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 48.0 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 
  <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">
B
Deployed 405a468 with MkDocs version: 1.0.4  
baiyfbupt 已提交
17 
  <link href="../../extra.css" rel="stylesheet">
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
18 19 20 
  
  <script>
    // Current page data
B
Deployed 0f9c3ce with MkDocs version: 1.0.4  
baiyfbupt 已提交
21 
    var mkdocs_page_name = "\u77e5\u8bc6\u84b8\u998f";
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
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 
    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">
B
Deployed 482d368 with MkDocs version: 1.0.4  
baiyfbupt 已提交
59 60 61 62 63 
    <a class="" href="../../model_zoo/">模型库</a>
	    </li>
          
            <li class="toctree-l1">
B
Deployed 55210fe with MkDocs version: 1.0.4  
baiyfbupt 已提交
64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 
    <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="">
W
Deployed 7790b33 with MkDocs version: 1.0.4  
wanghaoshuang 已提交
84 85 86 87 
    <a class="" href="../../tutorials/search_space.md">搜索空间</a>
                </li>
                <li class="">
B
Deployed 55210fe with MkDocs version: 1.0.4  
baiyfbupt 已提交
88 89 90 91 92 93 94 
    <a class="" href="../../tutorials/distillation_demo/">知识蒸馏</a>
                </li>
    </ul>
	    </li>
          
            <li class="toctree-l1">
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
95 96 97 98 99 100 101 102 
    <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
103 
    <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
104 105 106 
                </li>
                <li class="">
B
Deployed 0f9c3ce with MkDocs version: 1.0.4  
baiyfbupt 已提交
107 
    <a class="" href="../analysis_api/">模型分析</a>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
108 109 110 
                </li>
                <li class=" current">
B
Deployed 0f9c3ce with MkDocs version: 1.0.4  
baiyfbupt 已提交
111 
    <a class="current" href="./">知识蒸馏</a>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
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 
    <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="">
B
Deployed 0f9c3ce with MkDocs version: 1.0.4  
baiyfbupt 已提交
137 138 
    <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
139 140 141 
    </ul>
	    </li>
B
Deployed 0f9c3ce with MkDocs version: 1.0.4  
baiyfbupt 已提交
142 143 144 145 146 
            <li class="toctree-l1">
		
    <a class="" href="../../algo/algo/">算法原理</a>
	    </li>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
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 
        </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
173 
    <li>知识蒸馏</li>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
174 175 
    <li class="wy-breadcrumbs-aside">
B
Deployed 09a1807 with MkDocs version: 1.0.4  
baiyfbupt 已提交
176 
        <a href="https://github.com/PaddlePaddle/PaddleSlim/edit/master/docs/api/single_distiller_api.md"
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
177 178 179 180 181 182 183 184 185 186 187 
          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 09a1807 with MkDocs version: 1.0.4  
baiyfbupt 已提交
188 
<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
189 
<dd>
B
Deployed 09a1807 with MkDocs version: 1.0.4  
baiyfbupt 已提交
190 
<p>merge将teacher_program融合到student_program中。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
191 192 193 194 
</dd>
</dl>
<p><strong>参数：</strong></p>
<ul>
B
Deployed c953571 with MkDocs version: 1.0.4  
baiyfbupt 已提交
195 196 
<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
197 
<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
198 
<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
199 200 
<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
201 
</ul>
B
Deployed 09a1807 with MkDocs version: 1.0.4  
baiyfbupt 已提交
202 
<p><strong>返回：</strong> 无</p>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
203 204 205 206 
<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
207 
<p><strong>使用示例：</strong></p>
B
Deployed d95da67 with MkDocs version: 1.0.4  
baiyfbupt 已提交
208 
<div class="codehilite"><pre><span></span><span class="kn">import</span> <span class="nn">paddle.fluid</span> <span class="k">as</span> <span class="nn">fluid</span>
B
Deployed 405a468 with MkDocs version: 1.0.4  
baiyfbupt 已提交
209 
<span class="kn">import</span> <span class="nn">paddleslim.dist</span> <span class="k">as</span> <span class="nn">dist</span>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
210 211 212 
<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
213 214 215 216 217 218 219 220 221 
    <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>
B
Deployed 405a468 with MkDocs version: 1.0.4  
baiyfbupt 已提交
222 
<span class="n">USE_GPU</span> <span class="o">=</span> <span class="kc">False</span>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
223 
<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 09a1807 with MkDocs version: 1.0.4  
baiyfbupt 已提交
224 
<span class="hll"><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>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
225 226 
</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>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
227 228 229 

<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
230 
<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
231 
<dd>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
232 
<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
233 234 235 236 237 238 239 240 
</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
241 
<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
242 
</ul>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
243 
<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
244 
<p><strong>使用示例：</strong></p>
B
Deployed d95da67 with MkDocs version: 1.0.4  
baiyfbupt 已提交
245 
<div class="codehilite"><pre><span></span><span class="kn">import</span> <span class="nn">paddle.fluid</span> <span class="k">as</span> <span class="nn">fluid</span>
B
Deployed 405a468 with MkDocs version: 1.0.4  
baiyfbupt 已提交
246 
<span class="kn">import</span> <span class="nn">paddleslim.dist</span> <span class="k">as</span> <span class="nn">dist</span>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
247 248 249 250 251 252 253 254 255 256 257 258 
<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>
B
Deployed 405a468 with MkDocs version: 1.0.4  
baiyfbupt 已提交
259 
<span class="n">USE_GPU</span> <span class="o">=</span> <span class="kc">False</span>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
260 
<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>
W
Deployed 7790b33 with MkDocs version: 1.0.4  
wanghaoshuang 已提交
261 
<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 class="n">data_name_map</span><span class="p">,</span> <span class="n">place</span><span class="p">)</span>
B
Deployed 09a1807 with MkDocs version: 1.0.4  
baiyfbupt 已提交
262 
<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>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
263 
<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>
W
Deployed 7790b33 with MkDocs version: 1.0.4  
wanghaoshuang 已提交
264 
</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">student_program</span><span class="p">)</span>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
265 
</span></pre></div>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
266 267 268 

<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
269 
<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
270 271 272 273 274 275 
<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>
B
Deployed 09a1807 with MkDocs version: 1.0.4  
baiyfbupt 已提交
276 
<li><strong>teacher_var_name</strong>(str): teacher_var的名称.</li>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
277 
<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
278 
<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
279 
</ul>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
280 
<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
281 
<p><strong>使用示例：</strong></p>
B
Deployed d95da67 with MkDocs version: 1.0.4  
baiyfbupt 已提交
282 
<div class="codehilite"><pre><span></span><span class="kn">import</span> <span class="nn">paddle.fluid</span> <span class="k">as</span> <span class="nn">fluid</span>
B
Deployed 405a468 with MkDocs version: 1.0.4  
baiyfbupt 已提交
283 
<span class="kn">import</span> <span class="nn">paddleslim.dist</span> <span class="k">as</span> <span class="nn">dist</span>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
284 285 286 287 288 289 290 291 292 293 294 295 
<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>
B
Deployed 405a468 with MkDocs version: 1.0.4  
baiyfbupt 已提交
296 
<span class="n">USE_GPU</span> <span class="o">=</span> <span class="kc">False</span>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
297 
<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>
W
Deployed 7790b33 with MkDocs version: 1.0.4  
wanghaoshuang 已提交
298 
<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 class="n">data_name_map</span><span class="p">,</span> <span class="n">place</span><span class="p">)</span>
B
Deployed 09a1807 with MkDocs version: 1.0.4  
baiyfbupt 已提交
299 
<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>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
300 
<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>
W
Deployed 7790b33 with MkDocs version: 1.0.4  
wanghaoshuang 已提交
301 
</span><span class="hll">                                     <span class="n">student_program</span><span class="p">)</span>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
302 
</span></pre></div>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
303 304 305 

<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
306 
<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
307 
<dd>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
308 
<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
309 310 311 312 
</dd>
</dl>
<p><strong>参数：</strong></p>
<ul>
B
Deployed 09a1807 with MkDocs version: 1.0.4  
baiyfbupt 已提交
313 314 
<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 09a1807 with MkDocs version: 1.0.4  
baiyfbupt 已提交
316 317 
<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>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
318 
</ul>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
319 
<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
320 
<p><strong>使用示例：</strong></p>
B
Deployed d95da67 with MkDocs version: 1.0.4  
baiyfbupt 已提交
321 
<div class="codehilite"><pre><span></span><span class="kn">import</span> <span class="nn">paddle.fluid</span> <span class="k">as</span> <span class="nn">fluid</span>
B
Deployed 405a468 with MkDocs version: 1.0.4  
baiyfbupt 已提交
322 
<span class="kn">import</span> <span class="nn">paddleslim.dist</span> <span class="k">as</span> <span class="nn">dist</span>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
323 324 325 326 327 328 329 330 331 332 333 334 
<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>
B
Deployed 405a468 with MkDocs version: 1.0.4  
baiyfbupt 已提交
335 
<span class="n">USE_GPU</span> <span class="o">=</span> <span class="kc">False</span>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
336 
<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>
W
Deployed 7790b33 with MkDocs version: 1.0.4  
wanghaoshuang 已提交
337 
<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 class="n">data_name_map</span><span class="p">,</span> <span class="n">place</span><span class="p">)</span>
B
Deployed 09a1807 with MkDocs version: 1.0.4  
baiyfbupt 已提交
338 
<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>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
339 
<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>
W
Deployed 7790b33 with MkDocs version: 1.0.4  
wanghaoshuang 已提交
340 
</span><span class="hll">                                             <span class="s1">&#39;s2.tmp_1&#39;</span><span class="p">,</span> <span class="n">student_program</span><span class="p">,</span> <span class="mf">1.</span><span class="p">,</span> <span class="mf">1.</span><span class="p">)</span>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
341 
</span></pre></div>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
342 343 344 

<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
345 
<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
346 347 348 349 350 351 
<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>
B
Deployed 09a1807 with MkDocs version: 1.0.4  
baiyfbupt 已提交
352 
<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
353 
<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
354 355 356 357 
<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 d95da67 with MkDocs version: 1.0.4  
baiyfbupt 已提交
358 
<div class="codehilite"><pre><span></span><span class="kn">import</span> <span class="nn">paddle.fluid</span> <span class="k">as</span> <span class="nn">fluid</span>
B
Deployed 405a468 with MkDocs version: 1.0.4  
baiyfbupt 已提交
359 
<span class="kn">import</span> <span class="nn">paddleslim.dist</span> <span class="k">as</span> <span class="nn">dist</span>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
360 361 362 363 364 365 366 367 368 369 370 371 
<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>
B
Deployed 405a468 with MkDocs version: 1.0.4  
baiyfbupt 已提交
372 
<span class="n">USE_GPU</span> <span class="o">=</span> <span class="kc">False</span>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
373 
<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>
W
Deployed 7790b33 with MkDocs version: 1.0.4  
wanghaoshuang 已提交
374 
<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 class="n">data_name_map</span><span class="p">,</span> <span class="n">place</span><span class="p">)</span>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
375 376 377 378 379 
<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>
B
Deployed 09a1807 with MkDocs version: 1.0.4  
baiyfbupt 已提交
380 
<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>
W
Deployed 7790b33 with MkDocs version: 1.0.4  
wanghaoshuang 已提交
381 
<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">adaptation_loss</span><span class="p">,</span> <span class="n">student_program</span><span class="p">,</span>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
382 383 
</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>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
384
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
385 386 
<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
387 
<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
388 389 
<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
390 
<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
391 
</ol>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
392 
</div>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
393 394 395 396 397 398 399 400 401 402 
              
            </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
403 
        <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
404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 
      
    </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
441 
      <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
442 443 444 445 
      <script src="../../search/main.js" defer></script>

</body>
</html>