<!-- HTML header for doxygen 1.8.14-->
<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd">
<html xmlns="http://www.w3.org/1999/xhtml">
<head>
<meta http-equiv="Content-Type" content="text/xhtml;charset=UTF-8"/>
<meta http-equiv="X-UA-Compatible" content="IE=9"/>
<meta name="generator" content="Doxygen 1.8.11"/>
<meta name="viewport" content="width=device-width, initial-scale=1"/>
<title>Caffe2 - Python API: caffe2/python/mkl/convnet_benchmarks.py Source File</title>
<link href="tabs.css" rel="stylesheet" type="text/css"/>
<link rel="icon" href="/static/favicon.png" type="image/x-icon">
<script type="text/javascript" src="jquery.js"></script>
<script type="text/javascript" src="dynsections.js"></script>
<link href="search/search.css" rel="stylesheet" type="text/css"/>
<script type="text/javascript" src="search/searchdata.js"></script>
<script type="text/javascript" src="search/search.js"></script>
<script type="text/javascript">
$(document).ready(function() { init_search(); });
</script>
<link href="stylesheet.css" rel="stylesheet" type="text/css" />
<link href="main.css" rel="stylesheet" type="text/css"/>
</head>
<body>
<div id="top"><!-- do not remove this div, it is closed by doxygen! -->
<div id="titlearea">
<table cellspacing="0" cellpadding="0">
<tbody>
<tr style="height: 56px;">
<td id="projectlogo" width="56"><a href="/"><img alt="Logo" src="Caffe2-with-name-55-tall.png"/></a></td>
<td id="projectalign" style="padding-left: 0.5em;">
<div id="projectname">Caffe2 - Python API
</div>
<div id="projectbrief">A deep learning, cross platform ML framework</div>
</td>
</tr>
</tbody>
</table>
</div>
<!-- end header part -->
<!-- Generated by Doxygen 1.8.11 -->
<script type="text/javascript">
var searchBox = new SearchBox("searchBox", "search",false,'Search');
</script>
<div id="navrow1" class="tabs">
<ul class="tablist">
<li><a href="namespaces.html"><span>Packages</span></a></li>
<li><a href="annotated.html"><span>Classes</span></a></li>
<li class="current"><a href="files.html"><span>Files</span></a></li>
<li><a href="/doxygen-c/html/classes.html"><span>C++ API</span></a></li>
<li><a href="/doxygen-python/html/annotated.html"><span>Python API</span></a></li>
<li><a href="https://github.com/caffe2/caffe2"><span>GitHub</span></a></li>
<li>
<div id="MSearchBox" class="MSearchBoxInactive">
<span class="left">
<img id="MSearchSelect" src="search/mag_sel.png"
onmouseover="return searchBox.OnSearchSelectShow()"
onmouseout="return searchBox.OnSearchSelectHide()"
alt=""/>
<input type="text" id="MSearchField" value="Search" accesskey="S"
onfocus="searchBox.OnSearchFieldFocus(true)"
onblur="searchBox.OnSearchFieldFocus(false)"
onkeyup="searchBox.OnSearchFieldChange(event)"/>
</span><span class="right">
<a id="MSearchClose" href="javascript:searchBox.CloseResultsWindow()"><img id="MSearchCloseImg" border="0" src="search/close.png" alt=""/></a>
</span>
</div>
</li>
</ul>
</div>
<div id="navrow2" class="tabs2">
<ul class="tablist">
<li><a href="files.html"><span>File List</span></a></li>
</ul>
</div>
<!-- window showing the filter options -->
<div id="MSearchSelectWindow"
onmouseover="return searchBox.OnSearchSelectShow()"
onmouseout="return searchBox.OnSearchSelectHide()"
onkeydown="return searchBox.OnSearchSelectKey(event)">
</div>
<!-- iframe showing the search results (closed by default) -->
<div id="MSearchResultsWindow">
<iframe src="javascript:void(0)" frameborder="0"
name="MSearchResults" id="MSearchResults">
</iframe>
</div>
<div id="nav-path" class="navpath">
<ul>
<li class="navelem"><a class="el" href="dir_20697b8f204bdfcab31e6b1a416f3ab8.html">caffe2</a></li><li class="navelem"><a class="el" href="dir_f4fda25fc5253eea1ed54677ae5fa2de.html">python</a></li><li class="navelem"><a class="el" href="dir_66a75f7994308fa68b6557c2ea15a7a1.html">mkl</a></li> </ul>
</div>
</div><!-- top -->
<div class="header">
<div class="headertitle">
<div class="title">convnet_benchmarks.py</div> </div>
</div><!--header-->
<div class="contents">
<div class="fragment"><div class="line"><a name="l00001"></a><span class="lineno"> 1</span> <span class="comment">## @package convnet_benchmarks</span></div><div class="line"><a name="l00002"></a><span class="lineno"> 2</span> <span class="comment"># Module caffe2.python.convnet_benchmarks</span></div><div class="line"><a name="l00003"></a><span class="lineno"> 3</span> <span class="keyword">from</span> __future__ <span class="keyword">import</span> absolute_import</div><div class="line"><a name="l00004"></a><span class="lineno"> 4</span> <span class="keyword">from</span> __future__ <span class="keyword">import</span> division</div><div class="line"><a name="l00005"></a><span class="lineno"> 5</span> <span class="keyword">from</span> __future__ <span class="keyword">import</span> print_function</div><div class="line"><a name="l00006"></a><span class="lineno"> 6</span> <span class="keyword">from</span> __future__ <span class="keyword">import</span> unicode_literals</div><div class="line"><a name="l00007"></a><span class="lineno"> 7</span> <span class="stringliteral">"""</span></div><div class="line"><a name="l00008"></a><span class="lineno"> 8</span> <span class="stringliteral">Benchmark for common convnets.</span></div><div class="line"><a name="l00009"></a><span class="lineno"> 9</span> <span class="stringliteral"></span></div><div class="line"><a name="l00010"></a><span class="lineno"> 10</span> <span class="stringliteral">Speed on Titan X, with 10 warmup steps and 10 main steps and with different</span></div><div class="line"><a name="l00011"></a><span class="lineno"> 11</span> <span class="stringliteral">versions of cudnn, are as follows (time reported below is per-batch time,</span></div><div class="line"><a name="l00012"></a><span class="lineno"> 12</span> <span class="stringliteral">forward / forward+backward):</span></div><div class="line"><a name="l00013"></a><span class="lineno"> 13</span> <span class="stringliteral"></span></div><div class="line"><a name="l00014"></a><span class="lineno"> 14</span> <span class="stringliteral"> CuDNN V3 CuDNN v4</span></div><div class="line"><a name="l00015"></a><span class="lineno"> 15</span> <span class="stringliteral">AlexNet 32.5 / 108.0 27.4 / 90.1</span></div><div class="line"><a name="l00016"></a><span class="lineno"> 16</span> <span class="stringliteral">OverFeat 113.0 / 342.3 91.7 / 276.5</span></div><div class="line"><a name="l00017"></a><span class="lineno"> 17</span> <span class="stringliteral">Inception 134.5 / 485.8 125.7 / 450.6</span></div><div class="line"><a name="l00018"></a><span class="lineno"> 18</span> <span class="stringliteral">VGG (batch 64) 200.8 / 650.0 164.1 / 551.7</span></div><div class="line"><a name="l00019"></a><span class="lineno"> 19</span> <span class="stringliteral"></span></div><div class="line"><a name="l00020"></a><span class="lineno"> 20</span> <span class="stringliteral">Speed on Inception with varied batch sizes and CuDNN v4 is as follows:</span></div><div class="line"><a name="l00021"></a><span class="lineno"> 21</span> <span class="stringliteral"></span></div><div class="line"><a name="l00022"></a><span class="lineno"> 22</span> <span class="stringliteral">Batch Size Speed per batch Speed per image</span></div><div class="line"><a name="l00023"></a><span class="lineno"> 23</span> <span class="stringliteral"> 16 22.8 / 72.7 1.43 / 4.54</span></div><div class="line"><a name="l00024"></a><span class="lineno"> 24</span> <span class="stringliteral"> 32 38.0 / 127.5 1.19 / 3.98</span></div><div class="line"><a name="l00025"></a><span class="lineno"> 25</span> <span class="stringliteral"> 64 67.2 / 233.6 1.05 / 3.65</span></div><div class="line"><a name="l00026"></a><span class="lineno"> 26</span> <span class="stringliteral">128 125.7 / 450.6 0.98 / 3.52</span></div><div class="line"><a name="l00027"></a><span class="lineno"> 27</span> <span class="stringliteral"></span></div><div class="line"><a name="l00028"></a><span class="lineno"> 28</span> <span class="stringliteral">Speed on Tesla M40, which 10 warmup steps and 10 main steps and with cudnn</span></div><div class="line"><a name="l00029"></a><span class="lineno"> 29</span> <span class="stringliteral">v4, is as follows:</span></div><div class="line"><a name="l00030"></a><span class="lineno"> 30</span> <span class="stringliteral"></span></div><div class="line"><a name="l00031"></a><span class="lineno"> 31</span> <span class="stringliteral">AlexNet 68.4 / 218.1</span></div><div class="line"><a name="l00032"></a><span class="lineno"> 32</span> <span class="stringliteral">OverFeat 210.5 / 630.3</span></div><div class="line"><a name="l00033"></a><span class="lineno"> 33</span> <span class="stringliteral">Inception 300.2 / 1122.2</span></div><div class="line"><a name="l00034"></a><span class="lineno"> 34</span> <span class="stringliteral">VGG (batch 64) 405.8 / 1327.7</span></div><div class="line"><a name="l00035"></a><span class="lineno"> 35</span> <span class="stringliteral"></span></div><div class="line"><a name="l00036"></a><span class="lineno"> 36</span> <span class="stringliteral">(Note that these numbers involve a "full" backprop, i.e. the gradient</span></div><div class="line"><a name="l00037"></a><span class="lineno"> 37</span> <span class="stringliteral">with respect to the input image is also computed.)</span></div><div class="line"><a name="l00038"></a><span class="lineno"> 38</span> <span class="stringliteral"></span></div><div class="line"><a name="l00039"></a><span class="lineno"> 39</span> <span class="stringliteral">To get the numbers, simply run:</span></div><div class="line"><a name="l00040"></a><span class="lineno"> 40</span> <span class="stringliteral"></span></div><div class="line"><a name="l00041"></a><span class="lineno"> 41</span> <span class="stringliteral">for MODEL in AlexNet OverFeat Inception; do</span></div><div class="line"><a name="l00042"></a><span class="lineno"> 42</span> <span class="stringliteral"> PYTHONPATH=../gen:$PYTHONPATH python convnet_benchmarks.py \</span></div><div class="line"><a name="l00043"></a><span class="lineno"> 43</span> <span class="stringliteral"> --batch_size 128 --model $MODEL --forward_only True</span></div><div class="line"><a name="l00044"></a><span class="lineno"> 44</span> <span class="stringliteral">done</span></div><div class="line"><a name="l00045"></a><span class="lineno"> 45</span> <span class="stringliteral">for MODEL in AlexNet OverFeat Inception; do</span></div><div class="line"><a name="l00046"></a><span class="lineno"> 46</span> <span class="stringliteral"> PYTHONPATH=../gen:$PYTHONPATH python convnet_benchmarks.py \</span></div><div class="line"><a name="l00047"></a><span class="lineno"> 47</span> <span class="stringliteral"> --batch_size 128 --model $MODEL</span></div><div class="line"><a name="l00048"></a><span class="lineno"> 48</span> <span class="stringliteral">done</span></div><div class="line"><a name="l00049"></a><span class="lineno"> 49</span> <span class="stringliteral">PYTHONPATH=../gen:$PYTHONPATH python convnet_benchmarks.py \</span></div><div class="line"><a name="l00050"></a><span class="lineno"> 50</span> <span class="stringliteral"> --batch_size 64 --model VGGA --forward_only True</span></div><div class="line"><a name="l00051"></a><span class="lineno"> 51</span> <span class="stringliteral">PYTHONPATH=../gen:$PYTHONPATH python convnet_benchmarks.py \</span></div><div class="line"><a name="l00052"></a><span class="lineno"> 52</span> <span class="stringliteral"> --batch_size 64 --model VGGA</span></div><div class="line"><a name="l00053"></a><span class="lineno"> 53</span> <span class="stringliteral"></span></div><div class="line"><a name="l00054"></a><span class="lineno"> 54</span> <span class="stringliteral">for BS in 16 32 64 128; do</span></div><div class="line"><a name="l00055"></a><span class="lineno"> 55</span> <span class="stringliteral"> PYTHONPATH=../gen:$PYTHONPATH python convnet_benchmarks.py \</span></div><div class="line"><a name="l00056"></a><span class="lineno"> 56</span> <span class="stringliteral"> --batch_size $BS --model Inception --forward_only True</span></div><div class="line"><a name="l00057"></a><span class="lineno"> 57</span> <span class="stringliteral"> PYTHONPATH=../gen:$PYTHONPATH python convnet_benchmarks.py \</span></div><div class="line"><a name="l00058"></a><span class="lineno"> 58</span> <span class="stringliteral"> --batch_size $BS --model Inception</span></div><div class="line"><a name="l00059"></a><span class="lineno"> 59</span> <span class="stringliteral">done</span></div><div class="line"><a name="l00060"></a><span class="lineno"> 60</span> <span class="stringliteral"></span></div><div class="line"><a name="l00061"></a><span class="lineno"> 61</span> <span class="stringliteral">Note that VGG needs to be run at batch 64 due to memory limit on the backward</span></div><div class="line"><a name="l00062"></a><span class="lineno"> 62</span> <span class="stringliteral">pass.</span></div><div class="line"><a name="l00063"></a><span class="lineno"> 63</span> <span class="stringliteral">"""</span></div><div class="line"><a name="l00064"></a><span class="lineno"> 64</span> </div><div class="line"><a name="l00065"></a><span class="lineno"> 65</span> <span class="keyword">import</span> argparse</div><div class="line"><a name="l00066"></a><span class="lineno"> 66</span> </div><div class="line"><a name="l00067"></a><span class="lineno"> 67</span> <span class="keyword">from</span> <a class="code" href="namespacecaffe2_1_1python.html">caffe2.python</a> <span class="keyword">import</span> brew, cnn, workspace</div><div class="line"><a name="l00068"></a><span class="lineno"> 68</span> <span class="keyword">from</span> <a class="code" href="namespacecaffe2_1_1python_1_1model__helper.html">caffe2.python.model_helper</a> <span class="keyword">import</span> ModelHelper</div><div class="line"><a name="l00069"></a><span class="lineno"> 69</span> </div><div class="line"><a name="l00070"></a><span class="lineno"> 70</span> <span class="keyword">from</span> <a class="code" href="namespacecaffe2_1_1python_1_1models.html">caffe2.python.models</a> <span class="keyword">import</span> resnet</div><div class="line"><a name="l00071"></a><span class="lineno"> 71</span> <span class="keyword">import</span> numpy <span class="keyword">as</span> np</div><div class="line"><a name="l00072"></a><span class="lineno"> 72</span> </div><div class="line"><a name="l00073"></a><span class="lineno"> 73</span> <span class="keyword">def </span>MLP(order, cudnn_ws, mkl):</div><div class="line"><a name="l00074"></a><span class="lineno"> 74</span>  model = ModelHelper(name=<span class="stringliteral">"benchmark"</span>)</div><div class="line"><a name="l00075"></a><span class="lineno"> 75</span>  d = 256</div><div class="line"><a name="l00076"></a><span class="lineno"> 76</span>  depth = 20</div><div class="line"><a name="l00077"></a><span class="lineno"> 77</span>  width = 3</div><div class="line"><a name="l00078"></a><span class="lineno"> 78</span>  <span class="keywordflow">for</span> i <span class="keywordflow">in</span> range(depth):</div><div class="line"><a name="l00079"></a><span class="lineno"> 79</span>  <span class="keywordflow">for</span> j <span class="keywordflow">in</span> range(width):</div><div class="line"><a name="l00080"></a><span class="lineno"> 80</span>  current = <span class="stringliteral">"fc_{}_{}"</span>.format(i, j) <span class="keywordflow">if</span> i > 0 <span class="keywordflow">else</span> <span class="stringliteral">"data"</span></div><div class="line"><a name="l00081"></a><span class="lineno"> 81</span>  next_ = <span class="stringliteral">"fc_{}_{}"</span>.format(i + 1, j)</div><div class="line"><a name="l00082"></a><span class="lineno"> 82</span>  brew.fc(</div><div class="line"><a name="l00083"></a><span class="lineno"> 83</span>  model,</div><div class="line"><a name="l00084"></a><span class="lineno"> 84</span>  current, next_,</div><div class="line"><a name="l00085"></a><span class="lineno"> 85</span>  dim_in=d, dim_out=d,</div><div class="line"><a name="l00086"></a><span class="lineno"> 86</span>  weight_init=(<span class="stringliteral">'XavierFill'</span>, {}),</div><div class="line"><a name="l00087"></a><span class="lineno"> 87</span>  bias_init=(<span class="stringliteral">'XavierFill'</span>, {}))</div><div class="line"><a name="l00088"></a><span class="lineno"> 88</span> </div><div class="line"><a name="l00089"></a><span class="lineno"> 89</span>  brew.sum(model, [<span class="stringliteral">"fc_{}_{}"</span>.format(depth, j) <span class="keywordflow">for</span> j <span class="keywordflow">in</span> range(width)], [<span class="stringliteral">"sum"</span>])</div><div class="line"><a name="l00090"></a><span class="lineno"> 90</span>  brew.fc(model, <span class="stringliteral">"sum"</span>, <span class="stringliteral">"last"</span>,</div><div class="line"><a name="l00091"></a><span class="lineno"> 91</span>  dim_in=d, dim_out=1000,</div><div class="line"><a name="l00092"></a><span class="lineno"> 92</span>  weight_init=(<span class="stringliteral">'XavierFill'</span>, {}),</div><div class="line"><a name="l00093"></a><span class="lineno"> 93</span>  bias_init=(<span class="stringliteral">'XavierFill'</span>, {}))</div><div class="line"><a name="l00094"></a><span class="lineno"> 94</span>  xent = model.LabelCrossEntropy([<span class="stringliteral">"last"</span>, <span class="stringliteral">"label"</span>], <span class="stringliteral">"xent"</span>)</div><div class="line"><a name="l00095"></a><span class="lineno"> 95</span>  <span class="keywordflow">if</span> <span class="keywordflow">not</span> mkl:</div><div class="line"><a name="l00096"></a><span class="lineno"> 96</span>  model.AveragedLoss(xent, <span class="stringliteral">"loss"</span>)</div><div class="line"><a name="l00097"></a><span class="lineno"> 97</span>  <span class="keywordflow">return</span> model, d</div><div class="line"><a name="l00098"></a><span class="lineno"> 98</span> </div><div class="line"><a name="l00099"></a><span class="lineno"> 99</span> </div><div class="line"><a name="l00100"></a><span class="lineno"> 100</span> <span class="keyword">def </span>ResNet50(order, cudnn_ws, mkl):</div><div class="line"><a name="l00101"></a><span class="lineno"> 101</span>  my_arg_scope = {<span class="stringliteral">'order'</span>: order, <span class="stringliteral">'use_cudnn'</span>: <span class="keyword">True</span>,</div><div class="line"><a name="l00102"></a><span class="lineno"> 102</span>  <span class="stringliteral">'cudnn_exhaustive_search'</span>: <span class="keyword">True</span>,</div><div class="line"><a name="l00103"></a><span class="lineno"> 103</span>  <span class="stringliteral">'ws_nbytes_limit'</span>: str(cudnn_ws)}</div><div class="line"><a name="l00104"></a><span class="lineno"> 104</span>  model = ModelHelper(name=<span class="stringliteral">"alexnet"</span>, arg_scope=my_arg_scope)</div><div class="line"><a name="l00105"></a><span class="lineno"> 105</span>  resnet.create_resnet50(model, <span class="stringliteral">"data"</span>, 3, 1000, is_test=<span class="keyword">True</span>,</div><div class="line"><a name="l00106"></a><span class="lineno"> 106</span>  final_avg_kernel=14)</div><div class="line"><a name="l00107"></a><span class="lineno"> 107</span>  <span class="keywordflow">return</span> model, 448</div><div class="line"><a name="l00108"></a><span class="lineno"> 108</span> </div><div class="line"><a name="l00109"></a><span class="lineno"> 109</span> <span class="keyword">def </span>AlexNet(order, cudnn_ws, mkl):</div><div class="line"><a name="l00110"></a><span class="lineno"> 110</span>  my_arg_scope = {<span class="stringliteral">'order'</span>: order, <span class="stringliteral">'use_cudnn'</span>: <span class="keyword">True</span>,</div><div class="line"><a name="l00111"></a><span class="lineno"> 111</span>  <span class="stringliteral">'cudnn_exhaustive_search'</span>: <span class="keyword">True</span>,</div><div class="line"><a name="l00112"></a><span class="lineno"> 112</span>  <span class="stringliteral">'ws_nbytes_limit'</span>: str(cudnn_ws)}</div><div class="line"><a name="l00113"></a><span class="lineno"> 113</span>  model = ModelHelper(name=<span class="stringliteral">"alexnet"</span>, arg_scope=my_arg_scope)</div><div class="line"><a name="l00114"></a><span class="lineno"> 114</span>  conv1 = brew.conv(</div><div class="line"><a name="l00115"></a><span class="lineno"> 115</span>  model,</div><div class="line"><a name="l00116"></a><span class="lineno"> 116</span>  <span class="stringliteral">"data"</span>,</div><div class="line"><a name="l00117"></a><span class="lineno"> 117</span>  <span class="stringliteral">"conv1"</span>,</div><div class="line"><a name="l00118"></a><span class="lineno"> 118</span>  3,</div><div class="line"><a name="l00119"></a><span class="lineno"> 119</span>  64,</div><div class="line"><a name="l00120"></a><span class="lineno"> 120</span>  11,</div><div class="line"><a name="l00121"></a><span class="lineno"> 121</span>  (<span class="stringliteral">'XavierFill'</span>, {}),</div><div class="line"><a name="l00122"></a><span class="lineno"> 122</span>  (<span class="stringliteral">'ConstantFill'</span>, {}),</div><div class="line"><a name="l00123"></a><span class="lineno"> 123</span>  stride=4,</div><div class="line"><a name="l00124"></a><span class="lineno"> 124</span>  pad=2</div><div class="line"><a name="l00125"></a><span class="lineno"> 125</span>  )</div><div class="line"><a name="l00126"></a><span class="lineno"> 126</span>  relu1 = brew.relu(model, conv1, <span class="stringliteral">"conv1"</span>)</div><div class="line"><a name="l00127"></a><span class="lineno"> 127</span>  pool1 = brew.max_pool(model, relu1, <span class="stringliteral">"pool1"</span>, kernel=3, stride=2)</div><div class="line"><a name="l00128"></a><span class="lineno"> 128</span>  conv2 = brew.conv(</div><div class="line"><a name="l00129"></a><span class="lineno"> 129</span>  model,</div><div class="line"><a name="l00130"></a><span class="lineno"> 130</span>  pool1,</div><div class="line"><a name="l00131"></a><span class="lineno"> 131</span>  <span class="stringliteral">"conv2"</span>,</div><div class="line"><a name="l00132"></a><span class="lineno"> 132</span>  64,</div><div class="line"><a name="l00133"></a><span class="lineno"> 133</span>  192,</div><div class="line"><a name="l00134"></a><span class="lineno"> 134</span>  5,</div><div class="line"><a name="l00135"></a><span class="lineno"> 135</span>  (<span class="stringliteral">'XavierFill'</span>, {}),</div><div class="line"><a name="l00136"></a><span class="lineno"> 136</span>  (<span class="stringliteral">'ConstantFill'</span>, {}),</div><div class="line"><a name="l00137"></a><span class="lineno"> 137</span>  pad=2</div><div class="line"><a name="l00138"></a><span class="lineno"> 138</span>  )</div><div class="line"><a name="l00139"></a><span class="lineno"> 139</span>  relu2 = brew.relu(model, conv2, <span class="stringliteral">"conv2"</span>)</div><div class="line"><a name="l00140"></a><span class="lineno"> 140</span>  pool2 = brew.max_pool(model, relu2, <span class="stringliteral">"pool2"</span>, kernel=3, stride=2)</div><div class="line"><a name="l00141"></a><span class="lineno"> 141</span>  conv3 = brew.conv(</div><div class="line"><a name="l00142"></a><span class="lineno"> 142</span>  model,</div><div class="line"><a name="l00143"></a><span class="lineno"> 143</span>  pool2,</div><div class="line"><a name="l00144"></a><span class="lineno"> 144</span>  <span class="stringliteral">"conv3"</span>,</div><div class="line"><a name="l00145"></a><span class="lineno"> 145</span>  192,</div><div class="line"><a name="l00146"></a><span class="lineno"> 146</span>  384,</div><div class="line"><a name="l00147"></a><span class="lineno"> 147</span>  3,</div><div class="line"><a name="l00148"></a><span class="lineno"> 148</span>  (<span class="stringliteral">'XavierFill'</span>, {}),</div><div class="line"><a name="l00149"></a><span class="lineno"> 149</span>  (<span class="stringliteral">'ConstantFill'</span>, {}),</div><div class="line"><a name="l00150"></a><span class="lineno"> 150</span>  pad=1</div><div class="line"><a name="l00151"></a><span class="lineno"> 151</span>  )</div><div class="line"><a name="l00152"></a><span class="lineno"> 152</span>  relu3 = brew.relu(model, conv3, <span class="stringliteral">"conv3"</span>)</div><div class="line"><a name="l00153"></a><span class="lineno"> 153</span>  conv4 = brew.conv(</div><div class="line"><a name="l00154"></a><span class="lineno"> 154</span>  model,</div><div class="line"><a name="l00155"></a><span class="lineno"> 155</span>  relu3,</div><div class="line"><a name="l00156"></a><span class="lineno"> 156</span>  <span class="stringliteral">"conv4"</span>,</div><div class="line"><a name="l00157"></a><span class="lineno"> 157</span>  384,</div><div class="line"><a name="l00158"></a><span class="lineno"> 158</span>  256,</div><div class="line"><a name="l00159"></a><span class="lineno"> 159</span>  3,</div><div class="line"><a name="l00160"></a><span class="lineno"> 160</span>  (<span class="stringliteral">'XavierFill'</span>, {}),</div><div class="line"><a name="l00161"></a><span class="lineno"> 161</span>  (<span class="stringliteral">'ConstantFill'</span>, {}),</div><div class="line"><a name="l00162"></a><span class="lineno"> 162</span>  pad=1</div><div class="line"><a name="l00163"></a><span class="lineno"> 163</span>  )</div><div class="line"><a name="l00164"></a><span class="lineno"> 164</span>  relu4 = brew.relu(model, conv4, <span class="stringliteral">"conv4"</span>)</div><div class="line"><a name="l00165"></a><span class="lineno"> 165</span>  conv5 = brew.conv(</div><div class="line"><a name="l00166"></a><span class="lineno"> 166</span>  model,</div><div class="line"><a name="l00167"></a><span class="lineno"> 167</span>  relu4,</div><div class="line"><a name="l00168"></a><span class="lineno"> 168</span>  <span class="stringliteral">"conv5"</span>,</div><div class="line"><a name="l00169"></a><span class="lineno"> 169</span>  256,</div><div class="line"><a name="l00170"></a><span class="lineno"> 170</span>  256,</div><div class="line"><a name="l00171"></a><span class="lineno"> 171</span>  3,</div><div class="line"><a name="l00172"></a><span class="lineno"> 172</span>  (<span class="stringliteral">'XavierFill'</span>, {}),</div><div class="line"><a name="l00173"></a><span class="lineno"> 173</span>  (<span class="stringliteral">'ConstantFill'</span>, {}),</div><div class="line"><a name="l00174"></a><span class="lineno"> 174</span>  pad=1</div><div class="line"><a name="l00175"></a><span class="lineno"> 175</span>  )</div><div class="line"><a name="l00176"></a><span class="lineno"> 176</span>  relu5 = brew.relu(model, conv5, <span class="stringliteral">"conv5"</span>)</div><div class="line"><a name="l00177"></a><span class="lineno"> 177</span>  pool5 = brew.max_pool(model, relu5, <span class="stringliteral">"pool5"</span>, kernel=3, stride=2)</div><div class="line"><a name="l00178"></a><span class="lineno"> 178</span>  fc6 = brew.fc(</div><div class="line"><a name="l00179"></a><span class="lineno"> 179</span>  model, pool5, <span class="stringliteral">"fc6"</span>, 256 * 6 * 6, 4096, (<span class="stringliteral">'XavierFill'</span>, {}),</div><div class="line"><a name="l00180"></a><span class="lineno"> 180</span>  (<span class="stringliteral">'ConstantFill'</span>, {})</div><div class="line"><a name="l00181"></a><span class="lineno"> 181</span>  )</div><div class="line"><a name="l00182"></a><span class="lineno"> 182</span>  relu6 = brew.relu(model, fc6, <span class="stringliteral">"fc6"</span>)</div><div class="line"><a name="l00183"></a><span class="lineno"> 183</span>  fc7 = brew.fc(</div><div class="line"><a name="l00184"></a><span class="lineno"> 184</span>  model, relu6, <span class="stringliteral">"fc7"</span>, 4096, 4096, (<span class="stringliteral">'XavierFill'</span>, {}), (<span class="stringliteral">'ConstantFill'</span>, {})</div><div class="line"><a name="l00185"></a><span class="lineno"> 185</span>  )</div><div class="line"><a name="l00186"></a><span class="lineno"> 186</span>  relu7 = brew.relu(model, fc7, <span class="stringliteral">"fc7"</span>)</div><div class="line"><a name="l00187"></a><span class="lineno"> 187</span>  fc8 = brew.fc(</div><div class="line"><a name="l00188"></a><span class="lineno"> 188</span>  model, relu7, <span class="stringliteral">"fc8"</span>, 4096, 1000, (<span class="stringliteral">'XavierFill'</span>, {}), (<span class="stringliteral">'ConstantFill'</span>, {})</div><div class="line"><a name="l00189"></a><span class="lineno"> 189</span>  )</div><div class="line"><a name="l00190"></a><span class="lineno"> 190</span>  pred = brew.softmax(model, fc8, <span class="stringliteral">"pred"</span>)</div><div class="line"><a name="l00191"></a><span class="lineno"> 191</span>  xent = model.LabelCrossEntropy([pred, <span class="stringliteral">"label"</span>], <span class="stringliteral">"xent"</span>)</div><div class="line"><a name="l00192"></a><span class="lineno"> 192</span>  <span class="keywordflow">if</span> <span class="keywordflow">not</span> mkl:</div><div class="line"><a name="l00193"></a><span class="lineno"> 193</span>  loss = model.AveragedLoss(xent, <span class="stringliteral">"loss"</span>)</div><div class="line"><a name="l00194"></a><span class="lineno"> 194</span>  <span class="keywordflow">return</span> model, 224</div><div class="line"><a name="l00195"></a><span class="lineno"> 195</span> </div><div class="line"><a name="l00196"></a><span class="lineno"> 196</span> </div><div class="line"><a name="l00197"></a><span class="lineno"> 197</span> <span class="keyword">def </span>OverFeat(order, cudnn_ws, mkl):</div><div class="line"><a name="l00198"></a><span class="lineno"> 198</span>  my_arg_scope = {<span class="stringliteral">'order'</span>: order, <span class="stringliteral">'use_cudnn'</span>: <span class="keyword">True</span>,</div><div class="line"><a name="l00199"></a><span class="lineno"> 199</span>  <span class="stringliteral">'cudnn_exhaustive_search'</span>: <span class="keyword">True</span>,</div><div class="line"><a name="l00200"></a><span class="lineno"> 200</span>  <span class="stringliteral">'ws_nbytes_limit'</span>: str(cudnn_ws)}</div><div class="line"><a name="l00201"></a><span class="lineno"> 201</span>  model = ModelHelper(name=<span class="stringliteral">'overfeat'</span>, arg_scope=my_arg_scope)</div><div class="line"><a name="l00202"></a><span class="lineno"> 202</span>  conv1 = brew.conv(</div><div class="line"><a name="l00203"></a><span class="lineno"> 203</span>  model,</div><div class="line"><a name="l00204"></a><span class="lineno"> 204</span>  <span class="stringliteral">"data"</span>,</div><div class="line"><a name="l00205"></a><span class="lineno"> 205</span>  <span class="stringliteral">"conv1"</span>,</div><div class="line"><a name="l00206"></a><span class="lineno"> 206</span>  3,</div><div class="line"><a name="l00207"></a><span class="lineno"> 207</span>  96,</div><div class="line"><a name="l00208"></a><span class="lineno"> 208</span>  11,</div><div class="line"><a name="l00209"></a><span class="lineno"> 209</span>  (<span class="stringliteral">'XavierFill'</span>, {}),</div><div class="line"><a name="l00210"></a><span class="lineno"> 210</span>  (<span class="stringliteral">'ConstantFill'</span>, {}),</div><div class="line"><a name="l00211"></a><span class="lineno"> 211</span>  stride=4</div><div class="line"><a name="l00212"></a><span class="lineno"> 212</span>  )</div><div class="line"><a name="l00213"></a><span class="lineno"> 213</span>  relu1 = brew.relu(model, conv1, <span class="stringliteral">"conv1"</span>)</div><div class="line"><a name="l00214"></a><span class="lineno"> 214</span>  pool1 = brew.max_pool(model, relu1, <span class="stringliteral">"pool1"</span>, kernel=2, stride=2)</div><div class="line"><a name="l00215"></a><span class="lineno"> 215</span>  conv2 = brew.conv(</div><div class="line"><a name="l00216"></a><span class="lineno"> 216</span>  model, pool1, <span class="stringliteral">"conv2"</span>, 96, 256, 5, (<span class="stringliteral">'XavierFill'</span>, {}), (<span class="stringliteral">'ConstantFill'</span>, {})</div><div class="line"><a name="l00217"></a><span class="lineno"> 217</span>  )</div><div class="line"><a name="l00218"></a><span class="lineno"> 218</span>  relu2 = brew.relu(model, conv2, <span class="stringliteral">"conv2"</span>)</div><div class="line"><a name="l00219"></a><span class="lineno"> 219</span>  pool2 = brew.max_pool(model, relu2, <span class="stringliteral">"pool2"</span>, kernel=2, stride=2)</div><div class="line"><a name="l00220"></a><span class="lineno"> 220</span>  conv3 = brew.conv(</div><div class="line"><a name="l00221"></a><span class="lineno"> 221</span>  model,</div><div class="line"><a name="l00222"></a><span class="lineno"> 222</span>  pool2,</div><div class="line"><a name="l00223"></a><span class="lineno"> 223</span>  <span class="stringliteral">"conv3"</span>,</div><div class="line"><a name="l00224"></a><span class="lineno"> 224</span>  256,</div><div class="line"><a name="l00225"></a><span class="lineno"> 225</span>  512,</div><div class="line"><a name="l00226"></a><span class="lineno"> 226</span>  3,</div><div class="line"><a name="l00227"></a><span class="lineno"> 227</span>  (<span class="stringliteral">'XavierFill'</span>, {}),</div><div class="line"><a name="l00228"></a><span class="lineno"> 228</span>  (<span class="stringliteral">'ConstantFill'</span>, {}),</div><div class="line"><a name="l00229"></a><span class="lineno"> 229</span>  pad=1</div><div class="line"><a name="l00230"></a><span class="lineno"> 230</span>  )</div><div class="line"><a name="l00231"></a><span class="lineno"> 231</span>  relu3 = brew.relu(model, conv3, <span class="stringliteral">"conv3"</span>)</div><div class="line"><a name="l00232"></a><span class="lineno"> 232</span>  conv4 = brew.conv(</div><div class="line"><a name="l00233"></a><span class="lineno"> 233</span>  model,</div><div class="line"><a name="l00234"></a><span class="lineno"> 234</span>  relu3,</div><div class="line"><a name="l00235"></a><span class="lineno"> 235</span>  <span class="stringliteral">"conv4"</span>,</div><div class="line"><a name="l00236"></a><span class="lineno"> 236</span>  512,</div><div class="line"><a name="l00237"></a><span class="lineno"> 237</span>  1024,</div><div class="line"><a name="l00238"></a><span class="lineno"> 238</span>  3,</div><div class="line"><a name="l00239"></a><span class="lineno"> 239</span>  (<span class="stringliteral">'XavierFill'</span>, {}),</div><div class="line"><a name="l00240"></a><span class="lineno"> 240</span>  (<span class="stringliteral">'ConstantFill'</span>, {}),</div><div class="line"><a name="l00241"></a><span class="lineno"> 241</span>  pad=1</div><div class="line"><a name="l00242"></a><span class="lineno"> 242</span>  )</div><div class="line"><a name="l00243"></a><span class="lineno"> 243</span>  relu4 = brew.relu(model, conv4, <span class="stringliteral">"conv4"</span>)</div><div class="line"><a name="l00244"></a><span class="lineno"> 244</span>  conv5 = brew.conv(</div><div class="line"><a name="l00245"></a><span class="lineno"> 245</span>  model,</div><div class="line"><a name="l00246"></a><span class="lineno"> 246</span>  relu4,</div><div class="line"><a name="l00247"></a><span class="lineno"> 247</span>  <span class="stringliteral">"conv5"</span>,</div><div class="line"><a name="l00248"></a><span class="lineno"> 248</span>  1024,</div><div class="line"><a name="l00249"></a><span class="lineno"> 249</span>  1024,</div><div class="line"><a name="l00250"></a><span class="lineno"> 250</span>  3,</div><div class="line"><a name="l00251"></a><span class="lineno"> 251</span>  (<span class="stringliteral">'XavierFill'</span>, {}),</div><div class="line"><a name="l00252"></a><span class="lineno"> 252</span>  (<span class="stringliteral">'ConstantFill'</span>, {}),</div><div class="line"><a name="l00253"></a><span class="lineno"> 253</span>  pad=1</div><div class="line"><a name="l00254"></a><span class="lineno"> 254</span>  )</div><div class="line"><a name="l00255"></a><span class="lineno"> 255</span>  relu5 = brew.relu(model, conv5, <span class="stringliteral">"conv5"</span>)</div><div class="line"><a name="l00256"></a><span class="lineno"> 256</span>  pool5 = brew.max_pool(model, relu5, <span class="stringliteral">"pool5"</span>, kernel=2, stride=2)</div><div class="line"><a name="l00257"></a><span class="lineno"> 257</span>  fc6 = brew.fc(</div><div class="line"><a name="l00258"></a><span class="lineno"> 258</span>  model, pool5, <span class="stringliteral">"fc6"</span>, 1024 * 6 * 6, 3072, (<span class="stringliteral">'XavierFill'</span>, {}),</div><div class="line"><a name="l00259"></a><span class="lineno"> 259</span>  (<span class="stringliteral">'ConstantFill'</span>, {})</div><div class="line"><a name="l00260"></a><span class="lineno"> 260</span>  )</div><div class="line"><a name="l00261"></a><span class="lineno"> 261</span>  relu6 = brew.relu(model, fc6, <span class="stringliteral">"fc6"</span>)</div><div class="line"><a name="l00262"></a><span class="lineno"> 262</span>  fc7 = brew.fc(</div><div class="line"><a name="l00263"></a><span class="lineno"> 263</span>  model, relu6, <span class="stringliteral">"fc7"</span>, 3072, 4096, (<span class="stringliteral">'XavierFill'</span>, {}), (<span class="stringliteral">'ConstantFill'</span>, {})</div><div class="line"><a name="l00264"></a><span class="lineno"> 264</span>  )</div><div class="line"><a name="l00265"></a><span class="lineno"> 265</span>  relu7 = brew.relu(model, fc7, <span class="stringliteral">"fc7"</span>)</div><div class="line"><a name="l00266"></a><span class="lineno"> 266</span>  fc8 = brew.fc(</div><div class="line"><a name="l00267"></a><span class="lineno"> 267</span>  model, relu7, <span class="stringliteral">"fc8"</span>, 4096, 1000, (<span class="stringliteral">'XavierFill'</span>, {}), (<span class="stringliteral">'ConstantFill'</span>, {})</div><div class="line"><a name="l00268"></a><span class="lineno"> 268</span>  )</div><div class="line"><a name="l00269"></a><span class="lineno"> 269</span>  pred = brew.softmax(model, fc8, <span class="stringliteral">"pred"</span>)</div><div class="line"><a name="l00270"></a><span class="lineno"> 270</span>  xent = model.LabelCrossEntropy([pred, <span class="stringliteral">"label"</span>], <span class="stringliteral">"xent"</span>)</div><div class="line"><a name="l00271"></a><span class="lineno"> 271</span>  <span class="keywordflow">if</span> <span class="keywordflow">not</span> mkl:</div><div class="line"><a name="l00272"></a><span class="lineno"> 272</span>  loss = model.AveragedLoss(xent, <span class="stringliteral">"loss"</span>)</div><div class="line"><a name="l00273"></a><span class="lineno"> 273</span>  <span class="keywordflow">return</span> model, 231</div><div class="line"><a name="l00274"></a><span class="lineno"> 274</span> </div><div class="line"><a name="l00275"></a><span class="lineno"> 275</span> </div><div class="line"><a name="l00276"></a><span class="lineno"> 276</span> <span class="keyword">def </span>VGGA(order, cudnn_ws, mkl):</div><div class="line"><a name="l00277"></a><span class="lineno"> 277</span>  my_arg_scope = {<span class="stringliteral">'order'</span>: order, <span class="stringliteral">'use_cudnn'</span>: <span class="keyword">True</span>,</div><div class="line"><a name="l00278"></a><span class="lineno"> 278</span>  <span class="stringliteral">'cudnn_exhaustive_search'</span>: <span class="keyword">True</span>,</div><div class="line"><a name="l00279"></a><span class="lineno"> 279</span>  <span class="stringliteral">'ws_nbytes_limit'</span>: str(cudnn_ws)}</div><div class="line"><a name="l00280"></a><span class="lineno"> 280</span>  model = ModelHelper(name=<span class="stringliteral">'vgg-a'</span>, arg_scope=my_arg_scope)</div><div class="line"><a name="l00281"></a><span class="lineno"> 281</span>  conv1 = brew.conv(</div><div class="line"><a name="l00282"></a><span class="lineno"> 282</span>  model,</div><div class="line"><a name="l00283"></a><span class="lineno"> 283</span>  <span class="stringliteral">"data"</span>,</div><div class="line"><a name="l00284"></a><span class="lineno"> 284</span>  <span class="stringliteral">"conv1"</span>,</div><div class="line"><a name="l00285"></a><span class="lineno"> 285</span>  3,</div><div class="line"><a name="l00286"></a><span class="lineno"> 286</span>  64,</div><div class="line"><a name="l00287"></a><span class="lineno"> 287</span>  3,</div><div class="line"><a name="l00288"></a><span class="lineno"> 288</span>  (<span class="stringliteral">'XavierFill'</span>, {}),</div><div class="line"><a name="l00289"></a><span class="lineno"> 289</span>  (<span class="stringliteral">'ConstantFill'</span>, {}),</div><div class="line"><a name="l00290"></a><span class="lineno"> 290</span>  pad=1</div><div class="line"><a name="l00291"></a><span class="lineno"> 291</span>  )</div><div class="line"><a name="l00292"></a><span class="lineno"> 292</span>  relu1 = brew.relu(model, conv1, <span class="stringliteral">"conv1"</span>)</div><div class="line"><a name="l00293"></a><span class="lineno"> 293</span>  pool1 = brew.max_pool(model, relu1, <span class="stringliteral">"pool1"</span>, kernel=2, stride=2)</div><div class="line"><a name="l00294"></a><span class="lineno"> 294</span>  conv2 = brew.conv(</div><div class="line"><a name="l00295"></a><span class="lineno"> 295</span>  model,</div><div class="line"><a name="l00296"></a><span class="lineno"> 296</span>  pool1,</div><div class="line"><a name="l00297"></a><span class="lineno"> 297</span>  <span class="stringliteral">"conv2"</span>,</div><div class="line"><a name="l00298"></a><span class="lineno"> 298</span>  64,</div><div class="line"><a name="l00299"></a><span class="lineno"> 299</span>  128,</div><div class="line"><a name="l00300"></a><span class="lineno"> 300</span>  3,</div><div class="line"><a name="l00301"></a><span class="lineno"> 301</span>  (<span class="stringliteral">'XavierFill'</span>, {}),</div><div class="line"><a name="l00302"></a><span class="lineno"> 302</span>  (<span class="stringliteral">'ConstantFill'</span>, {}),</div><div class="line"><a name="l00303"></a><span class="lineno"> 303</span>  pad=1</div><div class="line"><a name="l00304"></a><span class="lineno"> 304</span>  )</div><div class="line"><a name="l00305"></a><span class="lineno"> 305</span>  relu2 = brew.relu(model, conv2, <span class="stringliteral">"conv2"</span>)</div><div class="line"><a name="l00306"></a><span class="lineno"> 306</span>  pool2 = brew.max_pool(model, relu2, <span class="stringliteral">"pool2"</span>, kernel=2, stride=2)</div><div class="line"><a name="l00307"></a><span class="lineno"> 307</span>  conv3 = brew.conv(</div><div class="line"><a name="l00308"></a><span class="lineno"> 308</span>  model,</div><div class="line"><a name="l00309"></a><span class="lineno"> 309</span>  pool2,</div><div class="line"><a name="l00310"></a><span class="lineno"> 310</span>  <span class="stringliteral">"conv3"</span>,</div><div class="line"><a name="l00311"></a><span class="lineno"> 311</span>  128,</div><div class="line"><a name="l00312"></a><span class="lineno"> 312</span>  256,</div><div class="line"><a name="l00313"></a><span class="lineno"> 313</span>  3,</div><div class="line"><a name="l00314"></a><span class="lineno"> 314</span>  (<span class="stringliteral">'XavierFill'</span>, {}),</div><div class="line"><a name="l00315"></a><span class="lineno"> 315</span>  (<span class="stringliteral">'ConstantFill'</span>, {}),</div><div class="line"><a name="l00316"></a><span class="lineno"> 316</span>  pad=1</div><div class="line"><a name="l00317"></a><span class="lineno"> 317</span>  )</div><div class="line"><a name="l00318"></a><span class="lineno"> 318</span>  relu3 = brew.relu(model, conv3, <span class="stringliteral">"conv3"</span>)</div><div class="line"><a name="l00319"></a><span class="lineno"> 319</span>  conv4 = brew.conv(</div><div class="line"><a name="l00320"></a><span class="lineno"> 320</span>  model,</div><div class="line"><a name="l00321"></a><span class="lineno"> 321</span>  relu3,</div><div class="line"><a name="l00322"></a><span class="lineno"> 322</span>  <span class="stringliteral">"conv4"</span>,</div><div class="line"><a name="l00323"></a><span class="lineno"> 323</span>  256,</div><div class="line"><a name="l00324"></a><span class="lineno"> 324</span>  256,</div><div class="line"><a name="l00325"></a><span class="lineno"> 325</span>  3,</div><div class="line"><a name="l00326"></a><span class="lineno"> 326</span>  (<span class="stringliteral">'XavierFill'</span>, {}),</div><div class="line"><a name="l00327"></a><span class="lineno"> 327</span>  (<span class="stringliteral">'ConstantFill'</span>, {}),</div><div class="line"><a name="l00328"></a><span class="lineno"> 328</span>  pad=1</div><div class="line"><a name="l00329"></a><span class="lineno"> 329</span>  )</div><div class="line"><a name="l00330"></a><span class="lineno"> 330</span>  relu4 = brew.relu(model, conv4, <span class="stringliteral">"conv4"</span>)</div><div class="line"><a name="l00331"></a><span class="lineno"> 331</span>  pool4 = brew.max_pool(model, relu4, <span class="stringliteral">"pool4"</span>, kernel=2, stride=2)</div><div class="line"><a name="l00332"></a><span class="lineno"> 332</span>  conv5 = brew.conv(</div><div class="line"><a name="l00333"></a><span class="lineno"> 333</span>  model,</div><div class="line"><a name="l00334"></a><span class="lineno"> 334</span>  pool4,</div><div class="line"><a name="l00335"></a><span class="lineno"> 335</span>  <span class="stringliteral">"conv5"</span>,</div><div class="line"><a name="l00336"></a><span class="lineno"> 336</span>  256,</div><div class="line"><a name="l00337"></a><span class="lineno"> 337</span>  512,</div><div class="line"><a name="l00338"></a><span class="lineno"> 338</span>  3,</div><div class="line"><a name="l00339"></a><span class="lineno"> 339</span>  (<span class="stringliteral">'XavierFill'</span>, {}),</div><div class="line"><a name="l00340"></a><span class="lineno"> 340</span>  (<span class="stringliteral">'ConstantFill'</span>, {}),</div><div class="line"><a name="l00341"></a><span class="lineno"> 341</span>  pad=1</div><div class="line"><a name="l00342"></a><span class="lineno"> 342</span>  )</div><div class="line"><a name="l00343"></a><span class="lineno"> 343</span>  relu5 = brew.relu(model, conv5, <span class="stringliteral">"conv5"</span>)</div><div class="line"><a name="l00344"></a><span class="lineno"> 344</span>  conv6 = brew.conv(</div><div class="line"><a name="l00345"></a><span class="lineno"> 345</span>  model,</div><div class="line"><a name="l00346"></a><span class="lineno"> 346</span>  relu5,</div><div class="line"><a name="l00347"></a><span class="lineno"> 347</span>  <span class="stringliteral">"conv6"</span>,</div><div class="line"><a name="l00348"></a><span class="lineno"> 348</span>  512,</div><div class="line"><a name="l00349"></a><span class="lineno"> 349</span>  512,</div><div class="line"><a name="l00350"></a><span class="lineno"> 350</span>  3,</div><div class="line"><a name="l00351"></a><span class="lineno"> 351</span>  (<span class="stringliteral">'XavierFill'</span>, {}),</div><div class="line"><a name="l00352"></a><span class="lineno"> 352</span>  (<span class="stringliteral">'ConstantFill'</span>, {}),</div><div class="line"><a name="l00353"></a><span class="lineno"> 353</span>  pad=1</div><div class="line"><a name="l00354"></a><span class="lineno"> 354</span>  )</div><div class="line"><a name="l00355"></a><span class="lineno"> 355</span>  relu6 = brew.relu(model, conv6, <span class="stringliteral">"conv6"</span>)</div><div class="line"><a name="l00356"></a><span class="lineno"> 356</span>  pool6 = brew.max_pool(model, relu6, <span class="stringliteral">"pool6"</span>, kernel=2, stride=2)</div><div class="line"><a name="l00357"></a><span class="lineno"> 357</span>  conv7 = brew.conv(</div><div class="line"><a name="l00358"></a><span class="lineno"> 358</span>  model,</div><div class="line"><a name="l00359"></a><span class="lineno"> 359</span>  pool6,</div><div class="line"><a name="l00360"></a><span class="lineno"> 360</span>  <span class="stringliteral">"conv7"</span>,</div><div class="line"><a name="l00361"></a><span class="lineno"> 361</span>  512,</div><div class="line"><a name="l00362"></a><span class="lineno"> 362</span>  512,</div><div class="line"><a name="l00363"></a><span class="lineno"> 363</span>  3,</div><div class="line"><a name="l00364"></a><span class="lineno"> 364</span>  (<span class="stringliteral">'XavierFill'</span>, {}),</div><div class="line"><a name="l00365"></a><span class="lineno"> 365</span>  (<span class="stringliteral">'ConstantFill'</span>, {}),</div><div class="line"><a name="l00366"></a><span class="lineno"> 366</span>  pad=1</div><div class="line"><a name="l00367"></a><span class="lineno"> 367</span>  )</div><div class="line"><a name="l00368"></a><span class="lineno"> 368</span>  relu7 = brew.relu(model, conv7, <span class="stringliteral">"conv7"</span>)</div><div class="line"><a name="l00369"></a><span class="lineno"> 369</span>  conv8 = brew.conv(</div><div class="line"><a name="l00370"></a><span class="lineno"> 370</span>  model,</div><div class="line"><a name="l00371"></a><span class="lineno"> 371</span>  relu7,</div><div class="line"><a name="l00372"></a><span class="lineno"> 372</span>  <span class="stringliteral">"conv8"</span>,</div><div class="line"><a name="l00373"></a><span class="lineno"> 373</span>  512,</div><div class="line"><a name="l00374"></a><span class="lineno"> 374</span>  512,</div><div class="line"><a name="l00375"></a><span class="lineno"> 375</span>  3,</div><div class="line"><a name="l00376"></a><span class="lineno"> 376</span>  (<span class="stringliteral">'XavierFill'</span>, {}),</div><div class="line"><a name="l00377"></a><span class="lineno"> 377</span>  (<span class="stringliteral">'ConstantFill'</span>, {}),</div><div class="line"><a name="l00378"></a><span class="lineno"> 378</span>  pad=1</div><div class="line"><a name="l00379"></a><span class="lineno"> 379</span>  )</div><div class="line"><a name="l00380"></a><span class="lineno"> 380</span>  relu8 = brew.relu(model, conv8, <span class="stringliteral">"conv8"</span>)</div><div class="line"><a name="l00381"></a><span class="lineno"> 381</span>  pool8 = brew.max_pool(model, relu8, <span class="stringliteral">"pool8"</span>, kernel=2, stride=2)</div><div class="line"><a name="l00382"></a><span class="lineno"> 382</span> </div><div class="line"><a name="l00383"></a><span class="lineno"> 383</span>  fcix = brew.fc(</div><div class="line"><a name="l00384"></a><span class="lineno"> 384</span>  model, pool8, <span class="stringliteral">"fcix"</span>, 512 * 7 * 7, 4096, (<span class="stringliteral">'XavierFill'</span>, {}),</div><div class="line"><a name="l00385"></a><span class="lineno"> 385</span>  (<span class="stringliteral">'ConstantFill'</span>, {})</div><div class="line"><a name="l00386"></a><span class="lineno"> 386</span>  )</div><div class="line"><a name="l00387"></a><span class="lineno"> 387</span>  reluix = brew.relu(model, fcix, <span class="stringliteral">"fcix"</span>)</div><div class="line"><a name="l00388"></a><span class="lineno"> 388</span>  fcx = brew.fc(</div><div class="line"><a name="l00389"></a><span class="lineno"> 389</span>  model, reluix, <span class="stringliteral">"fcx"</span>, 4096, 4096, (<span class="stringliteral">'XavierFill'</span>, {}), (<span class="stringliteral">'ConstantFill'</span>, {})</div><div class="line"><a name="l00390"></a><span class="lineno"> 390</span>  )</div><div class="line"><a name="l00391"></a><span class="lineno"> 391</span>  relux = brew.relu(model, fcx, <span class="stringliteral">"fcx"</span>)</div><div class="line"><a name="l00392"></a><span class="lineno"> 392</span>  fcxi = brew.fc(</div><div class="line"><a name="l00393"></a><span class="lineno"> 393</span>  model, relux, <span class="stringliteral">"fcxi"</span>, 4096, 1000, (<span class="stringliteral">'XavierFill'</span>, {}), (<span class="stringliteral">'ConstantFill'</span>, {})</div><div class="line"><a name="l00394"></a><span class="lineno"> 394</span>  )</div><div class="line"><a name="l00395"></a><span class="lineno"> 395</span>  pred = brew.softmax(model, fcxi, <span class="stringliteral">"pred"</span>)</div><div class="line"><a name="l00396"></a><span class="lineno"> 396</span>  xent = model.LabelCrossEntropy([pred, <span class="stringliteral">"label"</span>], <span class="stringliteral">"xent"</span>)</div><div class="line"><a name="l00397"></a><span class="lineno"> 397</span>  <span class="keywordflow">if</span> <span class="keywordflow">not</span> mkl:</div><div class="line"><a name="l00398"></a><span class="lineno"> 398</span>  loss = model.AveragedLoss(xent, <span class="stringliteral">"loss"</span>)</div><div class="line"><a name="l00399"></a><span class="lineno"> 399</span>  <span class="keywordflow">return</span> model, 231</div><div class="line"><a name="l00400"></a><span class="lineno"> 400</span> </div><div class="line"><a name="l00401"></a><span class="lineno"> 401</span> </div><div class="line"><a name="l00402"></a><span class="lineno"> 402</span> <span class="keyword">def </span>_InceptionModule(</div><div class="line"><a name="l00403"></a><span class="lineno"> 403</span>  model, input_blob, input_depth, output_name, conv1_depth, conv3_depths,</div><div class="line"><a name="l00404"></a><span class="lineno"> 404</span>  conv5_depths, pool_depth</div><div class="line"><a name="l00405"></a><span class="lineno"> 405</span> ):</div><div class="line"><a name="l00406"></a><span class="lineno"> 406</span>  <span class="comment"># path 1: 1x1 conv</span></div><div class="line"><a name="l00407"></a><span class="lineno"> 407</span>  conv1 = brew.conv(</div><div class="line"><a name="l00408"></a><span class="lineno"> 408</span>  model, input_blob, output_name + <span class="stringliteral">":conv1"</span>, input_depth, conv1_depth, 1,</div><div class="line"><a name="l00409"></a><span class="lineno"> 409</span>  (<span class="stringliteral">'XavierFill'</span>, {}), (<span class="stringliteral">'ConstantFill'</span>, {})</div><div class="line"><a name="l00410"></a><span class="lineno"> 410</span>  )</div><div class="line"><a name="l00411"></a><span class="lineno"> 411</span>  conv1 = brew.relu(model, conv1, conv1)</div><div class="line"><a name="l00412"></a><span class="lineno"> 412</span>  <span class="comment"># path 2: 1x1 conv + 3x3 conv</span></div><div class="line"><a name="l00413"></a><span class="lineno"> 413</span>  conv3_reduce = brew.conv(</div><div class="line"><a name="l00414"></a><span class="lineno"> 414</span>  model, input_blob, output_name + <span class="stringliteral">":conv3_reduce"</span>, input_depth,</div><div class="line"><a name="l00415"></a><span class="lineno"> 415</span>  conv3_depths[0], 1, (<span class="stringliteral">'XavierFill'</span>, {}), (<span class="stringliteral">'ConstantFill'</span>, {})</div><div class="line"><a name="l00416"></a><span class="lineno"> 416</span>  )</div><div class="line"><a name="l00417"></a><span class="lineno"> 417</span>  conv3_reduce = brew.relu(model, conv3_reduce, conv3_reduce)</div><div class="line"><a name="l00418"></a><span class="lineno"> 418</span>  conv3 = brew.conv(</div><div class="line"><a name="l00419"></a><span class="lineno"> 419</span>  model,</div><div class="line"><a name="l00420"></a><span class="lineno"> 420</span>  conv3_reduce,</div><div class="line"><a name="l00421"></a><span class="lineno"> 421</span>  output_name + <span class="stringliteral">":conv3"</span>,</div><div class="line"><a name="l00422"></a><span class="lineno"> 422</span>  conv3_depths[0],</div><div class="line"><a name="l00423"></a><span class="lineno"> 423</span>  conv3_depths[1],</div><div class="line"><a name="l00424"></a><span class="lineno"> 424</span>  3,</div><div class="line"><a name="l00425"></a><span class="lineno"> 425</span>  (<span class="stringliteral">'XavierFill'</span>, {}),</div><div class="line"><a name="l00426"></a><span class="lineno"> 426</span>  (<span class="stringliteral">'ConstantFill'</span>, {}),</div><div class="line"><a name="l00427"></a><span class="lineno"> 427</span>  pad=1</div><div class="line"><a name="l00428"></a><span class="lineno"> 428</span>  )</div><div class="line"><a name="l00429"></a><span class="lineno"> 429</span>  conv3 = brew.relu(model, conv3, conv3)</div><div class="line"><a name="l00430"></a><span class="lineno"> 430</span>  <span class="comment"># path 3: 1x1 conv + 5x5 conv</span></div><div class="line"><a name="l00431"></a><span class="lineno"> 431</span>  conv5_reduce = brew.conv(</div><div class="line"><a name="l00432"></a><span class="lineno"> 432</span>  model, input_blob, output_name + <span class="stringliteral">":conv5_reduce"</span>, input_depth,</div><div class="line"><a name="l00433"></a><span class="lineno"> 433</span>  conv5_depths[0], 1, (<span class="stringliteral">'XavierFill'</span>, {}), (<span class="stringliteral">'ConstantFill'</span>, {})</div><div class="line"><a name="l00434"></a><span class="lineno"> 434</span>  )</div><div class="line"><a name="l00435"></a><span class="lineno"> 435</span>  conv5_reduce = brew.relu(model, conv5_reduce, conv5_reduce)</div><div class="line"><a name="l00436"></a><span class="lineno"> 436</span>  conv5 = brew.conv(</div><div class="line"><a name="l00437"></a><span class="lineno"> 437</span>  model,</div><div class="line"><a name="l00438"></a><span class="lineno"> 438</span>  conv5_reduce,</div><div class="line"><a name="l00439"></a><span class="lineno"> 439</span>  output_name + <span class="stringliteral">":conv5"</span>,</div><div class="line"><a name="l00440"></a><span class="lineno"> 440</span>  conv5_depths[0],</div><div class="line"><a name="l00441"></a><span class="lineno"> 441</span>  conv5_depths[1],</div><div class="line"><a name="l00442"></a><span class="lineno"> 442</span>  5,</div><div class="line"><a name="l00443"></a><span class="lineno"> 443</span>  (<span class="stringliteral">'XavierFill'</span>, {}),</div><div class="line"><a name="l00444"></a><span class="lineno"> 444</span>  (<span class="stringliteral">'ConstantFill'</span>, {}),</div><div class="line"><a name="l00445"></a><span class="lineno"> 445</span>  pad=2</div><div class="line"><a name="l00446"></a><span class="lineno"> 446</span>  )</div><div class="line"><a name="l00447"></a><span class="lineno"> 447</span>  conv5 = brew.relu(model, conv5, conv5)</div><div class="line"><a name="l00448"></a><span class="lineno"> 448</span>  <span class="comment"># path 4: pool + 1x1 conv</span></div><div class="line"><a name="l00449"></a><span class="lineno"> 449</span>  pool = brew.max_pool(</div><div class="line"><a name="l00450"></a><span class="lineno"> 450</span>  model,</div><div class="line"><a name="l00451"></a><span class="lineno"> 451</span>  input_blob,</div><div class="line"><a name="l00452"></a><span class="lineno"> 452</span>  output_name + <span class="stringliteral">":pool"</span>,</div><div class="line"><a name="l00453"></a><span class="lineno"> 453</span>  kernel=3,</div><div class="line"><a name="l00454"></a><span class="lineno"> 454</span>  stride=1,</div><div class="line"><a name="l00455"></a><span class="lineno"> 455</span>  pad=1</div><div class="line"><a name="l00456"></a><span class="lineno"> 456</span>  )</div><div class="line"><a name="l00457"></a><span class="lineno"> 457</span>  pool_proj = brew.conv(</div><div class="line"><a name="l00458"></a><span class="lineno"> 458</span>  model, pool, output_name + <span class="stringliteral">":pool_proj"</span>, input_depth, pool_depth, 1,</div><div class="line"><a name="l00459"></a><span class="lineno"> 459</span>  (<span class="stringliteral">'XavierFill'</span>, {}), (<span class="stringliteral">'ConstantFill'</span>, {})</div><div class="line"><a name="l00460"></a><span class="lineno"> 460</span>  )</div><div class="line"><a name="l00461"></a><span class="lineno"> 461</span>  pool_proj = brew.relu(model, pool_proj, pool_proj)</div><div class="line"><a name="l00462"></a><span class="lineno"> 462</span>  output = brew.concat(model, [conv1, conv3, conv5, pool_proj], output_name)</div><div class="line"><a name="l00463"></a><span class="lineno"> 463</span>  <span class="keywordflow">return</span> output</div><div class="line"><a name="l00464"></a><span class="lineno"> 464</span> </div><div class="line"><a name="l00465"></a><span class="lineno"> 465</span> </div><div class="line"><a name="l00466"></a><span class="lineno"> 466</span> <span class="keyword">def </span>Inception(order, cudnn_ws, mkl):</div><div class="line"><a name="l00467"></a><span class="lineno"> 467</span>  my_arg_scope = {<span class="stringliteral">'order'</span>: order, <span class="stringliteral">'use_cudnn'</span>: <span class="keyword">True</span>,</div><div class="line"><a name="l00468"></a><span class="lineno"> 468</span>  <span class="stringliteral">'cudnn_exhaustive_search'</span>: <span class="keyword">True</span>,</div><div class="line"><a name="l00469"></a><span class="lineno"> 469</span>  <span class="stringliteral">'ws_nbytes_limit'</span>: str(cudnn_ws)}</div><div class="line"><a name="l00470"></a><span class="lineno"> 470</span>  model = ModelHelper(name=<span class="stringliteral">"inception"</span>, arg_scope=my_arg_scope)</div><div class="line"><a name="l00471"></a><span class="lineno"> 471</span>  conv1 = brew.conv(</div><div class="line"><a name="l00472"></a><span class="lineno"> 472</span>  model,</div><div class="line"><a name="l00473"></a><span class="lineno"> 473</span>  <span class="stringliteral">"data"</span>,</div><div class="line"><a name="l00474"></a><span class="lineno"> 474</span>  <span class="stringliteral">"conv1"</span>,</div><div class="line"><a name="l00475"></a><span class="lineno"> 475</span>  3,</div><div class="line"><a name="l00476"></a><span class="lineno"> 476</span>  64,</div><div class="line"><a name="l00477"></a><span class="lineno"> 477</span>  7,</div><div class="line"><a name="l00478"></a><span class="lineno"> 478</span>  (<span class="stringliteral">'XavierFill'</span>, {}),</div><div class="line"><a name="l00479"></a><span class="lineno"> 479</span>  (<span class="stringliteral">'ConstantFill'</span>, {}),</div><div class="line"><a name="l00480"></a><span class="lineno"> 480</span>  stride=2,</div><div class="line"><a name="l00481"></a><span class="lineno"> 481</span>  pad=3</div><div class="line"><a name="l00482"></a><span class="lineno"> 482</span>  )</div><div class="line"><a name="l00483"></a><span class="lineno"> 483</span>  relu1 = brew.relu(model, conv1, <span class="stringliteral">"conv1"</span>)</div><div class="line"><a name="l00484"></a><span class="lineno"> 484</span>  pool1 = brew.max_pool(model, relu1, <span class="stringliteral">"pool1"</span>, kernel=3, stride=2, pad=1)</div><div class="line"><a name="l00485"></a><span class="lineno"> 485</span>  conv2a = brew.conv(</div><div class="line"><a name="l00486"></a><span class="lineno"> 486</span>  model, pool1, <span class="stringliteral">"conv2a"</span>, 64, 64, 1,</div><div class="line"><a name="l00487"></a><span class="lineno"> 487</span>  (<span class="stringliteral">'XavierFill'</span>, {}), (<span class="stringliteral">'ConstantFill'</span>, {})</div><div class="line"><a name="l00488"></a><span class="lineno"> 488</span>  )</div><div class="line"><a name="l00489"></a><span class="lineno"> 489</span>  conv2a = brew.relu(model, conv2a, conv2a)</div><div class="line"><a name="l00490"></a><span class="lineno"> 490</span>  conv2 = brew.conv(</div><div class="line"><a name="l00491"></a><span class="lineno"> 491</span>  model,</div><div class="line"><a name="l00492"></a><span class="lineno"> 492</span>  conv2a,</div><div class="line"><a name="l00493"></a><span class="lineno"> 493</span>  <span class="stringliteral">"conv2"</span>,</div><div class="line"><a name="l00494"></a><span class="lineno"> 494</span>  64,</div><div class="line"><a name="l00495"></a><span class="lineno"> 495</span>  192,</div><div class="line"><a name="l00496"></a><span class="lineno"> 496</span>  3,</div><div class="line"><a name="l00497"></a><span class="lineno"> 497</span>  (<span class="stringliteral">'XavierFill'</span>, {}),</div><div class="line"><a name="l00498"></a><span class="lineno"> 498</span>  (<span class="stringliteral">'ConstantFill'</span>, {}),</div><div class="line"><a name="l00499"></a><span class="lineno"> 499</span>  pad=1</div><div class="line"><a name="l00500"></a><span class="lineno"> 500</span>  )</div><div class="line"><a name="l00501"></a><span class="lineno"> 501</span>  relu2 = brew.relu(model, conv2, <span class="stringliteral">"conv2"</span>)</div><div class="line"><a name="l00502"></a><span class="lineno"> 502</span>  pool2 = brew.max_pool(model, relu2, <span class="stringliteral">"pool2"</span>, kernel=3, stride=2, pad=1)</div><div class="line"><a name="l00503"></a><span class="lineno"> 503</span>  <span class="comment"># Inception modules</span></div><div class="line"><a name="l00504"></a><span class="lineno"> 504</span>  inc3 = _InceptionModule(</div><div class="line"><a name="l00505"></a><span class="lineno"> 505</span>  model, pool2, 192, <span class="stringliteral">"inc3"</span>, 64, [96, 128], [16, 32], 32</div><div class="line"><a name="l00506"></a><span class="lineno"> 506</span>  )</div><div class="line"><a name="l00507"></a><span class="lineno"> 507</span>  inc4 = _InceptionModule(</div><div class="line"><a name="l00508"></a><span class="lineno"> 508</span>  model, inc3, 256, <span class="stringliteral">"inc4"</span>, 128, [128, 192], [32, 96], 64</div><div class="line"><a name="l00509"></a><span class="lineno"> 509</span>  )</div><div class="line"><a name="l00510"></a><span class="lineno"> 510</span>  pool5 = brew.max_pool(model, inc4, <span class="stringliteral">"pool5"</span>, kernel=3, stride=2, pad=1)</div><div class="line"><a name="l00511"></a><span class="lineno"> 511</span>  inc5 = _InceptionModule(</div><div class="line"><a name="l00512"></a><span class="lineno"> 512</span>  model, pool5, 480, <span class="stringliteral">"inc5"</span>, 192, [96, 208], [16, 48], 64</div><div class="line"><a name="l00513"></a><span class="lineno"> 513</span>  )</div><div class="line"><a name="l00514"></a><span class="lineno"> 514</span>  inc6 = _InceptionModule(</div><div class="line"><a name="l00515"></a><span class="lineno"> 515</span>  model, inc5, 512, <span class="stringliteral">"inc6"</span>, 160, [112, 224], [24, 64], 64</div><div class="line"><a name="l00516"></a><span class="lineno"> 516</span>  )</div><div class="line"><a name="l00517"></a><span class="lineno"> 517</span>  inc7 = _InceptionModule(</div><div class="line"><a name="l00518"></a><span class="lineno"> 518</span>  model, inc6, 512, <span class="stringliteral">"inc7"</span>, 128, [128, 256], [24, 64], 64</div><div class="line"><a name="l00519"></a><span class="lineno"> 519</span>  )</div><div class="line"><a name="l00520"></a><span class="lineno"> 520</span>  inc8 = _InceptionModule(</div><div class="line"><a name="l00521"></a><span class="lineno"> 521</span>  model, inc7, 512, <span class="stringliteral">"inc8"</span>, 112, [144, 288], [32, 64], 64</div><div class="line"><a name="l00522"></a><span class="lineno"> 522</span>  )</div><div class="line"><a name="l00523"></a><span class="lineno"> 523</span>  inc9 = _InceptionModule(</div><div class="line"><a name="l00524"></a><span class="lineno"> 524</span>  model, inc8, 528, <span class="stringliteral">"inc9"</span>, 256, [160, 320], [32, 128], 128</div><div class="line"><a name="l00525"></a><span class="lineno"> 525</span>  )</div><div class="line"><a name="l00526"></a><span class="lineno"> 526</span>  pool9 = brew.max_pool(model, inc9, <span class="stringliteral">"pool9"</span>, kernel=3, stride=2, pad=1)</div><div class="line"><a name="l00527"></a><span class="lineno"> 527</span>  inc10 = _InceptionModule(</div><div class="line"><a name="l00528"></a><span class="lineno"> 528</span>  model, pool9, 832, <span class="stringliteral">"inc10"</span>, 256, [160, 320], [32, 128], 128</div><div class="line"><a name="l00529"></a><span class="lineno"> 529</span>  )</div><div class="line"><a name="l00530"></a><span class="lineno"> 530</span>  inc11 = _InceptionModule(</div><div class="line"><a name="l00531"></a><span class="lineno"> 531</span>  model, inc10, 832, <span class="stringliteral">"inc11"</span>, 384, [192, 384], [48, 128], 128</div><div class="line"><a name="l00532"></a><span class="lineno"> 532</span>  )</div><div class="line"><a name="l00533"></a><span class="lineno"> 533</span>  pool11 = brew.average_pool(model, inc11, <span class="stringliteral">"pool11"</span>, kernel=7, stride=1)</div><div class="line"><a name="l00534"></a><span class="lineno"> 534</span>  fc = brew.fc(</div><div class="line"><a name="l00535"></a><span class="lineno"> 535</span>  model, pool11, <span class="stringliteral">"fc"</span>, 1024, 1000,</div><div class="line"><a name="l00536"></a><span class="lineno"> 536</span>  (<span class="stringliteral">'XavierFill'</span>, {}), (<span class="stringliteral">'ConstantFill'</span>, {})</div><div class="line"><a name="l00537"></a><span class="lineno"> 537</span>  )</div><div class="line"><a name="l00538"></a><span class="lineno"> 538</span>  <span class="comment"># It seems that Soumith's benchmark does not have softmax on top</span></div><div class="line"><a name="l00539"></a><span class="lineno"> 539</span>  <span class="comment"># for Inception. We will add it anyway so we can have a proper</span></div><div class="line"><a name="l00540"></a><span class="lineno"> 540</span>  <span class="comment"># backward pass.</span></div><div class="line"><a name="l00541"></a><span class="lineno"> 541</span>  pred = brew.softmax(model, fc, <span class="stringliteral">"pred"</span>)</div><div class="line"><a name="l00542"></a><span class="lineno"> 542</span>  xent = model.LabelCrossEntropy([pred, <span class="stringliteral">"label"</span>], <span class="stringliteral">"xent"</span>)</div><div class="line"><a name="l00543"></a><span class="lineno"> 543</span>  <span class="keywordflow">if</span> <span class="keywordflow">not</span> mkl:</div><div class="line"><a name="l00544"></a><span class="lineno"> 544</span>  loss = model.AveragedLoss(xent, <span class="stringliteral">"loss"</span>)</div><div class="line"><a name="l00545"></a><span class="lineno"> 545</span>  <span class="keywordflow">return</span> model, 224</div><div class="line"><a name="l00546"></a><span class="lineno"> 546</span> </div><div class="line"><a name="l00547"></a><span class="lineno"> 547</span> </div><div class="line"><a name="l00548"></a><span class="lineno"> 548</span> <span class="keyword">def </span><a class="code" href="namespaceconvnet__benchmarks.html#ac095378d9a77f65064a88ed77e89b4a8">AddParameterUpdate</a>(model):</div><div class="line"><a name="l00549"></a><span class="lineno"> 549</span>  <span class="stringliteral">""" Simple plain SGD update -- not tuned to actually train the models """</span></div><div class="line"><a name="l00550"></a><span class="lineno"> 550</span>  ITER = brew.iter(model, <span class="stringliteral">"iter"</span>)</div><div class="line"><a name="l00551"></a><span class="lineno"> 551</span>  LR = model.LearningRate(</div><div class="line"><a name="l00552"></a><span class="lineno"> 552</span>  ITER, <span class="stringliteral">"LR"</span>, base_lr=-1e-8, policy=<span class="stringliteral">"step"</span>, stepsize=10000, gamma=0.999)</div><div class="line"><a name="l00553"></a><span class="lineno"> 553</span>  ONE = model.param_init_net.ConstantFill([], <span class="stringliteral">"ONE"</span>, shape=[1], value=1.0)</div><div class="line"><a name="l00554"></a><span class="lineno"> 554</span>  <span class="keywordflow">for</span> param <span class="keywordflow">in</span> model.params:</div><div class="line"><a name="l00555"></a><span class="lineno"> 555</span>  param_grad = model.param_to_grad[param]</div><div class="line"><a name="l00556"></a><span class="lineno"> 556</span>  model.WeightedSum([param, ONE, param_grad, LR], param)</div><div class="line"><a name="l00557"></a><span class="lineno"> 557</span> </div><div class="line"><a name="l00558"></a><span class="lineno"> 558</span> </div><div class="line"><a name="l00559"></a><span class="lineno"> 559</span> <span class="keyword">def </span>Benchmark(model_gen, arg):</div><div class="line"><a name="l00560"></a><span class="lineno"> 560</span>  model, input_size = model_gen(arg.order, arg.cudnn_ws, arg.mkl)</div><div class="line"><a name="l00561"></a><span class="lineno"> 561</span>  model.Proto().type = arg.net_type</div><div class="line"><a name="l00562"></a><span class="lineno"> 562</span>  model.Proto().num_workers = arg.num_workers</div><div class="line"><a name="l00563"></a><span class="lineno"> 563</span> </div><div class="line"><a name="l00564"></a><span class="lineno"> 564</span>  <span class="comment"># In order to be able to run everything without feeding more stuff, let's</span></div><div class="line"><a name="l00565"></a><span class="lineno"> 565</span>  <span class="comment"># add the data and label blobs to the parameter initialization net as well.</span></div><div class="line"><a name="l00566"></a><span class="lineno"> 566</span>  <span class="keywordflow">if</span> arg.order == <span class="stringliteral">"NCHW"</span>:</div><div class="line"><a name="l00567"></a><span class="lineno"> 567</span>  input_shape = [arg.batch_size, 3, input_size, input_size]</div><div class="line"><a name="l00568"></a><span class="lineno"> 568</span>  <span class="keywordflow">else</span>:</div><div class="line"><a name="l00569"></a><span class="lineno"> 569</span>  input_shape = [arg.batch_size, input_size, input_size, 3]</div><div class="line"><a name="l00570"></a><span class="lineno"> 570</span>  <span class="keywordflow">if</span> arg.model == <span class="stringliteral">"MLP"</span>:</div><div class="line"><a name="l00571"></a><span class="lineno"> 571</span>  input_shape = [arg.batch_size, input_size]</div><div class="line"><a name="l00572"></a><span class="lineno"> 572</span> </div><div class="line"><a name="l00573"></a><span class="lineno"> 573</span>  model.param_init_net.GaussianFill(</div><div class="line"><a name="l00574"></a><span class="lineno"> 574</span>  [],</div><div class="line"><a name="l00575"></a><span class="lineno"> 575</span>  <span class="stringliteral">"data"</span>,</div><div class="line"><a name="l00576"></a><span class="lineno"> 576</span>  shape=input_shape,</div><div class="line"><a name="l00577"></a><span class="lineno"> 577</span>  mean=0.0,</div><div class="line"><a name="l00578"></a><span class="lineno"> 578</span>  std=1.0</div><div class="line"><a name="l00579"></a><span class="lineno"> 579</span>  )</div><div class="line"><a name="l00580"></a><span class="lineno"> 580</span>  <span class="comment">#MKL doesn't support int, so have to use numpy</span></div><div class="line"><a name="l00581"></a><span class="lineno"> 581</span>  <span class="keywordflow">if</span> arg.mkl:</div><div class="line"><a name="l00582"></a><span class="lineno"> 582</span>  label = np.random.randint(low=0, high=1000, size=(arg.batch_size,)).astype(np.int32)</div><div class="line"><a name="l00583"></a><span class="lineno"> 583</span>  workspace.FeedBlob(<span class="stringliteral">"label"</span>, label)</div><div class="line"><a name="l00584"></a><span class="lineno"> 584</span>  <span class="keywordflow">else</span>:</div><div class="line"><a name="l00585"></a><span class="lineno"> 585</span>  model.param_init_net.UniformIntFill(</div><div class="line"><a name="l00586"></a><span class="lineno"> 586</span>  [],</div><div class="line"><a name="l00587"></a><span class="lineno"> 587</span>  <span class="stringliteral">"label"</span>,</div><div class="line"><a name="l00588"></a><span class="lineno"> 588</span>  shape=[arg.batch_size, ],</div><div class="line"><a name="l00589"></a><span class="lineno"> 589</span>  min=0,</div><div class="line"><a name="l00590"></a><span class="lineno"> 590</span>  max=999</div><div class="line"><a name="l00591"></a><span class="lineno"> 591</span>  )</div><div class="line"><a name="l00592"></a><span class="lineno"> 592</span> </div><div class="line"><a name="l00593"></a><span class="lineno"> 593</span>  <span class="keywordflow">if</span> arg.forward_only:</div><div class="line"><a name="l00594"></a><span class="lineno"> 594</span>  print(<span class="stringliteral">'{}: running forward only.'</span>.format(arg.model))</div><div class="line"><a name="l00595"></a><span class="lineno"> 595</span>  <span class="keywordflow">else</span>:</div><div class="line"><a name="l00596"></a><span class="lineno"> 596</span>  <span class="keywordflow">if</span> arg.mkl:</div><div class="line"><a name="l00597"></a><span class="lineno"> 597</span>  print(</div><div class="line"><a name="l00598"></a><span class="lineno"> 598</span>  <span class="stringliteral">'==WARNING==\n'</span></div><div class="line"><a name="l00599"></a><span class="lineno"> 599</span>  <span class="stringliteral">'forward-backward not supported yet in MKL, so exiting'</span></div><div class="line"><a name="l00600"></a><span class="lineno"> 600</span>  )</div><div class="line"><a name="l00601"></a><span class="lineno"> 601</span>  print(<span class="stringliteral">'{}: running forward-backward.'</span>.format(arg.model))</div><div class="line"><a name="l00602"></a><span class="lineno"> 602</span>  model.AddGradientOperators([<span class="stringliteral">"loss"</span>])</div><div class="line"><a name="l00603"></a><span class="lineno"> 603</span>  <a class="code" href="namespaceconvnet__benchmarks.html#ac095378d9a77f65064a88ed77e89b4a8">AddParameterUpdate</a>(model)</div><div class="line"><a name="l00604"></a><span class="lineno"> 604</span>  <span class="keywordflow">if</span> arg.order == <span class="stringliteral">'NHWC'</span>:</div><div class="line"><a name="l00605"></a><span class="lineno"> 605</span>  print(</div><div class="line"><a name="l00606"></a><span class="lineno"> 606</span>  <span class="stringliteral">'==WARNING==\n'</span></div><div class="line"><a name="l00607"></a><span class="lineno"> 607</span>  <span class="stringliteral">'NHWC order with CuDNN may not be supported yet, so I might\n'</span></div><div class="line"><a name="l00608"></a><span class="lineno"> 608</span>  <span class="stringliteral">'exit suddenly.'</span></div><div class="line"><a name="l00609"></a><span class="lineno"> 609</span>  )</div><div class="line"><a name="l00610"></a><span class="lineno"> 610</span> </div><div class="line"><a name="l00611"></a><span class="lineno"> 611</span>  <span class="keywordflow">if</span> <span class="keywordflow">not</span> arg.cpu:</div><div class="line"><a name="l00612"></a><span class="lineno"> 612</span>  <span class="keywordflow">if</span> arg.mkl:</div><div class="line"><a name="l00613"></a><span class="lineno"> 613</span>  model.param_init_net.RunAllOnMKL()</div><div class="line"><a name="l00614"></a><span class="lineno"> 614</span>  model.net.RunAllOnMKL()</div><div class="line"><a name="l00615"></a><span class="lineno"> 615</span>  <span class="keywordflow">else</span>:</div><div class="line"><a name="l00616"></a><span class="lineno"> 616</span>  model.param_init_net.RunAllOnGPU()</div><div class="line"><a name="l00617"></a><span class="lineno"> 617</span>  model.net.RunAllOnGPU()</div><div class="line"><a name="l00618"></a><span class="lineno"> 618</span> </div><div class="line"><a name="l00619"></a><span class="lineno"> 619</span>  <span class="keywordflow">if</span> arg.engine:</div><div class="line"><a name="l00620"></a><span class="lineno"> 620</span>  <span class="keywordflow">for</span> op <span class="keywordflow">in</span> model.net.Proto().op:</div><div class="line"><a name="l00621"></a><span class="lineno"> 621</span>  op.engine = arg.engine</div><div class="line"><a name="l00622"></a><span class="lineno"> 622</span> </div><div class="line"><a name="l00623"></a><span class="lineno"> 623</span>  <span class="keywordflow">if</span> arg.dump_model:</div><div class="line"><a name="l00624"></a><span class="lineno"> 624</span>  <span class="comment"># Writes out the pbtxt for benchmarks on e.g. Android</span></div><div class="line"><a name="l00625"></a><span class="lineno"> 625</span>  with open(</div><div class="line"><a name="l00626"></a><span class="lineno"> 626</span>  <span class="stringliteral">"{0}_init_batch_{1}.pbtxt"</span>.format(arg.model, arg.batch_size), <span class="stringliteral">"w"</span></div><div class="line"><a name="l00627"></a><span class="lineno"> 627</span>  ) <span class="keyword">as</span> fid:</div><div class="line"><a name="l00628"></a><span class="lineno"> 628</span>  fid.write(str(model.param_init_net.Proto()))</div><div class="line"><a name="l00629"></a><span class="lineno"> 629</span>  with open(<span class="stringliteral">"{0}.pbtxt"</span>.format(arg.model, arg.batch_size), <span class="stringliteral">"w"</span>) <span class="keyword">as</span> fid:</div><div class="line"><a name="l00630"></a><span class="lineno"> 630</span>  fid.write(str(model.net.Proto()))</div><div class="line"><a name="l00631"></a><span class="lineno"> 631</span> </div><div class="line"><a name="l00632"></a><span class="lineno"> 632</span>  workspace.RunNetOnce(model.param_init_net)</div><div class="line"><a name="l00633"></a><span class="lineno"> 633</span>  workspace.CreateNet(model.net)</div><div class="line"><a name="l00634"></a><span class="lineno"> 634</span>  workspace.BenchmarkNet(</div><div class="line"><a name="l00635"></a><span class="lineno"> 635</span>  model.net.Proto().name, arg.warmup_iterations, arg.iterations,</div><div class="line"><a name="l00636"></a><span class="lineno"> 636</span>  arg.layer_wise_benchmark)</div><div class="line"><a name="l00637"></a><span class="lineno"> 637</span> </div><div class="line"><a name="l00638"></a><span class="lineno"> 638</span> </div><div class="line"><a name="l00639"></a><span class="lineno"> 639</span> <span class="keyword">def </span>GetArgumentParser():</div><div class="line"><a name="l00640"></a><span class="lineno"> 640</span>  parser = argparse.ArgumentParser(description=<span class="stringliteral">"Caffe2 benchmark."</span>)</div><div class="line"><a name="l00641"></a><span class="lineno"> 641</span>  parser.add_argument(</div><div class="line"><a name="l00642"></a><span class="lineno"> 642</span>  <span class="stringliteral">"--batch_size"</span>,</div><div class="line"><a name="l00643"></a><span class="lineno"> 643</span>  type=int,</div><div class="line"><a name="l00644"></a><span class="lineno"> 644</span>  default=128,</div><div class="line"><a name="l00645"></a><span class="lineno"> 645</span>  help=<span class="stringliteral">"The batch size."</span></div><div class="line"><a name="l00646"></a><span class="lineno"> 646</span>  )</div><div class="line"><a name="l00647"></a><span class="lineno"> 647</span>  parser.add_argument(<span class="stringliteral">"--model"</span>, type=str, help=<span class="stringliteral">"The model to benchmark."</span>)</div><div class="line"><a name="l00648"></a><span class="lineno"> 648</span>  parser.add_argument(</div><div class="line"><a name="l00649"></a><span class="lineno"> 649</span>  <span class="stringliteral">"--order"</span>,</div><div class="line"><a name="l00650"></a><span class="lineno"> 650</span>  type=str,</div><div class="line"><a name="l00651"></a><span class="lineno"> 651</span>  default=<span class="stringliteral">"NCHW"</span>,</div><div class="line"><a name="l00652"></a><span class="lineno"> 652</span>  help=<span class="stringliteral">"The order to evaluate."</span></div><div class="line"><a name="l00653"></a><span class="lineno"> 653</span>  )</div><div class="line"><a name="l00654"></a><span class="lineno"> 654</span>  parser.add_argument(</div><div class="line"><a name="l00655"></a><span class="lineno"> 655</span>  <span class="stringliteral">"--cudnn_ws"</span>,</div><div class="line"><a name="l00656"></a><span class="lineno"> 656</span>  type=int,</div><div class="line"><a name="l00657"></a><span class="lineno"> 657</span>  help=<span class="stringliteral">"The cudnn workspace size."</span></div><div class="line"><a name="l00658"></a><span class="lineno"> 658</span>  )</div><div class="line"><a name="l00659"></a><span class="lineno"> 659</span>  parser.add_argument(</div><div class="line"><a name="l00660"></a><span class="lineno"> 660</span>  <span class="stringliteral">"--iterations"</span>,</div><div class="line"><a name="l00661"></a><span class="lineno"> 661</span>  type=int,</div><div class="line"><a name="l00662"></a><span class="lineno"> 662</span>  default=10,</div><div class="line"><a name="l00663"></a><span class="lineno"> 663</span>  help=<span class="stringliteral">"Number of iterations to run the network."</span></div><div class="line"><a name="l00664"></a><span class="lineno"> 664</span>  )</div><div class="line"><a name="l00665"></a><span class="lineno"> 665</span>  parser.add_argument(</div><div class="line"><a name="l00666"></a><span class="lineno"> 666</span>  <span class="stringliteral">"--warmup_iterations"</span>,</div><div class="line"><a name="l00667"></a><span class="lineno"> 667</span>  type=int,</div><div class="line"><a name="l00668"></a><span class="lineno"> 668</span>  default=10,</div><div class="line"><a name="l00669"></a><span class="lineno"> 669</span>  help=<span class="stringliteral">"Number of warm-up iterations before benchmarking."</span></div><div class="line"><a name="l00670"></a><span class="lineno"> 670</span>  )</div><div class="line"><a name="l00671"></a><span class="lineno"> 671</span>  parser.add_argument(</div><div class="line"><a name="l00672"></a><span class="lineno"> 672</span>  <span class="stringliteral">"--forward_only"</span>,</div><div class="line"><a name="l00673"></a><span class="lineno"> 673</span>  action=<span class="stringliteral">'store_true'</span>,</div><div class="line"><a name="l00674"></a><span class="lineno"> 674</span>  help=<span class="stringliteral">"If set, only run the forward pass."</span></div><div class="line"><a name="l00675"></a><span class="lineno"> 675</span>  )</div><div class="line"><a name="l00676"></a><span class="lineno"> 676</span>  parser.add_argument(</div><div class="line"><a name="l00677"></a><span class="lineno"> 677</span>  <span class="stringliteral">"--layer_wise_benchmark"</span>,</div><div class="line"><a name="l00678"></a><span class="lineno"> 678</span>  action=<span class="stringliteral">'store_true'</span>,</div><div class="line"><a name="l00679"></a><span class="lineno"> 679</span>  help=<span class="stringliteral">"If True, run the layer-wise benchmark as well."</span></div><div class="line"><a name="l00680"></a><span class="lineno"> 680</span>  )</div><div class="line"><a name="l00681"></a><span class="lineno"> 681</span>  parser.add_argument(</div><div class="line"><a name="l00682"></a><span class="lineno"> 682</span>  <span class="stringliteral">"--cpu"</span>,</div><div class="line"><a name="l00683"></a><span class="lineno"> 683</span>  action=<span class="stringliteral">'store_true'</span>,</div><div class="line"><a name="l00684"></a><span class="lineno"> 684</span>  help=<span class="stringliteral">"If True, run testing on CPU instead of GPU."</span></div><div class="line"><a name="l00685"></a><span class="lineno"> 685</span>  )</div><div class="line"><a name="l00686"></a><span class="lineno"> 686</span>  parser.add_argument(</div><div class="line"><a name="l00687"></a><span class="lineno"> 687</span>  <span class="stringliteral">"--mkl"</span>,</div><div class="line"><a name="l00688"></a><span class="lineno"> 688</span>  action=<span class="stringliteral">'store_true'</span>,</div><div class="line"><a name="l00689"></a><span class="lineno"> 689</span>  help=<span class="stringliteral">"If True, run testing on CPU-MKL instead of GPU."</span></div><div class="line"><a name="l00690"></a><span class="lineno"> 690</span>  )</div><div class="line"><a name="l00691"></a><span class="lineno"> 691</span>  parser.add_argument(</div><div class="line"><a name="l00692"></a><span class="lineno"> 692</span>  <span class="stringliteral">"--engine"</span>,</div><div class="line"><a name="l00693"></a><span class="lineno"> 693</span>  type=str,</div><div class="line"><a name="l00694"></a><span class="lineno"> 694</span>  default=<span class="stringliteral">""</span>,</div><div class="line"><a name="l00695"></a><span class="lineno"> 695</span>  help=<span class="stringliteral">"If set, blindly prefer the given engine(s) for every op."</span>)</div><div class="line"><a name="l00696"></a><span class="lineno"> 696</span>  parser.add_argument(</div><div class="line"><a name="l00697"></a><span class="lineno"> 697</span>  <span class="stringliteral">"--dump_model"</span>,</div><div class="line"><a name="l00698"></a><span class="lineno"> 698</span>  action=<span class="stringliteral">'store_true'</span>,</div><div class="line"><a name="l00699"></a><span class="lineno"> 699</span>  help=<span class="stringliteral">"If True, dump the model prototxts to disk."</span></div><div class="line"><a name="l00700"></a><span class="lineno"> 700</span>  )</div><div class="line"><a name="l00701"></a><span class="lineno"> 701</span>  parser.add_argument(<span class="stringliteral">"--net_type"</span>, type=str, default=<span class="stringliteral">"simple"</span>)</div><div class="line"><a name="l00702"></a><span class="lineno"> 702</span>  parser.add_argument(<span class="stringliteral">"--num_workers"</span>, type=int, default=2)</div><div class="line"><a name="l00703"></a><span class="lineno"> 703</span>  parser.add_argument(<span class="stringliteral">"--use-nvtx"</span>, default=<span class="keyword">False</span>, action=<span class="stringliteral">'store_true'</span>)</div><div class="line"><a name="l00704"></a><span class="lineno"> 704</span>  parser.add_argument(<span class="stringliteral">"--htrace_span_log_path"</span>, type=str)</div><div class="line"><a name="l00705"></a><span class="lineno"> 705</span>  <span class="keywordflow">return</span> parser</div><div class="line"><a name="l00706"></a><span class="lineno"> 706</span> </div><div class="line"><a name="l00707"></a><span class="lineno"> 707</span> </div><div class="line"><a name="l00708"></a><span class="lineno"> 708</span> <span class="keywordflow">if</span> __name__ == <span class="stringliteral">'__main__'</span>:</div><div class="line"><a name="l00709"></a><span class="lineno"> 709</span>  args, extra_args = GetArgumentParser().parse_known_args()</div><div class="line"><a name="l00710"></a><span class="lineno"> 710</span>  <span class="keywordflow">if</span> (</div><div class="line"><a name="l00711"></a><span class="lineno"> 711</span>  <span class="keywordflow">not</span> args.batch_size <span class="keywordflow">or</span> <span class="keywordflow">not</span> args.model <span class="keywordflow">or</span> <span class="keywordflow">not</span> args.order</div><div class="line"><a name="l00712"></a><span class="lineno"> 712</span>  ):</div><div class="line"><a name="l00713"></a><span class="lineno"> 713</span>  GetArgumentParser().print_help()</div><div class="line"><a name="l00714"></a><span class="lineno"> 714</span>  <span class="keywordflow">else</span>:</div><div class="line"><a name="l00715"></a><span class="lineno"> 715</span>  workspace.GlobalInit(</div><div class="line"><a name="l00716"></a><span class="lineno"> 716</span>  [<span class="stringliteral">'caffe2'</span>, <span class="stringliteral">'--caffe2_log_level=0'</span>] + extra_args +</div><div class="line"><a name="l00717"></a><span class="lineno"> 717</span>  ([<span class="stringliteral">'--caffe2_use_nvtx'</span>] <span class="keywordflow">if</span> args.use_nvtx <span class="keywordflow">else</span> []) +</div><div class="line"><a name="l00718"></a><span class="lineno"> 718</span>  ([<span class="stringliteral">'--caffe2_htrace_span_log_path='</span> + args.htrace_span_log_path]</div><div class="line"><a name="l00719"></a><span class="lineno"> 719</span>  <span class="keywordflow">if</span> args.htrace_span_log_path <span class="keywordflow">else</span> []))</div><div class="line"><a name="l00720"></a><span class="lineno"> 720</span> </div><div class="line"><a name="l00721"></a><span class="lineno"> 721</span>  model_map = {</div><div class="line"><a name="l00722"></a><span class="lineno"> 722</span>  <span class="stringliteral">'AlexNet'</span>: AlexNet,</div><div class="line"><a name="l00723"></a><span class="lineno"> 723</span>  <span class="stringliteral">'OverFeat'</span>: OverFeat,</div><div class="line"><a name="l00724"></a><span class="lineno"> 724</span>  <span class="stringliteral">'VGGA'</span>: VGGA,</div><div class="line"><a name="l00725"></a><span class="lineno"> 725</span>  <span class="stringliteral">'Inception'</span>: Inception,</div><div class="line"><a name="l00726"></a><span class="lineno"> 726</span>  <span class="stringliteral">'ResNet50'</span>: ResNet50,</div><div class="line"><a name="l00727"></a><span class="lineno"> 727</span>  <span class="stringliteral">'MLP'</span>: MLP,</div><div class="line"><a name="l00728"></a><span class="lineno"> 728</span>  }</div><div class="line"><a name="l00729"></a><span class="lineno"> 729</span>  Benchmark(model_map[args.model], args)</div><div class="ttc" id="namespacecaffe2_1_1python_html"><div class="ttname"><a href="namespacecaffe2_1_1python.html">caffe2.python</a></div><div class="ttdef"><b>Definition:</b> <a href="python_2____init_____8py_source.html#l00001">__init__.py:1</a></div></div>
<div class="ttc" id="namespacecaffe2_1_1python_1_1models_html"><div class="ttname"><a href="namespacecaffe2_1_1python_1_1models.html">caffe2.python.models</a></div><div class="ttdef"><b>Definition:</b> <a href="python_2models_2____init_____8py_source.html#l00001">__init__.py:1</a></div></div>
<div class="ttc" id="namespacecaffe2_1_1python_1_1model__helper_html"><div class="ttname"><a href="namespacecaffe2_1_1python_1_1model__helper.html">caffe2.python.model_helper</a></div><div class="ttdef"><b>Definition:</b> <a href="model__helper_8py_source.html#l00001">model_helper.py:1</a></div></div>
<div class="ttc" id="namespaceconvnet__benchmarks_html_ac095378d9a77f65064a88ed77e89b4a8"><div class="ttname"><a href="namespaceconvnet__benchmarks.html#ac095378d9a77f65064a88ed77e89b4a8">convnet_benchmarks.AddParameterUpdate</a></div><div class="ttdeci">def AddParameterUpdate(model)</div><div class="ttdef"><b>Definition:</b> <a href="experiments_2python_2convnet__benchmarks_8py_source.html#l00527">convnet_benchmarks.py:527</a></div></div>
</div><!-- fragment --></div><!-- contents -->
<!-- HTML footer for doxygen 1.8.14-->
<!-- start footer part -->
<hr class="footer"/><address class="footer"><small>
Generated on Thu Apr 19 2018 13:04:04 for Caffe2 - Python API by  <a href="http://www.doxygen.org/index.html">
<img class="footer" src="doxygen.png" alt="doxygen"/>
</a> 1.8.11
</small></address>
<div class="footerContainer">
<div id="footer_wrap" class="wrapper footerWrapper">
<div class="footerBlocks">
<div id="fb_oss" class="footerSection fbOpenSourceFooter">
<svg class="facebookOSSLogoSvg" viewBox="0 0 1133.9 1133.9" x="0px" y="0px" height=50 width=50>
<g>
<path class="logoRing outerRing" d="M 498.3 3.7 c 153.6 88.9 307.3 177.7 461.1 266.2 c 7.6 4.4 10.3 9.1 10.3 17.8 c -0.3 179.1 -0.2 358.3 0 537.4 c 0 8.1 -2.4 12.8 -9.7 17.1 c -154.5 88.9 -308.8 178.1 -462.9 267.5 c -9 5.2 -15.5 5.3 -24.6 0.1 c -153.9 -89.2 -307.9 -178 -462.1 -266.8 C 3 838.8 0 833.9 0 825.1 c 0.3 -179.1 0.2 -358.3 0 -537.4 c 0 -8.6 2.6 -13.6 10.2 -18 C 164.4 180.9 318.4 92 472.4 3 C 477 -1.5 494.3 -0.7 498.3 3.7 Z M 48.8 555.3 c 0 79.9 0.2 159.9 -0.2 239.8 c -0.1 10 3 15.6 11.7 20.6 c 137.2 78.8 274.2 157.8 411 237.3 c 9.9 5.7 17 5.7 26.8 0.1 c 137.5 -79.8 275.2 -159.2 412.9 -238.5 c 7.4 -4.3 10.5 -8.9 10.5 -17.8 c -0.3 -160.2 -0.3 -320.5 0 -480.7 c 0 -8.8 -2.8 -13.6 -10.3 -18 C 772.1 218 633.1 137.8 494.2 57.4 c -6.5 -3.8 -11.5 -4.5 -18.5 -0.5 C 336.8 137.4 197.9 217.7 58.8 297.7 c -7.7 4.4 -10.2 9.2 -10.2 17.9 C 48.9 395.5 48.8 475.4 48.8 555.3 Z" />
<path class="logoRing middleRing" d="M 184.4 555.9 c 0 -33.3 -1 -66.7 0.3 -100 c 1.9 -48 24.1 -86 64.7 -110.9 c 54.8 -33.6 110.7 -65.5 167 -96.6 c 45.7 -25.2 92.9 -24.7 138.6 1 c 54.4 30.6 108.7 61.5 162.2 93.7 c 44 26.5 67.3 66.8 68 118.4 c 0.9 63.2 0.9 126.5 0 189.7 c -0.7 50.6 -23.4 90.7 -66.6 116.9 c -55 33.4 -110.8 65.4 -167.1 96.5 c -43.4 24 -89 24.2 -132.3 0.5 c -57.5 -31.3 -114.2 -64 -170 -98.3 c -41 -25.1 -62.9 -63.7 -64.5 -112.2 C 183.5 621.9 184.3 588.9 184.4 555.9 Z M 232.9 556.3 c 0 29.5 0.5 59.1 -0.1 88.6 c -0.8 39.2 16.9 67.1 50.2 86.2 c 51.2 29.4 102.2 59.2 153.4 88.4 c 31.4 17.9 63.6 18.3 95 0.6 c 53.7 -30.3 107.1 -61.2 160.3 -92.5 c 29.7 -17.5 45 -44.5 45.3 -78.8 c 0.6 -61.7 0.5 -123.5 0 -185.2 c -0.3 -34.4 -15.3 -61.5 -44.9 -79 C 637.7 352.6 583 320.8 527.9 290 c -27.5 -15.4 -57.2 -16.1 -84.7 -0.7 c -56.9 31.6 -113.4 64 -169.1 97.6 c -26.4 15.9 -40.7 41.3 -41.1 72.9 C 232.6 491.9 232.9 524.1 232.9 556.3 Z" />
<path class="logoRing innerRing" d="M 484.9 424.4 c 69.8 -2.8 133.2 57.8 132.6 132 C 617 630 558.5 688.7 484.9 689.1 c -75.1 0.4 -132.6 -63.6 -132.7 -132.7 C 352.1 485 413.4 421.5 484.9 424.4 Z M 401.3 556.7 c -3.4 37.2 30.5 83.6 83 84.1 c 46.6 0.4 84.8 -37.6 84.9 -84 c 0.1 -46.6 -37.2 -84.4 -84.2 -84.6 C 432.2 472.1 397.9 518.3 401.3 556.7 Z" />
</g>
</svg>
<h2>Facebook Open Source</h2>
</div>
<div class="footerSection">
<a class="footerLink" href="https://code.facebook.com/projects/" target="_blank">Open Source Projects</a>
<a class="footerLink" href="https://github.com/facebook/" target="_blank">GitHub</a>
<a class="footerLink" href="https://twitter.com/fbOpenSource" target="_blank">Twitter</a>
</div>
<div class="footerSection rightAlign">
<a class="footerLink" href="https://github.com/caffe2/caffe2" target="_blank">Contribute to this project on GitHub</a>
</div>
</div>
</div>
</div>
<script type="text/javascript" src="/js/jekyll-link-anchors.js"></script>
<script>
(function(i,s,o,g,r,a,m){i['GoogleAnalyticsObject']=r;i[r]=i[r]||function(){
(i[r].q=i[r].q||[]).push(arguments)},i[r].l=1*new Date();a=s.createElement(o),
m=s.getElementsByTagName(o)[0];a.async=1;a.src=g;m.parentNode.insertBefore(a,m)
})(window,document,'script','//www.google-analytics.com/analytics.js','ga');
ga('create', '{{ site.gacode }}', 'auto');
ga('send', 'pageview');
</script>
</body>
</html>