p">,</span> <span class="n">name</span><span class="o">=</span><span class="s1">&#39;s2&#39;</span><span class="p">)</span>
<span class="n">teacher_program</span> <span class="o">=</span> <span class="n">fluid</span><span class="o">.</span><span class="n">Program</span><span class="p">()</span>
<span class="k">with</span> <span class="n">fluid</span><span class="o">.</span><span class="n">program_guard</span><span class="p">(</span><span class="n">teacher_program</span><span class="p">):</span>
    <span class="n">y</span> <span class="o">=</span> <span class="n">fluid</span><span class="o">.</span><span class="n">layers</span><span class="o">.</span><span class="n">data</span><span class="p">(</span><span class="n">name</span><span class="o">=</span><span class="s1">&#39;y&#39;</span><span class="p">,</span> <span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="mi">1</span><span class="p">,</span> <span class="mi">28</span><span class="p">,</span> <span class="mi">28</span><span class="p">])</span>
    <span class="n">conv</span> <span class="o">=</span> <span class="n">fluid</span><span class="o">.</span><span class="n">layers</span><span class="o">.</span><span class="n">conv2d</span><span class="p">(</span><span class="n">y</span><span class="p">,</span> <span class="mi">32</span><span class="p">,</span> <span class="mi">1</span><span class="p">,</span> <span class="n">name</span><span class="o">=</span><span class="s1">&#39;t1&#39;</span><span class="p">)</span>
    <span class="n">conv</span> <span class="o">=</span> <span class="n">fluid</span><span class="o">.</span><span class="n">layers</span><span class="o">.</span><span class="n">conv2d</span><span class="p">(</span><span class="n">conv</span><span class="p">,</span> <span class="mi">32</span><span class="p">,</span> <span class="mi">3</span><span class="p">,</span> <span class="n">padding</span><span class="o">=</span><span class="mi">1</span><span class="p">)</span>
    <span class="n">out</span> <span class="o">=</span> <span class="n">fluid</span><span class="o">.</span><span class="n">layers</span><span class="o">.</span><span class="n">conv2d</span><span class="p">(</span><span class="n">conv</span><span class="p">,</span> <span class="mi">64</span><span class="p">,</span> <span class="mi">3</span><span class="p">,</span> <span class="n">padding</span><span class="o">=</span><span class="mi">1</span><span class="p">,</span> <span class="n">name</span><span class="o">=</span><span class="s1">&#39;t2&#39;</span><span class="p">)</span>
<span class="n">data_name_map</span> <span class="o">=</span> <span class="p">{</span><span class="s1">&#39;y&#39;</span><span class="p">:</span><span class="s1">&#39;x&#39;</span><span class="p">}</span>
<span class="n">USE_GPU</span> <span class="o">=</span> <span class="bp">False</span>
<span class="n">place</span> <span class="o">=</span> <span class="n">fluid</span><span class="o">.</span><span class="n">CUDAPlace</span><span class="p">(</span><span class="mi">0</span><span class="p">)</span> <span class="k">if</span> <span class="n">USE_GPU</span> <span class="k">else</span> <span class="n">fluid</span><span class="o">.</span><span class="n">CPUPlace</span><span class="p">()</span>
<span class="n">main_program</span> <span class="o">=</span> <span class="n">merge</span><span class="p">(</span><span class="n">teacher_program</span><span class="p">,</span> <span class="n">student_program</span><span class="p">,</span> <span class="n">data_name_map</span><span class="p">,</span> <span class="n">place</span><span class="p">)</span>
<span class="k">with</span> <span class="n">fluid</span><span class="o">.</span><span class="n">program_guard</span><span class="p">(</span><span class="n">main_program</span><span class="p">):</span>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
299 300 301 302 
<span class="hll">    <span class="n">distillation_loss</span> <span class="o">=</span> <span class="n">dist</span><span class="o">.</span><span class="n">fsp_loss</span><span class="p">(</span><span class="s1">&#39;teacher_t1.tmp_1&#39;</span><span class="p">,</span> <span class="s1">&#39;teacher_t2.tmp_1&#39;</span><span class="p">,</span>
</span><span class="hll">                                      <span class="s1">&#39;s1.tmp_1&#39;</span><span class="p">,</span> <span class="s1">&#39;s2.tmp_1&#39;</span><span class="p">,</span> <span class="n">main_program</span><span class="p">)</span>
</span></pre></div>
</td></tr></table>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
303 304 305 

<h2 id="l2_loss">l2_loss<a class="headerlink" href="#l2_loss" title="Permanent link">#</a></h2>
<dl>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
306 
<dt>paddleslim.dist.l2_loss(teacher_var_name, student_var_name, program=fluid.default_main_program())<a href="https://github.com/PaddlePaddle/PaddleSlim/blob/develop/paddleslim/dist/single_distiller.py#L118">[源代码]</a></dt>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
307 308 309 310 311 312 313 314 
<dd>
<p>l2_loss为program内的teacher var和student var添加l2 loss</p>
</dd>
</dl>
<p><strong>参数：</strong></p>
<ul>
<li><strong>teacher_var_name</strong>(str): teacher_var的名称. </li>
<li><strong>student_var_name</strong>(str): student_var的名称.</li>
B
Deployed c953571 with MkDocs version: 1.0.4  
baiyfbupt 已提交
315 
<li><strong>program</strong>(Program): 用于蒸馏训练的fluid program。默认值：<a href="https://www.paddlepaddle.org.cn/documentation/docs/zh/1.3/api_cn/fluid_cn.html#default-main-program"><em>fluid.default_main_program()</em></a></li>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
316 
</ul>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
317 
<p><strong>返回：</strong> 由teacher_var, student_var组合得到的l2_loss</p>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
318 
<p><strong>使用示例：</strong></p>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 
<table class="codehilitetable"><tr><td class="linenos"><div class="linenodiv"><pre><span></span> 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20</pre></div></td><td class="code"><div class="codehilite"><pre><span></span><span class="kn">import</span> <span class="nn">paddle.fluid</span> <span class="kn">as</span> <span class="nn">fluid</span>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 
<span class="kn">import</span> <span class="nn">paddleslim.dist</span> <span class="kn">as</span> <span class="nn">dist</span>
<span class="n">student_program</span> <span class="o">=</span> <span class="n">fluid</span><span class="o">.</span><span class="n">Program</span><span class="p">()</span>
<span class="k">with</span> <span class="n">fluid</span><span class="o">.</span><span class="n">program_guard</span><span class="p">(</span><span class="n">student_program</span><span class="p">):</span>
    <span class="n">x</span> <span class="o">=</span> <span class="n">fluid</span><span class="o">.</span><span class="n">layers</span><span class="o">.</span><span class="n">data</span><span class="p">(</span><span class="n">name</span><span class="o">=</span><span class="s1">&#39;x&#39;</span><span class="p">,</span> <span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="mi">1</span><span class="p">,</span> <span class="mi">28</span><span class="p">,</span> <span class="mi">28</span><span class="p">])</span>
    <span class="n">conv</span> <span class="o">=</span> <span class="n">fluid</span><span class="o">.</span><span class="n">layers</span><span class="o">.</span><span class="n">conv2d</span><span class="p">(</span><span class="n">x</span><span class="p">,</span> <span class="mi">32</span><span class="p">,</span> <span class="mi">1</span><span class="p">,</span> <span class="n">name</span><span class="o">=</span><span class="s1">&#39;s1&#39;</span><span class="p">)</span>
    <span class="n">out</span> <span class="o">=</span> <span class="n">fluid</span><span class="o">.</span><span class="n">layers</span><span class="o">.</span><span class="n">conv2d</span><span class="p">(</span><span class="n">conv</span><span class="p">,</span> <span class="mi">64</span><span class="p">,</span> <span class="mi">3</span><span class="p">,</span> <span class="n">padding</span><span class="o">=</span><span class="mi">1</span><span class="p">,</span> <span class="n">name</span><span class="o">=</span><span class="s1">&#39;s2&#39;</span><span class="p">)</span>
<span class="n">teacher_program</span> <span class="o">=</span> <span class="n">fluid</span><span class="o">.</span><span class="n">Program</span><span class="p">()</span>
<span class="k">with</span> <span class="n">fluid</span><span class="o">.</span><span class="n">program_guard</span><span class="p">(</span><span class="n">teacher_program</span><span class="p">):</span>
    <span class="n">y</span> <span class="o">=</span> <span class="n">fluid</span><span class="o">.</span><span class="n">layers</span><span class="o">.</span><span class="n">data</span><span class="p">(</span><span class="n">name</span><span class="o">=</span><span class="s1">&#39;y&#39;</span><span class="p">,</span> <span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="mi">1</span><span class="p">,</span> <span class="mi">28</span><span class="p">,</span> <span class="mi">28</span><span class="p">])</span>
    <span class="n">conv</span> <span class="o">=</span> <span class="n">fluid</span><span class="o">.</span><span class="n">layers</span><span class="o">.</span><span class="n">conv2d</span><span class="p">(</span><span class="n">y</span><span class="p">,</span> <span class="mi">32</span><span class="p">,</span> <span class="mi">1</span><span class="p">,</span> <span class="n">name</span><span class="o">=</span><span class="s1">&#39;t1&#39;</span><span class="p">)</span>
    <span class="n">conv</span> <span class="o">=</span> <span class="n">fluid</span><span class="o">.</span><span class="n">layers</span><span class="o">.</span><span class="n">conv2d</span><span class="p">(</span><span class="n">conv</span><span class="p">,</span> <span class="mi">32</span><span class="p">,</span> <span class="mi">3</span><span class="p">,</span> <span class="n">padding</span><span class="o">=</span><span class="mi">1</span><span class="p">)</span>
    <span class="n">out</span> <span class="o">=</span> <span class="n">fluid</span><span class="o">.</span><span class="n">layers</span><span class="o">.</span><span class="n">conv2d</span><span class="p">(</span><span class="n">conv</span><span class="p">,</span> <span class="mi">64</span><span class="p">,</span> <span class="mi">3</span><span class="p">,</span> <span class="n">padding</span><span class="o">=</span><span class="mi">1</span><span class="p">,</span> <span class="n">name</span><span class="o">=</span><span class="s1">&#39;t2&#39;</span><span class="p">)</span>
<span class="n">data_name_map</span> <span class="o">=</span> <span class="p">{</span><span class="s1">&#39;y&#39;</span><span class="p">:</span><span class="s1">&#39;x&#39;</span><span class="p">}</span>
<span class="n">USE_GPU</span> <span class="o">=</span> <span class="bp">False</span>
<span class="n">place</span> <span class="o">=</span> <span class="n">fluid</span><span class="o">.</span><span class="n">CUDAPlace</span><span class="p">(</span><span class="mi">0</span><span class="p">)</span> <span class="k">if</span> <span class="n">USE_GPU</span> <span class="k">else</span> <span class="n">fluid</span><span class="o">.</span><span class="n">CPUPlace</span><span class="p">()</span>
<span class="n">main_program</span> <span class="o">=</span> <span class="n">merge</span><span class="p">(</span><span class="n">teacher_program</span><span class="p">,</span> <span class="n">student_program</span><span class="p">,</span> <span class="n">data_name_map</span><span class="p">,</span> <span class="n">place</span><span class="p">)</span>
<span class="k">with</span> <span class="n">fluid</span><span class="o">.</span><span class="n">program_guard</span><span class="p">(</span><span class="n">main_program</span><span class="p">):</span>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
356 357 358 359 
<span class="hll">    <span class="n">distillation_loss</span> <span class="o">=</span> <span class="n">dist</span><span class="o">.</span><span class="n">l2_loss</span><span class="p">(</span><span class="s1">&#39;teacher_t2.tmp_1&#39;</span><span class="p">,</span> <span class="s1">&#39;s2.tmp_1&#39;</span><span class="p">,</span>
</span><span class="hll">                                     <span class="n">main_program</span><span class="p">)</span>
</span></pre></div>
</td></tr></table>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
360 361 362 

<h2 id="soft_label_loss">soft_label_loss<a class="headerlink" href="#soft_label_loss" title="Permanent link">#</a></h2>
<dl>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
363 
<dt>paddleslim.dist.soft_label_loss(teacher_var_name, student_var_name, program=fluid.default_main_program(), teacher_temperature=1., student_temperature=1.)<a href="https://github.com/PaddlePaddle/PaddleSlim/blob/develop/paddleslim/dist/single_distiller.py#L136">[源代码]</a></dt>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
364 
<dd>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
365 
<p>soft_label_loss为program内的teacher var和student var添加soft label loss，出自论文<a href="https://arxiv.org/pdf/1503.02531.pdf">&lt;&lt;Distilling the Knowledge in a Neural Network>></a></p>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
366 367 368 369 370 371 
</dd>
</dl>
<p><strong>参数：</strong></p>
<ul>
<li><strong>teacher_var_name</strong>(str): teacher_var的名称. </li>
<li><strong>student_var_name</strong>(str): student_var的名称. </li>
B
Deployed c953571 with MkDocs version: 1.0.4  
baiyfbupt 已提交
372 
<li><strong>program</strong>(Program): 用于蒸馏训练的fluid program。默认值：<a href="https://www.paddlepaddle.org.cn/documentation/docs/zh/1.3/api_cn/fluid_cn.html#default-main-program"><em>fluid.default_main_program()</em></a></li>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
373 374 375 
<li><strong>teacher_temperature</strong>(float): 对teacher_var进行soft操作的温度值，温度值越大得到的特征图越平滑 </li>
<li><strong>student_temperature</strong>(float): 对student_var进行soft操作的温度值，温度值越大得到的特征图越平滑 </li>
</ul>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
376 
<p><strong>返回：</strong> 由teacher_var, student_var组合得到的soft_label_loss</p>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
377 
<p><strong>使用示例：</strong></p>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 
<table class="codehilitetable"><tr><td class="linenos"><div class="linenodiv"><pre><span></span> 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20</pre></div></td><td class="code"><div class="codehilite"><pre><span></span><span class="kn">import</span> <span class="nn">paddle.fluid</span> <span class="kn">as</span> <span class="nn">fluid</span>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 
<span class="kn">import</span> <span class="nn">paddleslim.dist</span> <span class="kn">as</span> <span class="nn">dist</span>
<span class="n">student_program</span> <span class="o">=</span> <span class="n">fluid</span><span class="o">.</span><span class="n">Program</span><span class="p">()</span>
<span class="k">with</span> <span class="n">fluid</span><span class="o">.</span><span class="n">program_guard</span><span class="p">(</span><span class="n">student_program</span><span class="p">):</span>
    <span class="n">x</span> <span class="o">=</span> <span class="n">fluid</span><span class="o">.</span><span class="n">layers</span><span class="o">.</span><span class="n">data</span><span class="p">(</span><span class="n">name</span><span class="o">=</span><span class="s1">&#39;x&#39;</span><span class="p">,</span> <span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="mi">1</span><span class="p">,</span> <span class="mi">28</span><span class="p">,</span> <span class="mi">28</span><span class="p">])</span>
    <span class="n">conv</span> <span class="o">=</span> <span class="n">fluid</span><span class="o">.</span><span class="n">layers</span><span class="o">.</span><span class="n">conv2d</span><span class="p">(</span><span class="n">x</span><span class="p">,</span> <span class="mi">32</span><span class="p">,</span> <span class="mi">1</span><span class="p">,</span> <span class="n">name</span><span class="o">=</span><span class="s1">&#39;s1&#39;</span><span class="p">)</span>
    <span class="n">out</span> <span class="o">=</span> <span class="n">fluid</span><span class="o">.</span><span class="n">layers</span><span class="o">.</span><span class="n">conv2d</span><span class="p">(</span><span class="n">conv</span><span class="p">,</span> <span class="mi">64</span><span class="p">,</span> <span class="mi">3</span><span class="p">,</span> <span class="n">padding</span><span class="o">=</span><span class="mi">1</span><span class="p">,</span> <span class="n">name</span><span class="o">=</span><span class="s1">&#39;s2&#39;</span><span class="p">)</span>
<span class="n">teacher_program</span> <span class="o">=</span> <span class="n">fluid</span><span class="o">.</span><span class="n">Program</span><span class="p">()</span>
<span class="k">with</span> <span class="n">fluid</span><span class="o">.</span><span class="n">program_guard</span><span class="p">(</span><span class="n">teacher_program</span><span class="p">):</span>
    <span class="n">y</span> <span class="o">=</span> <span class="n">fluid</span><span class="o">.</span><span class="n">layers</span><span class="o">.</span><span class="n">data</span><span class="p">(</span><span class="n">name</span><span class="o">=</span><span class="s1">&#39;y&#39;</span><span class="p">,</span> <span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="mi">1</span><span class="p">,</span> <span class="mi">28</span><span class="p">,</span> <span class="mi">28</span><span class="p">])</span>
    <span class="n">conv</span> <span class="o">=</span> <span class="n">fluid</span><span class="o">.</span><span class="n">layers</span><span class="o">.</span><span class="n">conv2d</span><span class="p">(</span><span class="n">y</span><span class="p">,</span> <span class="mi">32</span><span class="p">,</span> <span class="mi">1</span><span class="p">,</span> <span class="n">name</span><span class="o">=</span><span class="s1">&#39;t1&#39;</span><span class="p">)</span>
    <span class="n">conv</span> <span class="o">=</span> <span class="n">fluid</span><span class="o">.</span><span class="n">layers</span><span class="o">.</span><span class="n">conv2d</span><span class="p">(</span><span class="n">conv</span><span class="p">,</span> <span class="mi">32</span><span class="p">,</span> <span class="mi">3</span><span class="p">,</span> <span class="n">padding</span><span class="o">=</span><span class="mi">1</span><span class="p">)</span>
    <span class="n">out</span> <span class="o">=</span> <span class="n">fluid</span><span class="o">.</span><span class="n">layers</span><span class="o">.</span><span class="n">conv2d</span><span class="p">(</span><span class="n">conv</span><span class="p">,</span> <span class="mi">64</span><span class="p">,</span> <span class="mi">3</span><span class="p">,</span> <span class="n">padding</span><span class="o">=</span><span class="mi">1</span><span class="p">,</span> <span class="n">name</span><span class="o">=</span><span class="s1">&#39;t2&#39;</span><span class="p">)</span>
<span class="n">data_name_map</span> <span class="o">=</span> <span class="p">{</span><span class="s1">&#39;y&#39;</span><span class="p">:</span><span class="s1">&#39;x&#39;</span><span class="p">}</span>
<span class="n">USE_GPU</span> <span class="o">=</span> <span class="bp">False</span>
<span class="n">place</span> <span class="o">=</span> <span class="n">fluid</span><span class="o">.</span><span class="n">CUDAPlace</span><span class="p">(</span><span class="mi">0</span><span class="p">)</span> <span class="k">if</span> <span class="n">USE_GPU</span> <span class="k">else</span> <span class="n">fluid</span><span class="o">.</span><span class="n">CPUPlace</span><span class="p">()</span>
<span class="n">main_program</span> <span class="o">=</span> <span class="n">merge</span><span class="p">(</span><span class="n">teacher_program</span><span class="p">,</span> <span class="n">student_program</span><span class="p">,</span> <span class="n">data_name_map</span><span class="p">,</span> <span class="n">place</span><span class="p">)</span>
<span class="k">with</span> <span class="n">fluid</span><span class="o">.</span><span class="n">program_guard</span><span class="p">(</span><span class="n">main_program</span><span class="p">):</span>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
415 416 417 418 
<span class="hll">    <span class="n">distillation_loss</span> <span class="o">=</span> <span class="n">dist</span><span class="o">.</span><span class="n">soft_label_loss</span><span class="p">(</span><span class="s1">&#39;teacher_t2.tmp_1&#39;</span><span class="p">,</span>
</span><span class="hll">                                             <span class="s1">&#39;s2.tmp_1&#39;</span><span class="p">,</span> <span class="n">main_program</span><span class="p">,</span> <span class="mf">1.</span><span class="p">,</span> <span class="mf">1.</span><span class="p">)</span>
</span></pre></div>
</td></tr></table>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
419 420 421 

<h2 id="loss">loss<a class="headerlink" href="#loss" title="Permanent link">#</a></h2>
<dl>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
422 
<dt>paddleslim.dist.loss(loss_func, program=fluid.default_main_program(), **kwargs) <a href="https://github.com/PaddlePaddle/PaddleSlim/blob/develop/paddleslim/dist/single_distiller.py#L165">[源代码]</a></dt>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
423 424 425 426 427 428 429 
<dd>
<p>loss函数支持对任意多对teacher_var和student_var使用自定义损失函数</p>
</dd>
</dl>
<p><strong>参数：</strong></p>
<ul>
<li><strong>loss_func</strong>(python function): 自定义的损失函数，输入为teacher var和student var，输出为自定义的loss </li>
B
Deployed c953571 with MkDocs version: 1.0.4  
baiyfbupt 已提交
430 
<li><strong>program</strong>(Program): 用于蒸馏训练的fluid program。默认值：<a href="https://www.paddlepaddle.org.cn/documentation/docs/zh/1.3/api_cn/fluid_cn.html#default-main-program"><em>fluid.default_main_program()</em></a></li>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
431 432 433 434 
<li><strong>**kwargs</strong>: loss_func输入名与对应variable名称</li>
</ul>
<p><strong>返回</strong>：自定义的损失函数loss</p>
<p><strong>使用示例：</strong></p>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 
<table class="codehilitetable"><tr><td class="linenos"><div class="linenodiv"><pre><span></span> 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</pre></div></td><td class="code"><div class="codehilite"><pre><span></span><span class="kn">import</span> <span class="nn">paddle.fluid</span> <span class="kn">as</span> <span class="nn">fluid</span>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 
<span class="kn">import</span> <span class="nn">paddleslim.dist</span> <span class="kn">as</span> <span class="nn">dist</span>
<span class="n">student_program</span> <span class="o">=</span> <span class="n">fluid</span><span class="o">.</span><span class="n">Program</span><span class="p">()</span>
<span class="k">with</span> <span class="n">fluid</span><span class="o">.</span><span class="n">program_guard</span><span class="p">(</span><span class="n">student_program</span><span class="p">):</span>
    <span class="n">x</span> <span class="o">=</span> <span class="n">fluid</span><span class="o">.</span><span class="n">layers</span><span class="o">.</span><span class="n">data</span><span class="p">(</span><span class="n">name</span><span class="o">=</span><span class="s1">&#39;x&#39;</span><span class="p">,</span> <span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="mi">1</span><span class="p">,</span> <span class="mi">28</span><span class="p">,</span> <span class="mi">28</span><span class="p">])</span>
    <span class="n">conv</span> <span class="o">=</span> <span class="n">fluid</span><span class="o">.</span><span class="n">layers</span><span class="o">.</span><span class="n">conv2d</span><span class="p">(</span><span class="n">x</span><span class="p">,</span> <span class="mi">32</span><span class="p">,</span> <span class="mi">1</span><span class="p">,</span> <span class="n">name</span><span class="o">=</span><span class="s1">&#39;s1&#39;</span><span class="p">)</span>
    <span class="n">out</span> <span class="o">=</span> <span class="n">fluid</span><span class="o">.</span><span class="n">layers</span><span class="o">.</span><span class="n">conv2d</span><span class="p">(</span><span class="n">conv</span><span class="p">,</span> <span class="mi">64</span><span class="p">,</span> <span class="mi">3</span><span class="p">,</span> <span class="n">padding</span><span class="o">=</span><span class="mi">1</span><span class="p">,</span> <span class="n">name</span><span class="o">=</span><span class="s1">&#39;s2&#39;</span><span class="p">)</span>
<span class="n">teacher_program</span> <span class="o">=</span> <span class="n">fluid</span><span class="o">.</span><span class="n">Program</span><span class="p">()</span>
<span class="k">with</span> <span class="n">fluid</span><span class="o">.</span><span class="n">program_guard</span><span class="p">(</span><span class="n">teacher_program</span><span class="p">):</span>
    <span class="n">y</span> <span class="o">=</span> <span class="n">fluid</span><span class="o">.</span><span class="n">layers</span><span class="o">.</span><span class="n">data</span><span class="p">(</span><span class="n">name</span><span class="o">=</span><span class="s1">&#39;y&#39;</span><span class="p">,</span> <span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="mi">1</span><span class="p">,</span> <span class="mi">28</span><span class="p">,</span> <span class="mi">28</span><span class="p">])</span>
    <span class="n">conv</span> <span class="o">=</span> <span class="n">fluid</span><span class="o">.</span><span class="n">layers</span><span class="o">.</span><span class="n">conv2d</span><span class="p">(</span><span class="n">y</span><span class="p">,</span> <span class="mi">32</span><span class="p">,</span> <span class="mi">1</span><span class="p">,</span> <span class="n">name</span><span class="o">=</span><span class="s1">&#39;t1&#39;</span><span class="p">)</span>
    <span class="n">conv</span> <span class="o">=</span> <span class="n">fluid</span><span class="o">.</span><span class="n">layers</span><span class="o">.</span><span class="n">conv2d</span><span class="p">(</span><span class="n">conv</span><span class="p">,</span> <span class="mi">32</span><span class="p">,</span> <span class="mi">3</span><span class="p">,</span> <span class="n">padding</span><span class="o">=</span><span class="mi">1</span><span class="p">)</span>
    <span class="n">out</span> <span class="o">=</span> <span class="n">fluid</span><span class="o">.</span><span class="n">layers</span><span class="o">.</span><span class="n">conv2d</span><span class="p">(</span><span class="n">conv</span><span class="p">,</span> <span class="mi">64</span><span class="p">,</span> <span class="mi">3</span><span class="p">,</span> <span class="n">padding</span><span class="o">=</span><span class="mi">1</span><span class="p">,</span> <span class="n">name</span><span class="o">=</span><span class="s1">&#39;t2&#39;</span><span class="p">)</span>
<span class="n">data_name_map</span> <span class="o">=</span> <span class="p">{</span><span class="s1">&#39;y&#39;</span><span class="p">:</span><span class="s1">&#39;x&#39;</span><span class="p">}</span>
<span class="n">USE_GPU</span> <span class="o">=</span> <span class="bp">False</span>
<span class="n">place</span> <span class="o">=</span> <span class="n">fluid</span><span class="o">.</span><span class="n">CUDAPlace</span><span class="p">(</span><span class="mi">0</span><span class="p">)</span> <span class="k">if</span> <span class="n">USE_GPU</span> <span class="k">else</span> <span class="n">fluid</span><span class="o">.</span><span class="n">CPUPlace</span><span class="p">()</span>
<span class="n">main_program</span> <span class="o">=</span> <span class="n">merge</span><span class="p">(</span><span class="n">teacher_program</span><span class="p">,</span> <span class="n">student_program</span><span class="p">,</span> <span class="n">data_name_map</span><span class="p">,</span> <span class="n">place</span><span class="p">)</span>
<span class="k">def</span> <span class="nf">adaptation_loss</span><span class="p">(</span><span class="n">t_var</span><span class="p">,</span> <span class="n">s_var</span><span class="p">):</span>
    <span class="n">teacher_channel</span> <span class="o">=</span> <span class="n">t_var</span><span class="o">.</span><span class="n">shape</span><span class="p">[</span><span class="mi">1</span><span class="p">]</span>
    <span class="n">s_hint</span> <span class="o">=</span> <span class="n">fluid</span><span class="o">.</span><span class="n">layers</span><span class="o">.</span><span class="n">conv2d</span><span class="p">(</span><span class="n">s_var</span><span class="p">,</span> <span class="n">teacher_channel</span><span class="p">,</span> <span class="mi">1</span><span class="p">)</span>
    <span class="n">hint_loss</span> <span class="o">=</span> <span class="n">fluid</span><span class="o">.</span><span class="n">layers</span><span class="o">.</span><span class="n">reduce_mean</span><span class="p">(</span><span class="n">fluid</span><span class="o">.</span><span class="n">layers</span><span class="o">.</span><span class="n">square</span><span class="p">(</span><span class="n">s_hint</span> <span class="o">-</span> <span class="n">t_var</span><span class="p">))</span>
    <span class="k">return</span> <span class="n">hint_loss</span>
<span class="k">with</span> <span class="n">fluid</span><span class="o">.</span><span class="n">program_guard</span><span class="p">(</span><span class="n">main_program</span><span class="p">):</span>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
482 483 484 485 
<span class="hll">    <span class="n">distillation_loss</span> <span class="o">=</span> <span class="n">dist</span><span class="o">.</span><span class="n">loss</span><span class="p">(</span><span class="n">main_program</span><span class="p">,</span> <span class="n">adaptation_loss</span><span class="p">,</span>
</span><span class="hll">            <span class="n">t_var</span><span class="o">=</span><span class="s1">&#39;teacher_t2.tmp_1&#39;</span><span class="p">,</span> <span class="n">s_var</span><span class="o">=</span><span class="s1">&#39;s2.tmp_1&#39;</span><span class="p">)</span>
</span></pre></div>
</td></tr></table>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
486
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
487 488 
<div class="admonition note">
<p class="admonition-title">注意事项</p>
B
Deployed 76338d9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
489 
<p>在添加蒸馏loss时会引入新的variable，需要注意新引入的variable不要与student variables命名冲突。这里建议两种用法（两种方法任选其一即可）：</p>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
490 491 
<ol>
<li>建议与student_program使用同一个命名空间，以避免一些未指定名称的variables(例如tmp_0, tmp_1...)多次定义为同一名称出现命名冲突</li>
B
Deployed 76338d9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
492 
<li>建议在添加蒸馏loss时指定一个命名空间前缀，具体用法请参考Paddle官方文档<a href="https://www.paddlepaddle.org.cn/documentation/docs/zh/api_cn/fluid_cn/name_scope_cn.html#name-scope"><em>fluid.name_scope</em></a></li>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
493 
</ol>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
494 
</div>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
495 496 497 498 499 500 501 502 503 504 
              
            </div>
          </div>
          <footer>
  
    <div class="rst-footer-buttons" role="navigation" aria-label="footer navigation">
      
        <a href="../nas_api/" class="btn btn-neutral float-right" title="SA搜索">Next <span class="icon icon-circle-arrow-right"></span></a>
B
Deployed 0f9c3ce with MkDocs version: 1.0.4  
baiyfbupt 已提交
505 
        <a href="../analysis_api/" class="btn btn-neutral" title="模型分析"><span class="icon icon-circle-arrow-left"></span> Previous</a>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
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 
      
    </div>
  

  <hr/>

  <div role="contentinfo">
    <!-- Copyright etc -->
    
  </div>

  Built with <a href="http://www.mkdocs.org">MkDocs</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>

  <div class="rst-versions" role="note" style="cursor: pointer">
    <span class="rst-current-version" data-toggle="rst-current-version">
      
          <a href="https://github.com/PaddlePaddle/PaddleSlim/" class="fa fa-github" style="float: left; color: #fcfcfc"> GitHub</a>
      
      
        <span><a href="../analysis_api/" style="color: #fcfcfc;">&laquo; Previous</a></span>
      
      
        <span style="margin-left: 15px"><a href="../nas_api/" style="color: #fcfcfc">Next &raquo;</a></span>
      
    </span>
</div>
    <script>var base_url = '../..';</script>
    <script src="../../js/theme.js" defer></script>
      <script src="../../mathjax-config.js" defer></script>
B
Deployed 0f9c3ce with MkDocs version: 1.0.4  
baiyfbupt 已提交
543 
      <script src="https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.0/MathJax.js?config=TeX-AMS-MML_HTMLorMML" defer></script>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
544 545 546 547 
      <script src="../../search/main.js" defer></script>

</body>
</html>