-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathbuild_python_model.html
385 lines (263 loc) · 31 KB
/
build_python_model.html
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
<!DOCTYPE html>
<html class="writer-html5" lang="en" >
<head>
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>Build a SMAUG model with Python API — SMAUG: Simulating Machine Learning Applications Using gem5-Aladdin</title>
<link rel="stylesheet" href="_static/css/theme.css" type="text/css" />
<link rel="stylesheet" href="_static/pygments.css" type="text/css" />
<!--[if lt IE 9]>
<script src="_static/js/html5shiv.min.js"></script>
<![endif]-->
<script type="text/javascript" id="documentation_options" data-url_root="./" src="_static/documentation_options.js"></script>
<script src="_static/jquery.js"></script>
<script src="_static/underscore.js"></script>
<script src="_static/doctools.js"></script>
<script src="_static/language_data.js"></script>
<script type="text/javascript" src="_static/js/theme.js"></script>
<link rel="index" title="Index" href="genindex.html" />
<link rel="search" title="Search" href="search.html" />
<link rel="next" title="Run a model in gem5-Aladdin simulation" href="run_model_in_simulation.html" />
<link rel="prev" title="Tutorials" href="python_tutorials.html" />
</head>
<body class="wy-body-for-nav">
<div class="wy-grid-for-nav">
<nav data-toggle="wy-nav-shift" class="wy-nav-side">
<div class="wy-side-scroll">
<div class="wy-side-nav-search" >
<a href="index.html" class="icon icon-home" alt="Documentation Home"> SMAUG
</a>
<div role="search">
<form id="rtd-search-form" class="wy-form" action="search.html" method="get">
<input type="text" name="q" placeholder="Search docs" />
<input type="hidden" name="check_keywords" value="yes" />
<input type="hidden" name="area" value="default" />
</form>
</div>
</div>
<div class="wy-menu wy-menu-vertical" data-spy="affix" role="navigation" aria-label="main navigation">
<p class="caption"><span class="caption-text">Python API and tutorials</span></p>
<ul class="current">
<li class="toctree-l1"><a class="reference internal" href="python_api.html">SMAUG Python APIs</a></li>
<li class="toctree-l1 current"><a class="reference internal" href="python_tutorials.html">Tutorials</a><ul class="current">
<li class="toctree-l2 current"><a class="current reference internal" href="#">Build a SMAUG model with Python API</a></li>
<li class="toctree-l2"><a class="reference internal" href="run_model_in_simulation.html">Run a model in gem5-Aladdin simulation</a></li>
<li class="toctree-l2"><a class="reference internal" href="add_python_operator.html">Adding a new operator</a></li>
</ul>
</li>
</ul>
<p class="caption"><span class="caption-text">C++ docs</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="cpp_docs.html">C++ API and Tutorials</a></li>
</ul>
</div>
</div>
</nav>
<section data-toggle="wy-nav-shift" class="wy-nav-content-wrap">
<nav class="wy-nav-top" aria-label="top navigation">
<i data-toggle="wy-nav-top" class="fa fa-bars"></i>
<a href="index.html">SMAUG</a>
</nav>
<div class="wy-nav-content">
<div class="rst-content">
<div role="navigation" aria-label="breadcrumbs navigation">
<ul class="wy-breadcrumbs">
<li><a href="index.html" class="icon icon-home"></a> »</li>
<li><a href="python_tutorials.html">Tutorials</a> »</li>
<li>Build a SMAUG model with Python API</li>
<li class="wy-breadcrumbs-aside">
<a href="_sources/build_python_model.rst.txt" rel="nofollow"> View page source</a>
</li>
</ul>
<hr/>
</div>
<div role="main" class="document" itemscope="itemscope" itemtype="http://schema.org/Article">
<div itemprop="articleBody">
<div class="section" id="build-a-smaug-model-with-python-api">
<span id="label-build-python-model"></span><h1>Build a SMAUG model with Python API<a class="headerlink" href="#build-a-smaug-model-with-python-api" title="Permalink to this headline">¶</a></h1>
<p>SMAUG’s Python frontend provides easy APIs to build DL models. The created model
is a computational graph, which is serialized into two protobuf files, one for
the model topology and the other for the parameters. These two files are inputs
to SMAUG’s C++ runtime that performs the actual simulation. In this tutorial,
we will be using the SMAUG Python APIs to build new DL models.</p>
<p>Before building a model, we need to create a <cite>Graph</cite> context in which we will
add operators.</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="kn">import</span> <span class="nn">smaug</span> <span class="k">as</span> <span class="nn">sg</span>
<span class="k">with</span> <span class="n">sg</span><span class="o">.</span><span class="n">Graph</span><span class="p">(</span><span class="n">name</span><span class="o">=</span><span class="s2">"my_model"</span><span class="p">,</span> <span class="n">backend</span><span class="o">=</span><span class="s2">"SMV"</span><span class="p">)</span> <span class="k">as</span> <span class="n">graph</span><span class="p">:</span>
<span class="c1"># Any operators instantiated within the context will be added to `graph`.</span>
</pre></div>
</div>
<p>When using the <a class="reference internal" href="smaug.html#smaug.Graph" title="smaug.Graph"><code class="xref py py-class docutils literal notranslate"><span class="pre">smaug.Graph</span></code></a> API to create a graph context, we need to
give a name for the model and select the backend we want to use when running the
model through the C++ runtime. A backend in SMAUG is a logical combination of
hardware blocks that implements all the SMAUG operators. Refer to
<a class="reference external" href="doxygen_html/index.html">C++ docs</a> for more details of a backend
implementation in the C++ runtime. Here, we choose the <code class="code docutils literal notranslate"><span class="pre">SMV</span></code> backend that
comes with SMAUG, which is modeled after the NVDLA architecture. SMAUG also has
another backend named <code class="code docutils literal notranslate"><span class="pre">Reference</span></code>, which is a reference implementation
without having specific performance optimizations. Also, refer to
<a class="reference internal" href="smaug.html#smaug.Graph" title="smaug.Graph"><code class="xref py py-class docutils literal notranslate"><span class="pre">smaug.Graph</span></code></a> for a detailed description of the parameters.</p>
<p>Now we can start adding operators to the graph context. The following gives an
example of building a simple 3-layer model.</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="kn">import</span> <span class="nn">numpy</span> <span class="k">as</span> <span class="nn">np</span>
<span class="kn">import</span> <span class="nn">smaug</span> <span class="k">as</span> <span class="nn">sg</span>
<span class="k">def</span> <span class="nf">generate_random_data</span><span class="p">(</span><span class="n">shape</span><span class="p">):</span>
<span class="n">r</span> <span class="o">=</span> <span class="n">np</span><span class="o">.</span><span class="n">random</span><span class="o">.</span><span class="n">RandomState</span><span class="p">(</span><span class="mi">1234</span><span class="p">)</span>
<span class="k">return</span> <span class="p">(</span><span class="n">r</span><span class="o">.</span><span class="n">rand</span><span class="p">(</span><span class="o">*</span><span class="n">shape</span><span class="p">)</span> <span class="o">*</span> <span class="mf">0.005</span><span class="p">)</span><span class="o">.</span><span class="n">astype</span><span class="p">(</span><span class="n">np</span><span class="o">.</span><span class="n">float16</span><span class="p">)</span>
<span class="k">with</span> <span class="n">sg</span><span class="o">.</span><span class="n">Graph</span><span class="p">(</span><span class="n">name</span><span class="o">=</span><span class="s2">"my_model"</span><span class="p">,</span> <span class="n">backend</span><span class="o">=</span><span class="s2">"SMV"</span><span class="p">)</span> <span class="k">as</span> <span class="n">graph</span><span class="p">:</span>
<span class="n">input_tensor</span> <span class="o">=</span> <span class="n">sg</span><span class="o">.</span><span class="n">Tensor</span><span class="p">(</span>
<span class="n">data_layout</span><span class="o">=</span><span class="n">sg</span><span class="o">.</span><span class="n">NHWC</span><span class="p">,</span> <span class="n">tensor_data</span><span class="o">=</span><span class="n">generate_random_data</span><span class="p">((</span><span class="mi">1</span><span class="p">,</span> <span class="mi">28</span><span class="p">,</span> <span class="mi">28</span><span class="p">,</span> <span class="mi">1</span><span class="p">)))</span>
<span class="n">conv_weights</span> <span class="o">=</span> <span class="n">sg</span><span class="o">.</span><span class="n">Tensor</span><span class="p">(</span>
<span class="n">data_layout</span><span class="o">=</span><span class="n">sg</span><span class="o">.</span><span class="n">NHWC</span><span class="p">,</span> <span class="n">tensor_data</span><span class="o">=</span><span class="n">generate_random_data</span><span class="p">((</span><span class="mi">32</span><span class="p">,</span> <span class="mi">3</span><span class="p">,</span> <span class="mi">3</span><span class="p">,</span> <span class="mi">1</span><span class="p">)))</span>
<span class="n">fc_weights</span> <span class="o">=</span> <span class="n">sg</span><span class="o">.</span><span class="n">Tensor</span><span class="p">(</span>
<span class="n">data_layout</span><span class="o">=</span><span class="n">sg</span><span class="o">.</span><span class="n">NC</span><span class="p">,</span> <span class="n">tensor_data</span><span class="o">=</span><span class="n">generate_random_data</span><span class="p">((</span><span class="mi">10</span><span class="p">,</span> <span class="mi">6272</span><span class="p">)))</span>
<span class="c1"># Shape of act: [1, 28, 28, 1].</span>
<span class="n">act</span> <span class="o">=</span> <span class="n">sg</span><span class="o">.</span><span class="n">input_data</span><span class="p">(</span><span class="n">input_tensor</span><span class="p">)</span>
<span class="c1"># After the convolution, shape of act: [1, 32, 28, 28].</span>
<span class="n">act</span> <span class="o">=</span> <span class="n">sg</span><span class="o">.</span><span class="n">nn</span><span class="o">.</span><span class="n">convolution</span><span class="p">(</span>
<span class="n">act</span><span class="p">,</span> <span class="n">conv_weights</span><span class="p">,</span> <span class="n">stride</span><span class="o">=</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="n">padding</span><span class="o">=</span><span class="s2">"same"</span><span class="p">,</span> <span class="n">activation</span><span class="o">=</span><span class="s2">"relu"</span><span class="p">)</span>
<span class="c1"># After the max pooling, shape of act: [1, 32, 14, 14].</span>
<span class="n">act</span> <span class="o">=</span> <span class="n">sg</span><span class="o">.</span><span class="n">nn</span><span class="o">.</span><span class="n">max_pool</span><span class="p">(</span><span class="n">act</span><span class="p">,</span> <span class="n">pool_size</span><span class="o">=</span><span class="p">[</span><span class="mi">2</span><span class="p">,</span> <span class="mi">2</span><span class="p">],</span> <span class="n">stride</span><span class="o">=</span><span class="p">[</span><span class="mi">2</span><span class="p">,</span> <span class="mi">2</span><span class="p">])</span>
<span class="c1"># After the matrix multiply, shape of act: [1, 10].</span>
<span class="n">act</span> <span class="o">=</span> <span class="n">sg</span><span class="o">.</span><span class="n">nn</span><span class="o">.</span><span class="n">mat_mul</span><span class="p">(</span><span class="n">act</span><span class="p">,</span> <span class="n">fc_weights</span><span class="p">)</span>
</pre></div>
</div>
<p>As we create the first operator for the model, we need to first prepare an
input tensor and weight tensors that are used by the operator. Tensors are
represented by the <a class="reference internal" href="smaug.html#smaug.Tensor" title="smaug.Tensor"><code class="xref py py-class docutils literal notranslate"><span class="pre">smaug.Tensor</span></code></a>. In the example, we create an input
tensor <code class="code docutils literal notranslate"><span class="pre">input_tensor</span></code> using the API. Here, we specify the
<code class="code docutils literal notranslate"><span class="pre">data_layout</span></code> as <code class="code docutils literal notranslate"><span class="pre">sg.NHWC</span></code>, which stands for a 4D tensor shape with
the channel-major layout. We also specify the <code class="code docutils literal notranslate"><span class="pre">tensor_data</span></code> parameter
with a randomly generated NumPy array, with a shape of <code class="code docutils literal notranslate"><span class="pre">[1,</span> <span class="pre">28,</span> <span class="pre">28,</span> <span class="pre">1]</span></code>.
However, the user can use the real weights extracted from a pretrained model.
Likewise, we create two weight tensors that will be used by a convolution
operator and a matrix multiply operator, respectively.</p>
<p>A <code class="xref py py-func docutils literal notranslate"><span class="pre">smaug.data_op()</span></code>, which simply forwards an input tensor to its output, is
required for any tensor that is not the output of another operator. Here,
<code class="code docutils literal notranslate"><span class="pre">act</span></code> is a reference to code:<cite>input_tensor</cite>. Then, <code class="code docutils literal notranslate"><span class="pre">act</span></code> is
fed to a convolution operator that also takes <code class="code docutils literal notranslate"><span class="pre">conv_weights</span></code> as its
filter input. With more details provided in <a class="reference internal" href="nn.html#smaug.nn.convolution" title="smaug.nn.convolution"><code class="xref py py-func docutils literal notranslate"><span class="pre">smaug.nn.convolution()</span></code></a>, it
computes a 3D convolution given the 4D input and filter tensors, and we use 1x1
strides, the <code class="code docutils literal notranslate"><span class="pre">same</span></code> padding and a ReLU activation fused with the
convolution operation. The output of it then goes through a max pooling
operator with a 2x2 filter size, which in turn fans its output into the last
matrix multiply operator. Note that since the output of the max pooling
operator is a 4D tensor while <a class="reference internal" href="nn.html#smaug.nn.mat_mul" title="smaug.nn.mat_mul"><code class="xref py py-func docutils literal notranslate"><span class="pre">smaug.nn.mat_mul()</span></code></a> expects a 2D input
tensor, SMAUG will automatically add a layout transformation operator
<a class="reference internal" href="tensor.html#smaug.tensor.reorder" title="smaug.tensor.reorder"><code class="xref py py-func docutils literal notranslate"><span class="pre">smaug.tensor.reorder()</span></code></a> in between to make the data layout format
compatible. Thus, the 4D tensor of shape <code class="code docutils literal notranslate"><span class="pre">[1,</span> <span class="pre">32,</span> <span class="pre">14,</span> <span class="pre">14]</span></code> will be
flattened into a 2D tensor of shape <code class="code docutils literal notranslate"><span class="pre">[1,</span> <span class="pre">6272]</span></code> before running the matrix
multiply. Similarly, SMAUG will also perform the NHWC to NCHW layout
transformation or vice versa as per the expected layout format of the backend.</p>
<p>After finishing adding operators to the model, we can now take a look at the
summary of the model using the <a class="reference internal" href="smaug.html#smaug.Graph.print_summary" title="smaug.Graph.print_summary"><code class="xref py py-func docutils literal notranslate"><span class="pre">smaug.Graph.print_summary()</span></code></a> API.</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="n">graph</span><span class="o">.</span><span class="n">print_summary</span><span class="p">()</span>
</pre></div>
</div>
<p>This prints model-level information and operator-specific properties as below:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="o">=================================================================</span>
<span class="n">Summary</span> <span class="n">of</span> <span class="n">the</span> <span class="n">network</span><span class="p">:</span> <span class="n">my_model</span> <span class="p">(</span><span class="n">SMV</span><span class="p">)</span>
<span class="o">=================================================================</span>
<span class="n">Host</span> <span class="n">memory</span> <span class="n">access</span> <span class="n">policy</span><span class="p">:</span> <span class="n">AllDma</span><span class="o">.</span>
<span class="o">-----------------------------------------------------------------</span>
<span class="n">Name</span><span class="p">:</span> <span class="n">data</span> <span class="p">(</span><span class="n">Data</span><span class="p">)</span>
<span class="n">Parents</span><span class="p">:</span>
<span class="n">Children</span><span class="p">:</span><span class="n">conv</span>
<span class="n">Input</span> <span class="n">tensors</span><span class="p">:</span>
<span class="n">data</span><span class="o">/</span><span class="n">input0</span> <span class="n">Float16</span> <span class="p">[</span><span class="mi">1</span><span class="p">,</span> <span class="mi">28</span><span class="p">,</span> <span class="mi">28</span><span class="p">,</span> <span class="mi">1</span><span class="p">]</span> <span class="n">NHWC</span> <span class="n">alignment</span><span class="p">(</span><span class="mi">8</span><span class="p">)</span>
<span class="n">Output</span> <span class="n">tensors</span><span class="p">:</span>
<span class="n">data</span><span class="o">/</span><span class="n">output0</span> <span class="n">Float16</span> <span class="p">[</span><span class="mi">1</span><span class="p">,</span> <span class="mi">28</span><span class="p">,</span> <span class="mi">28</span><span class="p">,</span> <span class="mi">1</span><span class="p">]</span> <span class="n">NHWC</span> <span class="n">alignment</span><span class="p">(</span><span class="mi">8</span><span class="p">)</span>
<span class="o">-----------------------------------------------------------------</span>
<span class="n">Name</span><span class="p">:</span> <span class="n">data_1</span> <span class="p">(</span><span class="n">Data</span><span class="p">)</span>
<span class="n">Parents</span><span class="p">:</span>
<span class="n">Children</span><span class="p">:</span><span class="n">conv</span>
<span class="n">Input</span> <span class="n">tensors</span><span class="p">:</span>
<span class="n">data_1</span><span class="o">/</span><span class="n">input0</span> <span class="n">Float16</span> <span class="p">[</span><span class="mi">32</span><span class="p">,</span> <span class="mi">3</span><span class="p">,</span> <span class="mi">3</span><span class="p">,</span> <span class="mi">1</span><span class="p">]</span> <span class="n">NHWC</span> <span class="n">alignment</span><span class="p">(</span><span class="mi">8</span><span class="p">)</span>
<span class="n">Output</span> <span class="n">tensors</span><span class="p">:</span>
<span class="n">data_1</span><span class="o">/</span><span class="n">output0</span> <span class="n">Float16</span> <span class="p">[</span><span class="mi">32</span><span class="p">,</span> <span class="mi">3</span><span class="p">,</span> <span class="mi">3</span><span class="p">,</span> <span class="mi">1</span><span class="p">]</span> <span class="n">NHWC</span> <span class="n">alignment</span><span class="p">(</span><span class="mi">8</span><span class="p">)</span>
<span class="o">-----------------------------------------------------------------</span>
<span class="n">Name</span><span class="p">:</span> <span class="n">conv</span> <span class="p">(</span><span class="n">Convolution3d</span><span class="p">)</span>
<span class="n">Parents</span><span class="p">:</span><span class="n">data</span> <span class="n">data_1</span>
<span class="n">Children</span><span class="p">:</span><span class="n">max_pool</span>
<span class="n">Input</span> <span class="n">tensors</span><span class="p">:</span>
<span class="n">data</span><span class="o">/</span><span class="n">output0</span> <span class="n">Float16</span> <span class="p">[</span><span class="mi">1</span><span class="p">,</span> <span class="mi">28</span><span class="p">,</span> <span class="mi">28</span><span class="p">,</span> <span class="mi">1</span><span class="p">]</span> <span class="n">NHWC</span> <span class="n">alignment</span><span class="p">(</span><span class="mi">8</span><span class="p">)</span>
<span class="n">data_1</span><span class="o">/</span><span class="n">output0</span> <span class="n">Float16</span> <span class="p">[</span><span class="mi">32</span><span class="p">,</span> <span class="mi">3</span><span class="p">,</span> <span class="mi">3</span><span class="p">,</span> <span class="mi">1</span><span class="p">]</span> <span class="n">NHWC</span> <span class="n">alignment</span><span class="p">(</span><span class="mi">8</span><span class="p">)</span>
<span class="n">Output</span> <span class="n">tensors</span><span class="p">:</span>
<span class="n">conv</span><span class="o">/</span><span class="n">output0</span> <span class="n">Float16</span> <span class="p">[</span><span class="mi">1</span><span class="p">,</span> <span class="mi">28</span><span class="p">,</span> <span class="mi">28</span><span class="p">,</span> <span class="mi">32</span><span class="p">]</span> <span class="n">NHWC</span> <span class="n">alignment</span><span class="p">(</span><span class="mi">8</span><span class="p">)</span>
<span class="o">-----------------------------------------------------------------</span>
<span class="n">Name</span><span class="p">:</span> <span class="n">max_pool</span> <span class="p">(</span><span class="n">MaxPooling</span><span class="p">)</span>
<span class="n">Parents</span><span class="p">:</span><span class="n">conv</span>
<span class="n">Children</span><span class="p">:</span><span class="n">reorder</span>
<span class="n">Input</span> <span class="n">tensors</span><span class="p">:</span>
<span class="n">conv</span><span class="o">/</span><span class="n">output0</span> <span class="n">Float16</span> <span class="p">[</span><span class="mi">1</span><span class="p">,</span> <span class="mi">28</span><span class="p">,</span> <span class="mi">28</span><span class="p">,</span> <span class="mi">32</span><span class="p">]</span> <span class="n">NHWC</span> <span class="n">alignment</span><span class="p">(</span><span class="mi">8</span><span class="p">)</span>
<span class="n">Output</span> <span class="n">tensors</span><span class="p">:</span>
<span class="n">max_pool</span><span class="o">/</span><span class="n">output0</span> <span class="n">Float16</span> <span class="p">[</span><span class="mi">1</span><span class="p">,</span> <span class="mi">14</span><span class="p">,</span> <span class="mi">14</span><span class="p">,</span> <span class="mi">32</span><span class="p">]</span> <span class="n">NHWC</span> <span class="n">alignment</span><span class="p">(</span><span class="mi">8</span><span class="p">)</span>
<span class="o">-----------------------------------------------------------------</span>
<span class="n">Name</span><span class="p">:</span> <span class="n">reorder</span> <span class="p">(</span><span class="n">Reorder</span><span class="p">)</span>
<span class="n">Parents</span><span class="p">:</span><span class="n">max_pool</span>
<span class="n">Children</span><span class="p">:</span><span class="n">mat_mul</span>
<span class="n">Input</span> <span class="n">tensors</span><span class="p">:</span>
<span class="n">max_pool</span><span class="o">/</span><span class="n">output0</span> <span class="n">Float16</span> <span class="p">[</span><span class="mi">1</span><span class="p">,</span> <span class="mi">14</span><span class="p">,</span> <span class="mi">14</span><span class="p">,</span> <span class="mi">32</span><span class="p">]</span> <span class="n">NHWC</span> <span class="n">alignment</span><span class="p">(</span><span class="mi">8</span><span class="p">)</span>
<span class="n">Output</span> <span class="n">tensors</span><span class="p">:</span>
<span class="n">reorder</span><span class="o">/</span><span class="n">output0</span> <span class="n">Float16</span> <span class="p">[</span><span class="mi">1</span><span class="p">,</span> <span class="mi">6272</span><span class="p">]</span> <span class="n">NC</span> <span class="n">alignment</span><span class="p">(</span><span class="mi">8</span><span class="p">)</span>
<span class="o">-----------------------------------------------------------------</span>
<span class="n">Name</span><span class="p">:</span> <span class="n">data_2</span> <span class="p">(</span><span class="n">Data</span><span class="p">)</span>
<span class="n">Parents</span><span class="p">:</span>
<span class="n">Children</span><span class="p">:</span><span class="n">mat_mul</span>
<span class="n">Input</span> <span class="n">tensors</span><span class="p">:</span>
<span class="n">data_2</span><span class="o">/</span><span class="n">input0</span> <span class="n">Float16</span> <span class="p">[</span><span class="mi">10</span><span class="p">,</span> <span class="mi">6272</span><span class="p">]</span> <span class="n">NC</span> <span class="n">alignment</span><span class="p">(</span><span class="mi">8</span><span class="p">)</span>
<span class="n">Output</span> <span class="n">tensors</span><span class="p">:</span>
<span class="n">data_2</span><span class="o">/</span><span class="n">output0</span> <span class="n">Float16</span> <span class="p">[</span><span class="mi">10</span><span class="p">,</span> <span class="mi">6272</span><span class="p">]</span> <span class="n">NC</span> <span class="n">alignment</span><span class="p">(</span><span class="mi">8</span><span class="p">)</span>
<span class="o">-----------------------------------------------------------------</span>
<span class="n">Name</span><span class="p">:</span> <span class="n">mat_mul</span> <span class="p">(</span><span class="n">InnerProduct</span><span class="p">)</span>
<span class="n">Parents</span><span class="p">:</span><span class="n">reorder</span> <span class="n">data_2</span>
<span class="n">Children</span><span class="p">:</span>
<span class="n">Input</span> <span class="n">tensors</span><span class="p">:</span>
<span class="n">reorder</span><span class="o">/</span><span class="n">output0</span> <span class="n">Float16</span> <span class="p">[</span><span class="mi">1</span><span class="p">,</span> <span class="mi">6272</span><span class="p">]</span> <span class="n">NC</span> <span class="n">alignment</span><span class="p">(</span><span class="mi">8</span><span class="p">)</span>
<span class="n">data_2</span><span class="o">/</span><span class="n">output0</span> <span class="n">Float16</span> <span class="p">[</span><span class="mi">10</span><span class="p">,</span> <span class="mi">6272</span><span class="p">]</span> <span class="n">NC</span> <span class="n">alignment</span><span class="p">(</span><span class="mi">8</span><span class="p">)</span>
<span class="n">Output</span> <span class="n">tensors</span><span class="p">:</span>
<span class="n">mat_mul</span><span class="o">/</span><span class="n">output0</span> <span class="n">Float16</span> <span class="p">[</span><span class="mi">1</span><span class="p">,</span> <span class="mi">10</span><span class="p">]</span> <span class="n">NC</span> <span class="n">alignment</span><span class="p">(</span><span class="mi">8</span><span class="p">)</span>
<span class="o">-----------------------------------------------------------------</span>
</pre></div>
</div>
<p>Finally, we can export the model files using the
<a class="reference internal" href="smaug.html#smaug.Graph.write_graph" title="smaug.Graph.write_graph"><code class="xref py py-func docutils literal notranslate"><span class="pre">smaug.Graph.write_graph()</span></code></a> API.</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="n">graph</span><span class="o">.</span><span class="n">write_graph</span><span class="p">()</span>
</pre></div>
</div>
<p>This gives us two files named <code class="code docutils literal notranslate"><span class="pre">my_model_topo.pbtxt</span></code> and
<code class="code docutils literal notranslate"><span class="pre">my_model_params.pb</span></code>, where the former stores all the model information
except for the parameters, which are stored in the latter. This separation is
helpful for us to quickly check things in the human readable topology file
while still compressing as much as possible the oftentimes large paramaters.
We can now move on to the <a class="reference external" href="doxygen_html/index.html">C++ side tutorials</a> that
explain the details of using these two files to run the model.</p>
</div>
</div>
</div>
<footer>
<div class="rst-footer-buttons" role="navigation" aria-label="footer navigation">
<a href="run_model_in_simulation.html" class="btn btn-neutral float-right" title="Run a model in gem5-Aladdin simulation" accesskey="n" rel="next">Next <span class="fa fa-arrow-circle-right"></span></a>
<a href="python_tutorials.html" class="btn btn-neutral float-left" title="Tutorials" accesskey="p" rel="prev"><span class="fa fa-arrow-circle-left"></span> Previous</a>
</div>
<hr/>
<div role="contentinfo">
<p>
© Copyright 2020, SMAUG Contributors
</p>
</div>
Built with <a href="http://sphinx-doc.org/">Sphinx</a> using a
<a href="https://github.com/rtfd/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">
jQuery(function () {
SphinxRtdTheme.Navigation.enable(true);
});
</script>
</body>
</html>