Merge branch 'gh-pages' of https://github.com/PaddlePaddle/paddle into add-mobile-pages

css/home.css

@@ -2375,8 +2375,8 @@ header.site-header .logo {
   float: left;
 }
 header.site-header .logo > img {
-  height: 40px;
-  margin: 10px 10px 10px 10px;
+  height: 60px;
+  /*margin: 10px 10px 10px 10px;*/
 }
 header.site-header .top-nav {
   margin-left: 155px;
@@ -2397,20 +2397,23 @@ header.site-header .top-nav .site-links > li > a {
   height: 100%;
   line-height: 60px;
   padding: 0 20px;
-  color: rgba(255, 255, 255, 0.7);
+  color: #fff;
+  /*&.active,*/
 }
-header.site-header .top-nav .site-links > li > a.active,
 header.site-header .top-nav .site-links > li > a:hover {
-  color: #fff;
+  color: #0073eb;
+  background-color: #222222;
 }
 header.site-header .top-nav .version-switcher {
   position: relative;
   display: inline-block;
+  width: 100px;
   height: 60px;
   float: left;
 }
 header.site-header .top-nav .version-switcher:hover > a {
   color: #fff;
+  cursor: default;
 }
 header.site-header .top-nav .version-switcher:hover > a .fa::before {
   content: "\F106";
@@ -2427,10 +2430,10 @@ header.site-header .top-nav .version-switcher > a .fa::before {
 }
 header.site-header .top-nav .version-switcher ul {
   display: none;
-  top: 50px;
-  width: 60px;
+  top: 60px;
+  width: 100px;
   text-align: center;
-  line-height: 30px;
+  line-height: 50px;
   background: rgba(0, 0, 0, 0.85);
   color: #0073eb;
   position: absolute;
@@ -2438,9 +2441,13 @@ header.site-header .top-nav .version-switcher ul {
 }
 header.site-header .top-nav .version-switcher ul > li > a {
   color: #ababab;
+  width: 99px;
+  height: 50px;
+  display: block;
 }
 header.site-header .top-nav .version-switcher ul > li > a:hover {
   color: #0073eb;
+  background-color: #222;
 }
 header.site-header .top-nav .version-switcher:hover ul {
   display: inline-block;
@@ -2452,17 +2459,12 @@ header.site-header .right-nav {
 header.site-header .right-nav .language-switcher {
   position: relative;
   display: inline-block;
-  height: 60px;
   float: left;
 }
-header.site-header .right-nav .language-switcher:hover > a {
-  color: #fff;
-}
-header.site-header .right-nav .language-switcher:hover > a .fa::before {
-  content: "\F106";
-}
 header.site-header .right-nav .language-switcher > a {
-  color: rgba(255, 255, 255, 0.7);
+  color: #fff;
+  width: 80px;
+  text-align: center;
   line-height: 60px;
   display: inline-block;
   margin-right: 10px;
@@ -2471,35 +2473,19 @@ header.site-header .right-nav .language-switcher > a .fa::before {
   margin-left: 5px;
   content: "\F107";
 }
-header.site-header .right-nav .language-switcher ul {
-  display: none;
-  top: 50px;
-  width: 60px;
-  text-align: center;
-  line-height: 30px;
-  background: rgba(0, 0, 0, 0.85);
+header.site-header .right-nav .language-switcher > a:hover {
   color: #0073eb;
-  position: absolute;
-  left: 0;
-}
-header.site-header .right-nav .language-switcher ul > li > a {
-  color: #ababab;
-}
-header.site-header .right-nav .language-switcher ul > li > a:hover {
-  color: #0073eb;
-}
-header.site-header .right-nav .language-switcher:hover ul {
-  display: inline-block;
+  background-color: #222;
 }
 header.site-header .github-fork {
   float: left;
-  margin-left: 20px;
+  /* margin-left: 20px; */
 }
 header.site-header .github-fork > a {
   display: inline-block;
   line-height: 40px;
   width: 100px;
-  color: rgba(255, 255, 255, 0.7);
+  color: #fff;
   padding: 0 10px;
   margin: 10px 0;
   font-size: 14px;
@@ -2641,10 +2627,13 @@ header.site-header .github-fork > a:hover {
   color: #999;
 }
 .services .service-desc a.view-more {
-  color: #666;
+  color: #0073eb;
   font-size: 16px;
   display: inline-block;
 }
+.services .service-desc a.view-more:hover {
+  text-decoration: underline;
+}
 .features {
   background-color: #ffffff;
   padding-top: 80px;

develop/doc/_sources/design/cluster_train/save_model.md.txt

+# Design Doc: Save Model
+
+## Overview
+
+The model is the output of the training process. There are two
+ways from which user can obtain a model:
+
+- Save model triggered by user code: user code asks PaddlePaddle to
+  save a model.
+- Convert model from the checkpoint: model being converted from
+  pservers' periodic checkpoint. In this way, the user can cancel a
+  job at any time, and still have a relatively fresh model (we
+  checkpoint around every 5 minutes).
+
+### Trainer Saving Model vs. Pservers Saving Model
+
+Both trainers and pservers have access to the model. So the model can
+be saved from a trainer or pservers. We need to decide where the model
+is saved from.
+
+#### Dense Update vs. Sparse Update
+
+There are two types of model update methods: dense update and sparse
+update (when the model parameter is configured to be sparse).
+
+- Dense update
+
+  Every trainer has it's own full copy of the model. Every model
+  update will update the entire model.
+
+- Sparse update
+
+  The training input is sparse, and the trainer does not have the
+  entire model. It will only download the sub-model necessary related
+  to the input. When updating the model, only the sub-model related to
+  the training input is updated.
+
+
+#### Pservers Saving Model
+
+The benefit of letting pservers save model is they have the entire
+model all the time. However, since pservers are on different nodes, it
+requires a merging process to merge model shards into the same
+model. Thus requires the pservers to write models to a distributed
+filesystem, making the checkpoint shards visible to the merge program.
+
+#### Trainer Saving Model
+
+The benefit of letting one trainer to save the model is it does not
+require a distributed filesystem. And it's reusing the same save model
+logic when training locally - except when doing sparse update, the
+trainer needs to download the entire model during the saving process.
+
+#### Conclusion
+
+Given trainer saving model does not require a distributed filesystem,
+and is an intuitive extension to trainer saving model when training
+locally, we decide to let the trainer save the model when doing
+distributed training.
+
+
+### Convert Model from Checkpoint
+
+TODO
+
+
+## Timeline
+
+We first implement trainer save the model. Converting the latest
+snapshot to a model will be a TODO for future.
+
+
+## Trainer Save Model
+
+### Trainer Election
+
+One trainer will be elected as the one to save the model. When using
+etcd, trainer ID is a randomly generated UUID, we will utilize etcd to
+elect one trainer. When not using etcd, unique trainer IDs will be
+given by the administrator, the trainer whose ID is "0" is elected to
+save the model.
+
+### Model Save Path
+
+Each trainer will be given the directory to save the model. The
+elected trainer will save the model to
+`given-directory/trainerID`. Since the trainer ID is unique, this
+would prevent concurrent save to the same file when multiple trainers
+are elected to save the model when split-brain problem happens.
+
+### What Happens When Model Is Saving
+
+It takes some time to save model, we need to define what will happen
+when save model is taking place.
+
+When doing dense update, the trainer uses the local model. Pservers
+does not need to pause model update.
+
+When doing sparse update. The trainer needs to download the entire
+model while saving. To get the most accurate model, the model update
+needs to be paused before the download starts and resumed after the
+download finishes. Otherwise, the trainer gets a model that is
+"polluted": some part of the model is old, some part of the model is
+new.
+
+It's unclear that the "polluted" model will be inferior due to the
+stochastic nature of deep learning, and pausing the model update will
+add more complexity to the system. Since supporting sparse update is a
+TODO item. We defer the evaluation of pause the model update or not
+during saving model to the future.

develop/doc/_sources/design/simple_op_design.md.txt

+## Interaction between C++ and Python
+
+Users employ API in Python to describe their own network, however, the network construction actually happens in C++. so Protobuf is introduced to send the message between Python and C++. 
+
+The Interaction between Python and C++ can be simplified as two steps:
+
+1. C++ tells Python how many Ops there are, and what parameter do users need to offer to initialize a new Op. Python then builds API for each Op at compile time.
+
+2. Users invoke APIs built by Python and provide necessary parameters. These parameters will be sent to C++ fo finish Op construction task.
+
+### Message form C++ to Python
+
+We define a Protobuf message class `OpProto` to hold message needed in the first step. What should an `OpProto` contain? This question is equivalent to “What message do we need to offer, to build a Python API which is legal and user oriented and can use to describe a whole Op.”
+
+Following message are necessary:
+
+1. Op's name, and its simple comment.
+2. Input and output variable number; each variable's name, type, and comment.
+3. Op's attributes; each attribute includes name, type, comment, **default value** and **value range**.
+
+So `OpProto` can be defined as follows:
+
+```proto
+enum AttrType {
+	INT = 1;
+	FLOAT = 2;
+	STRING = 3;
+	INTS = 4;
+	FLOATS = 5;
+	STRINGS = 6;
+};
+
+message AttrValue {
+	AttrType type = 1;
+	optional int iv = 2;
+	optional float fv = 3;
+	optional string sv = 4;
+	repeated int ivs = 5;
+	repeated float fvs = 6;
+	repeated string svs = 7;
+};
+
+message AttrProto {
+	required string name = 1;
+	required string comment = 2;
+	required AttrType type = 3;
+};
+
+message VarProto {
+	required string name = 1;
+	required string comment = 2;
+};
+
+message OpProto {
+	repeated VarProto inputs = 1;
+	repeated VarProto outputs = 2;
+	repeated AttrProto attrs = 3;
+	required string type = 4;
+	required string comment = 5;
+};
+```
+
+To generate Python code automatically:
+
+```python 
+def create_python_ops_creatation_functions():
+	op_protos = paddle.framework.OpRegistry.get_all_op_proto()
+	for type_name in op_protos:
+		op_proto = op_protos[type_name]
+		def __impl__(**kwargs):  # User must use key word args in Paddle API
+			inputs = [kwargs.get(ipt.name, "") for ipt in op_proto.inputs]
+			outputs = [kwargs.get(opt.name, "") for opt in op_proto.outputs]
+			attrs = [cast_to_op_attr(attr, kwargs.get(attr.name, None)) for attr in op_proto.attrs]
+			opdesc = （input, outputs, type_name, attrs）
+			return paddle.framework.OpRegistry.CreateOp(opdesc)
+		__impl__.__doc__ = create_doc_string(op_proto)
+		globals()[type_name] = __impl__
+
+create_python_ops_creatation_functions()
+```
+
+### Message from Python to C++
+
+To hold message needed in the above second step, we define Protobuf message class `OpDesc`. It is used to hold user-specified parameters in Op describing.
+
+```proto
+message OpDesc {
+	required string type = 1;	
+	repeated string inputs = 2;
+	repeated string outputs = 3;
+	map<string, AttrValue> attrs = 4;
+};
+```
+
+## OpProto Register
+
+Every Op has its own `OpProto`. For using convenience, we need to register them and record all their messages. For each `Op` class, we define a corresponding `OpMaker` class, in whose constructor we implement the `OpProto`'s building process. `OpMaker`'s constructor will be invoked by another function `OpRegistry::RegisterOp()`.
+
+```cpp
+class OpProtoMaker {
+public:
+	OpProtoMaker(OpProto* proto): proto_(proto) {}
+protected:
+	OpProto* proto_;
+	void AddInput(const std::string& name, const std::string& desc) {...}
+	void AddAttr(const std::string& name, const std::string& desc, TypeId type) {...}
+	void AddComment(const std::string& comment) { ... }
+};
+
+class OpRegistry {
+public:
+	using OpCreator = std::function<OperatorBase* (OpDesc& desc)>;
+	
+	template <typename OpType, typename OpMaker>
+	static void RegisterOp(const std::string& name) {
+		gCreators_[name] = [](const OpDesc& desc) {
+			return new OpType(desc);
+		};
+		OpProto& opProto = gProtos_[name];
+		OpMaker()(&opProto);
+	}
+
+	static map<string, OpCreator> gCreators_;
+	static map<string, OpProto> gProtos_;
+};
+
+template <typename OpType, typename OpMaker>
+class OpRegister {
+  public:
+    OpRegister(std::string type) {
+        OpRegistry::RegisterOp<OpType, OpMaker>(type);
+    }
+};
+
+#define REGISTER_OP(op_class, op_maker_class, type_name)         \
+    class op_class##Register {                                   \
+      private:                                                   \
+        const static OpRegister<#op_class, #op_maker_class> reg; \
+    };                                                           \
+    const Register op_class##Register::reg(#type_name);
+    
+class CosineOp {
+// ...
+}
+
+struct CosineOpProtoMaker : public OpProtoMaker {
+	CosineOpProtoMaker(OpProto* proto) : OpProtoMaker(proto) {
+		AddInput("input", "input of cosine op");
+		AddAttr("scale", "scale of cosine op", float).Default(1.0).LargerThan(0.0);
+		AddType("cos");
+		AddComment("This is cos op");
+	}
+}
+
+REGISTER_OP(CosineOp, CosineOpProtoMaker, cos);
+```
+
+In `REGISTER_OP(CosineOp, CosineOpProtoMaker, cos)`, we register not only `CosineOp` but also `CosineOpProto`. As fields of `CosineOpProto`, the default value and value range of `scale` are also registered here. 
+
+## Python API
+
+Python  APIs are divided into two types, high-level API and low-level API.
+
+### High-Level API
+
+High-level API is called by users directly, so it should keep its style consistent with existing V2 APIs.
+
+Here is a sample about how a define a fc layer:
+
+```python
+hd = fc_layer(input=data, size=56, with_bias=True, activation="sigmoid");
+```
+
+`hd` is the output of `fc_layer` and it's a `variable`. It can be further sent into other layers as input.
+
+The definition of `fc_layer()`:
+
+```python
+def fc_layer(input, size, with_bias, activation):
+	attr_map = {"size":size}
+	check_attrs(attr_map)
+	w = make_variable('w')
+	if with_bias:
+		b = make_variable('b')
+	else:
+		b = None
+	fc_output = make_variable('fc_output');
+	fc_op(input, w, b, fc_output, attr_map)
+	act_output = make_variable('sigmod_output');
+	if activation == "sigmod":
+		sigmod_op(fc_output, act_output);
+	elif:
+		# ...
+	return act_output;
+``` 
+
+### Low Leval API
+
+In above sample, `fc_op` and `sigmod_op` are low-level API. They build `OpDesc` and invoke corresponding C++ code.
+
+*TODO*

develop/doc/design/cluster_train/save_model.html

+
+
+<!DOCTYPE html>
+<!--[if IE 8]><html class="no-js lt-ie9" lang="en" > <![endif]-->
+<!--[if gt IE 8]><!--> <html class="no-js" lang="en" > <!--<![endif]-->
+<head>
+  <meta charset="utf-8">
+  
+  <meta name="viewport" content="width=device-width, initial-scale=1.0">
+  
+  <title>Design Doc: Save Model &mdash; PaddlePaddle  documentation</title>
+  
+
+  
+  
+
+  
+
+  
+  
+    
+
+  
+
+  
+  
+    <link rel="stylesheet" href="../../_static/css/theme.css" type="text/css" />
+  
+
+  
+  
+        <link rel="index" title="Index"
+              href="../../genindex.html"/>
+        <link rel="search" title="Search" href="../../search.html"/>
+    <link rel="top" title="PaddlePaddle  documentation" href="../../index.html"/> 
+
+  <link rel="stylesheet" href="https://cdn.jsdelivr.net/perfect-scrollbar/0.6.14/css/perfect-scrollbar.min.css" type="text/css" />
+  <link rel="stylesheet" href="../../_static/css/override.css" type="text/css" />
+  <script>
+  var _hmt = _hmt || [];
+  (function() {
+    var hm = document.createElement("script");
+    hm.src = "//hm.baidu.com/hm.js?b9a314ab40d04d805655aab1deee08ba";
+    var s = document.getElementsByTagName("script")[0]; 
+    s.parentNode.insertBefore(hm, s);
+  })();
+  </script>
+
+  
+
+  
+  <script src="../../_static/js/modernizr.min.js"></script>
+
+</head>
+
+<body class="wy-body-for-nav" role="document">
+
+  
+  <header class="site-header">
+    <div class="site-logo">
+      <a href="/"><img src="../../_static/images/PP_w.png"></a>
+    </div>
+    <div class="site-nav-links">
+      <div class="site-menu">
+        <a class="fork-on-github" href="https://github.com/PaddlePaddle/Paddle" target="_blank"><i class="fa fa-github"></i>Folk me on Github</a>
+        <div class="language-switcher dropdown">
+          <a type="button" data-toggle="dropdown">
+            <span>English</span>
+            <i class="fa fa-angle-up"></i>
+            <i class="fa fa-angle-down"></i>
+          </a>
+          <ul class="dropdown-menu">
+            <li><a href="/doc_cn">中文</a></li>
+            <li><a href="/doc">English</a></li>
+          </ul>
+        </div>
+        <ul class="site-page-links">
+          <li><a href="/">Home</a></li>
+        </ul>
+      </div>
+      <div class="doc-module">
+        
+        <ul>
+<li class="toctree-l1"><a class="reference internal" href="../../getstarted/index_en.html">GET STARTED</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../../howto/index_en.html">HOW TO</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../../api/index_en.html">API</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../../about/index_en.html">ABOUT</a></li>
+</ul>
+
+        
+<div role="search">
+  <form id="rtd-search-form" class="wy-form" action="../../search.html" method="get">
+    <input type="text" name="q" placeholder="Search docs" />
+    <input type="hidden" name="check_keywords" value="yes" />
+    <input type="hidden" name="area" value="default" />
+  </form>
+</div>        
+      </div>
+    </div>
+  </header>
+  
+  <div class="main-content-wrap">
+
+    
+    <nav class="doc-menu-vertical" role="navigation">
+        
+          
+          <ul>
+<li class="toctree-l1"><a class="reference internal" href="../../getstarted/index_en.html">GET STARTED</a><ul>
+<li class="toctree-l2"><a class="reference internal" href="../../getstarted/build_and_install/index_en.html">Install and Build</a><ul>
+<li class="toctree-l3"><a class="reference internal" href="../../getstarted/build_and_install/docker_install_en.html">PaddlePaddle in Docker Containers</a></li>
+<li class="toctree-l3"><a class="reference internal" href="../../getstarted/build_and_install/ubuntu_install_en.html">Debian Package installation guide</a></li>
+<li class="toctree-l3"><a class="reference internal" href="../../getstarted/build_and_install/build_from_source_en.html">Installing from Sources</a></li>
+</ul>
+</li>
+</ul>
+</li>
+<li class="toctree-l1"><a class="reference internal" href="../../howto/index_en.html">HOW TO</a><ul>
+<li class="toctree-l2"><a class="reference internal" href="../../howto/usage/cmd_parameter/index_en.html">Set Command-line Parameters</a><ul>
+<li class="toctree-l3"><a class="reference internal" href="../../howto/usage/cmd_parameter/use_case_en.html">Use Case</a></li>
+<li class="toctree-l3"><a class="reference internal" href="../../howto/usage/cmd_parameter/arguments_en.html">Argument Outline</a></li>
+<li class="toctree-l3"><a class="reference internal" href="../../howto/usage/cmd_parameter/detail_introduction_en.html">Detail Description</a></li>
+</ul>
+</li>
+<li class="toctree-l2"><a class="reference internal" href="../../howto/usage/cluster/cluster_train_en.html">Run Distributed Training</a></li>
+<li class="toctree-l2"><a class="reference internal" href="../../howto/usage/k8s/k8s_en.html">Paddle On Kubernetes</a></li>
+<li class="toctree-l2"><a class="reference internal" href="../../howto/usage/k8s/k8s_aws_en.html">Distributed PaddlePaddle Training on AWS with Kubernetes</a></li>
+<li class="toctree-l2"><a class="reference internal" href="../../howto/dev/new_layer_en.html">Write New Layers</a></li>
+<li class="toctree-l2"><a class="reference internal" href="../../howto/dev/contribute_to_paddle_en.html">Contribute Code</a></li>
+<li class="toctree-l2"><a class="reference internal" href="../../howto/deep_model/rnn/index_en.html">RNN Models</a><ul>
+<li class="toctree-l3"><a class="reference internal" href="../../howto/deep_model/rnn/rnn_config_en.html">RNN Configuration</a></li>
+</ul>
+</li>
+<li class="toctree-l2"><a class="reference internal" href="../../howto/optimization/gpu_profiling_en.html">Tune GPU Performance</a></li>
+</ul>
+</li>
+<li class="toctree-l1"><a class="reference internal" href="../../api/index_en.html">API</a><ul>
+<li class="toctree-l2"><a class="reference internal" href="../../api/v2/model_configs.html">Model Configuration</a><ul>
+<li class="toctree-l3"><a class="reference internal" href="../../api/v2/config/activation.html">Activation</a></li>
+<li class="toctree-l3"><a class="reference internal" href="../../api/v2/config/layer.html">Layers</a></li>
+<li class="toctree-l3"><a class="reference internal" href="../../api/v2/config/evaluators.html">Evaluators</a></li>
+<li class="toctree-l3"><a class="reference internal" href="../../api/v2/config/optimizer.html">Optimizer</a></li>
+<li class="toctree-l3"><a class="reference internal" href="../../api/v2/config/pooling.html">Pooling</a></li>
+<li class="toctree-l3"><a class="reference internal" href="../../api/v2/config/networks.html">Networks</a></li>
+<li class="toctree-l3"><a class="reference internal" href="../../api/v2/config/attr.html">Parameter Attribute</a></li>
+</ul>
+</li>
+<li class="toctree-l2"><a class="reference internal" href="../../api/v2/data.html">Data Reader Interface and DataSets</a></li>
+<li class="toctree-l2"><a class="reference internal" href="../../api/v2/run_logic.html">Training and Inference</a></li>
+</ul>
+</li>
+<li class="toctree-l1"><a class="reference internal" href="../../about/index_en.html">ABOUT</a></li>
+</ul>
+
+        
+    </nav>
+    
+    <section class="doc-content-wrap">
+
+      
+
+ 
+
+
+
+
+
+
+
+<div role="navigation" aria-label="breadcrumbs navigation">
+  <ul class="wy-breadcrumbs">
+      
+    <li>Design Doc: Save Model</li>
+  </ul>
+</div>
+      
+      <div class="wy-nav-content" id="doc-content">
+        <div class="rst-content">
+          <div role="main" class="document" itemscope="itemscope" itemtype="http://schema.org/Article">
+           <div itemprop="articleBody">
+            
+  <div class="section" id="design-doc-save-model">
+<span id="design-doc-save-model"></span><h1>Design Doc: Save Model<a class="headerlink" href="#design-doc-save-model" title="Permalink to this headline">¶</a></h1>
+<div class="section" id="overview">
+<span id="overview"></span><h2>Overview<a class="headerlink" href="#overview" title="Permalink to this headline">¶</a></h2>
+<p>The model is the output of the training process. There are two
+ways from which user can obtain a model:</p>
+<ul class="simple">
+<li>Save model triggered by user code: user code asks PaddlePaddle to
+save a model.</li>
+<li>Convert model from the checkpoint: model being converted from
+pservers&#8217; periodic checkpoint. In this way, the user can cancel a
+job at any time, and still have a relatively fresh model (we
+checkpoint around every 5 minutes).</li>
+</ul>
+<div class="section" id="trainer-saving-model-vs-pservers-saving-model">
+<span id="trainer-saving-model-vs-pservers-saving-model"></span><h3>Trainer Saving Model vs. Pservers Saving Model<a class="headerlink" href="#trainer-saving-model-vs-pservers-saving-model" title="Permalink to this headline">¶</a></h3>
+<p>Both trainers and pservers have access to the model. So the model can
+be saved from a trainer or pservers. We need to decide where the model
+is saved from.</p>
+<div class="section" id="dense-update-vs-sparse-update">
+<span id="dense-update-vs-sparse-update"></span><h4>Dense Update vs. Sparse Update<a class="headerlink" href="#dense-update-vs-sparse-update" title="Permalink to this headline">¶</a></h4>
+<p>There are two types of model update methods: dense update and sparse
+update (when the model parameter is configured to be sparse).</p>
+<ul>
+<li><p class="first">Dense update</p>
+<p>Every trainer has it&#8217;s own full copy of the model. Every model
+update will update the entire model.</p>
+</li>
+<li><p class="first">Sparse update</p>
+<p>The training input is sparse, and the trainer does not have the
+entire model. It will only download the sub-model necessary related
+to the input. When updating the model, only the sub-model related to
+the training input is updated.</p>
+</li>
+</ul>
+</div>
+<div class="section" id="pservers-saving-model">
+<span id="pservers-saving-model"></span><h4>Pservers Saving Model<a class="headerlink" href="#pservers-saving-model" title="Permalink to this headline">¶</a></h4>
+<p>The benefit of letting pservers save model is they have the entire
+model all the time. However, since pservers are on different nodes, it
+requires a merging process to merge model shards into the same
+model. Thus requires the pservers to write models to a distributed
+filesystem, making the checkpoint shards visible to the merge program.</p>
+</div>
+<div class="section" id="trainer-saving-model">
+<span id="trainer-saving-model"></span><h4>Trainer Saving Model<a class="headerlink" href="#trainer-saving-model" title="Permalink to this headline">¶</a></h4>
+<p>The benefit of letting one trainer to save the model is it does not
+require a distributed filesystem. And it&#8217;s reusing the same save model
+logic when training locally - except when doing sparse update, the
+trainer needs to download the entire model during the saving process.</p>
+</div>
+<div class="section" id="conclusion">
+<span id="conclusion"></span><h4>Conclusion<a class="headerlink" href="#conclusion" title="Permalink to this headline">¶</a></h4>
+<p>Given trainer saving model does not require a distributed filesystem,
+and is an intuitive extension to trainer saving model when training
+locally, we decide to let the trainer save the model when doing
+distributed training.</p>
+</div>
+</div>
+<div class="section" id="convert-model-from-checkpoint">
+<span id="convert-model-from-checkpoint"></span><h3>Convert Model from Checkpoint<a class="headerlink" href="#convert-model-from-checkpoint" title="Permalink to this headline">¶</a></h3>
+<p>TODO</p>
+</div>
+</div>
+<div class="section" id="timeline">
+<span id="timeline"></span><h2>Timeline<a class="headerlink" href="#timeline" title="Permalink to this headline">¶</a></h2>
+<p>We first implement trainer save the model. Converting the latest
+snapshot to a model will be a TODO for future.</p>
+</div>
+<div class="section" id="trainer-save-model">
+<span id="trainer-save-model"></span><h2>Trainer Save Model<a class="headerlink" href="#trainer-save-model" title="Permalink to this headline">¶</a></h2>
+<div class="section" id="trainer-election">
+<span id="trainer-election"></span><h3>Trainer Election<a class="headerlink" href="#trainer-election" title="Permalink to this headline">¶</a></h3>
+<p>One trainer will be elected as the one to save the model. When using
+etcd, trainer ID is a randomly generated UUID, we will utilize etcd to
+elect one trainer. When not using etcd, unique trainer IDs will be
+given by the administrator, the trainer whose ID is &#8220;0&#8221; is elected to
+save the model.</p>
+</div>
+<div class="section" id="model-save-path">
+<span id="model-save-path"></span><h3>Model Save Path<a class="headerlink" href="#model-save-path" title="Permalink to this headline">¶</a></h3>
+<p>Each trainer will be given the directory to save the model. The
+elected trainer will save the model to
+<code class="docutils literal"><span class="pre">given-directory/trainerID</span></code>. Since the trainer ID is unique, this
+would prevent concurrent save to the same file when multiple trainers
+are elected to save the model when split-brain problem happens.</p>
+</div>
+<div class="section" id="what-happens-when-model-is-saving">
+<span id="what-happens-when-model-is-saving"></span><h3>What Happens When Model Is Saving<a class="headerlink" href="#what-happens-when-model-is-saving" title="Permalink to this headline">¶</a></h3>
+<p>It takes some time to save model, we need to define what will happen
+when save model is taking place.</p>
+<p>When doing dense update, the trainer uses the local model. Pservers
+does not need to pause model update.</p>
+<p>When doing sparse update. The trainer needs to download the entire
+model while saving. To get the most accurate model, the model update
+needs to be paused before the download starts and resumed after the
+download finishes. Otherwise, the trainer gets a model that is
+&#8220;polluted&#8221;: some part of the model is old, some part of the model is
+new.</p>
+<p>It&#8217;s unclear that the &#8220;polluted&#8221; model will be inferior due to the
+stochastic nature of deep learning, and pausing the model update will
+add more complexity to the system. Since supporting sparse update is a
+TODO item. We defer the evaluation of pause the model update or not
+during saving model to the future.</p>
+</div>
+</div>
+</div>
+
+
+           </div>
+          </div>
+          <footer>
+  
+
+  <hr/>
+
+  <div role="contentinfo">
+    <p>
+        &copy; Copyright 2016, PaddlePaddle developers.
+
+    </p>
+  </div>
+  Built with <a href="http://sphinx-doc.org/">Sphinx</a> using a <a href="https://github.com/snide/sphinx_rtd_theme">theme</a> provided by <a href="https://readthedocs.org">Read the Docs</a>. 
+
+</footer>
+
+        </div>
+      </div>
+
+    </section>
+
+  </div>
+  
+
+
+  
+
+    <script type="text/javascript">
+        var DOCUMENTATION_OPTIONS = {
+            URL_ROOT:'../../',
+            VERSION:'',
+            COLLAPSE_INDEX:false,
+            FILE_SUFFIX:'.html',
+            HAS_SOURCE:  true,
+            SOURCELINK_SUFFIX: ".txt",
+        };
+    </script>
+      <script type="text/javascript" src="../../_static/jquery.js"></script>
+      <script type="text/javascript" src="../../_static/underscore.js"></script>
+      <script type="text/javascript" src="../../_static/doctools.js"></script>
+      <script type="text/javascript" src="https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.0/MathJax.js?config=TeX-AMS-MML_HTMLorMML"></script>
+       
+  
+
+  
+  
+    <script type="text/javascript" src="../../_static/js/theme.js"></script>
+  
+  
+  <script src="https://maxcdn.bootstrapcdn.com/bootstrap/3.3.7/js/bootstrap.min.js" integrity="sha384-Tc5IQib027qvyjSMfHjOMaLkfuWVxZxUPnCJA7l2mCWNIpG9mGCD8wGNIcPD7Txa" crossorigin="anonymous"></script>
+  <script src="https://cdn.jsdelivr.net/perfect-scrollbar/0.6.14/js/perfect-scrollbar.jquery.min.js"></script>
+  <script src="../../_static/js/paddle_doc_init.js"></script> 
+
+</body>
+</html>
\ No newline at end of file

develop/doc_cn/_sources/design/cluster_train/save_model.md.txt

+# Design Doc: Save Model
+
+## Overview
+
+The model is the output of the training process. There are two
+ways from which user can obtain a model:
+
+- Save model triggered by user code: user code asks PaddlePaddle to
+  save a model.
+- Convert model from the checkpoint: model being converted from
+  pservers' periodic checkpoint. In this way, the user can cancel a
+  job at any time, and still have a relatively fresh model (we
+  checkpoint around every 5 minutes).
+
+### Trainer Saving Model vs. Pservers Saving Model
+
+Both trainers and pservers have access to the model. So the model can
+be saved from a trainer or pservers. We need to decide where the model
+is saved from.
+
+#### Dense Update vs. Sparse Update
+
+There are two types of model update methods: dense update and sparse
+update (when the model parameter is configured to be sparse).
+
+- Dense update
+
+  Every trainer has it's own full copy of the model. Every model
+  update will update the entire model.
+
+- Sparse update
+
+  The training input is sparse, and the trainer does not have the
+  entire model. It will only download the sub-model necessary related
+  to the input. When updating the model, only the sub-model related to
+  the training input is updated.
+
+
+#### Pservers Saving Model
+
+The benefit of letting pservers save model is they have the entire
+model all the time. However, since pservers are on different nodes, it
+requires a merging process to merge model shards into the same
+model. Thus requires the pservers to write models to a distributed
+filesystem, making the checkpoint shards visible to the merge program.
+
+#### Trainer Saving Model
+
+The benefit of letting one trainer to save the model is it does not
+require a distributed filesystem. And it's reusing the same save model
+logic when training locally - except when doing sparse update, the
+trainer needs to download the entire model during the saving process.
+
+#### Conclusion
+
+Given trainer saving model does not require a distributed filesystem,
+and is an intuitive extension to trainer saving model when training
+locally, we decide to let the trainer save the model when doing
+distributed training.
+
+
+### Convert Model from Checkpoint
+
+TODO
+
+
+## Timeline
+
+We first implement trainer save the model. Converting the latest
+snapshot to a model will be a TODO for future.
+
+
+## Trainer Save Model
+
+### Trainer Election
+
+One trainer will be elected as the one to save the model. When using
+etcd, trainer ID is a randomly generated UUID, we will utilize etcd to
+elect one trainer. When not using etcd, unique trainer IDs will be
+given by the administrator, the trainer whose ID is "0" is elected to
+save the model.
+
+### Model Save Path
+
+Each trainer will be given the directory to save the model. The
+elected trainer will save the model to
+`given-directory/trainerID`. Since the trainer ID is unique, this
+would prevent concurrent save to the same file when multiple trainers
+are elected to save the model when split-brain problem happens.
+
+### What Happens When Model Is Saving
+
+It takes some time to save model, we need to define what will happen
+when save model is taking place.
+
+When doing dense update, the trainer uses the local model. Pservers
+does not need to pause model update.
+
+When doing sparse update. The trainer needs to download the entire
+model while saving. To get the most accurate model, the model update
+needs to be paused before the download starts and resumed after the
+download finishes. Otherwise, the trainer gets a model that is
+"polluted": some part of the model is old, some part of the model is
+new.
+
+It's unclear that the "polluted" model will be inferior due to the
+stochastic nature of deep learning, and pausing the model update will
+add more complexity to the system. Since supporting sparse update is a
+TODO item. We defer the evaluation of pause the model update or not
+during saving model to the future.

develop/doc_cn/_sources/design/simple_op_design.md.txt

+## Interaction between C++ and Python
+
+Users employ API in Python to describe their own network, however, the network construction actually happens in C++. so Protobuf is introduced to send the message between Python and C++. 
+
+The Interaction between Python and C++ can be simplified as two steps:
+
+1. C++ tells Python how many Ops there are, and what parameter do users need to offer to initialize a new Op. Python then builds API for each Op at compile time.
+
+2. Users invoke APIs built by Python and provide necessary parameters. These parameters will be sent to C++ fo finish Op construction task.
+
+### Message form C++ to Python
+
+We define a Protobuf message class `OpProto` to hold message needed in the first step. What should an `OpProto` contain? This question is equivalent to “What message do we need to offer, to build a Python API which is legal and user oriented and can use to describe a whole Op.”
+
+Following message are necessary:
+
+1. Op's name, and its simple comment.
+2. Input and output variable number; each variable's name, type, and comment.
+3. Op's attributes; each attribute includes name, type, comment, **default value** and **value range**.
+
+So `OpProto` can be defined as follows:
+
+```proto
+enum AttrType {
+	INT = 1;
+	FLOAT = 2;
+	STRING = 3;
+	INTS = 4;
+	FLOATS = 5;
+	STRINGS = 6;
+};
+
+message AttrValue {
+	AttrType type = 1;
+	optional int iv = 2;
+	optional float fv = 3;
+	optional string sv = 4;
+	repeated int ivs = 5;
+	repeated float fvs = 6;
+	repeated string svs = 7;
+};
+
+message AttrProto {
+	required string name = 1;
+	required string comment = 2;
+	required AttrType type = 3;
+};
+
+message VarProto {
+	required string name = 1;
+	required string comment = 2;
+};
+
+message OpProto {
+	repeated VarProto inputs = 1;
+	repeated VarProto outputs = 2;
+	repeated AttrProto attrs = 3;
+	required string type = 4;
+	required string comment = 5;
+};
+```
+
+To generate Python code automatically:
+
+```python 
+def create_python_ops_creatation_functions():
+	op_protos = paddle.framework.OpRegistry.get_all_op_proto()
+	for type_name in op_protos:
+		op_proto = op_protos[type_name]
+		def __impl__(**kwargs):  # User must use key word args in Paddle API
+			inputs = [kwargs.get(ipt.name, "") for ipt in op_proto.inputs]
+			outputs = [kwargs.get(opt.name, "") for opt in op_proto.outputs]
+			attrs = [cast_to_op_attr(attr, kwargs.get(attr.name, None)) for attr in op_proto.attrs]
+			opdesc = （input, outputs, type_name, attrs）
+			return paddle.framework.OpRegistry.CreateOp(opdesc)
+		__impl__.__doc__ = create_doc_string(op_proto)
+		globals()[type_name] = __impl__
+
+create_python_ops_creatation_functions()
+```
+
+### Message from Python to C++
+
+To hold message needed in the above second step, we define Protobuf message class `OpDesc`. It is used to hold user-specified parameters in Op describing.
+
+```proto
+message OpDesc {
+	required string type = 1;	
+	repeated string inputs = 2;
+	repeated string outputs = 3;
+	map<string, AttrValue> attrs = 4;
+};
+```
+
+## OpProto Register
+
+Every Op has its own `OpProto`. For using convenience, we need to register them and record all their messages. For each `Op` class, we define a corresponding `OpMaker` class, in whose constructor we implement the `OpProto`'s building process. `OpMaker`'s constructor will be invoked by another function `OpRegistry::RegisterOp()`.
+
+```cpp
+class OpProtoMaker {
+public:
+	OpProtoMaker(OpProto* proto): proto_(proto) {}
+protected:
+	OpProto* proto_;
+	void AddInput(const std::string& name, const std::string& desc) {...}
+	void AddAttr(const std::string& name, const std::string& desc, TypeId type) {...}
+	void AddComment(const std::string& comment) { ... }
+};
+
+class OpRegistry {
+public:
+	using OpCreator = std::function<OperatorBase* (OpDesc& desc)>;
+	
+	template <typename OpType, typename OpMaker>
+	static void RegisterOp(const std::string& name) {
+		gCreators_[name] = [](const OpDesc& desc) {
+			return new OpType(desc);
+		};
+		OpProto& opProto = gProtos_[name];
+		OpMaker()(&opProto);
+	}
+
+	static map<string, OpCreator> gCreators_;
+	static map<string, OpProto> gProtos_;
+};
+
+template <typename OpType, typename OpMaker>
+class OpRegister {
+  public:
+    OpRegister(std::string type) {
+        OpRegistry::RegisterOp<OpType, OpMaker>(type);
+    }
+};
+
+#define REGISTER_OP(op_class, op_maker_class, type_name)         \
+    class op_class##Register {                                   \
+      private:                                                   \
+        const static OpRegister<#op_class, #op_maker_class> reg; \
+    };                                                           \
+    const Register op_class##Register::reg(#type_name);
+    
+class CosineOp {
+// ...
+}
+
+struct CosineOpProtoMaker : public OpProtoMaker {
+	CosineOpProtoMaker(OpProto* proto) : OpProtoMaker(proto) {
+		AddInput("input", "input of cosine op");
+		AddAttr("scale", "scale of cosine op", float).Default(1.0).LargerThan(0.0);
+		AddType("cos");
+		AddComment("This is cos op");
+	}
+}
+
+REGISTER_OP(CosineOp, CosineOpProtoMaker, cos);
+```
+
+In `REGISTER_OP(CosineOp, CosineOpProtoMaker, cos)`, we register not only `CosineOp` but also `CosineOpProto`. As fields of `CosineOpProto`, the default value and value range of `scale` are also registered here. 
+
+## Python API
+
+Python  APIs are divided into two types, high-level API and low-level API.
+
+### High-Level API
+
+High-level API is called by users directly, so it should keep its style consistent with existing V2 APIs.
+
+Here is a sample about how a define a fc layer:
+
+```python
+hd = fc_layer(input=data, size=56, with_bias=True, activation="sigmoid");
+```
+
+`hd` is the output of `fc_layer` and it's a `variable`. It can be further sent into other layers as input.
+
+The definition of `fc_layer()`:
+
+```python
+def fc_layer(input, size, with_bias, activation):
+	attr_map = {"size":size}
+	check_attrs(attr_map)
+	w = make_variable('w')
+	if with_bias:
+		b = make_variable('b')
+	else:
+		b = None
+	fc_output = make_variable('fc_output');
+	fc_op(input, w, b, fc_output, attr_map)
+	act_output = make_variable('sigmod_output');
+	if activation == "sigmod":
+		sigmod_op(fc_output, act_output);
+	elif:
+		# ...
+	return act_output;
+``` 
+
+### Low Leval API
+
+In above sample, `fc_op` and `sigmod_op` are low-level API. They build `OpDesc` and invoke corresponding C++ code.
+
+*TODO*

develop/doc_cn/design/cluster_train/save_model.html

+
+
+<!DOCTYPE html>
+<!--[if IE 8]><html class="no-js lt-ie9" lang="en" > <![endif]-->
+<!--[if gt IE 8]><!--> <html class="no-js" lang="en" > <!--<![endif]-->
+<head>
+  <meta charset="utf-8">
+  
+  <meta name="viewport" content="width=device-width, initial-scale=1.0">
+  
+  <title>Design Doc: Save Model &mdash; PaddlePaddle  文档</title>
+  
+
+  
+  
+
+  
+
+  
+  
+    
+
+  
+
+  
+  
+    <link rel="stylesheet" href="../../_static/css/theme.css" type="text/css" />
+  
+
+  
+  
+        <link rel="index" title="索引"
+              href="../../genindex.html"/>
+        <link rel="search" title="搜索" href="../../search.html"/>
+    <link rel="top" title="PaddlePaddle  文档" href="../../index.html"/> 
+
+  <link rel="stylesheet" href="https://cdn.jsdelivr.net/perfect-scrollbar/0.6.14/css/perfect-scrollbar.min.css" type="text/css" />
+  <link rel="stylesheet" href="../../_static/css/override.css" type="text/css" />
+  <script>
+  var _hmt = _hmt || [];
+  (function() {
+    var hm = document.createElement("script");
+    hm.src = "//hm.baidu.com/hm.js?b9a314ab40d04d805655aab1deee08ba";
+    var s = document.getElementsByTagName("script")[0]; 
+    s.parentNode.insertBefore(hm, s);
+  })();
+  </script>
+
+  
+
+  
+  <script src="../../_static/js/modernizr.min.js"></script>
+
+</head>
+
+<body class="wy-body-for-nav" role="document">
+
+  
+  <header class="site-header">
+    <div class="site-logo">
+      <a href="/"><img src="../../_static/images/PP_w.png"></a>
+    </div>
+    <div class="site-nav-links">
+      <div class="site-menu">
+        <a class="fork-on-github" href="https://github.com/PaddlePaddle/Paddle" target="_blank"><i class="fa fa-github"></i>Folk me on Github</a>
+        <div class="language-switcher dropdown">
+          <a type="button" data-toggle="dropdown">
+            <span>English</span>
+            <i class="fa fa-angle-up"></i>
+            <i class="fa fa-angle-down"></i>
+          </a>
+          <ul class="dropdown-menu">
+            <li><a href="/doc_cn">中文</a></li>
+            <li><a href="/doc">English</a></li>
+          </ul>
+        </div>
+        <ul class="site-page-links">
+          <li><a href="/">Home</a></li>
+        </ul>
+      </div>
+      <div class="doc-module">
+        
+        <ul>
+<li class="toctree-l1"><a class="reference internal" href="../../getstarted/index_cn.html">新手入门</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../../howto/index_cn.html">进阶指南</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../../api/index_cn.html">API</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../../faq/index_cn.html">FAQ</a></li>
+</ul>
+
+        
+<div role="search">
+  <form id="rtd-search-form" class="wy-form" action="../../search.html" method="get">
+    <input type="text" name="q" placeholder="Search docs" />
+    <input type="hidden" name="check_keywords" value="yes" />
+    <input type="hidden" name="area" value="default" />
+  </form>
+</div>        
+      </div>
+    </div>
+  </header>
+  
+  <div class="main-content-wrap">
+
+    
+    <nav class="doc-menu-vertical" role="navigation">
+        
+          
+          <ul>
+<li class="toctree-l1"><a class="reference internal" href="../../getstarted/index_cn.html">新手入门</a><ul>
+<li class="toctree-l2"><a class="reference internal" href="../../getstarted/build_and_install/index_cn.html">安装与编译</a><ul>
+<li class="toctree-l3"><a class="reference internal" href="../../getstarted/build_and_install/docker_install_cn.html">PaddlePaddle的Docker容器使用方式</a></li>
+<li class="toctree-l3"><a class="reference internal" href="../../getstarted/build_and_install/ubuntu_install_cn.html">Ubuntu部署PaddlePaddle</a></li>
+<li class="toctree-l3"><a class="reference internal" href="../../getstarted/build_and_install/cmake/build_from_source_cn.html">PaddlePaddle的编译选项</a></li>
+</ul>
+</li>
+<li class="toctree-l2"><a class="reference internal" href="../../getstarted/concepts/use_concepts_cn.html">基本使用概念</a></li>
+</ul>
+</li>
+<li class="toctree-l1"><a class="reference internal" href="../../howto/index_cn.html">进阶指南</a><ul>
+<li class="toctree-l2"><a class="reference internal" href="../../howto/usage/cmd_parameter/index_cn.html">设置命令行参数</a><ul>
+<li class="toctree-l3"><a class="reference internal" href="../../howto/usage/cmd_parameter/use_case_cn.html">使用案例</a></li>
+<li class="toctree-l3"><a class="reference internal" href="../../howto/usage/cmd_parameter/arguments_cn.html">参数概述</a></li>
+<li class="toctree-l3"><a class="reference internal" href="../../howto/usage/cmd_parameter/detail_introduction_cn.html">细节描述</a></li>
+</ul>
+</li>
+<li class="toctree-l2"><a class="reference internal" href="../../howto/usage/cluster/cluster_train_cn.html">运行分布式训练</a></li>
+<li class="toctree-l2"><a class="reference internal" href="../../howto/usage/k8s/k8s_basis_cn.html">Kubernetes 简介</a></li>
+<li class="toctree-l2"><a class="reference internal" href="../../howto/usage/k8s/k8s_cn.html">Kubernetes单机训练</a></li>
+<li class="toctree-l2"><a class="reference internal" href="../../howto/usage/k8s/k8s_distributed_cn.html">Kubernetes分布式训练</a></li>
+<li class="toctree-l2"><a class="reference internal" href="../../howto/dev/write_docs_cn.html">如何贡献/修改文档</a></li>
+<li class="toctree-l2"><a class="reference internal" href="../../howto/dev/contribute_to_paddle_cn.html">如何贡献代码</a></li>
+<li class="toctree-l2"><a class="reference internal" href="../../howto/deep_model/rnn/index_cn.html">RNN相关模型</a><ul>
+<li class="toctree-l3"><a class="reference internal" href="../../howto/deep_model/rnn/rnn_config_cn.html">RNN配置</a></li>
+<li class="toctree-l3"><a class="reference internal" href="../../howto/deep_model/rnn/recurrent_group_cn.html">Recurrent Group教程</a></li>
+<li class="toctree-l3"><a class="reference internal" href="../../howto/deep_model/rnn/hierarchical_layer_cn.html">支持双层序列作为输入的Layer</a></li>
+<li class="toctree-l3"><a class="reference internal" href="../../howto/deep_model/rnn/hrnn_rnn_api_compare_cn.html">单双层RNN API对比介绍</a></li>
+</ul>
+</li>
+<li class="toctree-l2"><a class="reference internal" href="../../howto/optimization/gpu_profiling_cn.html">GPU性能分析与调优</a></li>
+</ul>
+</li>
+<li class="toctree-l1"><a class="reference internal" href="../../api/index_cn.html">API</a><ul>
+<li class="toctree-l2"><a class="reference internal" href="../../api/v2/model_configs.html">模型配置</a><ul>
+<li class="toctree-l3"><a class="reference internal" href="../../api/v2/config/activation.html">Activation</a></li>
+<li class="toctree-l3"><a class="reference internal" href="../../api/v2/config/layer.html">Layers</a></li>
+<li class="toctree-l3"><a class="reference internal" href="../../api/v2/config/evaluators.html">Evaluators</a></li>
+<li class="toctree-l3"><a class="reference internal" href="../../api/v2/config/optimizer.html">Optimizer</a></li>
+<li class="toctree-l3"><a class="reference internal" href="../../api/v2/config/pooling.html">Pooling</a></li>
+<li class="toctree-l3"><a class="reference internal" href="../../api/v2/config/networks.html">Networks</a></li>
+<li class="toctree-l3"><a class="reference internal" href="../../api/v2/config/attr.html">Parameter Attribute</a></li>
+</ul>
+</li>
+<li class="toctree-l2"><a class="reference internal" href="../../api/v2/data.html">数据访问</a></li>
+<li class="toctree-l2"><a class="reference internal" href="../../api/v2/run_logic.html">训练与应用</a></li>
+</ul>
+</li>
+<li class="toctree-l1"><a class="reference internal" href="../../faq/index_cn.html">FAQ</a></li>
+</ul>
+
+        
+    </nav>
+    
+    <section class="doc-content-wrap">
+
+      
+
+ 
+
+
+
+
+
+
+
+<div role="navigation" aria-label="breadcrumbs navigation">
+  <ul class="wy-breadcrumbs">
+      
+    <li>Design Doc: Save Model</li>
+  </ul>
+</div>
+      
+      <div class="wy-nav-content" id="doc-content">
+        <div class="rst-content">
+          <div role="main" class="document" itemscope="itemscope" itemtype="http://schema.org/Article">
+           <div itemprop="articleBody">
+            
+  <div class="section" id="design-doc-save-model">
+<span id="design-doc-save-model"></span><h1>Design Doc: Save Model<a class="headerlink" href="#design-doc-save-model" title="永久链接至标题">¶</a></h1>
+<div class="section" id="overview">
+<span id="overview"></span><h2>Overview<a class="headerlink" href="#overview" title="永久链接至标题">¶</a></h2>
+<p>The model is the output of the training process. There are two
+ways from which user can obtain a model:</p>
+<ul class="simple">
+<li>Save model triggered by user code: user code asks PaddlePaddle to
+save a model.</li>
+<li>Convert model from the checkpoint: model being converted from
+pservers&#8217; periodic checkpoint. In this way, the user can cancel a
+job at any time, and still have a relatively fresh model (we
+checkpoint around every 5 minutes).</li>
+</ul>
+<div class="section" id="trainer-saving-model-vs-pservers-saving-model">
+<span id="trainer-saving-model-vs-pservers-saving-model"></span><h3>Trainer Saving Model vs. Pservers Saving Model<a class="headerlink" href="#trainer-saving-model-vs-pservers-saving-model" title="永久链接至标题">¶</a></h3>
+<p>Both trainers and pservers have access to the model. So the model can
+be saved from a trainer or pservers. We need to decide where the model
+is saved from.</p>
+<div class="section" id="dense-update-vs-sparse-update">
+<span id="dense-update-vs-sparse-update"></span><h4>Dense Update vs. Sparse Update<a class="headerlink" href="#dense-update-vs-sparse-update" title="永久链接至标题">¶</a></h4>
+<p>There are two types of model update methods: dense update and sparse
+update (when the model parameter is configured to be sparse).</p>
+<ul>
+<li><p class="first">Dense update</p>
+<p>Every trainer has it&#8217;s own full copy of the model. Every model
+update will update the entire model.</p>
+</li>
+<li><p class="first">Sparse update</p>
+<p>The training input is sparse, and the trainer does not have the
+entire model. It will only download the sub-model necessary related
+to the input. When updating the model, only the sub-model related to
+the training input is updated.</p>
+</li>
+</ul>
+</div>
+<div class="section" id="pservers-saving-model">
+<span id="pservers-saving-model"></span><h4>Pservers Saving Model<a class="headerlink" href="#pservers-saving-model" title="永久链接至标题">¶</a></h4>
+<p>The benefit of letting pservers save model is they have the entire
+model all the time. However, since pservers are on different nodes, it
+requires a merging process to merge model shards into the same
+model. Thus requires the pservers to write models to a distributed
+filesystem, making the checkpoint shards visible to the merge program.</p>
+</div>
+<div class="section" id="trainer-saving-model">
+<span id="trainer-saving-model"></span><h4>Trainer Saving Model<a class="headerlink" href="#trainer-saving-model" title="永久链接至标题">¶</a></h4>
+<p>The benefit of letting one trainer to save the model is it does not
+require a distributed filesystem. And it&#8217;s reusing the same save model
+logic when training locally - except when doing sparse update, the
+trainer needs to download the entire model during the saving process.</p>
+</div>
+<div class="section" id="conclusion">
+<span id="conclusion"></span><h4>Conclusion<a class="headerlink" href="#conclusion" title="永久链接至标题">¶</a></h4>
+<p>Given trainer saving model does not require a distributed filesystem,
+and is an intuitive extension to trainer saving model when training
+locally, we decide to let the trainer save the model when doing
+distributed training.</p>
+</div>
+</div>
+<div class="section" id="convert-model-from-checkpoint">
+<span id="convert-model-from-checkpoint"></span><h3>Convert Model from Checkpoint<a class="headerlink" href="#convert-model-from-checkpoint" title="永久链接至标题">¶</a></h3>
+<p>TODO</p>
+</div>
+</div>
+<div class="section" id="timeline">
+<span id="timeline"></span><h2>Timeline<a class="headerlink" href="#timeline" title="永久链接至标题">¶</a></h2>
+<p>We first implement trainer save the model. Converting the latest
+snapshot to a model will be a TODO for future.</p>
+</div>
+<div class="section" id="trainer-save-model">
+<span id="trainer-save-model"></span><h2>Trainer Save Model<a class="headerlink" href="#trainer-save-model" title="永久链接至标题">¶</a></h2>
+<div class="section" id="trainer-election">
+<span id="trainer-election"></span><h3>Trainer Election<a class="headerlink" href="#trainer-election" title="永久链接至标题">¶</a></h3>
+<p>One trainer will be elected as the one to save the model. When using
+etcd, trainer ID is a randomly generated UUID, we will utilize etcd to
+elect one trainer. When not using etcd, unique trainer IDs will be
+given by the administrator, the trainer whose ID is &#8220;0&#8221; is elected to
+save the model.</p>
+</div>
+<div class="section" id="model-save-path">
+<span id="model-save-path"></span><h3>Model Save Path<a class="headerlink" href="#model-save-path" title="永久链接至标题">¶</a></h3>
+<p>Each trainer will be given the directory to save the model. The
+elected trainer will save the model to
+<code class="docutils literal"><span class="pre">given-directory/trainerID</span></code>. Since the trainer ID is unique, this
+would prevent concurrent save to the same file when multiple trainers
+are elected to save the model when split-brain problem happens.</p>
+</div>
+<div class="section" id="what-happens-when-model-is-saving">
+<span id="what-happens-when-model-is-saving"></span><h3>What Happens When Model Is Saving<a class="headerlink" href="#what-happens-when-model-is-saving" title="永久链接至标题">¶</a></h3>
+<p>It takes some time to save model, we need to define what will happen
+when save model is taking place.</p>
+<p>When doing dense update, the trainer uses the local model. Pservers
+does not need to pause model update.</p>
+<p>When doing sparse update. The trainer needs to download the entire
+model while saving. To get the most accurate model, the model update
+needs to be paused before the download starts and resumed after the
+download finishes. Otherwise, the trainer gets a model that is
+&#8220;polluted&#8221;: some part of the model is old, some part of the model is
+new.</p>
+<p>It&#8217;s unclear that the &#8220;polluted&#8221; model will be inferior due to the
+stochastic nature of deep learning, and pausing the model update will
+add more complexity to the system. Since supporting sparse update is a
+TODO item. We defer the evaluation of pause the model update or not
+during saving model to the future.</p>
+</div>
+</div>
+</div>
+
+
+           </div>
+          </div>
+          <footer>
+  
+
+  <hr/>
+
+  <div role="contentinfo">
+    <p>
+        &copy; Copyright 2016, PaddlePaddle developers.
+
+    </p>
+  </div>
+  Built with <a href="http://sphinx-doc.org/">Sphinx</a> using a <a href="https://github.com/snide/sphinx_rtd_theme">theme</a> provided by <a href="https://readthedocs.org">Read the Docs</a>. 
+
+</footer>
+
+        </div>
+      </div>
+
+    </section>
+
+  </div>
+  
+
+
+  
+
+    <script type="text/javascript">
+        var DOCUMENTATION_OPTIONS = {
+            URL_ROOT:'../../',
+            VERSION:'',
+            COLLAPSE_INDEX:false,
+            FILE_SUFFIX:'.html',
+            HAS_SOURCE:  true,
+            SOURCELINK_SUFFIX: ".txt",
+        };
+    </script>
+      <script type="text/javascript" src="../../_static/jquery.js"></script>
+      <script type="text/javascript" src="../../_static/underscore.js"></script>
+      <script type="text/javascript" src="../../_static/doctools.js"></script>
+      <script type="text/javascript" src="../../_static/translations.js"></script>
+      <script type="text/javascript" src="https://cdn.bootcss.com/mathjax/2.7.0/MathJax.js"></script>
+       
+  
+
+  
+  
+    <script type="text/javascript" src="../../_static/js/theme.js"></script>
+  
+  
+  <script src="https://maxcdn.bootstrapcdn.com/bootstrap/3.3.7/js/bootstrap.min.js" integrity="sha384-Tc5IQib027qvyjSMfHjOMaLkfuWVxZxUPnCJA7l2mCWNIpG9mGCD8wGNIcPD7Txa" crossorigin="anonymous"></script>
+  <script src="https://cdn.jsdelivr.net/perfect-scrollbar/0.6.14/js/perfect-scrollbar.jquery.min.js"></script>
+  <script src="../../_static/js/paddle_doc_init.js"></script> 
+
+</body>
+</html>
\ No newline at end of file

develop/doc_cn/getstarted/concepts/use_concepts_cn.html

@@ -413,7 +413,7 @@ trainer.train<span class="o">(</span>
 <span class="c1"># define training dataset reader</span>
 <span class="k">def</span> <span class="nf">train_reader</span><span class="p">():</span>
     <span class="n">train_x</span> <span class="o">=</span> <span class="n">np</span><span class="o">.</span><span class="n">array</span><span class="p">([[</span><span class="mi">1</span><span class="p">,</span> <span class="mi">1</span><span class="p">],</span> <span class="p">[</span><span class="mi">1</span><span class="p">,</span> <span class="mi">2</span><span class="p">],</span> <span class="p">[</span><span class="mi">3</span><span class="p">,</span> <span class="mi">4</span><span class="p">],</span> <span class="p">[</span><span class="mi">5</span><span class="p">,</span> <span class="mi">2</span><span class="p">]])</span>
-    <span class="n">train_y</span> <span class="o">=</span> <span class="n">np</span><span class="o">.</span><span class="n">array</span><span class="p">([</span><span class="o">-</span><span class="mi">2</span><span class="p">,</span> <span class="o">-</span><span class="mi">3</span><span class="p">,</span> <span class="o">-</span><span class="mi">7</span><span class="p">,</span> <span class="o">-</span><span class="mi">7</span><span class="p">])</span>
+    <span class="n">train_y</span> <span class="o">=</span> <span class="n">np</span><span class="o">.</span><span class="n">array</span><span class="p">([[</span><span class="o">-</span><span class="mi">2</span><span class="p">],</span> <span class="p">[</span><span class="o">-</span><span class="mi">3</span><span class="p">],</span> <span class="p">[</span><span class="o">-</span><span class="mi">7</span><span class="p">],</span> <span class="p">[</span><span class="o">-</span><span class="mi">7</span><span class="p">]])</span>
 
     <span class="k">def</span> <span class="nf">reader</span><span class="p">():</span>
         <span class="k">for</span> <span class="n">i</span> <span class="ow">in</span> <span class="n">xrange</span><span class="p">(</span><span class="n">train_y</span><span class="o">.</span><span class="n">shape</span><span class="p">[</span><span class="mi">0</span><span class="p">]):</span>

index.cn.html

@@ -45,7 +45,7 @@
                 <li><a href="http://book.paddlepaddle.org/index.cn.html" target="_blank">快速开始</a></li>
                 <li><a href="http://www.paddlepaddle.org/doc_cn/howto/index_cn.html" target="_blank">文档中心</a></li>
                 <li class="version-switcher">
-                    <a>版本<i class="fa" aria-hidden="true"></i></a>
+                    <a>版本<i class="fa" aria-hidden="true"></i>&nbsp;&nbsp;&nbsp;&nbsp;</a>
                     <ul>
                         <li><a href="http://www.paddlepaddle.org/release/0.10.0/doc_cn/" target="_blank">0.10.0</a></li>
                         <li><a href="http://www.paddlepaddle.org/release_doc/0.9.0/doc_cn/" target="_blank">0.9.0</a></li>
@@ -88,7 +88,7 @@
         <div>
             <div class="service-desc">
                 <h3>机器视觉</h3>
-                <p>卷积神经网络可以识别图像中的主要对象,并输出分类结果</p>
+                <p>卷积神经网络可以识别图像中的主要对象，并输出分类结果</p>
                 <div>
                     <a role="button" class="view-more" href="http://book.paddlepaddle.org/03.image_classification/index.cn.html" target="_blank">查看更多 ></a>
                 </div>
@@ -116,7 +116,7 @@
         <div>
             <div class="service-desc">
                 <h3>推荐系统</h3>
-                <p>分析用户特征、电影特征、点评分数,预测新用户对不同电影的点评分数</p>
+                <p>分析用户特征、电影特征、点评分数，预测新用户对不同电影的打分</p>
                 <div>
                     <a role="button" class="view-more" href="http://book.paddlepaddle.org/05.recommender_system/index.cn.html" target="_blank">查看更多 ></a>
                 </div>
@@ -155,7 +155,7 @@
                 <img src="./images/feature-4.png">
             </div>
             <h3>扩展性</h3>
-            <p>全面支持多核、多GPU、多机环境。轻松应对大规模数据训练需求</p>
+            <p>全面支持多核、多GPU、多机环境。轻松应对大规模训练需求</p>
         </div>
     </div>
 </section>

index.html

 <!DOCTYPE html>
-<html lang="en">
+<html lang="zh-cn">
 <head>
     <meta charset="UTF-8">
     <title>PaddlePaddle</title>
@@ -43,7 +43,7 @@
             <ul class="site-links">
                 <li><a class="active">Home</a></li>
                 <li><a href="http://book.paddlepaddle.org/index.html" target="_blank">Quick Start</a></li>
-                <li><a href="http://www.paddlepaddle.org/doc/howto/index_en.html" target="_blank">Documents</a></li>
+                <li><a href="http://www.paddlepaddle.org/doc/howto/index_en.html" target="_blank">Documentation</a></li>
                 <li class="version-switcher">
                     <a>Version<i class="fa" aria-hidden="true"></i></a>
                     <ul>
@@ -79,7 +79,7 @@
 <section class="services">
     <div class="row">
         <h2><span>Diverse Application Scenarios</span></h2>
-        <p class="sub-title">Promoting 100+ new business products on the line</p>
+        <p class="sub-title">Powers hundreds of baidu and partner products</p>
     </div>
     <div class="row">
         <div>
@@ -88,7 +88,7 @@
         <div>
             <div class="service-desc">
                 <h3>Computer Vision</h3>
-                <p>Using Convolutional Neural Networks (CNN) for Image Recognition and Object Detection.</p>
+                <p>Using a Convolutional Neural Network (CNN) for Image Recognition and Object Detection.</p>
                 <div>
                     <a role="button" class="view-more" href="http://book.paddlepaddle.org/03.image_classification/index.html" target="_blank">Read more ></a>
                 </div>
@@ -99,7 +99,7 @@
         <div>
             <div class="service-desc">
                 <h3>Natural Language Understanding</h3>
-                <p>Using Recurrent Neural Network (RNN) for Sentiment Analysis.</p>
+                <p>Using a Recurrent Neural Network (RNN) for Sentiment Analysis.</p>
                 <div>
                     <a role="button" class="view-more" href="http://book.paddlepaddle.org/06.understand_sentiment/index.html" target="_blank">Read more ></a>
                 </div>