Skip to content

P
PaddleSlim

PaddlePaddle / PaddleSlim
大约 2 年 前同步成功

通知 51
Star 1434
Fork 344
P
PaddleSlim
提交 01c4cf93 编写于 作者: B baiyfbupt
浏览文件
操作

Deployed 09f99c9e with MkDocs version: 1.0.4

上级
展开全部 隐藏空白更改
内联 并排
Showing with 3496 addition and 0 deletion
404.html 0 → 100644
浏览文件 @ 01c4cf93
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="utf-8">
<meta http-equiv="X-UA-Compatible" content="IE=edge">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<link rel="shortcut icon" href="/img/favicon.ico">
<title>My Docs</title>
<link href="/css/bootstrap-custom.min.css" rel="stylesheet">
<link href="/css/font-awesome.min.css" rel="stylesheet">
<link href="/css/base.css" rel="stylesheet">
<link rel="stylesheet" href="https://cdnjs.cloudflare.com/ajax/libs/highlight.js/9.12.0/styles/github.min.css">
<!-- HTML5 shim and Respond.js IE8 support of HTML5 elements and media queries -->
<!--[if lt IE 9]>
<script src="https://oss.maxcdn.com/libs/html5shiv/3.7.0/html5shiv.js"></script>
<script src="https://oss.maxcdn.com/libs/respond.js/1.4.2/respond.min.js"></script>
<![endif]-->
<script src="/js/jquery-1.10.2.min.js" defer></script>
<script src="/js/bootstrap-3.0.3.min.js" defer></script>
<script src="https://cdnjs.cloudflare.com/ajax/libs/highlight.js/9.12.0/highlight.min.js"></script>
<script>hljs.initHighlightingOnLoad();</script>
</head>
<body>
<div class="navbar navbar-default navbar-fixed-top" role="navigation">
<div class="container">
<!-- Collapsed navigation -->
<div class="navbar-header">
<!-- Expander button -->
<button type="button" class="navbar-toggle" data-toggle="collapse" data-target=".navbar-collapse">
<span class="sr-only">Toggle navigation</span>
<span class="icon-bar"></span>
<span class="icon-bar"></span>
<span class="icon-bar"></span>
</button>
<a class="navbar-brand" href="/.">My Docs</a>
</div>
<!-- Expanded navigation -->
<div class="navbar-collapse collapse">
<!-- Main navigation -->
<ul class="nav navbar-nav">
<li >
<a href="/.">Welcome to MkDocs</a>
</li>
<li class="dropdown">
<a href="#" class="dropdown-toggle" data-toggle="dropdown">Docs <b class="caret"></b></a>
<ul class="dropdown-menu">
<li >
<a href="/docs/">PaddleSlim</a>
</li>
<li >
<a href="/docs/table_latency/">硬件延时评估表</a>
</li>
<li class="dropdown-submenu">
<a href="#">Api</a>
<ul class="dropdown-menu">
<li >
<a href="/docs/api/analysis_api/">模型分析API文档</a>
</li>
<li >
<a href="/docs/api/api_guide/">PaddleSlim API文档导航</a>
</li>
<li >
<a href="/docs/api/nas_api/">paddleslim.nas API文档</a>
</li>
<li >
<a href="/docs/api/prune_api/">卷积通道剪裁API文档</a>
</li>
<li >
<a href="/docs/api/quantization_api/">paddleslim.quant API文档</a>
</li>
<li >
<a href="/docs/api/search_space/">paddleslim.nas 提供的搜索空间：</a>
</li>
<li >
<a href="/docs/api/single_distiller_api/">paddleslim.dist API文档</a>
</li>
</ul>
</li>
<li class="dropdown-submenu">
<a href="#">Tutorials</a>
<ul class="dropdown-menu">
<li >
<a href="/docs/tutorials/demo_guide/">Demo guide</a>
</li>
<li >
<a href="/docs/tutorials/nas_demo/">网络结构搜索示例</a>
</li>
<li >
<a href="/docs/tutorials/quant_aware_demo/">在线量化示例</a>
</li>
<li >
<a href="/docs/tutorials/quant_embedding_demo/">Embedding量化示例</a>
</li>
<li >
<a href="/docs/tutorials/quant_post_demo/">离线量化示例</a>
</li>
</ul>
</li>
</ul>
</li>
</ul>
<ul class="nav navbar-nav navbar-right">
<li>
<a href="#" data-toggle="modal" data-target="#mkdocs_search_modal">
<i class="fa fa-search"></i> Search
</a>
</li>
</ul>
</div>
</div>
</div>
<div class="container">
<div class="row-fluid">
<div id="main-content" class="span12">
<h1 id="404-page-not-found" style="text-align: center">404</h1>
<p style="text-align: center"><strong>Page not found</strong></p>
</div>
</div>
</div>
<footer class="col-md-12">
<hr>
<p>Documentation built with <a href="https://www.mkdocs.org/">MkDocs</a>.</p>
</footer>
<script>
var base_url = "/",
shortcuts = {"help": 191, "next": 78, "previous": 80, "search": 83};
</script>
<script src="/js/base.js" defer></script>
<script src="/search/main.js" defer></script>
<div class="modal" id="mkdocs_search_modal" tabindex="-1" role="dialog" aria-labelledby="Search Modal" aria-hidden="true">
<div class="modal-dialog">
<div class="modal-content">
<div class="modal-header">
<button type="button" class="close" data-dismiss="modal"><span aria-hidden="true">&times;</span><span class="sr-only">Close</span></button>
<h4 class="modal-title" id="exampleModalLabel">Search</h4>
</div>
<div class="modal-body">
<p>
From here you can search these documents. Enter
your search terms below.
</p>
<form role="form">
<div class="form-group">
<input type="text" class="form-control" placeholder="Search..." id="mkdocs-search-query" title="Type search term here">
</div>
</form>
<div id="mkdocs-search-results"></div>
</div>
<div class="modal-footer">
</div>
</div>
</div>
</div><div class="modal" id="mkdocs_keyboard_modal" tabindex="-1" role="dialog" aria-labelledby="Keyboard Shortcuts Modal" aria-hidden="true">
<div class="modal-dialog">
<div class="modal-content">
<div class="modal-header">
<button type="button" class="close" data-dismiss="modal"><span aria-hidden="true">&times;</span><span class="sr-only">Close</span></button>
<h4 class="modal-title" id="exampleModalLabel">Keyboard Shortcuts</h4>
</div>
<div class="modal-body">
<table class="table">
<thead>
<tr>
<th style="width: 20%;">Keys</th>
<th>Action</th>
</tr>
</thead>
<tbody>
<tr>
<td class="help shortcut"><kbd>?</kbd></td>
<td>Open this help</td>
</tr>
<tr>
<td class="next shortcut"><kbd>n</kbd></td>
<td>Next page</td>
</tr>
<tr>
<td class="prev shortcut"><kbd>p</kbd></td>
<td>Previous page</td>
</tr>
<tr>
<td class="search shortcut"><kbd>s</kbd></td>
<td>Search</td>
</tr>
</tbody>
</table>
</div>
<div class="modal-footer">
</div>
</div>
</div>
</div>
</body>
</html>
css/base.css 0 → 100644
浏览文件 @ 01c4cf93
body {
padding-top: 70px;
}
/* Replacement for `body { background-attachment: fixed; }`, which has
performance issues when scrolling on large displays. See #1394. */
body::before {
content: ' ';
position: fixed;
width: 100%;
height: 100%;
top: 0;
left: 0;
background-color: #f8f8f8;
background: url(../img/grid.png) repeat-x;
will-change: transform;
z-index: -1;
}
body > .container {
min-height: 400px;
}
ul.nav .main {
font-weight: bold;
}
.col-md-3 {
padding-left: 0;
z-index: 1;
}
.col-md-9 {
padding-bottom: 100px;
}
.source-links {
float: right;
}
.col-md-9 img {
max-width: 100%;
display: inline-block;
padding: 4px;
line-height: 1.428571429;
background-color: #fff;
border: 1px solid #ddd;
border-radius: 4px;
margin: 20px auto 30px auto;
}
/*
* The code below adds some padding to the top of the current anchor target so
* that, when navigating to it, the header isn't hidden by the navbar at the
* top.
*/
:target::before {
content: "";
display: block;
margin-top: -75px;
height: 75px;
pointer-events: none;
}
h1 {
color: #444;
font-weight: 400;
font-size: 42px;
}
h2, h3, h4, h5, h6 {
color: #444;
font-weight: 300;
}
hr {
border-top: 1px solid #aaa;
}
pre, .rst-content tt {
max-width: 100%;
background: #fff;
border: solid 1px #e1e4e5;
color: #333;
overflow-x: auto;
}
code.code-large, .rst-content tt.code-large {
font-size: 90%;
}
code {
padding: 2px 5px;
background: #fff;
border: solid 1px #e1e4e5;
color: #333;
white-space: pre-wrap;
word-wrap: break-word;
}
pre code {
background: transparent;
border: none;
white-space: pre;
word-wrap: normal;
font-family: monospace,serif;
font-size: 12px;
}
kbd {
padding: 2px 4px;
font-size: 90%;
color: #fff;
background-color: #333;
border-radius: 3px;
-webkit-box-shadow: inset 0 -1px 0 rgba(0,0,0,.25);
box-shadow: inset 0 -1px 0 rgba(0,0,0,.25);
}
a code {
color: #2FA4E7;
}
a:hover code, a:focus code {
color: #157AB5;
}
footer {
margin-top: 30px;
margin-bottom: 10px;
text-align: center;
font-weight: 200;
}
.modal-dialog {
margin-top: 60px;
}
/*
* Side navigation
*
* Scrollspy and affixed enhanced navigation to highlight sections and secondary
* sections of docs content.
*/
/* By default it's not affixed in mobile views, so undo that */
.bs-sidebar.affix { /* csslint allow: adjoining-classes */
position: static;
}
.bs-sidebar.well { /* csslint allow: adjoining-classes */
padding: 0;
max-height: 90%;
overflow-y: auto;
}
/* First level of nav */
.bs-sidenav {
padding-top: 10px;
padding-bottom: 10px;
border-radius: 5px;
}
/* All levels of nav */
.bs-sidebar .nav > li > a {
display: block;
padding: 5px 20px;
z-index: 1;
}
.bs-sidebar .nav > li > a:hover,
.bs-sidebar .nav > li > a:focus {
text-decoration: none;
border-right: 1px solid;
}
.bs-sidebar .nav > .active > a,
.bs-sidebar .nav > .active:hover > a,
.bs-sidebar .nav > .active:focus > a {
font-weight: bold;
background-color: transparent;
border-right: 1px solid;
}
/* Nav: second level (shown on .active) */
.bs-sidebar .nav .nav {
display: none; /* Hide by default, but at >768px, show it */
margin-bottom: 8px;
}
.bs-sidebar .nav .nav > li > a {
padding-top: 3px;
padding-bottom: 3px;
padding-left: 30px;
font-size: 90%;
}
/* Show and affix the side nav when space allows it */
@media (min-width: 992px) {
/* Workaround a Safari bug when zooming to < 100%
https://github.com/mkdocs/mkdocs/issues/1050 */
.col-md-9 {
box-sizing: border-box; /* csslint allow: box-sizing */
padding-left: 25%;
width: 100%;
}
.bs-sidebar .nav > .active > ul {
display: block;
}
/* Widen the fixed sidebar */
.bs-sidebar.affix, /* csslint allow: adjoining-classes */
.bs-sidebar.affix-bottom { /* csslint allow: adjoining-classes */
width: 213px;
}
.bs-sidebar.affix { /* csslint allow: adjoining-classes */
position: fixed; /* Undo the static from mobile first approach */
top: 80px;
}
.bs-sidebar.affix-bottom { /* csslint allow: adjoining-classes */
position: absolute; /* Undo the static from mobile first approach */
}
.bs-sidebar.affix-bottom .bs-sidenav, /* csslint allow: adjoining-classes */
.bs-sidebar.affix .bs-sidenav { /* csslint allow: adjoining-classes */
margin-top: 0;
margin-bottom: 0;
}
}
@media (min-width: 1200px) {
/* Widen the fixed sidebar again */
.bs-sidebar.affix-bottom, /* csslint allow: adjoining-classes */
.bs-sidebar.affix { /* csslint allow: adjoining-classes */
width: 263px;
}
}
.headerlink {
font-family: FontAwesome;
font-size: 14px;
display: none;
padding-left: .5em;
}
h1:hover .headerlink, h2:hover .headerlink, h3:hover .headerlink, h4:hover .headerlink, h5:hover .headerlink, h6:hover .headerlink{
display:inline-block;
}
.admonition {
padding: 15px;
margin-bottom: 20px;
border: 1px solid transparent;
border-radius: 4px;
text-align: left;
}
.admonition.note { /* csslint allow: adjoining-classes */
color: #3a87ad;
background-color: #d9edf7;
border-color: #bce8f1;
}
.admonition.warning { /* csslint allow: adjoining-classes */
color: #c09853;
background-color: #fcf8e3;
border-color: #fbeed5;
}
.admonition.danger { /* csslint allow: adjoining-classes */
color: #b94a48;
background-color: #f2dede;
border-color: #eed3d7;
}
.admonition-title {
font-weight: bold;
text-align: left;
}
.dropdown-submenu {
position: relative;
}
.dropdown-submenu>.dropdown-menu {
top: 0;
left: 100%;
margin-top: -6px;
margin-left: -1px;
-webkit-border-radius: 0 6px 6px 6px;
-moz-border-radius: 0 6px 6px;
border-radius: 0 6px 6px 6px;
}
.dropdown-submenu:hover>.dropdown-menu {
display: block;
}
.dropdown-submenu>a:after {
display: block;
content: " ";
float: right;
width: 0;
height: 0;
border-color: transparent;
border-style: solid;
border-width: 5px 0 5px 5px;
border-left-color: #ccc;
margin-top: 5px;
margin-right: -10px;
}
.dropdown-submenu:hover>a:after {
border-left-color: #fff;
}
.dropdown-submenu.pull-left { /* csslint allow: adjoining-classes */
float: none;
}
.dropdown-submenu.pull-left>.dropdown-menu { /* csslint allow: adjoining-classes */
left: -100%;
margin-left: 10px;
-webkit-border-radius: 6px 0 6px 6px;
-moz-border-radius: 6px 0 6px 6px;
border-radius: 6px 0 6px 6px;
}
css/bootstrap-custom.min.css 0 → 100644
浏览文件 @ 01c4cf93
此差异已折叠。
css/font-awesome.min.css 0 → 100644
浏览文件 @ 01c4cf93
此差异已折叠。
docs/api/analysis_api/index.html 0 → 100644
浏览文件 @ 01c4cf93
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="utf-8">
<meta http-equiv="X-UA-Compatible" content="IE=edge">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<link rel="shortcut icon" href="../../../img/favicon.ico">
<title>模型分析API文档 - My Docs</title>
<link href="../../../css/bootstrap-custom.min.css" rel="stylesheet">
<link href="../../../css/font-awesome.min.css" rel="stylesheet">
<link href="../../../css/base.css" rel="stylesheet">
<link rel="stylesheet" href="https://cdnjs.cloudflare.com/ajax/libs/highlight.js/9.12.0/styles/github.min.css">
<!-- HTML5 shim and Respond.js IE8 support of HTML5 elements and media queries -->
<!--[if lt IE 9]>
<script src="https://oss.maxcdn.com/libs/html5shiv/3.7.0/html5shiv.js"></script>
<script src="https://oss.maxcdn.com/libs/respond.js/1.4.2/respond.min.js"></script>
<![endif]-->
<script src="../../../js/jquery-1.10.2.min.js" defer></script>
<script src="../../../js/bootstrap-3.0.3.min.js" defer></script>
<script src="https://cdnjs.cloudflare.com/ajax/libs/highlight.js/9.12.0/highlight.min.js"></script>
<script>hljs.initHighlightingOnLoad();</script>
</head>
<body>
<div class="navbar navbar-default navbar-fixed-top" role="navigation">
<div class="container">
<!-- Collapsed navigation -->
<div class="navbar-header">
<!-- Expander button -->
<button type="button" class="navbar-toggle" data-toggle="collapse" data-target=".navbar-collapse">
<span class="sr-only">Toggle navigation</span>
<span class="icon-bar"></span>
<span class="icon-bar"></span>
<span class="icon-bar"></span>
</button>
<a class="navbar-brand" href="../../..">My Docs</a>
</div>
<!-- Expanded navigation -->
<div class="navbar-collapse collapse">
<!-- Main navigation -->
<ul class="nav navbar-nav">
<li >
<a href="../../..">Welcome to MkDocs</a>
</li>
<li class="dropdown active">
<a href="#" class="dropdown-toggle" data-toggle="dropdown">Docs <b class="caret"></b></a>
<ul class="dropdown-menu">
<li >
<a href="../../">PaddleSlim</a>
</li>
<li >
<a href="../../table_latency/">硬件延时评估表</a>
</li>
<li class="dropdown-submenu">
<a href="#">Api</a>
<ul class="dropdown-menu">
<li class="active">
<a href="./">模型分析API文档</a>
</li>
<li >
<a href="../api_guide/">PaddleSlim API文档导航</a>
</li>
<li >
<a href="../nas_api/">paddleslim.nas API文档</a>
</li>
<li >
<a href="../prune_api/">卷积通道剪裁API文档</a>
</li>
<li >
<a href="../quantization_api/">paddleslim.quant API文档</a>
</li>
<li >
<a href="../search_space/">paddleslim.nas 提供的搜索空间：</a>
</li>
<li >
<a href="../single_distiller_api/">paddleslim.dist API文档</a>
</li>
</ul>
</li>
<li class="dropdown-submenu">
<a href="#">Tutorials</a>
<ul class="dropdown-menu">
<li >
<a href="../../tutorials/demo_guide/">Demo guide</a>
</li>
<li >
<a href="../../tutorials/nas_demo/">网络结构搜索示例</a>
</li>
<li >
<a href="../../tutorials/quant_aware_demo/">在线量化示例</a>
</li>
<li >
<a href="../../tutorials/quant_embedding_demo/">Embedding量化示例</a>
</li>
<li >
<a href="../../tutorials/quant_post_demo/">离线量化示例</a>
</li>
</ul>
</li>
</ul>
</li>
</ul>
<ul class="nav navbar-nav navbar-right">
<li>
<a href="#" data-toggle="modal" data-target="#mkdocs_search_modal">
<i class="fa fa-search"></i> Search
</a>
</li>
<li >
<a rel="next" href="../../table_latency/">
<i class="fa fa-arrow-left"></i> Previous
</a>
</li>
<li >
<a rel="prev" href="../api_guide/">
Next <i class="fa fa-arrow-right"></i>
</a>
</li>
</ul>
</div>
</div>
</div>
<div class="container">
<div class="col-md-3"><div class="bs-sidebar hidden-print affix well" role="complementary">
<ul class="nav bs-sidenav">
<li class="main active"><a href="#api">模型分析API文档</a></li>
<li><a href="#flops">flops</a></li>
<li><a href="#model_size">model_size</a></li>
<li><a href="#tablelatencyevaluator">TableLatencyEvaluator</a></li>
</ul>
</div></div>
<div class="col-md-9" role="main">
<h1 id="api">模型分析API文档</h1>
<h2 id="flops">flops</h2>
<blockquote>
<p>paddleslim.analysis.flops(program, detail=False) <a href="">源代码</a></p>
</blockquote>
<p>获得指定网络的每秒浮点运算次数(FLOPS)。</p>
<p><strong>参数：</strong></p>
<ul>
<li>
<p><strong>program(paddle.fluid.Program):</strong> 待分析的目标网络。更多关于Program的介绍请参考：<a href="https://www.paddlepaddle.org.cn/documentation/docs/zh/api_cn/fluid_cn/Program_cn.html#program">Program概念介绍</a>。</p>
</li>
<li>
<p><strong>detail(bool):</strong> 是否返回每个卷积层的FLOPS。默认为False。</p>
</li>
</ul>
<p><strong>返回值：</strong></p>
<ul>
<li>
<p><strong>flops(float):</strong> 整个网络的FLOPS。</p>
</li>
<li>
<p><strong>params2flops(dict):</strong> 每层卷积对应的FLOPS，其中key为卷积层参数名称，value为FLOPS值。</p>
</li>
</ul>
<p><strong>示例：</strong></p>
<pre><code>import paddle.fluid as fluid
from paddle.fluid.param_attr import ParamAttr
from paddleslim.analysis import flops
def conv_bn_layer(input,
num_filters,
filter_size,
name,
stride=1,
groups=1,
act=None):
conv = fluid.layers.conv2d(
input=input,
num_filters=num_filters,
filter_size=filter_size,
stride=stride,
padding=(filter_size - 1) // 2,
groups=groups,
act=None,
param_attr=ParamAttr(name=name + &quot;_weights&quot;),
bias_attr=False,
name=name + &quot;_out&quot;)
bn_name = name + &quot;_bn&quot;
return fluid.layers.batch_norm(
input=conv,
act=act,
name=bn_name + '_output',
param_attr=ParamAttr(name=bn_name + '_scale'),
bias_attr=ParamAttr(bn_name + '_offset'),
moving_mean_name=bn_name + '_mean',
moving_variance_name=bn_name + '_variance', )
main_program = fluid.Program()
startup_program = fluid.Program()
# X X O X O
# conv1--&gt;conv2--&gt;sum1--&gt;conv3--&gt;conv4--&gt;sum2--&gt;conv5--&gt;conv6
# | ^ | ^
# |____________| |____________________|
#
# X: prune output channels
# O: prune input channels
with fluid.program_guard(main_program, startup_program):
input = fluid.data(name=&quot;image&quot;, shape=[None, 3, 16, 16])
conv1 = conv_bn_layer(input, 8, 3, &quot;conv1&quot;)
conv2 = conv_bn_layer(conv1, 8, 3, &quot;conv2&quot;)
sum1 = conv1 + conv2
conv3 = conv_bn_layer(sum1, 8, 3, &quot;conv3&quot;)
conv4 = conv_bn_layer(conv3, 8, 3, &quot;conv4&quot;)
sum2 = conv4 + sum1
conv5 = conv_bn_layer(sum2, 8, 3, &quot;conv5&quot;)
conv6 = conv_bn_layer(conv5, 8, 3, &quot;conv6&quot;)
print(&quot;FLOPS: {}&quot;.format(flops(main_program)))
</code></pre>
<h2 id="model_size">model_size</h2>
<blockquote>
<p>paddleslim.analysis.model_size(program) <a href="">源代码</a></p>
</blockquote>
<p>获得指定网络的参数数量。</p>
<p><strong>参数：</strong></p>
<ul>
<li><strong>program(paddle.fluid.Program):</strong> 待分析的目标网络。更多关于Program的介绍请参考：<a href="https://www.paddlepaddle.org.cn/documentation/docs/zh/api_cn/fluid_cn/Program_cn.html#program">Program概念介绍</a>。</li>
</ul>
<p><strong>返回值：</strong></p>
<ul>
<li><strong>model_size(int):</strong> 整个网络的参数数量。</li>
</ul>
<p><strong>示例：</strong></p>
<pre><code>import paddle.fluid as fluid
from paddle.fluid.param_attr import ParamAttr
from paddleslim.analysis import model_size
def conv_layer(input,
num_filters,
filter_size,
name,
stride=1,
groups=1,
act=None):
conv = fluid.layers.conv2d(
input=input,
num_filters=num_filters,
filter_size=filter_size,
stride=stride,
padding=(filter_size - 1) // 2,
groups=groups,
act=None,
param_attr=ParamAttr(name=name + &quot;_weights&quot;),
bias_attr=False,
name=name + &quot;_out&quot;)
return conv
main_program = fluid.Program()
startup_program = fluid.Program()
# X X O X O
# conv1--&gt;conv2--&gt;sum1--&gt;conv3--&gt;conv4--&gt;sum2--&gt;conv5--&gt;conv6
# | ^ | ^
# |____________| |____________________|
#
# X: prune output channels
# O: prune input channels
with fluid.program_guard(main_program, startup_program):
input = fluid.data(name=&quot;image&quot;, shape=[None, 3, 16, 16])
conv1 = conv_layer(input, 8, 3, &quot;conv1&quot;)
conv2 = conv_layer(conv1, 8, 3, &quot;conv2&quot;)
sum1 = conv1 + conv2
conv3 = conv_layer(sum1, 8, 3, &quot;conv3&quot;)
conv4 = conv_layer(conv3, 8, 3, &quot;conv4&quot;)
sum2 = conv4 + sum1
conv5 = conv_layer(sum2, 8, 3, &quot;conv5&quot;)
conv6 = conv_layer(conv5, 8, 3, &quot;conv6&quot;)
print(&quot;FLOPS: {}&quot;.format(model_size(main_program)))
</code></pre>
<h2 id="tablelatencyevaluator">TableLatencyEvaluator</h2>
<blockquote>
<p>paddleslim.analysis.TableLatencyEvaluator(table_file, delimiter=",") <a href="">源代码</a></p>
</blockquote>
<p>基于硬件延时表的模型延时评估器。</p>
<p><strong>参数：</strong></p>
<ul>
<li>
<p><strong>table_file(str):</strong> 所使用的延时评估表的绝对路径。关于演示评估表格式请参考：<a href="../paddleslim/analysis/table_latency.md">PaddleSlim硬件延时评估表格式</a></p>
</li>
<li>
<p><strong>delimiter(str):</strong> 硬件延时评估表中，操作信息之前所使用的分割符，默认为英文字符逗号。</p>
</li>
</ul>
<p><strong>返回值：</strong></p>
<ul>
<li><strong>Evaluator:</strong> 硬件延时评估器的实例。</li>
</ul>
<blockquote>
<p>paddleslim.analysis.TableLatencyEvaluator.latency(graph) <a href="">源代码</a></p>
</blockquote>
<p>获得指定网络的预估延时。</p>
<p><strong>参数：</strong></p>
<ul>
<li><strong>graph(Program):</strong> 待预估的目标网络。</li>
</ul>
<p><strong>返回值：</strong></p>
<ul>
<li><strong>latency:</strong> 目标网络的预估延时。</li>
</ul></div>
</div>
<footer class="col-md-12">
<hr>
<p>Documentation built with <a href="https://www.mkdocs.org/">MkDocs</a>.</p>
</footer>
<script>
var base_url = "../../..",
shortcuts = {"help": 191, "next": 78, "previous": 80, "search": 83};
</script>
<script src="../../../js/base.js" defer></script>
<script src="../../../search/main.js" defer></script>
<div class="modal" id="mkdocs_search_modal" tabindex="-1" role="dialog" aria-labelledby="Search Modal" aria-hidden="true">
<div class="modal-dialog">
<div class="modal-content">
<div class="modal-header">
<button type="button" class="close" data-dismiss="modal"><span aria-hidden="true">&times;</span><span class="sr-only">Close</span></button>
<h4 class="modal-title" id="exampleModalLabel">Search</h4>
</div>
<div class="modal-body">
<p>
From here you can search these documents. Enter
your search terms below.
</p>
<form role="form">
<div class="form-group">
<input type="text" class="form-control" placeholder="Search..." id="mkdocs-search-query" title="Type search term here">
</div>
</form>
<div id="mkdocs-search-results"></div>
</div>
<div class="modal-footer">
</div>
</div>
</div>
</div><div class="modal" id="mkdocs_keyboard_modal" tabindex="-1" role="dialog" aria-labelledby="Keyboard Shortcuts Modal" aria-hidden="true">
<div class="modal-dialog">
<div class="modal-content">
<div class="modal-header">
<button type="button" class="close" data-dismiss="modal"><span aria-hidden="true">&times;</span><span class="sr-only">Close</span></button>
<h4 class="modal-title" id="exampleModalLabel">Keyboard Shortcuts</h4>
</div>
<div class="modal-body">
<table class="table">
<thead>
<tr>
<th style="width: 20%;">Keys</th>
<th>Action</th>
</tr>
</thead>
<tbody>
<tr>
<td class="help shortcut"><kbd>?</kbd></td>
<td>Open this help</td>
</tr>
<tr>
<td class="next shortcut"><kbd>n</kbd></td>
<td>Next page</td>
</tr>
<tr>
<td class="prev shortcut"><kbd>p</kbd></td>
<td>Previous page</td>
</tr>
<tr>
<td class="search shortcut"><kbd>s</kbd></td>
<td>Search</td>
</tr>
</tbody>
</table>
</div>
<div class="modal-footer">
</div>
</div>
</div>
</div>
</body>
</html>
docs/api/api_guide/index.html 0 → 100644
浏览文件 @ 01c4cf93
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="utf-8">
<meta http-equiv="X-UA-Compatible" content="IE=edge">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<link rel="shortcut icon" href="../../../img/favicon.ico">
<title>PaddleSlim API文档导航 - My Docs</title>
<link href="../../../css/bootstrap-custom.min.css" rel="stylesheet">
<link href="../../../css/font-awesome.min.css" rel="stylesheet">
<link href="../../../css/base.css" rel="stylesheet">
<link rel="stylesheet" href="https://cdnjs.cloudflare.com/ajax/libs/highlight.js/9.12.0/styles/github.min.css">
<!-- HTML5 shim and Respond.js IE8 support of HTML5 elements and media queries -->
<!--[if lt IE 9]>
<script src="https://oss.maxcdn.com/libs/html5shiv/3.7.0/html5shiv.js"></script>
<script src="https://oss.maxcdn.com/libs/respond.js/1.4.2/respond.min.js"></script>
<![endif]-->
<script src="../../../js/jquery-1.10.2.min.js" defer></script>
<script src="../../../js/bootstrap-3.0.3.min.js" defer></script>
<script src="https://cdnjs.cloudflare.com/ajax/libs/highlight.js/9.12.0/highlight.min.js"></script>
<script>hljs.initHighlightingOnLoad();</script>
</head>
<body>
<div class="navbar navbar-default navbar-fixed-top" role="navigation">
<div class="container">
<!-- Collapsed navigation -->
<div class="navbar-header">
<!-- Expander button -->
<button type="button" class="navbar-toggle" data-toggle="collapse" data-target=".navbar-collapse">
<span class="sr-only">Toggle navigation</span>
<span class="icon-bar"></span>
<span class="icon-bar"></span>
<span class="icon-bar"></span>
</button>
<a class="navbar-brand" href="../../..">My Docs</a>
</div>
<!-- Expanded navigation -->
<div class="navbar-collapse collapse">
<!-- Main navigation -->
<ul class="nav navbar-nav">
<li >
<a href="../../..">Welcome to MkDocs</a>
</li>
<li class="dropdown active">
<a href="#" class="dropdown-toggle" data-toggle="dropdown">Docs <b class="caret"></b></a>
<ul class="dropdown-menu">
<li >
<a href="../../">PaddleSlim</a>
</li>
<li >
<a href="../../table_latency/">硬件延时评估表</a>
</li>
<li class="dropdown-submenu">
<a href="#">Api</a>
<ul class="dropdown-menu">
<li >
<a href="../analysis_api/">模型分析API文档</a>
</li>
<li class="active">
<a href="./">PaddleSlim API文档导航</a>
</li>
<li >
<a href="../nas_api/">paddleslim.nas API文档</a>
</li>
<li >
<a href="../prune_api/">卷积通道剪裁API文档</a>
</li>
<li >
<a href="../quantization_api/">paddleslim.quant API文档</a>
</li>
<li >
<a href="../search_space/">paddleslim.nas 提供的搜索空间：</a>
</li>
<li >
<a href="../single_distiller_api/">paddleslim.dist API文档</a>
</li>
</ul>
</li>
<li class="dropdown-submenu">
<a href="#">Tutorials</a>
<ul class="dropdown-menu">
<li >
<a href="../../tutorials/demo_guide/">Demo guide</a>
</li>
<li >
<a href="../../tutorials/nas_demo/">网络结构搜索示例</a>
</li>
<li >
<a href="../../tutorials/quant_aware_demo/">在线量化示例</a>
</li>
<li >
<a href="../../tutorials/quant_embedding_demo/">Embedding量化示例</a>
</li>
<li >
<a href="../../tutorials/quant_post_demo/">离线量化示例</a>
</li>
</ul>
</li>
</ul>
</li>
</ul>
<ul class="nav navbar-nav navbar-right">
<li>
<a href="#" data-toggle="modal" data-target="#mkdocs_search_modal">
<i class="fa fa-search"></i> Search
</a>
</li>
<li >
<a rel="next" href="../analysis_api/">
<i class="fa fa-arrow-left"></i> Previous
</a>
</li>
<li >
<a rel="prev" href="../nas_api/">
Next <i class="fa fa-arrow-right"></i>
</a>
</li>
</ul>
</div>
</div>
</div>
<div class="container">
<div class="col-md-3"><div class="bs-sidebar hidden-print affix well" role="complementary">
<ul class="nav bs-sidenav">
<li class="main active"><a href="#paddleslim-api">PaddleSlim API文档导航</a></li>
<li><a href="#_1">模型分析</a></li>
<li><a href="#_2">卷积通道剪裁</a></li>
<li><a href="#_3">蒸馏</a></li>
<li><a href="#_5">小模型结构搜索</a></li>
</ul>
</div></div>
<div class="col-md-9" role="main">
<h1 id="paddleslim-api">PaddleSlim API文档导航</h1>
<h2 id="_1"><a href="../analysis_api/">模型分析</a></h2>
<h2 id="_2"><a href="../prune_api/">卷积通道剪裁</a></h2>
<h2 id="_3"><a href="">蒸馏</a></h2>
<ul>
<li>
<p><a href="../single_distiller_api/">单进程蒸馏</a></p>
</li>
<li>
<p><a href="../prune_api/">通道剪裁</a></p>
</li>
</ul>
<h3 id="_4"><a href="../quantization_api/">量化</a></h3>
<ul>
<li>
<p><a href="../quantization_api/#量化训练API">量化训练</a></p>
</li>
<li>
<p><a href="../quantization_api/#离线量化API">离线量化</a></p>
</li>
<li>
<p><a href="../quantization_api/#Embedding量化API">embedding量化</a></p>
</li>
</ul>
<h2 id="_5"><a href="">小模型结构搜索</a></h2>
<ul>
<li><a href="../nas_api/">nas API</a></li>
<li><a href="../search_space/">SearchSpace</a></li>
</ul></div>
</div>
<footer class="col-md-12">
<hr>
<p>Documentation built with <a href="https://www.mkdocs.org/">MkDocs</a>.</p>
</footer>
<script>
var base_url = "../../..",
shortcuts = {"help": 191, "next": 78, "previous": 80, "search": 83};
</script>
<script src="../../../js/base.js" defer></script>
<script src="../../../search/main.js" defer></script>
<div class="modal" id="mkdocs_search_modal" tabindex="-1" role="dialog" aria-labelledby="Search Modal" aria-hidden="true">
<div class="modal-dialog">
<div class="modal-content">
<div class="modal-header">
<button type="button" class="close" data-dismiss="modal"><span aria-hidden="true">&times;</span><span class="sr-only">Close</span></button>
<h4 class="modal-title" id="exampleModalLabel">Search</h4>
</div>
<div class="modal-body">
<p>
From here you can search these documents. Enter
your search terms below.
</p>
<form role="form">
<div class="form-group">
<input type="text" class="form-control" placeholder="Search..." id="mkdocs-search-query" title="Type search term here">
</div>
</form>
<div id="mkdocs-search-results"></div>
</div>
<div class="modal-footer">
</div>
</div>
</div>
</div><div class="modal" id="mkdocs_keyboard_modal" tabindex="-1" role="dialog" aria-labelledby="Keyboard Shortcuts Modal" aria-hidden="true">
<div class="modal-dialog">
<div class="modal-content">
<div class="modal-header">
<button type="button" class="close" data-dismiss="modal"><span aria-hidden="true">&times;</span><span class="sr-only">Close</span></button>
<h4 class="modal-title" id="exampleModalLabel">Keyboard Shortcuts</h4>
</div>
<div class="modal-body">
<table class="table">
<thead>
<tr>
<th style="width: 20%;">Keys</th>
<th>Action</th>
</tr>
</thead>
<tbody>
<tr>
<td class="help shortcut"><kbd>?</kbd></td>
<td>Open this help</td>
</tr>
<tr>
<td class="next shortcut"><kbd>n</kbd></td>
<td>Next page</td>
</tr>
<tr>
<td class="prev shortcut"><kbd>p</kbd></td>
<td>Previous page</td>
</tr>
<tr>
<td class="search shortcut"><kbd>s</kbd></td>
<td>Search</td>
</tr>
</tbody>
</table>
</div>
<div class="modal-footer">
</div>
</div>
</div>
</div>
</body>
</html>
docs/api/nas_api/index.html 0 → 100644
浏览文件 @ 01c4cf93
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="utf-8">
<meta http-equiv="X-UA-Compatible" content="IE=edge">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<link rel="shortcut icon" href="../../../img/favicon.ico">
<title>paddleslim.nas API文档 - My Docs</title>
<link href="../../../css/bootstrap-custom.min.css" rel="stylesheet">
<link href="../../../css/font-awesome.min.css" rel="stylesheet">
<link href="../../../css/base.css" rel="stylesheet">
<link rel="stylesheet" href="https://cdnjs.cloudflare.com/ajax/libs/highlight.js/9.12.0/styles/github.min.css">
<!-- HTML5 shim and Respond.js IE8 support of HTML5 elements and media queries -->
<!--[if lt IE 9]>
<script src="https://oss.maxcdn.com/libs/html5shiv/3.7.0/html5shiv.js"></script>
<script src="https://oss.maxcdn.com/libs/respond.js/1.4.2/respond.min.js"></script>
<![endif]-->
<script src="../../../js/jquery-1.10.2.min.js" defer></script>
<script src="../../../js/bootstrap-3.0.3.min.js" defer></script>
<script src="https://cdnjs.cloudflare.com/ajax/libs/highlight.js/9.12.0/highlight.min.js"></script>
<script>hljs.initHighlightingOnLoad();</script>
</head>
<body>
<div class="navbar navbar-default navbar-fixed-top" role="navigation">
<div class="container">
<!-- Collapsed navigation -->
<div class="navbar-header">
<!-- Expander button -->
<button type="button" class="navbar-toggle" data-toggle="collapse" data-target=".navbar-collapse">
<span class="sr-only">Toggle navigation</span>
<span class="icon-bar"></span>
<span class="icon-bar"></span>
<span class="icon-bar"></span>
</button>
<a class="navbar-brand" href="../../..">My Docs</a>
</div>
<!-- Expanded navigation -->
<div class="navbar-collapse collapse">
<!-- Main navigation -->
<ul class="nav navbar-nav">
<li >
<a href="../../..">Welcome to MkDocs</a>
</li>
<li class="dropdown active">
<a href="#" class="dropdown-toggle" data-toggle="dropdown">Docs <b class="caret"></b></a>
<ul class="dropdown-menu">
<li >
<a href="../../">PaddleSlim</a>
</li>
<li >
<a href="../../table_latency/">硬件延时评估表</a>
</li>
<li class="dropdown-submenu">
<a href="#">Api</a>
<ul class="dropdown-menu">
<li >
<a href="../analysis_api/">模型分析API文档</a>
</li>
<li >
<a href="../api_guide/">PaddleSlim API文档导航</a>
</li>
<li class="active">
<a href="./">paddleslim.nas API文档</a>
</li>
<li >
<a href="../prune_api/">卷积通道剪裁API文档</a>
</li>
<li >
<a href="../quantization_api/">paddleslim.quant API文档</a>
</li>
<li >
<a href="../search_space/">paddleslim.nas 提供的搜索空间：</a>
</li>
<li >
<a href="../single_distiller_api/">paddleslim.dist API文档</a>
</li>
</ul>
</li>
<li class="dropdown-submenu">
<a href="#">Tutorials</a>
<ul class="dropdown-menu">
<li >
<a href="../../tutorials/demo_guide/">Demo guide</a>
</li>
<li >
<a href="../../tutorials/nas_demo/">网络结构搜索示例</a>
</li>
<li >
<a href="../../tutorials/quant_aware_demo/">在线量化示例</a>
</li>
<li >
<a href="../../tutorials/quant_embedding_demo/">Embedding量化示例</a>
</li>
<li >
<a href="../../tutorials/quant_post_demo/">离线量化示例</a>
</li>
</ul>
</li>
</ul>
</li>
</ul>
<ul class="nav navbar-nav navbar-right">
<li>
<a href="#" data-toggle="modal" data-target="#mkdocs_search_modal">
<i class="fa fa-search"></i> Search
</a>
</li>
<li >
<a rel="next" href="../api_guide/">
<i class="fa fa-arrow-left"></i> Previous
</a>
</li>
<li >
<a rel="prev" href="../prune_api/">
Next <i class="fa fa-arrow-right"></i>
</a>
</li>
</ul>
</div>
</div>
</div>
<div class="container">
<div class="col-md-3"><div class="bs-sidebar hidden-print affix well" role="complementary">
<ul class="nav bs-sidenav">
<li class="main active"><a href="#paddleslimnas-api">paddleslim.nas API文档</a></li>
<li><a href="#sanas-api">SANAS API文档</a></li>
<li><a href="#class-sanas">class SANAS</a></li>
</ul>
</div></div>
<div class="col-md-9" role="main">
<h1 id="paddleslimnas-api">paddleslim.nas API文档</h1>
<h2 id="sanas-api">SANAS API文档</h2>
<h2 id="class-sanas">class SANAS</h2>
<p>SANAS（Simulated Annealing Neural Architecture Search）是基于模拟退火算法进行模型结构搜索的算法，一般用于离散搜索任务。</p>
<hr />
<blockquote>
<p>paddleslim.nas.SANAS(configs, server_addr, init_temperature, reduce_rate, search_steps, save_checkpoint, load_checkpoint, is_server)</p>
</blockquote>
<p><strong>参数：</strong>
- <strong>configs(list<tuple>):</strong> 搜索空间配置列表，格式是<code>[(key, {input_size, output_size, block_num, block_mask})]</code>或者<code>[(key)]</code>（MobileNetV2、MobilenetV1和ResNet的搜索空间使用和原本网络结构相同的搜索空间，所以仅需指定<code>key</code>即可）, <code>input_size</code> 和<code>output_size</code>表示输入和输出的特征图的大小，<code>block_num</code>是指搜索网络中的block数量，<code>block_mask</code>是一组由0和1组成的列表，0代表不进行下采样的block，1代表下采样的block。 更多paddleslim提供的搜索空间配置可以参考。
- <strong>server_addr(tuple):</strong> SANAS的地址，包括server的ip地址和端口号，如果ip地址为None或者为""的话则默认使用本机ip。默认：（"", 8881）。
- <strong>init_temperature(float):</strong> 基于模拟退火进行搜索的初始温度。默认：100。
- <strong>reduce_rate(float):</strong> 基于模拟退火进行搜索的衰减率。默认：0.85。
- <strong>search_steps(int):</strong> 搜索过程迭代的次数。默认：300。
- <strong>save_checkpoint(str|None):</strong> 保存checkpoint的文件目录，如果设置为None的话则不保存checkpoint。默认：<code>./nas_checkpoint</code>。
- <strong>load_checkpoint(str|None):</strong> 加载checkpoint的文件目录，如果设置为None的话则不加载checkpoint。默认：None。
- <strong>is_server(bool):</strong> 当前实例是否要启动一个server。默认：True。</p>
<p><strong>返回：</strong>
一个SANAS类的实例</p>
<p><strong>示例代码：</strong></p>
<pre><code>from paddleslim.nas import SANAS
config = [('MobileNetV2Space')]
sanas = SANAS(config=config)
</code></pre>
<hr />
<blockquote>
<p>tokens2arch(tokens)
通过一组token得到实际的模型结构，一般用来把搜索到最优的token转换为模型结构用来做最后的训练。</p>
</blockquote>
<p><strong>参数：</strong>
- <strong>tokens(list):</strong> 一组token。</p>
<p><strong>返回</strong>
返回一个模型结构实例。</p>
<p><strong>示例代码：</strong></p>
<pre><code>import paddle.fluid as fluid
input = fluid.data(name='input', shape=[None, 3, 32, 32], dtype='float32')
archs = sanas.token2arch(tokens)
for arch in archs:
output = arch(input)
input = output
</code></pre>
<hr />
<blockquote>
<p>next_archs():
获取下一组模型结构。</p>
</blockquote>
<p><strong>返回</strong>
返回模型结构实例的列表，形式为list。</p>
<p><strong>示例代码：</strong></p>
<pre><code>import paddle.fluid as fluid
input = fluid.data(name='input', shape=[None, 3, 32, 32], dtype='float32')
archs = sanas.next_archs()
for arch in archs:
output = arch(input)
input = output
</code></pre>
<hr />
<blockquote>
<p>reward(score):
把当前模型结构的得分情况回传。</p>
</blockquote>
<p><strong>参数：</strong>
<strong>score<float>:</strong> 当前模型的得分，分数越大越好。</p>
<p><strong>返回</strong>
模型结构更新成功或者失败，成功则返回<code>True</code>，失败则返回<code>False</code>。</p>
<p><strong>代码示例</strong></p>
<pre><code class="python">import numpy as np
import paddle
import paddle.fluid as fluid
from paddleslim.nas import SANAS
from paddleslim.analysis import flops
max_flops = 321208544
batch_size = 256
# 搜索空间配置
config=[('MobileNetV2Space')]
# 实例化SANAS
sa_nas = SANAS(config, server_addr=(&quot;&quot;, 8887), init_temperature=10.24, reduce_rate=0.85, search_steps=100, is_server=True)
for step in range(100):
archs = sa_nas.next_archs()
train_program = fluid.Program()
test_program = fluid.Program()
startup_program = fluid.Program()
### 构造训练program
with fluid.program_guard(train_program, startup_program):
image = fluid.data(name='image', shape=[None, 3, 32, 32], dtype='float32')
label = fluid.data(name='label', shape=[None, 1], dtype='int64')
for arch in archs:
output = arch(image)
out = fluid.layers.fc(output, size=10, act=&quot;softmax&quot;)
softmax_out = fluid.layers.softmax(input=out, use_cudnn=False)
cost = fluid.layers.cross_entropy(input=softmax_out, label=label)
avg_cost = fluid.layers.mean(cost)
acc_top1 = fluid.layers.accuracy(input=softmax_out, label=label, k=1)
### 构造测试program
test_program = train_program.clone(for_test=True)
### 定义优化器
sgd = fluid.optimizer.SGD(learning_rate=1e-3)
sgd.minimize(avg_cost)
### 增加限制条件，如果没有则进行无限制搜索
if flops(train_program) &gt; max_flops:
continue
### 定义代码是在cpu上运行
place = fluid.CPUPlace()
exe = fluid.Executor(place)
exe.run(startup_program)
### 定义训练输入数据
train_reader = paddle.batch(
paddle.reader.shuffle(
paddle.dataset.cifar.train10(cycle=False), buf_size=1024),
batch_size=batch_size,
drop_last=True)
### 定义预测输入数据
test_reader = paddle.batch(
paddle.dataset.cifar.test10(cycle=False),
batch_size=batch_size,
drop_last=False)
train_feeder = fluid.DataFeeder([image, label], place, program=train_program)
test_feeder = fluid.DataFeeder([image, label], place, program=test_program)
### 开始训练，每个搜索结果训练5个epoch
for epoch_id in range(5):
for batch_id, data in enumerate(train_reader()):
fetches = [avg_cost.name]
outs = exe.run(train_program,
feed=train_feeder.feed(data),
fetch_list=fetches)[0]
if batch_id % 10 == 0:
print('TRAIN: steps: {}, epoch: {}, batch: {}, cost: {}'.format(step, epoch_id, batch_id, outs[0]))
### 开始预测，得到最终的测试结果作为score回传给sa_nas
reward = []
for batch_id, data in enumerate(test_reader()):
test_fetches = [
avg_cost.name, acc_top1.name
]
batch_reward = exe.run(test_program,
feed=test_feeder.feed(data),
fetch_list=test_fetches)
reward_avg = np.mean(np.array(batch_reward), axis=1)
reward.append(reward_avg)
print('TEST: step: {}, batch: {}, avg_cost: {}, acc_top1: {}'.
format(step, batch_id, batch_reward[0],batch_reward[1]))
finally_reward = np.mean(np.array(reward), axis=0)
print(
'FINAL TEST: avg_cost: {}, acc_top1: {}'.format(
finally_reward[0], finally_reward[1]))
### 回传score
sa_nas.reward(float(finally_reward[1]))
</code></pre></div>
</div>
<footer class="col-md-12">
<hr>
<p>Documentation built with <a href="https://www.mkdocs.org/">MkDocs</a>.</p>
</footer>
<script>
var base_url = "../../..",
shortcuts = {"help": 191, "next": 78, "previous": 80, "search": 83};
</script>
<script src="../../../js/base.js" defer></script>
<script src="../../../search/main.js" defer></script>
<div class="modal" id="mkdocs_search_modal" tabindex="-1" role="dialog" aria-labelledby="Search Modal" aria-hidden="true">
<div class="modal-dialog">
<div class="modal-content">
<div class="modal-header">
<button type="button" class="close" data-dismiss="modal"><span aria-hidden="true">&times;</span><span class="sr-only">Close</span></button>
<h4 class="modal-title" id="exampleModalLabel">Search</h4>
</div>
<div class="modal-body">
<p>
From here you can search these documents. Enter
your search terms below.
</p>
<form role="form">
<div class="form-group">
<input type="text" class="form-control" placeholder="Search..." id="mkdocs-search-query" title="Type search term here">
</div>
</form>
<div id="mkdocs-search-results"></div>
</div>
<div class="modal-footer">
</div>
</div>
</div>
</div><div class="modal" id="mkdocs_keyboard_modal" tabindex="-1" role="dialog" aria-labelledby="Keyboard Shortcuts Modal" aria-hidden="true">
<div class="modal-dialog">
<div class="modal-content">
<div class="modal-header">
<button type="button" class="close" data-dismiss="modal"><span aria-hidden="true">&times;</span><span class="sr-only">Close</span></button>
<h4 class="modal-title" id="exampleModalLabel">Keyboard Shortcuts</h4>
</div>
<div class="modal-body">
<table class="table">
<thead>
<tr>
<th style="width: 20%;">Keys</th>
<th>Action</th>
</tr>
</thead>
<tbody>
<tr>
<td class="help shortcut"><kbd>?</kbd></td>
<td>Open this help</td>
</tr>
<tr>
<td class="next shortcut"><kbd>n</kbd></td>
<td>Next page</td>
</tr>
<tr>
<td class="prev shortcut"><kbd>p</kbd></td>
<td>Previous page</td>
</tr>
<tr>
<td class="search shortcut"><kbd>s</kbd></td>
<td>Search</td>
</tr>
</tbody>
</table>
</div>
<div class="modal-footer">
</div>
</div>
</div>
</div>
</body>
</html>
docs/api/prune_api/index.html 0 → 100644
浏览文件 @ 01c4cf93
此差异已折叠。
docs/api/quantization_api/index.html 0 → 100644
浏览文件 @ 01c4cf93
此差异已折叠。
docs/api/search_space/index.html 0 → 100644
浏览文件 @ 01c4cf93
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="utf-8">
<meta http-equiv="X-UA-Compatible" content="IE=edge">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<link rel="shortcut icon" href="../../../img/favicon.ico">
<title>paddleslim.nas 提供的搜索空间： - My Docs</title>
<link href="../../../css/bootstrap-custom.min.css" rel="stylesheet">
<link href="../../../css/font-awesome.min.css" rel="stylesheet">
<link href="../../../css/base.css" rel="stylesheet">
<link rel="stylesheet" href="https://cdnjs.cloudflare.com/ajax/libs/highlight.js/9.12.0/styles/github.min.css">
<!-- HTML5 shim and Respond.js IE8 support of HTML5 elements and media queries -->
<!--[if lt IE 9]>
<script src="https://oss.maxcdn.com/libs/html5shiv/3.7.0/html5shiv.js"></script>
<script src="https://oss.maxcdn.com/libs/respond.js/1.4.2/respond.min.js"></script>
<![endif]-->
<script src="../../../js/jquery-1.10.2.min.js" defer></script>
<script src="../../../js/bootstrap-3.0.3.min.js" defer></script>
<script src="https://cdnjs.cloudflare.com/ajax/libs/highlight.js/9.12.0/highlight.min.js"></script>
<script>hljs.initHighlightingOnLoad();</script>
</head>
<body>
<div class="navbar navbar-default navbar-fixed-top" role="navigation">
<div class="container">
<!-- Collapsed navigation -->
<div class="navbar-header">
<!-- Expander button -->
<button type="button" class="navbar-toggle" data-toggle="collapse" data-target=".navbar-collapse">
<span class="sr-only">Toggle navigation</span>
<span class="icon-bar"></span>
<span class="icon-bar"></span>
<span class="icon-bar"></span>
</button>
<a class="navbar-brand" href="../../..">My Docs</a>
</div>
<!-- Expanded navigation -->
<div class="navbar-collapse collapse">
<!-- Main navigation -->
<ul class="nav navbar-nav">
<li >
<a href="../../..">Welcome to MkDocs</a>
</li>
<li class="dropdown active">
<a href="#" class="dropdown-toggle" data-toggle="dropdown">Docs <b class="caret"></b></a>
<ul class="dropdown-menu">
<li >
<a href="../../">PaddleSlim</a>
</li>
<li >
<a href="../../table_latency/">硬件延时评估表</a>
</li>
<li class="dropdown-submenu">
<a href="#">Api</a>
<ul class="dropdown-menu">
<li >
<a href="../analysis_api/">模型分析API文档</a>
</li>
<li >
<a href="../api_guide/">PaddleSlim API文档导航</a>
</li>
<li >
<a href="../nas_api/">paddleslim.nas API文档</a>
</li>
<li >
<a href="../prune_api/">卷积通道剪裁API文档</a>
</li>
<li >
<a href="../quantization_api/">paddleslim.quant API文档</a>
</li>
<li class="active">
<a href="./">paddleslim.nas 提供的搜索空间：</a>
</li>
<li >
<a href="../single_distiller_api/">paddleslim.dist API文档</a>
</li>
</ul>
</li>
<li class="dropdown-submenu">
<a href="#">Tutorials</a>
<ul class="dropdown-menu">
<li >
<a href="../../tutorials/demo_guide/">Demo guide</a>
</li>
<li >
<a href="../../tutorials/nas_demo/">网络结构搜索示例</a>
</li>
<li >
<a href="../../tutorials/quant_aware_demo/">在线量化示例</a>
</li>
<li >
<a href="../../tutorials/quant_embedding_demo/">Embedding量化示例</a>
</li>
<li >
<a href="../../tutorials/quant_post_demo/">离线量化示例</a>
</li>
</ul>
</li>
</ul>
</li>
</ul>
<ul class="nav navbar-nav navbar-right">
<li>
<a href="#" data-toggle="modal" data-target="#mkdocs_search_modal">
<i class="fa fa-search"></i> Search
</a>
</li>
<li >
<a rel="next" href="../quantization_api/">
<i class="fa fa-arrow-left"></i> Previous
</a>
</li>
<li >
<a rel="prev" href="../single_distiller_api/">
Next <i class="fa fa-arrow-right"></i>
</a>
</li>
</ul>
</div>
</div>
</div>
<div class="container">
<div class="col-md-3"><div class="bs-sidebar hidden-print affix well" role="complementary">
<ul class="nav bs-sidenav">
<li class="main active"><a href="#paddleslimnas">paddleslim.nas 提供的搜索空间：</a></li>
<li><a href="#_1">搜索空间的配置介绍：</a></li>
<li><a href="#_2">搜索空间示例：</a></li>
<li class="main "><a href="#search-space">自定义搜索空间(search space)</a></li>
</ul>
</div></div>
<div class="col-md-9" role="main">
<h1 id="paddleslimnas">paddleslim.nas 提供的搜索空间：</h1>
<ol>
<li>根据原本模型结构构造搜索空间：</li>
</ol>
<p>1.1 MobileNetV2Space</p>
<p>1.2 MobileNetV1Space</p>
<p>1.3 ResNetSpace</p>
<ol>
<li>根据相应模型的block构造搜索空间</li>
</ol>
<p>2.1 MobileNetV1BlockSpace</p>
<p>2.2 MobileNetV2BlockSpace</p>
<p>2.3 ResNetBlockSpace</p>
<p>2.4 InceptionABlockSpace</p>
<p>2.5 InceptionCBlockSpace</p>
<h2 id="_1">搜索空间的配置介绍：</h2>
<p><strong>input_size(int|None)</strong>：<code>input_size</code>表示输入feature map的大小。
<strong>output_size(int|None)</strong>：<code>output_size</code>表示输出feature map的大小。
<strong>block_num(int|None)</strong>：<code>block_num</code>表示搜索空间中block的数量。
<strong>block_mask(list|None)</strong>：<code>block_mask</code>表示当前的block是一个reduction block还是一个normal block，是一组由0、1组成的列表，0表示当前block是normal block，1表示当前block是reduction block。如果设置了<code>block_mask</code>，则主要以<code>block_mask</code>为主要配置，<code>input_size</code>，<code>output_size</code>和<code>block_num</code>三种配置是无效的。</p>
<p><strong>Note:</strong>
1. reduction block表示经过这个block之后的feature map大小下降为之前的一半，normal block表示经过这个block之后feature map大小不变。
2. <code>input_size</code>和<code>output_size</code>用来计算整个模型结构中reduction block数量。</p>
<h2 id="_2">搜索空间示例：</h2>
<ol>
<li>使用paddleslim中提供用原本的模型结构来构造搜索空间的话，仅需要指定搜索空间名字即可。例如：如果使用原本的MobileNetV2的搜索空间进行搜索的话，传入SANAS中的config直接指定为[('MobileNetV2Space')]。</li>
<li>使用paddleslim中提供的block搜索空间构造搜索空间：
2.1 使用<code>input_size</code>, <code>output_size</code>和<code>block_num</code>来构造搜索空间。例如：传入SANAS的config可以指定为[('MobileNetV2BlockSpace', {'input_size': 224, 'output_size': 32, 'block_num': 10})]。
2.2 使用<code>block_mask</code>构造搜索空间。例如：传入SANAS的config可以指定为[('MobileNetV2BlockSpace', {'block_mask': [0, 1, 1, 1, 1, 0, 1, 0]})]。</li>
</ol>
<h1 id="search-space">自定义搜索空间(search space)</h1>
<p>自定义搜索空间类需要继承搜索空间基类并重写以下几部分：
1. 初始化的tokens(<code>init_tokens</code>函数)，可以设置为自己想要的tokens列表, tokens列表中的每个数字指的是当前数字在相应的搜索列表中的索引。例如本示例中若tokens=[0, 3, 5]，则代表当前模型结构搜索到的通道数为[8, 40, 128]。
2. token中每个数字的搜索列表长度(<code>range_table</code>函数)，tokens中每个token的索引范围。
3. 根据token产生模型结构(<code>token2arch</code>函数)，根据搜索到的tokens列表产生模型结构。</p>
<p>以新增reset block为例说明如何构造自己的search space。自定义的search space不能和已有的search space同名。</p>
<pre><code class="python">### 引入搜索空间基类函数和search space的注册类函数
from .search_space_base import SearchSpaceBase
from .search_space_registry import SEARCHSPACE
import numpy as np
### 需要调用注册函数把自定义搜索空间注册到space space中
@SEARCHSPACE.register
### 定义一个继承SearchSpaceBase基类的搜索空间的类函数
class ResNetBlockSpace2(SearchSpaceBase):
def __init__(self, input_size, output_size, block_num, block_mask):
### 定义一些实际想要搜索的内容，例如：通道数、每个卷积的重复次数、卷积核大小等等
### self.filter_num 代表通道数的搜索列表
self.filter_num = np.array([8, 16, 32, 40, 64, 128, 256, 512])
### 定义初始化token，初始化token的长度根据传入的block_num或者block_mask的长度来得到的
def init_tokens(self):
return [0] * 3 * len(self.block_mask)
### 定义
def range_table(self):
return [len(self.filter_num)] * 3 * len(self.block_mask)
def token2arch(self, tokens=None):
if tokens == None:
tokens = self.init_tokens()
self.bottleneck_params_list = []
for i in range(len(self.block_mask)):
self.bottleneck_params_list.append(self.filter_num[tokens[i * 3 + 0]],
self.filter_num[tokens[i * 3 + 1]],
self.filter_num[tokens[i * 3 + 2]],
2 if self.block_mask[i] == 1 else 1)
def net_arch(input):
for i, layer_setting in enumerate(self.bottleneck_params_list):
channel_num, stride = layer_setting[:-1], layer_setting[-1]
input = self._resnet_block(input, channel_num, stride, name='resnet_layer{}'.format(i+1))
return input
return net_arch
### 构造具体block的操作
def _resnet_block(self, input, channel_num, stride, name=None):
shortcut_conv = self._shortcut(input, channel_num[2], stride, name=name)
input = self._conv_bn_layer(input=input, num_filters=channel_num[0], filter_size=1, act='relu', name=name + '_conv0')
input = self._conv_bn_layer(input=input, num_filters=channel_num[1], filter_size=3, stride=stride, act='relu', name=name + '_conv1')
input = self._conv_bn_layer(input=input, num_filters=channel_num[2], filter_size=1, name=name + '_conv2')
return fluid.layers.elementwise_add(x=shortcut_conv, y=input, axis=0, name=name+'_elementwise_add')
def _shortcut(self, input, channel_num, stride, name=None):
channel_in = input.shape[1]
if channel_in != channel_num or stride != 1:
return self.conv_bn_layer(input, num_filters=channel_num, filter_size=1, stride=stride, name=name+'_shortcut')
else:
return input
def _conv_bn_layer(self, input, num_filters, filter_size, stride=1, padding='SAME', act=None, name=None):
conv = fluid.layers.conv2d(input, num_filters, filter_size, stride, name=name+'_conv')
bn = fluid.layers.batch_norm(conv, act=act, name=name+'_bn')
return bn
</code></pre></div>
</div>
<footer class="col-md-12">
<hr>
<p>Documentation built with <a href="https://www.mkdocs.org/">MkDocs</a>.</p>
</footer>
<script>
var base_url = "../../..",
shortcuts = {"help": 191, "next": 78, "previous": 80, "search": 83};
</script>
<script src="../../../js/base.js" defer></script>
<script src="../../../search/main.js" defer></script>
<div class="modal" id="mkdocs_search_modal" tabindex="-1" role="dialog" aria-labelledby="Search Modal" aria-hidden="true">
<div class="modal-dialog">
<div class="modal-content">
<div class="modal-header">
<button type="button" class="close" data-dismiss="modal"><span aria-hidden="true">&times;</span><span class="sr-only">Close</span></button>
<h4 class="modal-title" id="exampleModalLabel">Search</h4>
</div>
<div class="modal-body">
<p>
From here you can search these documents. Enter
your search terms below.
</p>
<form role="form">
<div class="form-group">
<input type="text" class="form-control" placeholder="Search..." id="mkdocs-search-query" title="Type search term here">
</div>
</form>
<div id="mkdocs-search-results"></div>
</div>
<div class="modal-footer">
</div>
</div>
</div>
</div><div class="modal" id="mkdocs_keyboard_modal" tabindex="-1" role="dialog" aria-labelledby="Keyboard Shortcuts Modal" aria-hidden="true">
<div class="modal-dialog">
<div class="modal-content">
<div class="modal-header">
<button type="button" class="close" data-dismiss="modal"><span aria-hidden="true">&times;</span><span class="sr-only">Close</span></button>
<h4 class="modal-title" id="exampleModalLabel">Keyboard Shortcuts</h4>
</div>
<div class="modal-body">
<table class="table">
<thead>
<tr>
<th style="width: 20%;">Keys</th>
<th>Action</th>
</tr>
</thead>
<tbody>
<tr>
<td class="help shortcut"><kbd>?</kbd></td>
<td>Open this help</td>
</tr>
<tr>
<td class="next shortcut"><kbd>n</kbd></td>
<td>Next page</td>
</tr>
<tr>
<td class="prev shortcut"><kbd>p</kbd></td>
<td>Previous page</td>
</tr>
<tr>
<td class="search shortcut"><kbd>s</kbd></td>
<td>Search</td>
</tr>
</tbody>
</table>
</div>
<div class="modal-footer">
</div>
</div>
</div>
</div>
</body>
</html>
docs/api/single_distiller_api/index.html 0 → 100644
浏览文件 @ 01c4cf93
此差异已折叠。
docs/index.html 0 → 100644
浏览文件 @ 01c4cf93
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="utf-8">
<meta http-equiv="X-UA-Compatible" content="IE=edge">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<link rel="shortcut icon" href="../img/favicon.ico">
<title>PaddleSlim - My Docs</title>
<link href="../css/bootstrap-custom.min.css" rel="stylesheet">
<link href="../css/font-awesome.min.css" rel="stylesheet">
<link href="../css/base.css" rel="stylesheet">
<link rel="stylesheet" href="https://cdnjs.cloudflare.com/ajax/libs/highlight.js/9.12.0/styles/github.min.css">
<!-- HTML5 shim and Respond.js IE8 support of HTML5 elements and media queries -->
<!--[if lt IE 9]>
<script src="https://oss.maxcdn.com/libs/html5shiv/3.7.0/html5shiv.js"></script>
<script src="https://oss.maxcdn.com/libs/respond.js/1.4.2/respond.min.js"></script>
<![endif]-->
<script src="../js/jquery-1.10.2.min.js" defer></script>
<script src="../js/bootstrap-3.0.3.min.js" defer></script>
<script src="https://cdnjs.cloudflare.com/ajax/libs/highlight.js/9.12.0/highlight.min.js"></script>
<script>hljs.initHighlightingOnLoad();</script>
</head>
<body>
<div class="navbar navbar-default navbar-fixed-top" role="navigation">
<div class="container">
<!-- Collapsed navigation -->
<div class="navbar-header">
<!-- Expander button -->
<button type="button" class="navbar-toggle" data-toggle="collapse" data-target=".navbar-collapse">
<span class="sr-only">Toggle navigation</span>
<span class="icon-bar"></span>
<span class="icon-bar"></span>
<span class="icon-bar"></span>
</button>
<a class="navbar-brand" href="..">My Docs</a>
</div>
<!-- Expanded navigation -->
<div class="navbar-collapse collapse">
<!-- Main navigation -->
<ul class="nav navbar-nav">
<li >
<a href="..">Welcome to MkDocs</a>
</li>
<li class="dropdown active">
<a href="#" class="dropdown-toggle" data-toggle="dropdown">Docs <b class="caret"></b></a>
<ul class="dropdown-menu">
<li class="active">
<a href="./">PaddleSlim</a>
</li>
<li >
<a href="table_latency/">硬件延时评估表</a>
</li>
<li class="dropdown-submenu">
<a href="#">Api</a>
<ul class="dropdown-menu">
<li >
<a href="api/analysis_api/">模型分析API文档</a>
</li>
<li >
<a href="api/api_guide/">PaddleSlim API文档导航</a>
</li>
<li >
<a href="api/nas_api/">paddleslim.nas API文档</a>
</li>
<li >
<a href="api/prune_api/">卷积通道剪裁API文档</a>
</li>
<li >
<a href="api/quantization_api/">paddleslim.quant API文档</a>
</li>
<li >
<a href="api/search_space/">paddleslim.nas 提供的搜索空间：</a>
</li>
<li >
<a href="api/single_distiller_api/">paddleslim.dist API文档</a>
</li>
</ul>
</li>
<li class="dropdown-submenu">
<a href="#">Tutorials</a>
<ul class="dropdown-menu">
<li >
<a href="tutorials/demo_guide/">Demo guide</a>
</li>
<li >
<a href="tutorials/nas_demo/">网络结构搜索示例</a>
</li>
<li >
<a href="tutorials/quant_aware_demo/">在线量化示例</a>
</li>
<li >
<a href="tutorials/quant_embedding_demo/">Embedding量化示例</a>
</li>
<li >
<a href="tutorials/quant_post_demo/">离线量化示例</a>
</li>
</ul>
</li>
</ul>
</li>
</ul>
<ul class="nav navbar-nav navbar-right">
<li>
<a href="#" data-toggle="modal" data-target="#mkdocs_search_modal">
<i class="fa fa-search"></i> Search
</a>
</li>
<li >
<a rel="next" href="..">
<i class="fa fa-arrow-left"></i> Previous
</a>
</li>
<li >
<a rel="prev" href="table_latency/">
Next <i class="fa fa-arrow-right"></i>
</a>
</li>
</ul>
</div>
</div>
</div>
<div class="container">
<div class="col-md-3"><div class="bs-sidebar hidden-print affix well" role="complementary">
<ul class="nav bs-sidenav">
<li class="main active"><a href="#paddleslim">PaddleSlim</a></li>
<li><a href="#_1">功能</a></li>
<li><a href="#_2">安装</a></li>
<li><a href="#_3">使用</a></li>
<li><a href="#_4">贡献与反馈</a></li>
</ul>
</div></div>
<div class="col-md-9" role="main">
<h1 id="paddleslim">PaddleSlim</h1>
<p>PaddleSlim是PaddlePaddle框架的一个子模块，主要用于压缩图像领域模型。在PaddleSlim中，不仅实现了目前主流的网络剪枝、量化、蒸馏三种压缩策略，还实现了超参数搜索和小模型网络结构搜索功能。在后续版本中，会添加更多的压缩策略，以及完善对NLP领域模型的支持。</p>
<h2 id="_1">功能</h2>
<ul>
<li>模型剪裁</li>
<li>支持通道均匀模型剪裁（uniform pruning)</li>
<li>基于敏感度的模型剪裁</li>
<li>
<p>基于进化算法的自动模型剪裁三种方式</p>
</li>
<li>
<p>量化训练</p>
</li>
<li>在线量化训练（training aware）</li>
<li>离线量化（post training）</li>
<li>
<p>支持对权重全局量化和Channel-Wise量化</p>
</li>
<li>
<p>蒸馏</p>
</li>
<li>
<p>轻量神经网络结构自动搜索（Light-NAS）</p>
</li>
<li>支持基于进化算法的轻量神经网络结构自动搜索（Light-NAS）</li>
<li>支持 FLOPS / 硬件延时约束</li>
<li>支持多平台模型延时评估</li>
</ul>
<h2 id="_2">安装</h2>
<p>安装PaddleSlim前，请确认已正确安装Paddle1.6版本或更新版本。Paddle安装请参考：<a href="https://www.paddlepaddle.org.cn/install/quick">Paddle安装教程</a>。</p>
<ul>
<li>安装develop版本</li>
</ul>
<pre><code>git clone https://github.com/PaddlePaddle/PaddleSlim.git
cd PaddleSlim
python setup.py install
</code></pre>
<ul>
<li>安装官方发布的最新版本</li>
</ul>
<pre><code>pip install paddleslim -i https://pypi.org/simple
</code></pre>
<ul>
<li>安装历史版本</li>
</ul>
<p>请点击<a href="https://pypi.org/project/paddleslim/#history">pypi.org</a>查看可安装历史版本。</p>
<h2 id="_3">使用</h2>
<ul>
<li><a href="doc/api_guide.md">API文档</a>：API使用介绍，包括<a href="">蒸馏</a>、<a href="">剪裁</a>、<a href="">量化</a>和<a href="">模型结构搜索</a>。</li>
<li><a href="doc/demo_guide.md">示例</a>：基于mnist和cifar10等简单分类任务的模型压缩示例，您可以通过该部分快速体验和了解PaddleSlim的功能。</li>
<li><a href="">实践教程</a>：经典模型的分析和压缩实验教程。</li>
<li><a href="">模型库</a>：经过压缩的分类、检测、语义分割模型，包括权重文件、网络结构文件和性能数据。</li>
<li><a href="">Paddle检测库</a>：介绍如何在检测库中使用PaddleSlim。</li>
<li><a href="">Paddle分割库</a>：介绍如何在分割库中使用PaddleSlim。</li>
<li><a href="">PaddleLite</a>：介绍如何使用预测库PaddleLite部署PaddleSlim产出的模型。</li>
</ul>
<h2 id="_4">贡献与反馈</h2></div>
</div>
<footer class="col-md-12">
<hr>
<p>Documentation built with <a href="https://www.mkdocs.org/">MkDocs</a>.</p>
</footer>
<script>
var base_url = "..",
shortcuts = {"help": 191, "next": 78, "previous": 80, "search": 83};
</script>
<script src="../js/base.js" defer></script>
<script src="../search/main.js" defer></script>
<div class="modal" id="mkdocs_search_modal" tabindex="-1" role="dialog" aria-labelledby="Search Modal" aria-hidden="true">
<div class="modal-dialog">
<div class="modal-content">
<div class="modal-header">
<button type="button" class="close" data-dismiss="modal"><span aria-hidden="true">&times;</span><span class="sr-only">Close</span></button>
<h4 class="modal-title" id="exampleModalLabel">Search</h4>
</div>
<div class="modal-body">
<p>
From here you can search these documents. Enter
your search terms below.
</p>
<form role="form">
<div class="form-group">
<input type="text" class="form-control" placeholder="Search..." id="mkdocs-search-query" title="Type search term here">
</div>
</form>
<div id="mkdocs-search-results"></div>
</div>
<div class="modal-footer">
</div>
</div>
</div>
</div><div class="modal" id="mkdocs_keyboard_modal" tabindex="-1" role="dialog" aria-labelledby="Keyboard Shortcuts Modal" aria-hidden="true">
<div class="modal-dialog">
<div class="modal-content">
<div class="modal-header">
<button type="button" class="close" data-dismiss="modal"><span aria-hidden="true">&times;</span><span class="sr-only">Close</span></button>
<h4 class="modal-title" id="exampleModalLabel">Keyboard Shortcuts</h4>
</div>
<div class="modal-body">
<table class="table">
<thead>
<tr>
<th style="width: 20%;">Keys</th>
<th>Action</th>
</tr>
</thead>
<tbody>
<tr>
<td class="help shortcut"><kbd>?</kbd></td>
<td>Open this help</td>
</tr>
<tr>
<td class="next shortcut"><kbd>n</kbd></td>
<td>Next page</td>
</tr>
<tr>
<td class="prev shortcut"><kbd>p</kbd></td>
<td>Previous page</td>
</tr>
<tr>
<td class="search shortcut"><kbd>s</kbd></td>
<td>Search</td>
</tr>
</tbody>
</table>
</div>
<div class="modal-footer">
</div>
</div>
</div>
</div>
</body>
</html>
docs/table_latency/index.html 0 → 100644
浏览文件 @ 01c4cf93
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="utf-8">
<meta http-equiv="X-UA-Compatible" content="IE=edge">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<link rel="shortcut icon" href="../../img/favicon.ico">
<title>硬件延时评估表 - My Docs</title>
<link href="../../css/bootstrap-custom.min.css" rel="stylesheet">
<link href="../../css/font-awesome.min.css" rel="stylesheet">
<link href="../../css/base.css" rel="stylesheet">
<link rel="stylesheet" href="https://cdnjs.cloudflare.com/ajax/libs/highlight.js/9.12.0/styles/github.min.css">
<!-- HTML5 shim and Respond.js IE8 support of HTML5 elements and media queries -->
<!--[if lt IE 9]>
<script src="https://oss.maxcdn.com/libs/html5shiv/3.7.0/html5shiv.js"></script>
<script src="https://oss.maxcdn.com/libs/respond.js/1.4.2/respond.min.js"></script>
<![endif]-->
<script src="../../js/jquery-1.10.2.min.js" defer></script>
<script src="../../js/bootstrap-3.0.3.min.js" defer></script>
<script src="https://cdnjs.cloudflare.com/ajax/libs/highlight.js/9.12.0/highlight.min.js"></script>
<script>hljs.initHighlightingOnLoad();</script>
</head>
<body>
<div class="navbar navbar-default navbar-fixed-top" role="navigation">
<div class="container">
<!-- Collapsed navigation -->
<div class="navbar-header">
<!-- Expander button -->
<button type="button" class="navbar-toggle" data-toggle="collapse" data-target=".navbar-collapse">
<span class="sr-only">Toggle navigation</span>
<span class="icon-bar"></span>
<span class="icon-bar"></span>
<span class="icon-bar"></span>
</button>
<a class="navbar-brand" href="../..">My Docs</a>
</div>
<!-- Expanded navigation -->
<div class="navbar-collapse collapse">
<!-- Main navigation -->
<ul class="nav navbar-nav">
<li >
<a href="../..">Welcome to MkDocs</a>
</li>
<li class="dropdown active">
<a href="#" class="dropdown-toggle" data-toggle="dropdown">Docs <b class="caret"></b></a>
<ul class="dropdown-menu">
<li >
<a href="../">PaddleSlim</a>
</li>
<li class="active">
<a href="./">硬件延时评估表</a>
</li>
<li class="dropdown-submenu">
<a href="#">Api</a>
<ul class="dropdown-menu">
<li >
<a href="../api/analysis_api/">模型分析API文档</a>
</li>
<li >
<a href="../api/api_guide/">PaddleSlim API文档导航</a>
</li>
<li >
<a href="../api/nas_api/">paddleslim.nas API文档</a>
</li>
<li >
<a href="../api/prune_api/">卷积通道剪裁API文档</a>
</li>
<li >
<a href="../api/quantization_api/">paddleslim.quant API文档</a>
</li>
<li >
<a href="../api/search_space/">paddleslim.nas 提供的搜索空间：</a>
</li>
<li >
<a href="../api/single_distiller_api/">paddleslim.dist API文档</a>
</li>
</ul>
</li>
<li class="dropdown-submenu">
<a href="#">Tutorials</a>
<ul class="dropdown-menu">
<li >
<a href="../tutorials/demo_guide/">Demo guide</a>
</li>
<li >
<a href="../tutorials/nas_demo/">网络结构搜索示例</a>
</li>
<li >
<a href="../tutorials/quant_aware_demo/">在线量化示例</a>
</li>
<li >
<a href="../tutorials/quant_embedding_demo/">Embedding量化示例</a>
</li>
<li >
<a href="../tutorials/quant_post_demo/">离线量化示例</a>
</li>
</ul>
</li>
</ul>
</li>
</ul>
<ul class="nav navbar-nav navbar-right">
<li>
<a href="#" data-toggle="modal" data-target="#mkdocs_search_modal">
<i class="fa fa-search"></i> Search
</a>
</li>
<li >
<a rel="next" href="../">
<i class="fa fa-arrow-left"></i> Previous
</a>
</li>
<li >
<a rel="prev" href="../api/analysis_api/">
Next <i class="fa fa-arrow-right"></i>
</a>
</li>
</ul>
</div>
</div>
</div>
<div class="container">
<div class="col-md-3"><div class="bs-sidebar hidden-print affix well" role="complementary">
<ul class="nav bs-sidenav">
<li class="main active"><a href="#_1">硬件延时评估表</a></li>
<li><a href="#_2">概述</a></li>
<li><a href="#_3">整体格式</a></li>
<li><a href="#_4">版本信息</a></li>
<li><a href="#_5">操作信息</a></li>
</ul>
</div></div>
<div class="col-md-9" role="main">
<h1 id="_1">硬件延时评估表</h1>
<p>硬件延时评估表用于快速评估一个模型在特定硬件环境和推理引擎上的推理速度。
该文档主要用于定义PaddleSlim支持的硬件延时评估表的格式。</p>
<h2 id="_2">概述</h2>
<p>硬件延时评估表中存放着所有可能的操作对应的延时信息，该表中的一个操作包括操作类型和操作参数，比如：操作类型可以是<code>conv2d</code>，对应的操作参数有输入特征图的大小、卷积核个数、卷积核大小等。
给定操作的延时依赖于硬件环境和推理引擎。</p>
<h2 id="_3">整体格式</h2>
<p>硬件延时评估表以文件或多行字符串的形式保存。</p>
<p>硬件延时评估表第一行保存版本信息，后续每行为一个操作和对应的延时信息。</p>
<h2 id="_4">版本信息</h2>
<p>版本信息以英文字符逗号分割，内容依次为硬件环境名称、推理引擎名称和时间戳。</p>
<ul>
<li>
<p><strong>硬件环境名称：</strong> 用于标识硬件环境，可以包含计算架构类型、版本号等信息。</p>
</li>
<li>
<p><strong>推理引擎名称：</strong> 用于标识推理引擎，可以包含推理引擎名称、版本号、优化选项等信息。</p>
</li>
<li>
<p><strong>时间戳：</strong> 该评估表的创建时间。</p>
</li>
</ul>
<h2 id="_5">操作信息</h2>
<p>操作信息字段之间以逗号分割。操作信息与延迟信息之间以制表符分割。</p>
<h3 id="conv2d">conv2d</h3>
<p><strong>格式</strong></p>
<pre><code>op_type,flag_bias,flag_relu,n_in,c_in,h_in,w_in,c_out,groups,kernel,padding,stride,dilation\tlatency
</code></pre>
<p><strong>字段解释</strong></p>
<ul>
<li><strong>op_type(str)</strong> - 当前op类型。</li>
<li><strong>flag_bias (int)</strong> - 是否有 bias（0：无，1：有）。</li>
<li><strong>flag_relu (int)</strong> - 是否有 relu（0：无，1：有）。</li>
<li><strong>n_in (int)</strong> - 输入 Tensor 的批尺寸 (batch size)。</li>
<li><strong>c_in (int)</strong> - 输入 Tensor 的通道 (channel) 数。</li>
<li><strong>h_in (int)</strong> - 输入 Tensor 的特征高度。</li>
<li><strong>w_in (int)</strong> - 输入 Tensor 的特征宽度。</li>
<li><strong>c_out (int)</strong> - 输出 Tensor 的通道 (channel) 数。</li>
<li><strong>groups (int)</strong> - 卷积二维层（Conv2D Layer）的组数。</li>
<li><strong>kernel (int)</strong> - 卷积核大小。</li>
<li><strong>padding (int)</strong> - 填充 (padding) 大小。</li>
<li><strong>stride (int)</strong> - 步长 (stride) 大小。</li>
<li><strong>dilation (int)</strong> - 膨胀 (dilation) 大小。</li>
<li><strong>latency (float)</strong> - 当前op的延时时间</li>
</ul>
<h3 id="activation">activation</h3>
<p><strong>格式</strong></p>
<pre><code>op_type,n_in,c_in,h_in,w_in\tlatency
</code></pre>
<p><strong>字段解释</strong></p>
<ul>
<li><strong>op_type(str)</strong> - 当前op类型。</li>
<li><strong>n_in (int)</strong> - 输入 Tensor 的批尺寸 (batch size)。</li>
<li><strong>c_in (int)</strong> - 输入 Tensor 的通道 (channel) 数。</li>
<li><strong>h_in (int)</strong> - 输入 Tensor 的特征高度。</li>
<li><strong>w_in (int)</strong> - 输入 Tensor 的特征宽度。</li>
<li><strong>latency (float)</strong> - 当前op的延时时间</li>
</ul>
<h3 id="batch_norm">batch_norm</h3>
<p><strong>格式</strong></p>
<pre><code>op_type,active_type,n_in,c_in,h_in,w_in\tlatency
</code></pre>
<p><strong>字段解释</strong></p>
<ul>
<li><strong>op_type(str)</strong> - 当前op类型。</li>
<li><strong>active_type (string)</strong> - 激活函数类型，包含：relu, prelu, sigmoid, relu6, tanh。</li>
<li><strong>n_in (int)</strong> - 输入 Tensor 的批尺寸 (batch size)。</li>
<li><strong>c_in (int)</strong> - 输入 Tensor 的通道 (channel) 数。</li>
<li><strong>h_in (int)</strong> - 输入 Tensor 的特征高度。</li>
<li><strong>w_in (int)</strong> - 输入 Tensor 的特征宽度。</li>
<li><strong>latency (float)</strong> - 当前op的延时时间</li>
</ul>
<h3 id="eltwise">eltwise</h3>
<p><strong>格式</strong></p>
<pre><code>op_type,n_in,c_in,h_in,w_in\tlatency
</code></pre>
<p><strong>字段解释</strong></p>
<ul>
<li><strong>op_type(str)</strong> - 当前op类型。</li>
<li><strong>n_in (int)</strong> - 输入 Tensor 的批尺寸 (batch size)。</li>
<li><strong>c_in (int)</strong> - 输入 Tensor 的通道 (channel) 数。</li>
<li><strong>h_in (int)</strong> - 输入 Tensor 的特征高度。</li>
<li><strong>w_in (int)</strong> - 输入 Tensor 的特征宽度。</li>
<li><strong>latency (float)</strong> - 当前op的延时时间</li>
</ul>
<h3 id="pooling">pooling</h3>
<p><strong>格式</strong></p>
<pre><code>op_type,flag_global_pooling,n_in,c_in,h_in,w_in,kernel,padding,stride,ceil_mode,pool_type\tlatency
</code></pre>
<p><strong>字段解释</strong></p>
<ul>
<li><strong>op_type(str)</strong> - 当前op类型。</li>
<li><strong>flag_global_pooling (int)</strong> - 是否为全局池化（0：不是，1：是）。</li>
<li><strong>n_in (int)</strong> - 输入 Tensor 的批尺寸 (batch size)。</li>
<li><strong>c_in (int)</strong> - 输入 Tensor 的通道 (channel) 数。</li>
<li><strong>h_in (int)</strong> - 输入 Tensor 的特征高度。</li>
<li><strong>w_in (int)</strong> - 输入 Tensor 的特征宽度。</li>
<li><strong>kernel (int)</strong> - 卷积核大小。</li>
<li><strong>padding (int)</strong> - 填充 (padding) 大小。</li>
<li><strong>stride (int)</strong> - 步长 (stride) 大小。</li>
<li><strong>ceil_mode (int)</strong> - 是否用 ceil 函数计算输出高度和宽度。0 表示使用 floor 函数，1 表示使用 ceil 函数。</li>
<li><strong>pool_type (int)</strong> - 池化类型，其中 1 表示 pooling_max，2 表示 pooling_average_include_padding，3 表示 pooling_average_exclude_padding。</li>
<li><strong>latency (float)</strong> - 当前op的延时时间</li>
</ul>
<h3 id="softmax">softmax</h3>
<p><strong>格式</strong></p>
<pre><code>op_type,axis,n_in,c_in,h_in,w_in\tlatency
</code></pre>
<p><strong>字段解释</strong></p>
<ul>
<li><strong>op_type(str)</strong> - 当前op类型。</li>
<li><strong>axis (int)</strong> - 执行 softmax 计算的维度索引，应该在 [−1，rank − 1] 范围内，其中 rank 是输入变量的秩。</li>
<li><strong>n_in (int)</strong> - 输入 Tensor 的批尺寸 (batch size)。</li>
<li><strong>c_in (int)</strong> - 输入 Tensor 的通道 (channel) 数。</li>
<li><strong>h_in (int)</strong> - 输入 Tensor 的特征高度。</li>
<li><strong>w_in (int)</strong> - 输入 Tensor 的特征宽度。</li>
<li><strong>latency (float)</strong> - 当前op的延时时间</li>
</ul></div>
</div>
<footer class="col-md-12">
<hr>
<p>Documentation built with <a href="https://www.mkdocs.org/">MkDocs</a>.</p>
</footer>
<script>
var base_url = "../..",
shortcuts = {"help": 191, "next": 78, "previous": 80, "search": 83};
</script>
<script src="../../js/base.js" defer></script>
<script src="../../search/main.js" defer></script>
<div class="modal" id="mkdocs_search_modal" tabindex="-1" role="dialog" aria-labelledby="Search Modal" aria-hidden="true">
<div class="modal-dialog">
<div class="modal-content">
<div class="modal-header">
<button type="button" class="close" data-dismiss="modal"><span aria-hidden="true">&times;</span><span class="sr-only">Close</span></button>
<h4 class="modal-title" id="exampleModalLabel">Search</h4>
</div>
<div class="modal-body">
<p>
From here you can search these documents. Enter
your search terms below.
</p>
<form role="form">
<div class="form-group">
<input type="text" class="form-control" placeholder="Search..." id="mkdocs-search-query" title="Type search term here">
</div>
</form>
<div id="mkdocs-search-results"></div>
</div>
<div class="modal-footer">
</div>
</div>
</div>
</div><div class="modal" id="mkdocs_keyboard_modal" tabindex="-1" role="dialog" aria-labelledby="Keyboard Shortcuts Modal" aria-hidden="true">
<div class="modal-dialog">
<div class="modal-content">
<div class="modal-header">
<button type="button" class="close" data-dismiss="modal"><span aria-hidden="true">&times;</span><span class="sr-only">Close</span></button>
<h4 class="modal-title" id="exampleModalLabel">Keyboard Shortcuts</h4>
</div>
<div class="modal-body">
<table class="table">
<thead>
<tr>
<th style="width: 20%;">Keys</th>
<th>Action</th>
</tr>
</thead>
<tbody>
<tr>
<td class="help shortcut"><kbd>?</kbd></td>
<td>Open this help</td>
</tr>
<tr>
<td class="next shortcut"><kbd>n</kbd></td>
<td>Next page</td>
</tr>
<tr>
<td class="prev shortcut"><kbd>p</kbd></td>
<td>Previous page</td>
</tr>
<tr>
<td class="search shortcut"><kbd>s</kbd></td>
<td>Search</td>
</tr>
</tbody>
</table>
</div>
<div class="modal-footer">
</div>
</div>
</div>
</div>
</body>
</html>
docs/tutorials/demo_guide/index.html 0 → 100644
浏览文件 @ 01c4cf93
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="utf-8">
<meta http-equiv="X-UA-Compatible" content="IE=edge">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<link rel="shortcut icon" href="../../../img/favicon.ico">
<title>Demo guide - My Docs</title>
<link href="../../../css/bootstrap-custom.min.css" rel="stylesheet">
<link href="../../../css/font-awesome.min.css" rel="stylesheet">
<link href="../../../css/base.css" rel="stylesheet">
<link rel="stylesheet" href="https://cdnjs.cloudflare.com/ajax/libs/highlight.js/9.12.0/styles/github.min.css">
<!-- HTML5 shim and Respond.js IE8 support of HTML5 elements and media queries -->
<!--[if lt IE 9]>
<script src="https://oss.maxcdn.com/libs/html5shiv/3.7.0/html5shiv.js"></script>
<script src="https://oss.maxcdn.com/libs/respond.js/1.4.2/respond.min.js"></script>
<![endif]-->
<script src="../../../js/jquery-1.10.2.min.js" defer></script>
<script src="../../../js/bootstrap-3.0.3.min.js" defer></script>
<script src="https://cdnjs.cloudflare.com/ajax/libs/highlight.js/9.12.0/highlight.min.js"></script>
<script>hljs.initHighlightingOnLoad();</script>
</head>
<body>
<div class="navbar navbar-default navbar-fixed-top" role="navigation">
<div class="container">
<!-- Collapsed navigation -->
<div class="navbar-header">
<!-- Expander button -->
<button type="button" class="navbar-toggle" data-toggle="collapse" data-target=".navbar-collapse">
<span class="sr-only">Toggle navigation</span>
<span class="icon-bar"></span>
<span class="icon-bar"></span>
<span class="icon-bar"></span>
</button>
<a class="navbar-brand" href="../../..">My Docs</a>
</div>
<!-- Expanded navigation -->
<div class="navbar-collapse collapse">
<!-- Main navigation -->
<ul class="nav navbar-nav">
<li >
<a href="../../..">Welcome to MkDocs</a>
</li>
<li class="dropdown active">
<a href="#" class="dropdown-toggle" data-toggle="dropdown">Docs <b class="caret"></b></a>
<ul class="dropdown-menu">
<li >
<a href="../../">PaddleSlim</a>
</li>
<li >
<a href="../../table_latency/">硬件延时评估表</a>
</li>
<li class="dropdown-submenu">
<a href="#">Api</a>
<ul class="dropdown-menu">
<li >
<a href="../../api/analysis_api/">模型分析API文档</a>
</li>
<li >
<a href="../../api/api_guide/">PaddleSlim API文档导航</a>
</li>
<li >
<a href="../../api/nas_api/">paddleslim.nas API文档</a>
</li>
<li >
<a href="../../api/prune_api/">卷积通道剪裁API文档</a>
</li>
<li >
<a href="../../api/quantization_api/">paddleslim.quant API文档</a>
</li>
<li >
<a href="../../api/search_space/">paddleslim.nas 提供的搜索空间：</a>
</li>
<li >
<a href="../../api/single_distiller_api/">paddleslim.dist API文档</a>
</li>
</ul>
</li>
<li class="dropdown-submenu">
<a href="#">Tutorials</a>
<ul class="dropdown-menu">
<li class="active">
<a href="./">Demo guide</a>
</li>
<li >
<a href="../nas_demo/">网络结构搜索示例</a>
</li>
<li >
<a href="../quant_aware_demo/">在线量化示例</a>
</li>
<li >
<a href="../quant_embedding_demo/">Embedding量化示例</a>
</li>
<li >
<a href="../quant_post_demo/">离线量化示例</a>
</li>
</ul>
</li>
</ul>
</li>
</ul>
<ul class="nav navbar-nav navbar-right">
<li>
<a href="#" data-toggle="modal" data-target="#mkdocs_search_modal">
<i class="fa fa-search"></i> Search
</a>
</li>
<li >
<a rel="next" href="../../api/single_distiller_api/">
<i class="fa fa-arrow-left"></i> Previous
</a>
</li>
<li >
<a rel="prev" href="../nas_demo/">
Next <i class="fa fa-arrow-right"></i>
</a>
</li>
</ul>
</div>
</div>
</div>
<div class="container">
<div class="col-md-3"><div class="bs-sidebar hidden-print affix well" role="complementary">
<ul class="nav bs-sidenav">
<li class="main active"><a href="#_1">蒸馏</a></li>
<li class="main "><a href="#_2">量化</a></li>
<li><a href="#demo">量化训练demo文档</a></li>
<li><a href="#demo_1">离线量化demo文档</a></li>
<li><a href="#embeddingdemo">Embedding量化demo文档</a></li>
<li class="main "><a href="#nas">NAS</a></li>
<li><a href="#nas_1">NAS示例</a></li>
</ul>
</div></div>
<div class="col-md-9" role="main">
<h2 id="_1"><a href="../demo/distillation/distillation_demo.py">蒸馏</a></h2>
<p>蒸馏demo默认使用ResNet50作为teacher网络，MobileNet作为student网络，此外还支持将teacher和student换成<a href="../demo/models">models目录</a>支持的任意模型。</p>
<p>demo中对teahcer模型和student模型的一层特征图添加了l2_loss的蒸馏损失函数，使用时也可根据需要选择fsp_loss, soft_label_loss以及自定义的loss函数。</p>
<p>训练默认使用的是cifar10数据集，piecewise_decay学习率衰减策略，momentum优化器进行120轮蒸馏训练。使用者也可以简单地用args参数切换为使用ImageNet数据集，cosine_decay学习率衰减策略等其他训练配置。</p>
<h2 id="_2">量化</h2>
<h3 id="demo"><a href="../quant_aware_demo/">量化训练demo文档</a></h3>
<h3 id="demo_1"><a href="../quant_post_demo/">离线量化demo文档</a></h3>
<h3 id="embeddingdemo"><a href="../quant_embedding_demo/">Embedding量化demo文档</a></h3>
<h2 id="nas">NAS</h2>
<h3 id="nas_1"><a href="../nas_demo/">NAS示例</a></h3></div>
</div>
<footer class="col-md-12">
<hr>
<p>Documentation built with <a href="https://www.mkdocs.org/">MkDocs</a>.</p>
</footer>
<script>
var base_url = "../../..",
shortcuts = {"help": 191, "next": 78, "previous": 80, "search": 83};
</script>
<script src="../../../js/base.js" defer></script>
<script src="../../../search/main.js" defer></script>
<div class="modal" id="mkdocs_search_modal" tabindex="-1" role="dialog" aria-labelledby="Search Modal" aria-hidden="true">
<div class="modal-dialog">
<div class="modal-content">
<div class="modal-header">
<button type="button" class="close" data-dismiss="modal"><span aria-hidden="true">&times;</span><span class="sr-only">Close</span></button>
<h4 class="modal-title" id="exampleModalLabel">Search</h4>
</div>
<div class="modal-body">
<p>
From here you can search these documents. Enter
your search terms below.
</p>
<form role="form">
<div class="form-group">
<input type="text" class="form-control" placeholder="Search..." id="mkdocs-search-query" title="Type search term here">
</div>
</form>
<div id="mkdocs-search-results"></div>
</div>
<div class="modal-footer">
</div>
</div>
</div>
</div><div class="modal" id="mkdocs_keyboard_modal" tabindex="-1" role="dialog" aria-labelledby="Keyboard Shortcuts Modal" aria-hidden="true">
<div class="modal-dialog">
<div class="modal-content">
<div class="modal-header">
<button type="button" class="close" data-dismiss="modal"><span aria-hidden="true">&times;</span><span class="sr-only">Close</span></button>
<h4 class="modal-title" id="exampleModalLabel">Keyboard Shortcuts</h4>
</div>
<div class="modal-body">
<table class="table">
<thead>
<tr>
<th style="width: 20%;">Keys</th>
<th>Action</th>
</tr>
</thead>
<tbody>
<tr>
<td class="help shortcut"><kbd>?</kbd></td>
<td>Open this help</td>
</tr>
<tr>
<td class="next shortcut"><kbd>n</kbd></td>
<td>Next page</td>
</tr>
<tr>
<td class="prev shortcut"><kbd>p</kbd></td>
<td>Previous page</td>
</tr>
<tr>
<td class="search shortcut"><kbd>s</kbd></td>
<td>Search</td>
</tr>
</tbody>
</table>
</div>
<div class="modal-footer">
</div>
</div>
</div>
</div>
</body>
</html>
docs/tutorials/nas_demo/index.html 0 → 100644
浏览文件 @ 01c4cf93
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="utf-8">
<meta http-equiv="X-UA-Compatible" content="IE=edge">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<link rel="shortcut icon" href="../../../img/favicon.ico">
<title>网络结构搜索示例 - My Docs</title>
<link href="../../../css/bootstrap-custom.min.css" rel="stylesheet">
<link href="../../../css/font-awesome.min.css" rel="stylesheet">
<link href="../../../css/base.css" rel="stylesheet">
<link rel="stylesheet" href="https://cdnjs.cloudflare.com/ajax/libs/highlight.js/9.12.0/styles/github.min.css">
<!-- HTML5 shim and Respond.js IE8 support of HTML5 elements and media queries -->
<!--[if lt IE 9]>
<script src="https://oss.maxcdn.com/libs/html5shiv/3.7.0/html5shiv.js"></script>
<script src="https://oss.maxcdn.com/libs/respond.js/1.4.2/respond.min.js"></script>
<![endif]-->
<script src="../../../js/jquery-1.10.2.min.js" defer></script>
<script src="../../../js/bootstrap-3.0.3.min.js" defer></script>
<script src="https://cdnjs.cloudflare.com/ajax/libs/highlight.js/9.12.0/highlight.min.js"></script>
<script>hljs.initHighlightingOnLoad();</script>
</head>
<body>
<div class="navbar navbar-default navbar-fixed-top" role="navigation">
<div class="container">
<!-- Collapsed navigation -->
<div class="navbar-header">
<!-- Expander button -->
<button type="button" class="navbar-toggle" data-toggle="collapse" data-target=".navbar-collapse">
<span class="sr-only">Toggle navigation</span>
<span class="icon-bar"></span>
<span class="icon-bar"></span>
<span class="icon-bar"></span>
</button>
<a class="navbar-brand" href="../../..">My Docs</a>
</div>
<!-- Expanded navigation -->
<div class="navbar-collapse collapse">
<!-- Main navigation -->
<ul class="nav navbar-nav">
<li >
<a href="../../..">Welcome to MkDocs</a>
</li>
<li class="dropdown active">
<a href="#" class="dropdown-toggle" data-toggle="dropdown">Docs <b class="caret"></b></a>
<ul class="dropdown-menu">
<li >
<a href="../../">PaddleSlim</a>
</li>
<li >
<a href="../../table_latency/">硬件延时评估表</a>
</li>
<li class="dropdown-submenu">
<a href="#">Api</a>
<ul class="dropdown-menu">
<li >
<a href="../../api/analysis_api/">模型分析API文档</a>
</li>
<li >
<a href="../../api/api_guide/">PaddleSlim API文档导航</a>
</li>
<li >
<a href="../../api/nas_api/">paddleslim.nas API文档</a>
</li>
<li >
<a href="../../api/prune_api/">卷积通道剪裁API文档</a>
</li>
<li >
<a href="../../api/quantization_api/">paddleslim.quant API文档</a>
</li>
<li >
<a href="../../api/search_space/">paddleslim.nas 提供的搜索空间：</a>
</li>
<li >
<a href="../../api/single_distiller_api/">paddleslim.dist API文档</a>
</li>
</ul>
</li>
<li class="dropdown-submenu">
<a href="#">Tutorials</a>
<ul class="dropdown-menu">
<li >
<a href="../demo_guide/">Demo guide</a>
</li>
<li class="active">
<a href="./">网络结构搜索示例</a>
</li>
<li >
<a href="../quant_aware_demo/">在线量化示例</a>
</li>
<li >
<a href="../quant_embedding_demo/">Embedding量化示例</a>
</li>
<li >
<a href="../quant_post_demo/">离线量化示例</a>
</li>
</ul>
</li>
</ul>
</li>
</ul>
<ul class="nav navbar-nav navbar-right">
<li>
<a href="#" data-toggle="modal" data-target="#mkdocs_search_modal">
<i class="fa fa-search"></i> Search
</a>
</li>
<li >
<a rel="next" href="../demo_guide/">
<i class="fa fa-arrow-left"></i> Previous
</a>
</li>
<li >
<a rel="prev" href="../quant_aware_demo/">
Next <i class="fa fa-arrow-right"></i>
</a>
</li>
</ul>
</div>
</div>
</div>
<div class="container">
<div class="col-md-3"><div class="bs-sidebar hidden-print affix well" role="complementary">
<ul class="nav bs-sidenav">
<li class="main active"><a href="#_1">网络结构搜索示例</a></li>
<li><a href="#_2">接口介绍</a></li>
</ul>
</div></div>
<div class="col-md-9" role="main">
<h1 id="_1">网络结构搜索示例</h1>
<p>本示例介绍如何使用网络结构搜索接口，搜索到一个更小或者精度更高的模型，该文档仅介绍paddleslim中SANAS的使用及如何利用SANAS得到模型结构，完整示例代码请参考sa_nas_mobilenetv2.py或者block_sa_nas_mobilenetv2.py。</p>
<h2 id="_2">接口介绍</h2>
<p>请参考。</p>
<h3 id="1">1. 配置搜索空间</h3>
<p>详细的搜索空间配置可以参考<a href='../../../paddleslim/nas/nas_api.md'>神经网络搜索API文档</a>。</p>
<pre><code>config = [('MobileNetV2Space')]
</code></pre>
<h3 id="2-sanas">2. 利用搜索空间初始化SANAS实例</h3>
<pre><code>from paddleslim.nas import SANAS
sa_nas = SANAS(
config,
server_addr=(&quot;&quot;, 8881),
init_temperature=10.24,
reduce_rate=0.85,
search_steps=300,
is_server=True)
</code></pre>
<h3 id="3-nas">3. 根据实例化的NAS得到当前的网络结构</h3>
<pre><code>archs = sa_nas.next_archs()
</code></pre>
<h3 id="4-program">4. 根据得到的网络结构和输入构造训练和测试program</h3>
<pre><code>import paddle.fluid as fluid
train_program = fluid.Program()
test_program = fluid.Program()
startup_program = fluid.Program()
with fluid.program_guard(train_program, startup_program):
data = fluid.data(name='data', shape=[None, 3, 32, 32], dtype='float32')
label = fluid.data(name='label', shape=[None, 1], dtype='int64')
for arch in archs:
data = arch(data)
output = fluid.layers.fc(data, 10)
softmax_out = fluid.layers.softmax(input=output, use_cudnn=False)
cost = fluid.layers.cross_entropy(input=softmax_out, label=label)
avg_cost = fluid.layers.mean(cost)
acc_top1 = fluid.layers.accuracy(input=softmax_out, label=label, k=1)
test_program = train_program.clone(for_test=True)
sgd = fluid.optimizer.SGD(learning_rate=1e-3)
sgd.minimize(avg_cost)
</code></pre>
<h3 id="5-program">5. 根据构造的训练program添加限制条件</h3>
<pre><code>from paddleslim.analysis import flops
if flops(train_program) &gt; 321208544:
continue
</code></pre>
<h3 id="6-score">6. 回传score</h3>
<pre><code>sa_nas.reward(score)
</code></pre></div>
</div>
<footer class="col-md-12">
<hr>
<p>Documentation built with <a href="https://www.mkdocs.org/">MkDocs</a>.</p>
</footer>
<script>
var base_url = "../../..",
shortcuts = {"help": 191, "next": 78, "previous": 80, "search": 83};
</script>
<script src="../../../js/base.js" defer></script>
<script src="../../../search/main.js" defer></script>
<div class="modal" id="mkdocs_search_modal" tabindex="-1" role="dialog" aria-labelledby="Search Modal" aria-hidden="true">
<div class="modal-dialog">
<div class="modal-content">
<div class="modal-header">
<button type="button" class="close" data-dismiss="modal"><span aria-hidden="true">&times;</span><span class="sr-only">Close</span></button>
<h4 class="modal-title" id="exampleModalLabel">Search</h4>
</div>
<div class="modal-body">
<p>
From here you can search these documents. Enter
your search terms below.
</p>
<form role="form">
<div class="form-group">
<input type="text" class="form-control" placeholder="Search..." id="mkdocs-search-query" title="Type search term here">
</div>
</form>
<div id="mkdocs-search-results"></div>
</div>
<div class="modal-footer">
</div>
</div>
</div>
</div><div class="modal" id="mkdocs_keyboard_modal" tabindex="-1" role="dialog" aria-labelledby="Keyboard Shortcuts Modal" aria-hidden="true">
<div class="modal-dialog">
<div class="modal-content">
<div class="modal-header">
<button type="button" class="close" data-dismiss="modal"><span aria-hidden="true">&times;</span><span class="sr-only">Close</span></button>
<h4 class="modal-title" id="exampleModalLabel">Keyboard Shortcuts</h4>
</div>
<div class="modal-body">
<table class="table">
<thead>
<tr>
<th style="width: 20%;">Keys</th>
<th>Action</th>
</tr>
</thead>
<tbody>
<tr>
<td class="help shortcut"><kbd>?</kbd></td>
<td>Open this help</td>
</tr>
<tr>
<td class="next shortcut"><kbd>n</kbd></td>
<td>Next page</td>
</tr>
<tr>
<td class="prev shortcut"><kbd>p</kbd></td>
<td>Previous page</td>
</tr>
<tr>
<td class="search shortcut"><kbd>s</kbd></td>
<td>Search</td>
</tr>
</tbody>
</table>
</div>
<div class="modal-footer">
</div>
</div>
</div>
</div>
</body>
</html>
docs/tutorials/quant_aware_demo/index.html 0 → 100644
浏览文件 @ 01c4cf93
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="utf-8">
<meta http-equiv="X-UA-Compatible" content="IE=edge">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<link rel="shortcut icon" href="../../../img/favicon.ico">
<title>在线量化示例 - My Docs</title>
<link href="../../../css/bootstrap-custom.min.css" rel="stylesheet">
<link href="../../../css/font-awesome.min.css" rel="stylesheet">
<link href="../../../css/base.css" rel="stylesheet">
<link rel="stylesheet" href="https://cdnjs.cloudflare.com/ajax/libs/highlight.js/9.12.0/styles/github.min.css">
<!-- HTML5 shim and Respond.js IE8 support of HTML5 elements and media queries -->
<!--[if lt IE 9]>
<script src="https://oss.maxcdn.com/libs/html5shiv/3.7.0/html5shiv.js"></script>
<script src="https://oss.maxcdn.com/libs/respond.js/1.4.2/respond.min.js"></script>
<![endif]-->
<script src="../../../js/jquery-1.10.2.min.js" defer></script>
<script src="../../../js/bootstrap-3.0.3.min.js" defer></script>
<script src="https://cdnjs.cloudflare.com/ajax/libs/highlight.js/9.12.0/highlight.min.js"></script>
<script>hljs.initHighlightingOnLoad();</script>
</head>
<body>
<div class="navbar navbar-default navbar-fixed-top" role="navigation">
<div class="container">
<!-- Collapsed navigation -->
<div class="navbar-header">
<!-- Expander button -->
<button type="button" class="navbar-toggle" data-toggle="collapse" data-target=".navbar-collapse">
<span class="sr-only">Toggle navigation</span>
<span class="icon-bar"></span>
<span class="icon-bar"></span>
<span class="icon-bar"></span>
</button>
<a class="navbar-brand" href="../../..">My Docs</a>
</div>
<!-- Expanded navigation -->
<div class="navbar-collapse collapse">
<!-- Main navigation -->
<ul class="nav navbar-nav">
<li >
<a href="../../..">Welcome to MkDocs</a>
</li>
<li class="dropdown active">
<a href="#" class="dropdown-toggle" data-toggle="dropdown">Docs <b class="caret"></b></a>
<ul class="dropdown-menu">
<li >
<a href="../../">PaddleSlim</a>
</li>
<li >
<a href="../../table_latency/">硬件延时评估表</a>
</li>
<li class="dropdown-submenu">
<a href="#">Api</a>
<ul class="dropdown-menu">
<li >
<a href="../../api/analysis_api/">模型分析API文档</a>
</li>
<li >
<a href="../../api/api_guide/">PaddleSlim API文档导航</a>
</li>
<li >
<a href="../../api/nas_api/">paddleslim.nas API文档</a>
</li>
<li >
<a href="../../api/prune_api/">卷积通道剪裁API文档</a>
</li>
<li >
<a href="../../api/quantization_api/">paddleslim.quant API文档</a>
</li>
<li >
<a href="../../api/search_space/">paddleslim.nas 提供的搜索空间：</a>
</li>
<li >
<a href="../../api/single_distiller_api/">paddleslim.dist API文档</a>
</li>
</ul>
</li>
<li class="dropdown-submenu">
<a href="#">Tutorials</a>
<ul class="dropdown-menu">
<li >
<a href="../demo_guide/">Demo guide</a>
</li>
<li >
<a href="../nas_demo/">网络结构搜索示例</a>
</li>
<li class="active">
<a href="./">在线量化示例</a>
</li>
<li >
<a href="../quant_embedding_demo/">Embedding量化示例</a>
</li>
<li >
<a href="../quant_post_demo/">离线量化示例</a>
</li>
</ul>
</li>
</ul>
</li>
</ul>
<ul class="nav navbar-nav navbar-right">
<li>
<a href="#" data-toggle="modal" data-target="#mkdocs_search_modal">
<i class="fa fa-search"></i> Search
</a>
</li>
<li >
<a rel="next" href="../nas_demo/">
<i class="fa fa-arrow-left"></i> Previous
</a>
</li>
<li >
<a rel="prev" href="../quant_embedding_demo/">
Next <i class="fa fa-arrow-right"></i>
</a>
</li>
</ul>
</div>
</div>
</div>
<div class="container">
<div class="col-md-3"><div class="bs-sidebar hidden-print affix well" role="complementary">
<ul class="nav bs-sidenav">
<li class="main active"><a href="#_1">在线量化示例</a></li>
<li><a href="#_2">接口介绍</a></li>
<li><a href="#_3">分类模型的离线量化流程</a></li>
</ul>
</div></div>
<div class="col-md-9" role="main">
<h1 id="_1">在线量化示例</h1>
<p>本示例介绍如何使用在线量化接口，来对训练好的分类模型进行量化, 可以减少模型的存储空间和显存占用。</p>
<h2 id="_2">接口介绍</h2>
<p>请参考 <a href='../../../paddleslim/quant/quantization_api_doc.md'>量化API文档</a>。</p>
<h2 id="_3">分类模型的离线量化流程</h2>
<h3 id="1">1. 配置量化参数</h3>
<pre><code>quant_config = {
'weight_quantize_type': 'abs_max',
'activation_quantize_type': 'moving_average_abs_max',
'weight_bits': 8,
'activation_bits': 8,
'not_quant_pattern': ['skip_quant'],
'quantize_op_types': ['conv2d', 'depthwise_conv2d', 'mul'],
'dtype': 'int8',
'window_size': 10000,
'moving_rate': 0.9,
'quant_weight_only': False
}
</code></pre>
<h3 id="2-programop">2. 对训练和测试program插入可训练量化op</h3>
<pre><code>val_program = quant_aware(val_program, place, quant_config, scope=None, for_test=True)
compiled_train_prog = quant_aware(train_prog, place, quant_config, scope=None, for_test=False)
</code></pre>
<h3 id="3build">3.关掉指定build策略</h3>
<pre><code>build_strategy = fluid.BuildStrategy()
build_strategy.fuse_all_reduce_ops = False
build_strategy.sync_batch_norm = False
exec_strategy = fluid.ExecutionStrategy()
compiled_train_prog = compiled_train_prog.with_data_parallel(
loss_name=avg_cost.name,
build_strategy=build_strategy,
exec_strategy=exec_strategy)
</code></pre>
<h3 id="4-freeze-program">4. freeze program</h3>
<pre><code>float_program, int8_program = convert(val_program,
place,
quant_config,
scope=None,
save_int8=True)
</code></pre>
<h3 id="5">5.保存预测模型</h3>
<pre><code>fluid.io.save_inference_model(
dirname=float_path,
feeded_var_names=[image.name],
target_vars=[out], executor=exe,
main_program=float_program,
model_filename=float_path + '/model',
params_filename=float_path + '/params')
fluid.io.save_inference_model(
dirname=int8_path,
feeded_var_names=[image.name],
target_vars=[out], executor=exe,
main_program=int8_program,
model_filename=int8_path + '/model',
params_filename=int8_path + '/params')
</code></pre></div>
</div>
<footer class="col-md-12">
<hr>
<p>Documentation built with <a href="https://www.mkdocs.org/">MkDocs</a>.</p>
</footer>
<script>
var base_url = "../../..",
shortcuts = {"help": 191, "next": 78, "previous": 80, "search": 83};
</script>
<script src="../../../js/base.js" defer></script>
<script src="../../../search/main.js" defer></script>
<div class="modal" id="mkdocs_search_modal" tabindex="-1" role="dialog" aria-labelledby="Search Modal" aria-hidden="true">
<div class="modal-dialog">
<div class="modal-content">
<div class="modal-header">
<button type="button" class="close" data-dismiss="modal"><span aria-hidden="true">&times;</span><span class="sr-only">Close</span></button>
<h4 class="modal-title" id="exampleModalLabel">Search</h4>
</div>
<div class="modal-body">
<p>
From here you can search these documents. Enter
your search terms below.
</p>
<form role="form">
<div class="form-group">
<input type="text" class="form-control" placeholder="Search..." id="mkdocs-search-query" title="Type search term here">
</div>
</form>
<div id="mkdocs-search-results"></div>
</div>
<div class="modal-footer">
</div>
</div>
</div>
</div><div class="modal" id="mkdocs_keyboard_modal" tabindex="-1" role="dialog" aria-labelledby="Keyboard Shortcuts Modal" aria-hidden="true">
<div class="modal-dialog">
<div class="modal-content">
<div class="modal-header">
<button type="button" class="close" data-dismiss="modal"><span aria-hidden="true">&times;</span><span class="sr-only">Close</span></button>
<h4 class="modal-title" id="exampleModalLabel">Keyboard Shortcuts</h4>
</div>
<div class="modal-body">
<table class="table">
<thead>
<tr>
<th style="width: 20%;">Keys</th>
<th>Action</th>
</tr>
</thead>
<tbody>
<tr>
<td class="help shortcut"><kbd>?</kbd></td>
<td>Open this help</td>
</tr>
<tr>
<td class="next shortcut"><kbd>n</kbd></td>
<td>Next page</td>
</tr>
<tr>
<td class="prev shortcut"><kbd>p</kbd></td>
<td>Previous page</td>
</tr>
<tr>
<td class="search shortcut"><kbd>s</kbd></td>
<td>Search</td>
</tr>
</tbody>
</table>
</div>
<div class="modal-footer">
</div>
</div>
</div>
</div>
</body>
</html>
docs/tutorials/quant_embedding_demo/index.html 0 → 100644
浏览文件 @ 01c4cf93
此差异已折叠。
docs/tutorials/quant_post_demo/index.html 0 → 100644
浏览文件 @ 01c4cf93
此差异已折叠。
fonts/fontawesome-webfont.svg 0 → 100644
浏览文件 @ 01c4cf93
此差异已折叠。
fonts/glyphicons-halflings-regular.svg 0 → 100644
浏览文件 @ 01c4cf93
此差异已折叠。
index.html 0 → 100644
浏览文件 @ 01c4cf93
此差异已折叠。
js/base.js 0 → 100644
浏览文件 @ 01c4cf93
此差异已折叠。
js/bootstrap-3.0.3.min.js 0 → 100644
浏览文件 @ 01c4cf93
此差异已折叠。
js/jquery-1.10.2.min.js 0 → 100644
浏览文件 @ 01c4cf93
此差异已折叠。
mkdocs.yml 0 → 100644
浏览文件 @ 01c4cf93
site_name: PaddleSlim Docs
nav:
- Home: index.md
- 教程:
- 离线量化: tutorials/quant_post_demo.md
- 量化训练: tutorials/quant_aware_demo.md
- Embedding量化: tutorials/quant_embedding_demo.md
- SA搜索: tutorials/nas_demo.md
- API:
- 量化: api/quantization_api.md
- 剪枝: api/prune_api.md
- 敏感度分析: api/analysis_api.md
- 蒸馏: api/single_distiller_api.md
- SA搜索: api/nas_api.md
- 搜索空间: api/search_space.md
theme: readthedocs
markdown_extensions:
- pymdownx.arithmatex
- admonition
- codehilite
extra_javascript:
- mathjax-config.js
- MathJax.js?config=TeX-AMS-MML_HTMLorMML
search/lunr.js 0 → 100644
浏览文件 @ 01c4cf93
此差异已折叠。
search/main.js 0 → 100644
浏览文件 @ 01c4cf93
此差异已折叠。
search/search_index.json 0 → 100644
浏览文件 @ 01c4cf93
此差异已折叠。
search/worker.js 0 → 100644
浏览文件 @ 01c4cf93
此差异已折叠。
sitemap.xml 0 → 100644
浏览文件 @ 01c4cf93
此差异已折叠。
sitemap.xml.gz 0 → 100644
浏览文件 @ 01c4cf93
文件已添加
Markdown is supported
0% 或 .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑！
想要评论请 注册 或