Google Git
Sign in
apache / singa-site / 253615c20ca64588c0eee4b9048ffc8e2c96390f / . / content / docs / next / software-stack.html
blob: 7d283ed6d18f26e20fa2ecda8ac6c02278828d7d [file] [log] [blame]
<!DOCTYPE html><html lang="en"><head><meta charSet="utf-8"/><meta http-equiv="X-UA-Compatible" content="IE=edge"/><title>Software Stack · Apache SINGA</title><meta name="viewport" content="width=device-width"/><meta name="generator" content="Docusaurus"/><meta name="description" content="&lt;!--- Licensed to the Apache Software Foundation (ASF) under one or more contributor license agreements. See the NOTICE file distributed with this work for additional information regarding copyright ownership. The ASF licenses this file to you under the Apache License, Version 2.0 (the &quot;License&quot;); you may not use this file except in compliance with the License. You may obtain a copy of the License at http://www.apache.org/licenses/LICENSE-2.0 Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an &quot;AS IS&quot; BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License. --&gt;"/><meta name="docsearch:version" content="next"/><meta name="docsearch:language" content="en"/><meta property="og:title" content="Software Stack · Apache SINGA"/><meta property="og:type" content="website"/><meta property="og:url" content="https://feynmandna.github.io/"/><meta property="og:description" content="&lt;!--- Licensed to the Apache Software Foundation (ASF) under one or more contributor license agreements. See the NOTICE file distributed with this work for additional information regarding copyright ownership. The ASF licenses this file to you under the Apache License, Version 2.0 (the &quot;License&quot;); you may not use this file except in compliance with the License. You may obtain a copy of the License at http://www.apache.org/licenses/LICENSE-2.0 Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an &quot;AS IS&quot; BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License. --&gt;"/><meta property="og:image" content="https://feynmandna.github.io/img/singa_twitter_banner.jpeg"/><meta name="twitter:card" content="summary"/><meta name="twitter:image" content="https://feynmandna.github.io/img/singa_twitter_banner.jpeg"/><link rel="shortcut icon" href="/img/favicon.ico"/><link rel="stylesheet" href="//cdnjs.cloudflare.com/ajax/libs/highlight.js/9.12.0/styles/atom-one-dark.min.css"/><link rel="alternate" type="application/atom+xml" href="https://feynmandna.github.io/blog/atom.xml" title="Apache SINGA Blog ATOM Feed"/><link rel="alternate" type="application/rss+xml" href="https://feynmandna.github.io/blog/feed.xml" title="Apache SINGA Blog RSS Feed"/><script type="text/javascript" src="https://buttons.github.io/buttons.js"></script><script src="https://unpkg.com/vanilla-back-to-top@7.1.14/dist/vanilla-back-to-top.min.js"></script><script>
document.addEventListener('DOMContentLoaded', function() {
addBackToTop(
{"zIndex":100}
)
});
</script><script src="/js/scrollSpy.js"></script><link rel="stylesheet" href="/css/main.css"/><script src="/js/codetabs.js"></script></head><body class="sideNavVisible separateOnPageNav"><div class="fixedHeaderContainer"><div class="headerWrapper wrapper"><header><a href="/"><img class="logo" src="/img/singa.png" alt="Apache SINGA"/></a><a href="/versions"><h3>next</h3></a><div class="navigationWrapper navigationSlider"><nav class="slidingNav"><ul class="nav-site nav-site-internal"><li class="siteNavGroupActive"><a href="/docs/next/installation" target="_self">Docs</a></li><li class=""><a href="/docs/next/source-repository" target="_self">Community</a></li><li class=""><a href="/blog/" target="_self">News</a></li><li class=""><a href="https://apache-singa.readthedocs.io/en/latest/" target="_self">API</a></li><li class=""><a target="_self"></a></li><li class=""><a href="https://github.com/apache/singa-doc" target="_self">GitHub</a></li></ul></nav></div></header></div></div><div class="navPusher"><div class="docMainWrapper wrapper"><div class="docsNavContainer" id="docsNav"><nav class="toc"><div class="toggleNav"><section class="navWrapper wrapper"><div class="navBreadcrumb wrapper"><div class="navToggle" id="navToggler"><div class="hamburger-menu"><div class="line1"></div><div class="line2"></div><div class="line3"></div></div></div><h2><i>›</i><span>Getting Started</span></h2><div class="tocToggler" id="tocToggler"><i class="icon-toc"></i></div></div><div class="navGroups"><div class="navGroup"><h3 class="navGroupCategoryTitle">Getting Started</h3><ul class=""><li class="navListItem"><a class="navItem" href="/docs/next/installation">Installation</a></li><li class="navListItem navListItemActive"><a class="navItem" href="/docs/next/software-stack">Software Stack</a></li><li class="navListItem"><a class="navItem" href="/docs/next/examples">Examples</a></li></ul></div><div class="navGroup"><h3 class="navGroupCategoryTitle">Guides</h3><ul class=""><li class="navListItem"><a class="navItem" href="/docs/next/device">Device</a></li><li class="navListItem"><a class="navItem" href="/docs/next/tensor">Tensor</a></li><li class="navListItem"><a class="navItem" href="/docs/next/autograd">Autograd</a></li><li class="navListItem"><a class="navItem" href="/docs/next/graph">Computational Graph</a></li><li class="navListItem"><a class="navItem" href="/docs/next/dist-train">Distributed Training</a></li></ul></div><div class="navGroup"><h3 class="navGroupCategoryTitle">Development</h3><ul class=""><li class="navListItem"><a class="navItem" href="/docs/next/download-singa">Download SINGA</a></li><li class="navListItem"><a class="navItem" href="/docs/next/build">Build SINGA from Source</a></li><li class="navListItem"><a class="navItem" href="/docs/next/contribute-code">How to Contribute Code</a></li><li class="navListItem"><a class="navItem" href="/docs/next/contribute-docs">How to Contribute to Documentation</a></li><li class="navListItem"><a class="navItem" href="/docs/next/how-to-release">How to Prepare a Release</a></li><li class="navListItem"><a class="navItem" href="/docs/next/git-workflow">Git Workflow</a></li></ul></div></div></section></div><script>
var coll = document.getElementsByClassName('collapsible');
var checkActiveCategory = true;
for (var i = 0; i < coll.length; i++) {
var links = coll[i].nextElementSibling.getElementsByTagName('*');
if (checkActiveCategory){
for (var j = 0; j < links.length; j++) {
if (links[j].classList.contains('navListItemActive')){
coll[i].nextElementSibling.classList.toggle('hide');
coll[i].childNodes[1].classList.toggle('rotate');
checkActiveCategory = false;
break;
}
}
}
coll[i].addEventListener('click', function() {
var arrow = this.childNodes[1];
arrow.classList.toggle('rotate');
var content = this.nextElementSibling;
content.classList.toggle('hide');
});
}
document.addEventListener('DOMContentLoaded', function() {
createToggler('#navToggler', '#docsNav', 'docsSliderActive');
createToggler('#tocToggler', 'body', 'tocActive');
var headings = document.querySelector('.toc-headings');
headings && headings.addEventListener('click', function(event) {
var el = event.target;
while(el !== headings){
if (el.tagName === 'A') {
document.body.classList.remove('tocActive');
break;
} else{
el = el.parentNode;
}
}
}, false);
function createToggler(togglerSelector, targetSelector, className) {
var toggler = document.querySelector(togglerSelector);
var target = document.querySelector(targetSelector);
if (!toggler) {
return;
}
toggler.onclick = function(event) {
event.preventDefault();
target.classList.toggle(className);
};
}
});
</script></nav></div><div class="container mainContainer docsContainer"><div class="wrapper"><div class="post"><header class="postHeader"><a class="edit-page-link button" href="https://github.com/apache/singa-doc/blob/master/docs/software-stack.md" target="_blank" rel="noreferrer noopener">Edit</a><h1 id="__docusaurus" class="postHeaderTitle">Software Stack</h1></header><article><div><span><!--- Licensed to the Apache Software Foundation (ASF) under one or more contributor license agreements. See the NOTICE file distributed with this work for additional information regarding copyright ownership. The ASF licenses this file to you under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at http://www.apache.org/licenses/LICENSE-2.0 Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License. -->
<p>SINGA's software stack includes two major levels, the low level backend classes
and the Python interface level. Figure 1 illustrates them together with the
hardware. The backend components provides basic data structures for deep
learning models, hardware abstractions for scheduling and executing operations,
and communication components for distributed training. The Python interface
wraps some CPP data structures and provides additional high-level classes for
neural network training, which makes it convenient to implement complex neural
network models. Next, we introduce the software stack in a bottom-up manner.</p>
<p><img src="/docs/assets/singav3-sw.png" alt="SINGA V3 software stack"> <br/> <strong>Figure 1 - SINGA V3
software stack.</strong></p>
<h2><a class="anchor" aria-hidden="true" id="low-level-backend"></a><a href="#low-level-backend" aria-hidden="true" class="hash-link"><svg class="hash-link-icon" aria-hidden="true" height="16" version="1.1" viewBox="0 0 16 16" width="16"><path fill-rule="evenodd" d="M4 9h1v1H4c-1.5 0-3-1.69-3-3.5S2.55 3 4 3h4c1.45 0 3 1.69 3 3.5 0 1.41-.91 2.72-2 3.25V8.59c.58-.45 1-1.27 1-2.09C10 5.22 8.98 4 8 4H4c-.98 0-2 1.22-2 2.5S3 9 4 9zm9-3h-1v1h1c1 0 2 1.22 2 2.5S13.98 12 13 12H9c-.98 0-2-1.22-2-2.5 0-.83.42-1.64 1-2.09V6.25c-1.09.53-2 1.84-2 3.25C6 11.31 7.55 13 9 13h4c1.45 0 3-1.69 3-3.5S14.5 6 13 6z"></path></svg></a>Low-level Backend</h2>
<h3><a class="anchor" aria-hidden="true" id="device"></a><a href="#device" aria-hidden="true" class="hash-link"><svg class="hash-link-icon" aria-hidden="true" height="16" version="1.1" viewBox="0 0 16 16" width="16"><path fill-rule="evenodd" d="M4 9h1v1H4c-1.5 0-3-1.69-3-3.5S2.55 3 4 3h4c1.45 0 3 1.69 3 3.5 0 1.41-.91 2.72-2 3.25V8.59c.58-.45 1-1.27 1-2.09C10 5.22 8.98 4 8 4H4c-.98 0-2 1.22-2 2.5S3 9 4 9zm9-3h-1v1h1c1 0 2 1.22 2 2.5S13.98 12 13 12H9c-.98 0-2-1.22-2-2.5 0-.83.42-1.64 1-2.09V6.25c-1.09.53-2 1.84-2 3.25C6 11.31 7.55 13 9 13h4c1.45 0 3-1.69 3-3.5S14.5 6 13 6z"></path></svg></a>Device</h3>
<p>Each <code>Device</code> instance, i.e., a device, is created against one hardware device,
e.g. a GPU or a CPU. <code>Device</code> manages the memory of the data structures, and
schedules the operations for executing, e.g., on CUDA streams or CPU threads.
Depending on the hardware and its programming language, SINGA have implemented
the following specific device classes:</p>
<ul>
<li><strong>CudaGPU</strong> represents an Nvidia GPU card. The execution units are the CUDA
streams.</li>
<li><strong>CppCPU</strong> represents a normal CPU. The execution units are the CPU threads.</li>
<li><strong>OpenclGPU</strong> represents normal GPU card from both Nvidia and AMD. The
execution units are the CommandQueues. Given that OpenCL is compatible with
many hardware devices, e.g. FPGA and ARM, the OpenclGPU has the potential to
be extended for other devices.</li>
</ul>
<h3><a class="anchor" aria-hidden="true" id="tensor"></a><a href="#tensor" aria-hidden="true" class="hash-link"><svg class="hash-link-icon" aria-hidden="true" height="16" version="1.1" viewBox="0 0 16 16" width="16"><path fill-rule="evenodd" d="M4 9h1v1H4c-1.5 0-3-1.69-3-3.5S2.55 3 4 3h4c1.45 0 3 1.69 3 3.5 0 1.41-.91 2.72-2 3.25V8.59c.58-.45 1-1.27 1-2.09C10 5.22 8.98 4 8 4H4c-.98 0-2 1.22-2 2.5S3 9 4 9zm9-3h-1v1h1c1 0 2 1.22 2 2.5S13.98 12 13 12H9c-.98 0-2-1.22-2-2.5 0-.83.42-1.64 1-2.09V6.25c-1.09.53-2 1.84-2 3.25C6 11.31 7.55 13 9 13h4c1.45 0 3-1.69 3-3.5S14.5 6 13 6z"></path></svg></a>Tensor</h3>
<p><code>Tensor</code> class represents a multi-dimensional array, which stores model
variables, e.g., the input images and feature maps from the convolution layer.
Each <code>Tensor</code> instance (i.e. a tensor) is allocated on a a device, which manages
the memory of the tensor and schedules the (computation) operations against
tensors. Most machine learning algorithms could be expressed using (dense or
sparse) the tensor abstraction and its operations. Therefore, SINGA would be
able to run a wide range of models, including deep learning models and other
traditional machine learning models.</p>
<h3><a class="anchor" aria-hidden="true" id="operator"></a><a href="#operator" aria-hidden="true" class="hash-link"><svg class="hash-link-icon" aria-hidden="true" height="16" version="1.1" viewBox="0 0 16 16" width="16"><path fill-rule="evenodd" d="M4 9h1v1H4c-1.5 0-3-1.69-3-3.5S2.55 3 4 3h4c1.45 0 3 1.69 3 3.5 0 1.41-.91 2.72-2 3.25V8.59c.58-.45 1-1.27 1-2.09C10 5.22 8.98 4 8 4H4c-.98 0-2 1.22-2 2.5S3 9 4 9zm9-3h-1v1h1c1 0 2 1.22 2 2.5S13.98 12 13 12H9c-.98 0-2-1.22-2-2.5 0-.83.42-1.64 1-2.09V6.25c-1.09.53-2 1.84-2 3.25C6 11.31 7.55 13 9 13h4c1.45 0 3-1.69 3-3.5S14.5 6 13 6z"></path></svg></a>Operator</h3>
<p>There are two types of operators against tensors, linear algebra operators like
matrix multiplication, and neural network specific operators like convolution
and pooling. The linear algebra operators are provided as <code>Tensor</code> functions and
are implemented separately for different hardware devices</p>
<ul>
<li>CppMath (tensor_math_cpp.h) implements the tensor operations using Cpp for
CppCPU</li>
<li>CudaMath (tensor_math_cuda.h) implements the tensor operations using CUDA for
CudaGPU</li>
<li>OpenclMath (tensor_math_opencl.h) implements the tensor operations using
OpenCL for OpenclGPU</li>
</ul>
<p>The neural network specific operators are also implemented separately, e.g.,</p>
<ul>
<li>GpuConvFoward (convolution.h) implements the forward function of convolution
via CuDNN on Nvidia GPU.</li>
<li>CpuConvForward (convolution.h) implements the forward function of convolution
using CPP on CPU.</li>
</ul>
<p>Typically, users create a <code>Device</code> instance and use it to create multiple
<code>Tensor</code> instances. When users call the Tensor functions or neural network
operations, the corresponding implementation for the resident device will be
invoked In other words, the implementation of operators is transparent to users.</p>
<p>The Tensor and Device abstractions are extensible to support a wide range of
hardware device using different programming languages. A new hardware device
would be supported by adding a new Device subclass and the corresponding
implementation of the operators.</p>
<p>Optimizations in terms of speed and memory are done by the <code>Scheduler</code> and
<code>MemPool</code> of the <code>Device</code>. For example, the <code>Scheduler</code> creates a
<a href="./graph">computational graph</a> according to the dependency of the operators.
Then it can optimize the execution order of the operators for parallelism and
memory sharing.</p>
<h3><a class="anchor" aria-hidden="true" id="communicator"></a><a href="#communicator" aria-hidden="true" class="hash-link"><svg class="hash-link-icon" aria-hidden="true" height="16" version="1.1" viewBox="0 0 16 16" width="16"><path fill-rule="evenodd" d="M4 9h1v1H4c-1.5 0-3-1.69-3-3.5S2.55 3 4 3h4c1.45 0 3 1.69 3 3.5 0 1.41-.91 2.72-2 3.25V8.59c.58-.45 1-1.27 1-2.09C10 5.22 8.98 4 8 4H4c-.98 0-2 1.22-2 2.5S3 9 4 9zm9-3h-1v1h1c1 0 2 1.22 2 2.5S13.98 12 13 12H9c-.98 0-2-1.22-2-2.5 0-.83.42-1.64 1-2.09V6.25c-1.09.53-2 1.84-2 3.25C6 11.31 7.55 13 9 13h4c1.45 0 3-1.69 3-3.5S14.5 6 13 6z"></path></svg></a>Communicator</h3>
<p><code>Communicator</code> is to support <a href="./dist-train">distributed training</a>. It implements
the communication protocols using sockets, MPI and NCCL. Typically users only
need to call the high-level APIs like <code>put()</code> and <code>get()</code> for sending and
receiving tensors. Communication optimization for the topology, message size,
etc. is done internally.</p>
<h2><a class="anchor" aria-hidden="true" id="python-interface"></a><a href="#python-interface" aria-hidden="true" class="hash-link"><svg class="hash-link-icon" aria-hidden="true" height="16" version="1.1" viewBox="0 0 16 16" width="16"><path fill-rule="evenodd" d="M4 9h1v1H4c-1.5 0-3-1.69-3-3.5S2.55 3 4 3h4c1.45 0 3 1.69 3 3.5 0 1.41-.91 2.72-2 3.25V8.59c.58-.45 1-1.27 1-2.09C10 5.22 8.98 4 8 4H4c-.98 0-2 1.22-2 2.5S3 9 4 9zm9-3h-1v1h1c1 0 2 1.22 2 2.5S13.98 12 13 12H9c-.98 0-2-1.22-2-2.5 0-.83.42-1.64 1-2.09V6.25c-1.09.53-2 1.84-2 3.25C6 11.31 7.55 13 9 13h4c1.45 0 3-1.69 3-3.5S14.5 6 13 6z"></path></svg></a>Python Interface</h2>
<p>All the backend components are exposed as Python modules via SWIG. In addition,
the following classes are added to support the implementation of complex neural
networks.</p>
<h3><a class="anchor" aria-hidden="true" id="opt"></a><a href="#opt" aria-hidden="true" class="hash-link"><svg class="hash-link-icon" aria-hidden="true" height="16" version="1.1" viewBox="0 0 16 16" width="16"><path fill-rule="evenodd" d="M4 9h1v1H4c-1.5 0-3-1.69-3-3.5S2.55 3 4 3h4c1.45 0 3 1.69 3 3.5 0 1.41-.91 2.72-2 3.25V8.59c.58-.45 1-1.27 1-2.09C10 5.22 8.98 4 8 4H4c-.98 0-2 1.22-2 2.5S3 9 4 9zm9-3h-1v1h1c1 0 2 1.22 2 2.5S13.98 12 13 12H9c-.98 0-2-1.22-2-2.5 0-.83.42-1.64 1-2.09V6.25c-1.09.53-2 1.84-2 3.25C6 11.31 7.55 13 9 13h4c1.45 0 3-1.69 3-3.5S14.5 6 13 6z"></path></svg></a>Opt</h3>
<p><code>Opt</code> and its subclasses implement the methods (such as SGD) for updating model
parameter values using parameter gradients. A subclass <a href="./dist-train">DistOpt</a>
synchronizes the gradients across the workers for distributed training by
calling methods from <code>Communicator</code>.</p>
<h3><a class="anchor" aria-hidden="true" id="operator-1"></a><a href="#operator-1" aria-hidden="true" class="hash-link"><svg class="hash-link-icon" aria-hidden="true" height="16" version="1.1" viewBox="0 0 16 16" width="16"><path fill-rule="evenodd" d="M4 9h1v1H4c-1.5 0-3-1.69-3-3.5S2.55 3 4 3h4c1.45 0 3 1.69 3 3.5 0 1.41-.91 2.72-2 3.25V8.59c.58-.45 1-1.27 1-2.09C10 5.22 8.98 4 8 4H4c-.98 0-2 1.22-2 2.5S3 9 4 9zm9-3h-1v1h1c1 0 2 1.22 2 2.5S13.98 12 13 12H9c-.98 0-2-1.22-2-2.5 0-.83.42-1.64 1-2.09V6.25c-1.09.53-2 1.84-2 3.25C6 11.31 7.55 13 9 13h4c1.45 0 3-1.69 3-3.5S14.5 6 13 6z"></path></svg></a>Operator</h3>
<p><code>Operator</code> wraps multiple functions implemented using the Tensor or neural
network operators from the backend. For example, the forward function and
backward function <code>ReLU</code> compose the <code>ReLU</code> operator.</p>
<h3><a class="anchor" aria-hidden="true" id="layer"></a><a href="#layer" aria-hidden="true" class="hash-link"><svg class="hash-link-icon" aria-hidden="true" height="16" version="1.1" viewBox="0 0 16 16" width="16"><path fill-rule="evenodd" d="M4 9h1v1H4c-1.5 0-3-1.69-3-3.5S2.55 3 4 3h4c1.45 0 3 1.69 3 3.5 0 1.41-.91 2.72-2 3.25V8.59c.58-.45 1-1.27 1-2.09C10 5.22 8.98 4 8 4H4c-.98 0-2 1.22-2 2.5S3 9 4 9zm9-3h-1v1h1c1 0 2 1.22 2 2.5S13.98 12 13 12H9c-.98 0-2-1.22-2-2.5 0-.83.42-1.64 1-2.09V6.25c-1.09.53-2 1.84-2 3.25C6 11.31 7.55 13 9 13h4c1.45 0 3-1.69 3-3.5S14.5 6 13 6z"></path></svg></a>Layer</h3>
<p><code>Layer</code> and its subclasses wraps the operators with parameters. For instance,
convolution and linear operators<br>
have weight and bias parameters. The parameters are maintained by the
corresponding <code>Layer</code> class.</p>
<h3><a class="anchor" aria-hidden="true" id="autograd"></a><a href="#autograd" aria-hidden="true" class="hash-link"><svg class="hash-link-icon" aria-hidden="true" height="16" version="1.1" viewBox="0 0 16 16" width="16"><path fill-rule="evenodd" d="M4 9h1v1H4c-1.5 0-3-1.69-3-3.5S2.55 3 4 3h4c1.45 0 3 1.69 3 3.5 0 1.41-.91 2.72-2 3.25V8.59c.58-.45 1-1.27 1-2.09C10 5.22 8.98 4 8 4H4c-.98 0-2 1.22-2 2.5S3 9 4 9zm9-3h-1v1h1c1 0 2 1.22 2 2.5S13.98 12 13 12H9c-.98 0-2-1.22-2-2.5 0-.83.42-1.64 1-2.09V6.25c-1.09.53-2 1.84-2 3.25C6 11.31 7.55 13 9 13h4c1.45 0 3-1.69 3-3.5S14.5 6 13 6z"></path></svg></a>Autograd</h3>
<p><a href="./autograd">Autograd</a> implements the
<a href="https://rufflewind.com/2016-12-30/reverse-mode-automatic-differentiation">reverse-mode automatic differentiation</a>
by recording the execution of the forward functions of the operators calling the
backward functions automatically in the reverse order. All functions can be
buffered by the <code>Scheduler</code> to create a <a href="./graph">computational graph</a> for
efficiency and memory optimization.</p>
<h3><a class="anchor" aria-hidden="true" id="module"></a><a href="#module" aria-hidden="true" class="hash-link"><svg class="hash-link-icon" aria-hidden="true" height="16" version="1.1" viewBox="0 0 16 16" width="16"><path fill-rule="evenodd" d="M4 9h1v1H4c-1.5 0-3-1.69-3-3.5S2.55 3 4 3h4c1.45 0 3 1.69 3 3.5 0 1.41-.91 2.72-2 3.25V8.59c.58-.45 1-1.27 1-2.09C10 5.22 8.98 4 8 4H4c-.98 0-2 1.22-2 2.5S3 9 4 9zm9-3h-1v1h1c1 0 2 1.22 2 2.5S13.98 12 13 12H9c-.98 0-2-1.22-2-2.5 0-.83.42-1.64 1-2.09V6.25c-1.09.53-2 1.84-2 3.25C6 11.31 7.55 13 9 13h4c1.45 0 3-1.69 3-3.5S14.5 6 13 6z"></path></svg></a>Module</h3>
<p><code>Module</code> provides an easy interface to implement new network models. You just
need to inherit <code>Module</code> and define the forward propagation of the model by
creating and calling the layers or operators. <code>Module</code> will do autograd and
update the parameters via <code>Opt</code> automatically when training data is fed into it.</p>
<h3><a class="anchor" aria-hidden="true" id="onnx"></a><a href="#onnx" aria-hidden="true" class="hash-link"><svg class="hash-link-icon" aria-hidden="true" height="16" version="1.1" viewBox="0 0 16 16" width="16"><path fill-rule="evenodd" d="M4 9h1v1H4c-1.5 0-3-1.69-3-3.5S2.55 3 4 3h4c1.45 0 3 1.69 3 3.5 0 1.41-.91 2.72-2 3.25V8.59c.58-.45 1-1.27 1-2.09C10 5.22 8.98 4 8 4H4c-.98 0-2 1.22-2 2.5S3 9 4 9zm9-3h-1v1h1c1 0 2 1.22 2 2.5S13.98 12 13 12H9c-.98 0-2-1.22-2-2.5 0-.83.42-1.64 1-2.09V6.25c-1.09.53-2 1.84-2 3.25C6 11.31 7.55 13 9 13h4c1.45 0 3-1.69 3-3.5S14.5 6 13 6z"></path></svg></a>ONNX</h3>
<p>To support ONNX, SINGA implmenets a <a href="./onnx">sonnx</a> module, which includes</p>
<ul>
<li>SingaFrontend for saving SINGA model into onnx format.</li>
<li>SingaBackend for loading onnx format model into SINGA for training and
inference.</li>
</ul>
</span></div></article></div><div class="docLastUpdate"><em>Last updated on 4/9/2020</em></div><div class="docs-prevnext"><a class="docs-prev button" href="/docs/next/installation"><span class="arrow-prev">← </span><span>Installation</span></a><a class="docs-next button" href="/docs/next/examples"><span>Examples</span><span class="arrow-next"> →</span></a></div></div></div><nav class="onPageNav"><ul class="toc-headings"><li><a href="#low-level-backend">Low-level Backend</a><ul class="toc-headings"><li><a href="#device">Device</a></li><li><a href="#tensor">Tensor</a></li><li><a href="#operator">Operator</a></li><li><a href="#communicator">Communicator</a></li></ul></li><li><a href="#python-interface">Python Interface</a><ul class="toc-headings"><li><a href="#opt">Opt</a></li><li><a href="#operator-1">Operator</a></li><li><a href="#layer">Layer</a></li><li><a href="#autograd">Autograd</a></li><li><a href="#module">Module</a></li><li><a href="#onnx">ONNX</a></li></ul></li></ul></nav></div><footer class="nav-footer" id="footer"><section class="sitemap"><a href="/" class="nav-home"><img src="/img/singa-logo-square.png" alt="Apache SINGA" width="66" height="58"/></a><div><h5>Docs</h5><a href="/docs/installation">Getting Started</a><a href="/docs/device">Guides</a><a href="/en/#">API Reference (coming soon)</a><a href="/docs/model-zoo-cnn-cifar10">Model Zoo</a><a href="/docs/download-singa">Development</a></div><div><h5>Community</h5><a href="/en/users.html">User Showcase</a><a href="/docs/history-singa">SINGA History</a><a href="/docs/team-list">SINGA Team</a><a href="/news">SINGA News</a><a href="https://github.com/apache/singa-doc">GitHub</a><div class="social"><a class="github-button" href="https://github.com/apache/singa-doc" data-count-href="/apache/singa/stargazers" data-show-count="true" data-count-aria-label="# stargazers on GitHub" aria-label="Star this project on GitHub">apache/singa-doc</a></div><div class="social"><a href="https://twitter.com/ApacheSINGA" class="twitter-follow-button">Follow @ApacheSINGA</a></div></div><div><h5>Apache Software Foundation</h5><a href="https://apache.org/" target="_blank" rel="noreferrer noopener">Foundation</a><a href="http://www.apache.org/licenses/" target="_blank" rel="noreferrer noopener">License</a><a href="http://www.apache.org/foundation/sponsorship.html" target="_blank" rel="noreferrer noopener">Sponsorship</a><a href="http://www.apache.org/foundation/thanks.html" target="_blank" rel="noreferrer noopener">Thanks</a><a href="http://www.apache.org/events/current-event" target="_blank" rel="noreferrer noopener">Events</a><a href="http://www.apache.org/security/" target="_blank" rel="noreferrer noopener">Security</a></div></section><div style="width:100%;text-align:center"><a href="https://apache.org/" target="_blank" rel="noreferrer noopener" class="ApacheOpenSource"><img src="/img/asf_logo_wide.svg" alt="Apache Open Source"/></a><section class="copyright" style="max-width:60%;margin:0 auto">Copyright © 2020
The Apache Software Foundation. All rights reserved.
Apache SINGA, Apache, the Apache feather logo, and
the Apache SINGA project logos are trademarks of The
Apache Software Foundation. All other marks mentioned
may be trademarks or registered trademarks of their
respective owners.</section></div></footer></div><script>window.twttr=(function(d,s, id){var js,fjs=d.getElementsByTagName(s)[0],t=window.twttr||{};if(d.getElementById(id))return t;js=d.createElement(s);js.id=id;js.src='https://platform.twitter.com/widgets.js';fjs.parentNode.insertBefore(js, fjs);t._e = [];t.ready = function(f) {t._e.push(f);};return t;}(document, 'script', 'twitter-wjs'));</script></body></html>