forked from pytorch/pytorch.github.io
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathjit.html
1535 lines (1327 loc) · 126 KB
/
jit.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
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
<!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>Torch Script — PyTorch master documentation</title>
<link rel="canonical" href="https://pytorch.org/docs/stable/jit.html"/>
<link rel="stylesheet" href="_static/css/theme.css" type="text/css" />
<!-- <link rel="stylesheet" href="_static/pygments.css" type="text/css" /> -->
<link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/katex.min.css" type="text/css" />
<link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/katex.min.css" type="text/css" />
<link rel="stylesheet" href="_static/katex-math.css" type="text/css" />
<link rel="index" title="Index" href="genindex.html" />
<link rel="search" title="Search" href="search.html" />
<link rel="next" title="Multiprocessing package - torch.multiprocessing" href="multiprocessing.html" />
<link rel="prev" title="Probability distributions - torch.distributions" href="distributions.html" />
<script src="_static/js/modernizr.min.js"></script>
</head>
<div class="container-fluid header-holder tutorials-header" id="header-holder">
<div class="container">
<div class="header-container">
<a class="header-logo" href="https://pytorch.org/" aria-label="PyTorch"></a>
<div class="main-menu">
<ul>
<li>
<a href="https://pytorch.org/get-started">Get Started</a>
</li>
<li>
<a href="https://pytorch.org/features">Features</a>
</li>
<li>
<a href="https://pytorch.org/ecosystem">Ecosystem</a>
</li>
<li>
<a href="https://pytorch.org/blog/">Blog</a>
</li>
<li>
<a href="https://pytorch.org/tutorials">Tutorials</a>
</li>
<li class="active">
<a href="https://pytorch.org/docs/stable/index.html">Docs</a>
</li>
<li>
<a href="https://pytorch.org/resources">Resources</a>
</li>
<li>
<a href="https://github.com/pytorch/pytorch">Github</a>
</li>
</ul>
</div>
<a class="main-menu-open-button" href="#" data-behavior="open-mobile-menu"></a>
</div>
</div>
</div>
<body class="pytorch-body">
<div class="table-of-contents-link-wrapper">
<span>Table of Contents</span>
<a href="#" class="toggle-table-of-contents" data-behavior="toggle-table-of-contents"></a>
</div>
<nav data-toggle="wy-nav-shift" class="pytorch-left-menu" id="pytorch-left-menu">
<div class="pytorch-side-scroll">
<div class="pytorch-menu pytorch-menu-vertical" data-spy="affix" role="navigation" aria-label="main navigation">
<div class="pytorch-left-menu-search">
<div class="version">
<a href="http://pytorch.org/docs/versions.html"> 1.0.0 ▼</a>
</div>
<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>
<p class="caption"><span class="caption-text">Notes</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="notes/autograd.html">Autograd mechanics</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/broadcasting.html">Broadcasting semantics</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/cuda.html">CUDA semantics</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/extending.html">Extending PyTorch</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/faq.html">Frequently Asked Questions</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/multiprocessing.html">Multiprocessing best practices</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/randomness.html">Reproducibility</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/serialization.html">Serialization semantics</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/windows.html">Windows FAQ</a></li>
</ul>
<p class="caption"><span class="caption-text">Package Reference</span></p>
<ul class="current">
<li class="toctree-l1"><a class="reference internal" href="torch.html">torch</a></li>
<li class="toctree-l1"><a class="reference internal" href="tensors.html">torch.Tensor</a></li>
<li class="toctree-l1"><a class="reference internal" href="tensor_attributes.html">Tensor Attributes</a></li>
<li class="toctree-l1"><a class="reference internal" href="type_info.html">Type Info</a></li>
<li class="toctree-l1"><a class="reference internal" href="sparse.html">torch.sparse</a></li>
<li class="toctree-l1"><a class="reference internal" href="cuda.html">torch.cuda</a></li>
<li class="toctree-l1"><a class="reference internal" href="storage.html">torch.Storage</a></li>
<li class="toctree-l1"><a class="reference internal" href="nn.html">torch.nn</a></li>
<li class="toctree-l1"><a class="reference internal" href="nn.html#torch-nn-functional">torch.nn.functional</a></li>
<li class="toctree-l1"><a class="reference internal" href="nn.html#torch-nn-init">torch.nn.init</a></li>
<li class="toctree-l1"><a class="reference internal" href="optim.html">torch.optim</a></li>
<li class="toctree-l1"><a class="reference internal" href="autograd.html">torch.autograd</a></li>
<li class="toctree-l1"><a class="reference internal" href="distributed.html">torch.distributed</a></li>
<li class="toctree-l1"><a class="reference internal" href="distributions.html">torch.distributions</a></li>
<li class="toctree-l1 current"><a class="current reference internal" href="#">torch.jit</a></li>
<li class="toctree-l1"><a class="reference internal" href="multiprocessing.html">torch.multiprocessing</a></li>
<li class="toctree-l1"><a class="reference internal" href="bottleneck.html">torch.utils.bottleneck</a></li>
<li class="toctree-l1"><a class="reference internal" href="checkpoint.html">torch.utils.checkpoint</a></li>
<li class="toctree-l1"><a class="reference internal" href="cpp_extension.html">torch.utils.cpp_extension</a></li>
<li class="toctree-l1"><a class="reference internal" href="data.html">torch.utils.data</a></li>
<li class="toctree-l1"><a class="reference internal" href="dlpack.html">torch.utils.dlpack</a></li>
<li class="toctree-l1"><a class="reference internal" href="ffi.html">torch.utils.ffi</a></li>
<li class="toctree-l1"><a class="reference internal" href="hub.html">torch.hub</a></li>
<li class="toctree-l1"><a class="reference internal" href="model_zoo.html">torch.utils.model_zoo</a></li>
<li class="toctree-l1"><a class="reference internal" href="onnx.html">torch.onnx</a></li>
<li class="toctree-l1"><a class="reference internal" href="distributed_deprecated.html">torch.distributed.deprecated</a></li>
<li class="toctree-l1"><a class="reference internal" href="legacy.html">torch.legacy</a></li>
</ul>
<p class="caption"><span class="caption-text">torchvision Reference</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="torchvision/index.html">torchvision</a></li>
</ul>
</div>
</div>
</nav>
<div class="pytorch-container">
<div class="pytorch-page-level-bar" id="pytorch-page-level-bar">
<div class="pytorch-breadcrumbs-wrapper">
<div role="navigation" aria-label="breadcrumbs navigation">
<ul class="pytorch-breadcrumbs">
<li>
<a href="index.html">
Docs
</a> >
</li>
<li>Torch Script</li>
<li class="pytorch-breadcrumbs-aside">
<a href="_sources/jit.rst.txt" rel="nofollow"><img src="_static/images/view-page-source-icon.svg"></a>
</li>
</ul>
</div>
</div>
<div class="pytorch-shortcuts-wrapper" id="pytorch-shortcuts-wrapper">
Shortcuts
</div>
</div>
<section data-toggle="wy-nav-shift" id="pytorch-content-wrap" class="pytorch-content-wrap">
<div class="pytorch-content-left">
<div class="rst-content">
<div role="main" class="main-content" itemscope="itemscope" itemtype="http://schema.org/Article">
<article itemprop="articleBody" id="pytorch-article" class="pytorch-article">
<div class="section" id="torch-script">
<h1>Torch Script<a class="headerlink" href="#torch-script" title="Permalink to this headline">¶</a></h1>
<div class="contents local topic" id="contents">
<ul class="simple">
<li><a class="reference internal" href="#creating-torch-script-code" id="id1">Creating Torch Script Code</a></li>
<li><a class="reference internal" href="#mixing-tracing-and-scripting" id="id2">Mixing Tracing and Scripting</a></li>
<li><a class="reference internal" href="#torch-script-language-reference" id="id3">Torch Script Language Reference</a><ul>
<li><a class="reference internal" href="#types" id="id4">Types</a></li>
<li><a class="reference internal" href="#expressions" id="id5">Expressions</a></li>
<li><a class="reference internal" href="#statements" id="id6">Statements</a></li>
<li><a class="reference internal" href="#variable-resolution" id="id7">Variable Resolution</a></li>
<li><a class="reference internal" href="#use-of-python-values" id="id8">Use of Python Values</a></li>
<li><a class="reference internal" href="#debugging" id="id9">Debugging</a></li>
<li><a class="reference internal" href="#builtin-functions" id="id10">Builtin Functions</a></li>
</ul>
</li>
</ul>
</div>
<span class="target" id="module-torch.jit"></span><p>Torch Script is a way to create serializable and optimizable models from PyTorch code.
Any code written in Torch Script can be saved from your Python
process and loaded in a process where there is no Python dependency.</p>
<p>We provide tools to incrementally transition a model from being a pure Python program
to a Torch Script program that can be run independently from Python, for instance, in a standalone C++ program.
This makes it possible to train models in PyTorch using familiar tools and then export
the model to a production environment where it is not a good idea to run models as Python programs
for performance and multi-threading reasons.</p>
<div class="section" id="creating-torch-script-code">
<h2><a class="toc-backref" href="#id1">Creating Torch Script Code</a><a class="headerlink" href="#creating-torch-script-code" title="Permalink to this headline">¶</a></h2>
<dl class="class">
<dt id="torch.jit.ScriptModule">
<em class="property">class </em><code class="descclassname">torch.jit.</code><code class="descname">ScriptModule</code><span class="sig-paren">(</span><em>optimize=True</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/jit.html#ScriptModule"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.jit.ScriptModule" title="Permalink to this definition">¶</a></dt>
<dd><p>The core data structure in Torch Script is the <code class="docutils literal notranslate"><span class="pre">ScriptModule</span></code>. It is an
analogue of torch’s nn.Module and represents an entire model as a tree of
submodules. Like normal modules, each individual module in a ScriptModule can
have submodules, parameters, and methods. In nn.Modules methods are implemented
as Python functions, but in ScriptModules methods typically implemented as
<em>Torch Script</em> functions, a statically-typed subset of Python that contains all
of PyTorch’s built-in Tensor operations. This difference allows your
ScriptModules code to run without the need for a Python interpreter.</p>
<p>ScriptModules and the Torch Script functions inside of them can be created in
two ways:</p>
<p><strong>Tracing:</strong></p>
<blockquote>
<div><p>Using <code class="docutils literal notranslate"><span class="pre">torch.jit.trace</span></code>, you can take an existing module or python
function, provide example inputs, and we run the function, recording the
operations performed on all the tensors. We turn the resulting recording
into a Torch Script method that is installed as the <code class="docutils literal notranslate"><span class="pre">forward</span></code> method of a
ScriptModule. This module also contains any parameters that the original
module had as well.</p>
<p>Example:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="kn">import</span> <span class="nn">torch</span>
<span class="k">def</span> <span class="nf">foo</span><span class="p">(</span><span class="n">x</span><span class="p">,</span> <span class="n">y</span><span class="p">):</span>
<span class="k">return</span> <span class="mi">2</span><span class="o">*</span><span class="n">x</span> <span class="o">+</span> <span class="n">y</span>
<span class="n">traced_foo</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">jit</span><span class="o">.</span><span class="n">trace</span><span class="p">(</span><span class="n">foo</span><span class="p">,</span> <span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">rand</span><span class="p">(</span><span class="mi">3</span><span class="p">),</span> <span class="n">torch</span><span class="o">.</span><span class="n">rand</span><span class="p">(</span><span class="mi">3</span><span class="p">)))</span>
</pre></div>
</div>
<div class="admonition note">
<p class="first admonition-title">Note</p>
<p class="last">Tracing a <em>function</em> will produce a <code class="docutils literal notranslate"><span class="pre">ScriptModule</span></code> with a single
<code class="docutils literal notranslate"><span class="pre">forward</span></code> method that implements that function, and that contains
no parameters.</p>
</div>
<p>Example:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="kn">import</span> <span class="nn">torch</span>
<span class="kn">import</span> <span class="nn">torchvision</span>
<span class="n">traced_net</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">jit</span><span class="o">.</span><span class="n">trace</span><span class="p">(</span><span class="n">torchvision</span><span class="o">.</span><span class="n">models</span><span class="o">.</span><span class="n">resnet18</span><span class="p">(),</span>
<span class="n">torch</span><span class="o">.</span><span class="n">rand</span><span class="p">(</span><span class="mi">1</span><span class="p">,</span> <span class="mi">3</span><span class="p">,</span> <span class="mi">224</span><span class="p">,</span> <span class="mi">224</span><span class="p">))</span>
</pre></div>
</div>
<div class="admonition note">
<p class="first admonition-title">Note</p>
<p>Tracing only records operations done when the given function is run on the given
tensors. Therefore, the returned <code class="docutils literal notranslate"><span class="pre">ScriptModule</span></code> will always run the same traced
graph on any input. This has some important implications when your module is
expected to run different sets of operations, depending on the input and/or the
module state. For example,</p>
<blockquote>
<div><ul class="simple">
<li>Tracing will not record any control-flow like if statements or loops. When
this control-flow is constant across your module, this is fine and it often
just inlines configuration decisions. But sometimes the control-flow is
actually part of the model itself. For instance, a beam search in
sequence-to-sequence translation is a loop over the (varying) sequence
length of inputs.</li>
<li>In the returned <code class="docutils literal notranslate"><span class="pre">ScriptModule</span></code>, operations that have different behaviors
in <code class="docutils literal notranslate"><span class="pre">training</span></code> and <code class="docutils literal notranslate"><span class="pre">eval</span></code> modes will always behave as if it is in the
mode it was in during tracing, no matter which mode the <code class="docutils literal notranslate"><span class="pre">ScriptModule</span></code>
is in.</li>
</ul>
</div></blockquote>
<p class="last">In cases like these, tracing would not be appropriate and scripting is a better
choice.</p>
</div>
</div></blockquote>
<p><strong>Scripting:</strong></p>
<blockquote>
<div><p>You can write Torch Script code directly using Python syntax. You do this
using the <code class="docutils literal notranslate"><span class="pre">torch.jit.script</span></code> annotation (for functions) or
<code class="docutils literal notranslate"><span class="pre">torch.jit.script_method</span></code> annotation (for methods) on subclasses of
ScriptModule. With this annotation the body of the annotated function is
directly translated into Torch Script. Torch Script itself is a subset of
the Python language, so not all features in python work, but we provide
enough functionality to compute on tensors and do control-dependent
operations.</p>
<p>Example:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="kn">import</span> <span class="nn">torch</span>
<span class="nd">@torch</span><span class="o">.</span><span class="n">jit</span><span class="o">.</span><span class="n">script</span>
<span class="k">def</span> <span class="nf">foo</span><span class="p">(</span><span class="n">x</span><span class="p">,</span> <span class="n">y</span><span class="p">):</span>
<span class="k">if</span> <span class="n">x</span><span class="o">.</span><span class="n">max</span><span class="p">()</span> <span class="o">></span> <span class="n">y</span><span class="o">.</span><span class="n">max</span><span class="p">():</span>
<span class="n">r</span> <span class="o">=</span> <span class="n">x</span>
<span class="k">else</span><span class="p">:</span>
<span class="n">r</span> <span class="o">=</span> <span class="n">y</span>
<span class="k">return</span> <span class="n">r</span>
</pre></div>
</div>
<div class="admonition note">
<p class="first admonition-title">Note</p>
<p class="last">A script <em>function</em> annotation will construct a ScriptModule
with a single <code class="docutils literal notranslate"><span class="pre">forward</span></code> method that implements that function,
and that contains no parameters.</p>
</div>
<p>Example:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="kn">import</span> <span class="nn">torch</span>
<span class="k">class</span> <span class="nc">MyModule</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">jit</span><span class="o">.</span><span class="n">ScriptModule</span><span class="p">):</span>
<span class="k">def</span> <span class="nf">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">N</span><span class="p">,</span> <span class="n">M</span><span class="p">):</span>
<span class="nb">super</span><span class="p">(</span><span class="n">MyModule</span><span class="p">,</span> <span class="bp">self</span><span class="p">)</span><span class="o">.</span><span class="fm">__init__</span><span class="p">()</span>
<span class="bp">self</span><span class="o">.</span><span class="n">weight</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">nn</span><span class="o">.</span><span class="n">Parameter</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">rand</span><span class="p">(</span><span class="n">N</span><span class="p">,</span> <span class="n">M</span><span class="p">))</span>
<span class="nd">@torch</span><span class="o">.</span><span class="n">jit</span><span class="o">.</span><span class="n">script_method</span>
<span class="k">def</span> <span class="nf">forward</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="nb">input</span><span class="p">):</span>
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">weight</span><span class="o">.</span><span class="n">mv</span><span class="p">(</span><span class="nb">input</span><span class="p">)</span>
</pre></div>
</div>
<p>Example:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="kn">import</span> <span class="nn">torch</span>
<span class="kn">import</span> <span class="nn">torch.nn</span> <span class="k">as</span> <span class="nn">nn</span>
<span class="kn">import</span> <span class="nn">torch.nn.functional</span> <span class="k">as</span> <span class="nn">F</span>
<span class="kn">from</span> <span class="nn">torch.jit</span> <span class="k">import</span> <span class="n">ScriptModule</span><span class="p">,</span> <span class="n">script_method</span><span class="p">,</span> <span class="n">trace</span>
<span class="k">class</span> <span class="nc">MyScriptModule</span><span class="p">(</span><span class="n">ScriptModule</span><span class="p">):</span>
<span class="k">def</span> <span class="nf">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="nb">super</span><span class="p">(</span><span class="n">MyScriptModule</span><span class="p">,</span> <span class="bp">self</span><span class="p">)</span><span class="o">.</span><span class="fm">__init__</span><span class="p">()</span>
<span class="c1"># trace produces a ScriptModule's conv1 and conv2</span>
<span class="bp">self</span><span class="o">.</span><span class="n">conv1</span> <span class="o">=</span> <span class="n">trace</span><span class="p">(</span><span class="n">nn</span><span class="o">.</span><span class="n">Conv2d</span><span class="p">(</span><span class="mi">1</span><span class="p">,</span> <span class="mi">20</span><span class="p">,</span> <span class="mi">5</span><span class="p">),</span> <span class="n">torch</span><span class="o">.</span><span class="n">rand</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="mi">16</span><span class="p">,</span> <span class="mi">16</span><span class="p">))</span>
<span class="bp">self</span><span class="o">.</span><span class="n">conv2</span> <span class="o">=</span> <span class="n">trace</span><span class="p">(</span><span class="n">nn</span><span class="o">.</span><span class="n">Conv2d</span><span class="p">(</span><span class="mi">20</span><span class="p">,</span> <span class="mi">20</span><span class="p">,</span> <span class="mi">5</span><span class="p">),</span> <span class="n">torch</span><span class="o">.</span><span class="n">rand</span><span class="p">(</span><span class="mi">1</span><span class="p">,</span> <span class="mi">20</span><span class="p">,</span> <span class="mi">16</span><span class="p">,</span> <span class="mi">16</span><span class="p">))</span>
<span class="nd">@script_method</span>
<span class="k">def</span> <span class="nf">forward</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="nb">input</span><span class="p">):</span>
<span class="nb">input</span> <span class="o">=</span> <span class="n">F</span><span class="o">.</span><span class="n">relu</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">conv1</span><span class="p">(</span><span class="nb">input</span><span class="p">))</span>
<span class="nb">input</span> <span class="o">=</span> <span class="n">F</span><span class="o">.</span><span class="n">relu</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">conv2</span><span class="p">(</span><span class="nb">input</span><span class="p">))</span>
<span class="k">return</span> <span class="nb">input</span>
</pre></div>
</div>
</div></blockquote>
<dl class="method">
<dt id="torch.jit.ScriptModule.save">
<code class="descname">save</code><span class="sig-paren">(</span><em>filename</em><span class="sig-paren">)</span><a class="headerlink" href="#torch.jit.ScriptModule.save" title="Permalink to this definition">¶</a></dt>
<dd><p>Save an offline version of this module for use in a separate process. The saved
module serializes all of the methods and parameters of this module. It can be
loaded into the C++ API using <code class="docutils literal notranslate"><span class="pre">torch::jit::load(filename)</span></code> or into the Python
API with <code class="docutils literal notranslate"><span class="pre">torch.jit.load(filename)</span></code>.</p>
<p>To be able to save a module, it must not make any calls to native python functions.
This means that all submodules must be subclasses of ScriptModules as well.</p>
<div class="admonition danger">
<p class="first admonition-title">Danger</p>
<p class="last">All modules, no matter their device, are always loaded onto the CPU during loading.
This is different from <a class="reference internal" href="torch.html#torch.load" title="torch.load"><code class="xref py py-func docutils literal notranslate"><span class="pre">torch.load()</span></code></a>’s semantics and may change in the future.</p>
</div>
</dd></dl>
</dd></dl>
<dl class="function">
<dt id="torch.jit.load">
<code class="descclassname">torch.jit.</code><code class="descname">load</code><span class="sig-paren">(</span><em>f</em>, <em>map_location=None</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/jit.html#load"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.jit.load" title="Permalink to this definition">¶</a></dt>
<dd><p>Load a <code class="docutils literal notranslate"><span class="pre">ScriptModule</span></code> previously saved with <code class="xref py py-func docutils literal notranslate"><span class="pre">save</span></code></p>
<p>All previously saved modules, no matter their device, are first loaded onto CPU,
and then are moved to the devices they were saved from. If this fails (e.g. because
the run time system doesn’t have certain devices), an exception is raised.
However, storages can be dynamically remapped to an alternative set of devices
using the <cite>map_location</cite> argument. Comparing to <a class="reference internal" href="torch.html#torch.load" title="torch.load"><code class="xref py py-func docutils literal notranslate"><span class="pre">torch.load()</span></code></a>, <cite>map_location</cite>
in this function is simplified, which only accepts a string (e.g., ‘cpu’, ‘cuda:0’),
or torch.device (e.g., torch.device(‘cpu’))</p>
<table class="docutils field-list" frame="void" rules="none">
<col class="field-name" />
<col class="field-body" />
<tbody valign="top">
<tr class="field-odd field"><th class="field-name">Parameters:</th><td class="field-body"><ul class="first simple">
<li><strong>f</strong> – a file-like object (has to implement read, readline, tell, and seek),
or a string containing a file name</li>
<li><strong>map_location</strong> – can a string (e.g., ‘cpu’, ‘cuda:0’), a device (e.g.,
torch.device(‘cpu’))</li>
</ul>
</td>
</tr>
<tr class="field-even field"><th class="field-name">Returns:</th><td class="field-body"><p class="first last">A <code class="docutils literal notranslate"><span class="pre">ScriptModule</span></code> object.</p>
</td>
</tr>
</tbody>
</table>
<p class="rubric">Example</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="n">torch</span><span class="o">.</span><span class="n">jit</span><span class="o">.</span><span class="n">load</span><span class="p">(</span><span class="s1">'scriptmodule.pt'</span><span class="p">)</span>
<span class="go"># Load ScriptModule from io.BytesIO object</span>
<span class="gp">>>> </span><span class="k">with</span> <span class="nb">open</span><span class="p">(</span><span class="s1">'scriptmodule.pt'</span><span class="p">,</span> <span class="s1">'rb'</span><span class="p">)</span> <span class="k">as</span> <span class="n">f</span><span class="p">:</span>
<span class="go"> buffer = io.BytesIO(f.read())</span>
<span class="go"># Load all tensors to the original device</span>
<span class="gp">>>> </span><span class="n">torch</span><span class="o">.</span><span class="n">jit</span><span class="o">.</span><span class="n">load</span><span class="p">(</span><span class="n">buffer</span><span class="p">)</span>
<span class="go"># Load all tensors onto CPU, using a device</span>
<span class="gp">>>> </span><span class="n">torch</span><span class="o">.</span><span class="n">jit</span><span class="o">.</span><span class="n">load</span><span class="p">(</span><span class="n">buffer</span><span class="p">,</span> <span class="n">map_location</span><span class="o">=</span><span class="n">torch</span><span class="o">.</span><span class="n">device</span><span class="p">(</span><span class="s1">'cpu'</span><span class="p">))</span>
<span class="go"># Load all tensors onto CPU, using a string</span>
<span class="gp">>>> </span><span class="n">torch</span><span class="o">.</span><span class="n">jit</span><span class="o">.</span><span class="n">load</span><span class="p">(</span><span class="n">buffer</span><span class="p">,</span> <span class="n">map_location</span><span class="o">=</span><span class="s1">'cpu'</span><span class="p">)</span>
</pre></div>
</div>
</dd></dl>
<dl class="function">
<dt id="torch.jit.trace">
<code class="descclassname">torch.jit.</code><code class="descname">trace</code><span class="sig-paren">(</span><em>func</em>, <em>example_inputs</em>, <em>optimize=True</em>, <em>check_trace=True</em>, <em>check_inputs=None</em>, <em>check_tolerance=1e-05</em>, <em>_force_outplace=False</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/jit.html#trace"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.jit.trace" title="Permalink to this definition">¶</a></dt>
<dd><p>Trace a function and return an executable trace that will be optimized
using just-in-time compilation.</p>
<div class="admonition warning">
<p class="first admonition-title">Warning</p>
<p class="last">Tracing only correctly records functions and modules which are not data
dependent (e.g., have conditionals on data in tensors) and do not have
any untracked external dependencies (e.g., perform input/output or
access global variables). If you trace such models, you may silently get
incorrect results on subsequent invocations of the model. The tracer
will try to emit warnings when doing something that may cause an
incorrect trace to be produced.</p>
</div>
<table class="docutils field-list" frame="void" rules="none">
<col class="field-name" />
<col class="field-body" />
<tbody valign="top">
<tr class="field-odd field"><th class="field-name">Parameters:</th><td class="field-body"><ul class="first simple">
<li><strong>func</strong> (<em>callable</em><em> or </em><a class="reference internal" href="nn.html#torch.nn.Module" title="torch.nn.Module"><em>torch.nn.Module</em></a>) – a python function or torch.nn.Module
that will be run with example_inputs.
arguments and returns to func must be Tensors
or (possibly nested) tuples that
contain tensors.</li>
<li><strong>example_inputs</strong> (<a class="reference external" href="https://docs.python.org/3/library/stdtypes.html#tuple" title="(in Python v3.7)"><em>tuple</em></a>) – a tuple of example inputs that will be passed to the function
while tracing. The resulting trace can be run with
inputs of different types and shapes assuming the traced operations
support those types and shapes. example_inputs may also be a single
Tensor in which case it is automatically wrapped in a tuple</li>
</ul>
</td>
</tr>
<tr class="field-even field"><th class="field-name" colspan="2">Keyword Arguments:</th></tr>
<tr class="field-even field"><td> </td><td class="field-body"><ul class="first simple">
<li><strong>optimize</strong> (<a class="reference external" href="https://docs.python.org/3/library/functions.html#bool" title="(in Python v3.7)"><em>bool</em></a><em>, </em><em>optional</em>) – whether or not to apply optimizations. Default: <code class="docutils literal notranslate"><span class="pre">True</span></code>.</li>
<li><strong>check_trace</strong> (<a class="reference external" href="https://docs.python.org/3/library/functions.html#bool" title="(in Python v3.7)"><em>bool</em></a><em>, </em><em>optional</em>) – check if the same inputs run through
traced code produce the same outputs. Default: <code class="docutils literal notranslate"><span class="pre">True</span></code>. You might want
to disable this if, for example, your network contains non-
deterministic ops or if you are sure that the network is correct despite
a checker failure.</li>
<li><strong>check_inputs</strong> (<em>list of tuples</em><em>, </em><em>optional</em>) – A list of tuples of input arguments that should be used
to check the trace against what is expected. Each tuple
is equivalent to a seet of input arguments that would
be specified in <code class="docutils literal notranslate"><span class="pre">args</span></code>. For best results, pass in a
set of checking inputs representative of the space of
shapes and types of inputs you expect the network to see.
If not specified, the original <code class="docutils literal notranslate"><span class="pre">args</span></code> is used for checking</li>
<li><strong>check_tolerance</strong> (<a class="reference external" href="https://docs.python.org/3/library/functions.html#float" title="(in Python v3.7)"><em>float</em></a><em>, </em><em>optional</em>) – Floating-point comparison tolerance to use in the checker procedure.
This can be used to relax the checker strictness in the event that
results diverge numerically for a known reason, such as operator fusion.</li>
</ul>
</td>
</tr>
<tr class="field-odd field"><th class="field-name">Returns:</th><td class="field-body"><p class="first last">A <code class="docutils literal notranslate"><span class="pre">ScriptModule</span></code> object with a single <code class="docutils literal notranslate"><span class="pre">forward()</span></code> method containing the traced code.
When func is a <code class="docutils literal notranslate"><span class="pre">torch.nn.Module</span></code>, the returned <code class="docutils literal notranslate"><span class="pre">ScriptModule</span></code> will have the same set of
sub-modules and parameters as func.</p>
</td>
</tr>
</tbody>
</table>
<p class="rubric">Example</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="k">def</span> <span class="nf">f</span><span class="p">(</span><span class="n">x</span><span class="p">):</span>
<span class="gp">... </span> <span class="k">return</span> <span class="n">x</span> <span class="o">*</span> <span class="mi">2</span>
<span class="gp">>>> </span><span class="n">traced_f</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">jit</span><span class="o">.</span><span class="n">trace</span><span class="p">(</span><span class="n">f</span><span class="p">,</span> <span class="n">torch</span><span class="o">.</span><span class="n">rand</span><span class="p">(</span><span class="mi">1</span><span class="p">))</span>
</pre></div>
</div>
</dd></dl>
</div>
<div class="section" id="mixing-tracing-and-scripting">
<h2><a class="toc-backref" href="#id2">Mixing Tracing and Scripting</a><a class="headerlink" href="#mixing-tracing-and-scripting" title="Permalink to this headline">¶</a></h2>
<p>In many cases either tracing or script is an easier approach for converting a model.
We allow you to compose tracing and scripting to suit the particular requirements
of a part of a model.</p>
<p>Scripted functions can call traced ones. This is particularly useful when you need
to use control-flow around a simple feed-forward model. For instance the beam search
of a sequence to sequence model will typically be written in script but can call an
encoder module generated using tracing.</p>
<p>Example:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="kn">import</span> <span class="nn">torch</span>
<span class="k">def</span> <span class="nf">foo</span><span class="p">(</span><span class="n">x</span><span class="p">,</span> <span class="n">y</span><span class="p">):</span>
<span class="k">return</span> <span class="mi">2</span> <span class="o">*</span> <span class="n">x</span> <span class="o">+</span> <span class="n">y</span>
<span class="n">traced_foo</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">jit</span><span class="o">.</span><span class="n">trace</span><span class="p">(</span><span class="n">foo</span><span class="p">,</span> <span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">rand</span><span class="p">(</span><span class="mi">3</span><span class="p">),</span> <span class="n">torch</span><span class="o">.</span><span class="n">rand</span><span class="p">(</span><span class="mi">3</span><span class="p">)))</span>
<span class="nd">@torch</span><span class="o">.</span><span class="n">jit</span><span class="o">.</span><span class="n">script</span>
<span class="k">def</span> <span class="nf">bar</span><span class="p">(</span><span class="n">x</span><span class="p">):</span>
<span class="k">return</span> <span class="n">traced_foo</span><span class="p">(</span><span class="n">x</span><span class="p">,</span> <span class="n">x</span><span class="p">)</span>
</pre></div>
</div>
<p>Traced functions can call script functions. This is useful when a small part of
a model requires some control-flow even though most of the model is just a feed-forward
network. Control-flow inside of a script function called by a traced function is
preserved correctly:</p>
<p>Example:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="kn">import</span> <span class="nn">torch</span>
<span class="nd">@torch</span><span class="o">.</span><span class="n">jit</span><span class="o">.</span><span class="n">script</span>
<span class="k">def</span> <span class="nf">foo</span><span class="p">(</span><span class="n">x</span><span class="p">,</span> <span class="n">y</span><span class="p">):</span>
<span class="k">if</span> <span class="n">x</span><span class="o">.</span><span class="n">max</span><span class="p">()</span> <span class="o">></span> <span class="n">y</span><span class="o">.</span><span class="n">max</span><span class="p">():</span>
<span class="n">r</span> <span class="o">=</span> <span class="n">x</span>
<span class="k">else</span><span class="p">:</span>
<span class="n">r</span> <span class="o">=</span> <span class="n">y</span>
<span class="k">return</span> <span class="n">r</span>
<span class="k">def</span> <span class="nf">bar</span><span class="p">(</span><span class="n">x</span><span class="p">,</span> <span class="n">y</span><span class="p">,</span> <span class="n">z</span><span class="p">):</span>
<span class="k">return</span> <span class="n">foo</span><span class="p">(</span><span class="n">x</span><span class="p">,</span> <span class="n">y</span><span class="p">)</span> <span class="o">+</span> <span class="n">z</span>
<span class="n">traced_bar</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">jit</span><span class="o">.</span><span class="n">trace</span><span class="p">(</span><span class="n">bar</span><span class="p">,</span> <span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">rand</span><span class="p">(</span><span class="mi">3</span><span class="p">),</span> <span class="n">torch</span><span class="o">.</span><span class="n">rand</span><span class="p">(</span><span class="mi">3</span><span class="p">),</span> <span class="n">torch</span><span class="o">.</span><span class="n">rand</span><span class="p">(</span><span class="mi">3</span><span class="p">))</span>
</pre></div>
</div>
<p>This composition also works for modules as well, where it can be used to generate
a submodule using tracing that can be called from the methods of a script module:</p>
<p>Example:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="kn">import</span> <span class="nn">torch</span>
<span class="kn">import</span> <span class="nn">torchvision</span>
<span class="k">class</span> <span class="nc">MyScriptModule</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">jit</span><span class="o">.</span><span class="n">ScriptModule</span><span class="p">):</span>
<span class="k">def</span> <span class="nf">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="nb">super</span><span class="p">(</span><span class="n">MyScriptModule</span><span class="p">,</span> <span class="bp">self</span><span class="p">)</span><span class="o">.</span><span class="fm">__init__</span><span class="p">()</span>
<span class="bp">self</span><span class="o">.</span><span class="n">means</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">nn</span><span class="o">.</span><span class="n">Parameter</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">tensor</span><span class="p">([</span><span class="mf">103.939</span><span class="p">,</span> <span class="mf">116.779</span><span class="p">,</span> <span class="mf">123.68</span><span class="p">])</span>
<span class="o">.</span><span class="n">resize_</span><span class="p">(</span><span class="mi">1</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="mi">1</span><span class="p">))</span>
<span class="bp">self</span><span class="o">.</span><span class="n">resnet</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">jit</span><span class="o">.</span><span class="n">trace</span><span class="p">(</span><span class="n">torchvision</span><span class="o">.</span><span class="n">models</span><span class="o">.</span><span class="n">resnet18</span><span class="p">(),</span>
<span class="n">torch</span><span class="o">.</span><span class="n">rand</span><span class="p">(</span><span class="mi">1</span><span class="p">,</span> <span class="mi">3</span><span class="p">,</span> <span class="mi">224</span><span class="p">,</span> <span class="mi">224</span><span class="p">))</span>
<span class="nd">@torch</span><span class="o">.</span><span class="n">jit</span><span class="o">.</span><span class="n">script_method</span>
<span class="k">def</span> <span class="nf">forward</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="nb">input</span><span class="p">):</span>
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">resnet</span><span class="p">(</span><span class="nb">input</span> <span class="o">-</span> <span class="bp">self</span><span class="o">.</span><span class="n">means</span><span class="p">)</span>
</pre></div>
</div>
</div>
<div class="section" id="torch-script-language-reference">
<h2><a class="toc-backref" href="#id3">Torch Script Language Reference</a><a class="headerlink" href="#torch-script-language-reference" title="Permalink to this headline">¶</a></h2>
<p>Torch Script is a subset of Python that can either be written directly (using
the @script annotations) or generated automatically from Python code via
tracing. When using tracing, code is automatically converted into this subset of
Python by recording only the actual operators on tensors and simply executing and
discarding the other surrounding Python code.</p>
<p>When writing Torch Script directly using @script annotations, the programmer must
only use the subset of Python supported in Torch Script. This section documents
what is supported in Torch Script as if it were a language reference for a stand
alone language. Any features of Python not mentioned in this reference are not
part of Torch Script.</p>
<p>As a subset of Python any valid Torch Script function is also a valid Python
function. This makes it possible to remove the @script annotations and debug the
function using standard Python tools like pdb. The reverse is not true: there
are many valid python programs that are not valid Torch Script programs.
Instead, Torch Script focuses specifically on the features of Python that are
needed to represent neural network models in Torch.</p>
<dl class="envvar">
<dt id="envvar-PYTORCH_JIT=1">
<code class="descname">PYTORCH_JIT=1</code><a class="headerlink" href="#envvar-PYTORCH_JIT=1" title="Permalink to this definition">¶</a></dt>
<dd><p>Setting the environment variable <code class="docutils literal notranslate"><span class="pre">PYTORCH_JIT=0</span></code> will disable all script
and tracing annotations. If there is hard-to-debug error in one of your
ScriptModules, you can use this flag to force everything to run using native
Python. This allows the use of tools like <code class="docutils literal notranslate"><span class="pre">pdb</span></code> to debug code.</p>
</dd></dl>
<div class="section" id="types">
<h3><a class="toc-backref" href="#id4">Types</a><a class="headerlink" href="#types" title="Permalink to this headline">¶</a></h3>
<p>The largest difference between Torch Script and the full Python language is that
Torch Script only support a small set of types that are needed to express neural
net models. In particular Torch Script supports:</p>
<dl class="docutils">
<dt><code class="docutils literal notranslate"><span class="pre">Tensor</span></code></dt>
<dd>A PyTorch tensor of any dtype, dimension, or backend.</dd>
<dt><code class="docutils literal notranslate"><span class="pre">Tuple[T0,</span> <span class="pre">T1,</span> <span class="pre">...]</span></code></dt>
<dd>A tuple containing subtypes <code class="docutils literal notranslate"><span class="pre">T0</span></code>, <code class="docutils literal notranslate"><span class="pre">T1</span></code>, etc. (e.g. <code class="docutils literal notranslate"><span class="pre">Tuple[Tensor,</span> <span class="pre">Tensor]</span></code>)</dd>
<dt><code class="docutils literal notranslate"><span class="pre">int</span></code></dt>
<dd>A scalar integer</dd>
<dt><code class="docutils literal notranslate"><span class="pre">float</span></code></dt>
<dd>A scalar floating point number</dd>
<dt><code class="docutils literal notranslate"><span class="pre">List[T]</span></code></dt>
<dd>A list of which all members are type <code class="docutils literal notranslate"><span class="pre">T</span></code></dd>
</dl>
<p>Unlike Python, each variable in Torch Script function must have a single static type.
This makes it easier to optimize Torch Script functions.</p>
<p>Example:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="nd">@torch</span><span class="o">.</span><span class="n">jit</span><span class="o">.</span><span class="n">script</span>
<span class="k">def</span> <span class="nf">an_error</span><span class="p">(</span><span class="n">x</span><span class="p">):</span>
<span class="k">if</span> <span class="n">x</span><span class="p">:</span>
<span class="n">r</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">rand</span><span class="p">(</span><span class="mi">1</span><span class="p">)</span>
<span class="k">else</span><span class="p">:</span>
<span class="n">r</span> <span class="o">=</span> <span class="mi">4</span>
<span class="k">return</span> <span class="n">r</span> <span class="c1"># Type mismatch: r is set to type Tensor in the true branch</span>
<span class="c1"># and type int in the false branch</span>
</pre></div>
</div>
<p>By default, all parameters to a Torch Script function are assumed to be Tensor
because this is the most common type used in modules. To specify that an
argument to a Torch Script function is another type, it is possible to use
MyPy-style type annotations using the types listed above:</p>
<p>Example:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="nd">@torch</span><span class="o">.</span><span class="n">jit</span><span class="o">.</span><span class="n">script</span>
<span class="k">def</span> <span class="nf">foo</span><span class="p">(</span><span class="n">x</span><span class="p">,</span> <span class="n">tup</span><span class="p">):</span>
<span class="c1"># type: (int, Tuple[Tensor, Tensor]) -> Tensor</span>
<span class="n">t0</span><span class="p">,</span> <span class="n">t1</span> <span class="o">=</span> <span class="n">tup</span>
<span class="k">return</span> <span class="n">t0</span> <span class="o">+</span> <span class="n">t1</span> <span class="o">+</span> <span class="n">x</span>
<span class="nb">print</span><span class="p">(</span><span class="n">foo</span><span class="p">(</span><span class="mi">3</span><span class="p">,</span> <span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">rand</span><span class="p">(</span><span class="mi">3</span><span class="p">),</span> <span class="n">torch</span><span class="o">.</span><span class="n">rand</span><span class="p">(</span><span class="mi">3</span><span class="p">))))</span>
</pre></div>
</div>
<div class="admonition note">
<p class="first admonition-title">Note</p>
<p class="last">It is also possible to annotate types with Python 3 type annotations.
In our examples, we use comment-based annotations to ensure Python 2
compatibility as well.</p>
</div>
</div>
<div class="section" id="expressions">
<h3><a class="toc-backref" href="#id5">Expressions</a><a class="headerlink" href="#expressions" title="Permalink to this headline">¶</a></h3>
<p>The following Python Expressions are supported</p>
<dl class="docutils">
<dt>Literals</dt>
<dd><code class="docutils literal notranslate"><span class="pre">True</span></code>, <code class="docutils literal notranslate"><span class="pre">False</span></code>, <code class="docutils literal notranslate"><span class="pre">None</span></code>, <code class="docutils literal notranslate"><span class="pre">'string</span> <span class="pre">literals'</span></code>, <code class="docutils literal notranslate"><span class="pre">"string</span> <span class="pre">literals"</span></code>,
number literals <code class="docutils literal notranslate"><span class="pre">3</span></code> (interpreted as int) <code class="docutils literal notranslate"><span class="pre">3.4</span></code> (interpreter as a float)</dd>
<dt>Variables</dt>
<dd><p class="first"><code class="docutils literal notranslate"><span class="pre">a</span></code></p>
<div class="last admonition note">
<p class="first admonition-title">Note</p>
<p class="last">See <a class="reference internal" href="#variable-resolution">Variable Resolution</a> for how variables are resolved.</p>
</div>
</dd>
<dt>Tuple Construction</dt>
<dd><code class="docutils literal notranslate"><span class="pre">(3,</span> <span class="pre">4)</span></code>, <code class="docutils literal notranslate"><span class="pre">(3,)</span></code></dd>
<dt>List Construction</dt>
<dd><p class="first"><code class="docutils literal notranslate"><span class="pre">[3,</span> <span class="pre">4]</span></code>, <code class="docutils literal notranslate"><span class="pre">[]</span></code>, <code class="docutils literal notranslate"><span class="pre">[torch.rand(3),</span> <span class="pre">torch.rand(4)]</span></code></p>
<div class="last admonition note">
<p class="first admonition-title">Note</p>
<p class="last">an empty list is assumed have type <code class="docutils literal notranslate"><span class="pre">List[Tensor]</span></code>.
The types of other list literals are derived from the type of the members.</p>
</div>
</dd>
<dt>Arithmetic Operators</dt>
<dd><code class="docutils literal notranslate"><span class="pre">a</span> <span class="pre">+</span> <span class="pre">b</span></code>
<code class="docutils literal notranslate"><span class="pre">a</span> <span class="pre">-</span> <span class="pre">b</span></code>
<code class="docutils literal notranslate"><span class="pre">a</span> <span class="pre">*</span> <span class="pre">b</span></code>
<code class="docutils literal notranslate"><span class="pre">a</span> <span class="pre">/</span> <span class="pre">b</span></code>
<code class="docutils literal notranslate"><span class="pre">a</span> <span class="pre">^</span> <span class="pre">b</span></code>
<code class="docutils literal notranslate"><span class="pre">a</span> <span class="pre">@</span> <span class="pre">b</span></code></dd>
<dt>Comparison Operators</dt>
<dd><code class="docutils literal notranslate"><span class="pre">a</span> <span class="pre">==</span> <span class="pre">b</span></code>
<code class="docutils literal notranslate"><span class="pre">a</span> <span class="pre">!=</span> <span class="pre">b</span></code>
<code class="docutils literal notranslate"><span class="pre">a</span> <span class="pre"><</span> <span class="pre">b</span></code>
<code class="docutils literal notranslate"><span class="pre">a</span> <span class="pre">></span> <span class="pre">b</span></code>
<code class="docutils literal notranslate"><span class="pre">a</span> <span class="pre"><=</span> <span class="pre">b</span></code>
<code class="docutils literal notranslate"><span class="pre">a</span> <span class="pre">>=</span> <span class="pre">b</span></code></dd>
<dt>Logical Operators</dt>
<dd><code class="docutils literal notranslate"><span class="pre">a</span> <span class="pre">and</span> <span class="pre">b</span></code>
<code class="docutils literal notranslate"><span class="pre">a</span> <span class="pre">or</span> <span class="pre">b</span></code>
<code class="docutils literal notranslate"><span class="pre">not</span> <span class="pre">b</span></code></dd>
<dt>Subscripts</dt>
<dd><p class="first"><code class="docutils literal notranslate"><span class="pre">t[0]</span></code>
<code class="docutils literal notranslate"><span class="pre">t[-1]</span></code>
<code class="docutils literal notranslate"><span class="pre">t[0:2]</span></code>
<code class="docutils literal notranslate"><span class="pre">t[1:]</span></code>
<code class="docutils literal notranslate"><span class="pre">t[:1]</span></code>
<code class="docutils literal notranslate"><span class="pre">t[:]</span></code>
<code class="docutils literal notranslate"><span class="pre">t[0,</span> <span class="pre">1]</span></code>
<code class="docutils literal notranslate"><span class="pre">t[0,</span> <span class="pre">1:2]</span></code>
<code class="docutils literal notranslate"><span class="pre">t[0,</span> <span class="pre">:1]</span></code>
<code class="docutils literal notranslate"><span class="pre">t[-1,</span> <span class="pre">1:,</span> <span class="pre">0]</span></code>
<code class="docutils literal notranslate"><span class="pre">t[1:,</span> <span class="pre">-1,</span> <span class="pre">0]</span></code>
<code class="docutils literal notranslate"><span class="pre">t[i:j,</span> <span class="pre">i]</span></code></p>
<div class="last admonition note">
<p class="first admonition-title">Note</p>
<p class="last">Torch Script currently does not support mutating tensors in place, so any
tensor indexing can only appear on the right-hand size of an expression.</p>
</div>
</dd>
<dt>Function calls</dt>
<dd><p class="first">Calls to built-in functions: <code class="docutils literal notranslate"><span class="pre">torch.rand(3,</span> <span class="pre">dtype=torch.int)</span></code></p>
<p>Calls to other script functions:</p>
<div class="last highlight-default notranslate"><div class="highlight"><pre><span></span><span class="kn">import</span> <span class="nn">torch</span>
<span class="nd">@torch</span><span class="o">.</span><span class="n">jit</span><span class="o">.</span><span class="n">script</span>
<span class="k">def</span> <span class="nf">foo</span><span class="p">(</span><span class="n">x</span><span class="p">):</span>
<span class="k">return</span> <span class="n">x</span> <span class="o">+</span> <span class="mi">1</span>
<span class="nd">@torch</span><span class="o">.</span><span class="n">jit</span><span class="o">.</span><span class="n">script</span>
<span class="k">def</span> <span class="nf">bar</span><span class="p">(</span><span class="n">x</span><span class="p">):</span>
<span class="k">return</span> <span class="n">foo</span><span class="p">(</span><span class="n">x</span><span class="p">)</span>
</pre></div>
</div>
</dd>
<dt>Method calls</dt>
<dd><p class="first">Calls to methods of builtin types like tensor: <code class="docutils literal notranslate"><span class="pre">x.mm(y)</span></code></p>
<p>When defining a Script method inside of a ScriptModule, the <code class="docutils literal notranslate"><span class="pre">@script_method</span></code>
annotation is used. Inside of these methods it is possible to call other methods
of this class or access methods on the submodules.</p>
<p>Calling a submodule directly (e.g. <code class="docutils literal notranslate"><span class="pre">self.resnet(input)</span></code>) is equivalent to
calling its <code class="docutils literal notranslate"><span class="pre">forward</span></code> method (e.g. <code class="docutils literal notranslate"><span class="pre">self.resnet.forward(input)</span></code>)</p>
<div class="last highlight-default notranslate"><div class="highlight"><pre><span></span><span class="kn">import</span> <span class="nn">torch</span>
<span class="k">class</span> <span class="nc">MyScriptModule</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">jit</span><span class="o">.</span><span class="n">ScriptModule</span><span class="p">):</span>
<span class="k">def</span> <span class="nf">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="nb">super</span><span class="p">(</span><span class="n">MyScriptModule</span><span class="p">,</span> <span class="bp">self</span><span class="p">)</span><span class="o">.</span><span class="fm">__init__</span><span class="p">()</span>
<span class="bp">self</span><span class="o">.</span><span class="n">means</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">nn</span><span class="o">.</span><span class="n">Parameter</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">tensor</span><span class="p">([</span><span class="mf">103.939</span><span class="p">,</span> <span class="mf">116.779</span><span class="p">,</span> <span class="mf">123.68</span><span class="p">])</span>
<span class="o">.</span><span class="n">resize_</span><span class="p">(</span><span class="mi">1</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="mi">1</span><span class="p">))</span>
<span class="bp">self</span><span class="o">.</span><span class="n">resnet</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">jit</span><span class="o">.</span><span class="n">trace</span><span class="p">(</span><span class="n">torchvision</span><span class="o">.</span><span class="n">models</span><span class="o">.</span><span class="n">resnet18</span><span class="p">(),</span>
<span class="n">torch</span><span class="o">.</span><span class="n">rand</span><span class="p">(</span><span class="mi">1</span><span class="p">,</span> <span class="mi">3</span><span class="p">,</span> <span class="mi">224</span><span class="p">,</span> <span class="mi">224</span><span class="p">))</span>
<span class="nd">@torch</span><span class="o">.</span><span class="n">jit</span><span class="o">.</span><span class="n">script_method</span>
<span class="k">def</span> <span class="nf">helper</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="nb">input</span><span class="p">):</span>
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">resnet</span><span class="p">(</span><span class="nb">input</span> <span class="o">-</span> <span class="bp">self</span><span class="o">.</span><span class="n">means</span><span class="p">)</span>
<span class="nd">@torch</span><span class="o">.</span><span class="n">jit</span><span class="o">.</span><span class="n">script_method</span>
<span class="k">def</span> <span class="nf">forward</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="nb">input</span><span class="p">):</span>
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">helper</span><span class="p">(</span><span class="nb">input</span><span class="p">)</span>
</pre></div>
</div>
</dd>
<dt>If expressions</dt>
<dd><code class="docutils literal notranslate"><span class="pre">x</span> <span class="pre">if</span> <span class="pre">x</span> <span class="pre">></span> <span class="pre">y</span> <span class="pre">else</span> <span class="pre">y</span></code></dd>
<dt>Casts</dt>
<dd><code class="docutils literal notranslate"><span class="pre">float(ten)</span></code>, <code class="docutils literal notranslate"><span class="pre">int(3.5)</span></code>, <code class="docutils literal notranslate"><span class="pre">bool(ten)</span></code></dd>
<dt>Accessing Module Parameters</dt>
<dd><code class="docutils literal notranslate"><span class="pre">self.my_parameter</span></code> <code class="docutils literal notranslate"><span class="pre">self.my_submodule.my_parameter</span></code></dd>
</dl>
</div>
<div class="section" id="statements">
<h3><a class="toc-backref" href="#id6">Statements</a><a class="headerlink" href="#statements" title="Permalink to this headline">¶</a></h3>
<p>Torch Script supports the following types of statements:</p>
<p>Simple Assignments</p>
<blockquote>
<div><div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">a</span> <span class="o">=</span> <span class="n">b</span>
<span class="n">a</span> <span class="o">+=</span> <span class="n">b</span> <span class="c1"># short-hand for a = a + b, does not operate in-place on a</span>
<span class="n">a</span> <span class="o">-=</span> <span class="n">b</span>
</pre></div>
</div>
</div></blockquote>
<p>Pattern Matching Assignments</p>
<blockquote>
<div><div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">a</span><span class="p">,</span> <span class="n">b</span> <span class="o">=</span> <span class="n">tuple_or_list</span>
<span class="n">a</span><span class="p">,</span> <span class="n">b</span><span class="p">,</span> <span class="o">*</span><span class="n">c</span> <span class="o">=</span> <span class="n">a_tuple</span>
</pre></div>
</div>
</div></blockquote>
<p>Print Statements</p>
<blockquote>
<div><code class="docutils literal notranslate"><span class="pre">print("the</span> <span class="pre">result</span> <span class="pre">of</span> <span class="pre">an</span> <span class="pre">add:",</span> <span class="pre">a</span> <span class="pre">+</span> <span class="pre">b)</span></code></div></blockquote>
<p>If Statements</p>
<blockquote>
<div><div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="k">if</span> <span class="n">a</span> <span class="o"><</span> <span class="mi">4</span><span class="p">:</span>
<span class="n">r</span> <span class="o">=</span> <span class="o">-</span><span class="n">a</span>
<span class="k">elif</span> <span class="n">a</span> <span class="o"><</span> <span class="mi">3</span><span class="p">:</span>
<span class="n">r</span> <span class="o">=</span> <span class="n">a</span> <span class="o">+</span> <span class="n">a</span>
<span class="k">else</span><span class="p">:</span>
<span class="n">r</span> <span class="o">=</span> <span class="mi">3</span> <span class="o">*</span> <span class="n">a</span>
</pre></div>
</div>
</div></blockquote>
<p>While Loops</p>
<blockquote>
<div><div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">a</span> <span class="o">=</span> <span class="mi">0</span>
<span class="k">while</span> <span class="n">a</span> <span class="o"><</span> <span class="mi">4</span><span class="p">:</span>
<span class="nb">print</span><span class="p">(</span><span class="n">a</span><span class="p">)</span>
<span class="n">a</span> <span class="o">+=</span> <span class="mi">1</span>
</pre></div>
</div>
</div></blockquote>
<p>For loops with <code class="docutils literal notranslate"><span class="pre">range</span></code></p>
<blockquote>
<div><div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">x</span> <span class="o">=</span> <span class="mi">0</span>
<span class="k">for</span> <span class="n">i</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="mi">10</span><span class="p">):</span>
<span class="n">x</span> <span class="o">*=</span> <span class="n">i</span>
</pre></div>
</div>
<div class="admonition note">
<p class="first admonition-title">Note</p>
<p class="last">Script currently does not support iterating over generic iterable
objects like lists or tensors. Script currently does not support start or
increment parameters to range. These will be added in a future version.</p>
</div>
</div></blockquote>
<p>For loops over tuples:</p>
<blockquote>
<div><div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">tup</span> <span class="o">=</span> <span class="p">(</span><span class="mi">3</span><span class="p">,</span> <span class="n">torch</span><span class="o">.</span><span class="n">rand</span><span class="p">(</span><span class="mi">4</span><span class="p">))</span>
<span class="k">for</span> <span class="n">x</span> <span class="ow">in</span> <span class="n">tup</span><span class="p">:</span>
<span class="nb">print</span><span class="p">(</span><span class="n">x</span><span class="p">)</span>
</pre></div>
</div>
<div class="admonition note">
<p class="first admonition-title">Note</p>
<p class="last">for loops over tuples will unroll the loop, generating a body for
each member of the tuple. The body must type-check correctly for each member.</p>
</div>
</div></blockquote>
<p>For loops over constant <code class="docutils literal notranslate"><span class="pre">torch.nn.ModuleList</span></code></p>
<blockquote>
<div><div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="k">class</span> <span class="nc">SubModule</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">jit</span><span class="o">.</span><span class="n">ScriptModule</span><span class="p">):</span>
<span class="k">def</span> <span class="nf">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="nb">super</span><span class="p">(</span><span class="n">Sub</span><span class="p">,</span> <span class="bp">self</span><span class="p">)</span><span class="o">.</span><span class="fm">__init__</span><span class="p">()</span>
<span class="bp">self</span><span class="o">.</span><span class="n">weight</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">Parameter</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">randn</span><span class="p">(</span><span class="mi">2</span><span class="p">))</span>
<span class="nd">@torch</span><span class="o">.</span><span class="n">jit</span><span class="o">.</span><span class="n">script_method</span>
<span class="k">def</span> <span class="nf">forward</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="nb">input</span><span class="p">):</span>
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">weight</span> <span class="o">+</span> <span class="nb">input</span>
<span class="k">class</span> <span class="nc">MyModule</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">jit</span><span class="o">.</span><span class="n">ScriptModule</span><span class="p">):</span>
<span class="n">__constants__</span> <span class="o">=</span> <span class="p">[</span><span class="s1">'mods'</span><span class="p">]</span>
<span class="k">def</span> <span class="nf">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="nb">super</span><span class="p">(</span><span class="n">MyModule</span><span class="p">,</span> <span class="bp">self</span><span class="p">)</span><span class="o">.</span><span class="fm">__init__</span><span class="p">()</span>
<span class="bp">self</span><span class="o">.</span><span class="n">mods</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">nn</span><span class="o">.</span><span class="n">ModuleList</span><span class="p">([</span><span class="n">SubModule</span><span class="p">()</span> <span class="k">for</span> <span class="n">i</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="mi">10</span><span class="p">)])</span>
<span class="nd">@torch</span><span class="o">.</span><span class="n">jit</span><span class="o">.</span><span class="n">script_method</span>
<span class="k">def</span> <span class="nf">forward</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">v</span><span class="p">):</span>
<span class="k">for</span> <span class="n">module</span> <span class="ow">in</span> <span class="bp">self</span><span class="o">.</span><span class="n">mods</span><span class="p">:</span>
<span class="n">v</span> <span class="o">=</span> <span class="n">m</span><span class="p">(</span><span class="n">v</span><span class="p">)</span>
<span class="k">return</span> <span class="n">v</span>
</pre></div>
</div>
<div class="admonition note">
<p class="first admonition-title">Note</p>
<p class="last">To use a module list inside a <code class="docutils literal notranslate"><span class="pre">@script_method</span></code> it must be marked
constant by adding the name of the attribute to the <code class="docutils literal notranslate"><span class="pre">__constants__</span></code>
list for the type. For loops over a ModuleList will unroll the body of the
loop at compile time, with each member of the constant module list.</p>
</div>
</div></blockquote>
<dl class="docutils">
<dt>Return</dt>
<dd><p class="first"><code class="docutils literal notranslate"><span class="pre">return</span> <span class="pre">a,</span> <span class="pre">b</span></code></p>
<div class="last admonition note">
<p class="first admonition-title">Note</p>
<p class="last">there must be a return statement as the last member of the function
and return statements cannot appear anywhere else in the function. This
restriction will be removed in the future.</p>
</div>
</dd>
</dl>
</div>
<div class="section" id="variable-resolution">
<h3><a class="toc-backref" href="#id7">Variable Resolution</a><a class="headerlink" href="#variable-resolution" title="Permalink to this headline">¶</a></h3>
<p>Torch Script supports a subset of Python’s variable resolution (i.e. scoping)
rules. Local variables behave the same as in Python, except for the restriction
that a variable must have the same type along all paths through a function.
If a variable has a different type on different sides of an if statement, it
is an error to use it after the end of the if statement.</p>
<p>Similarly, a variable is not allowed to be used if it is only <em>defined</em> along some
paths through the function.</p>
<p>Example:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="nd">@torch</span><span class="o">.</span><span class="n">jit</span><span class="o">.</span><span class="n">script</span>
<span class="k">def</span> <span class="nf">foo</span><span class="p">(</span><span class="n">x</span><span class="p">):</span>
<span class="k">if</span> <span class="n">x</span> <span class="o"><</span> <span class="mi">0</span><span class="p">:</span>
<span class="n">y</span> <span class="o">=</span> <span class="mi">4</span>
<span class="nb">print</span><span class="p">(</span><span class="n">y</span><span class="p">)</span> <span class="c1"># Error: undefined value y</span>
</pre></div>
</div>
<p>Non-local variables are resolved to Python values at compile time when the
function is defined. These values are then converted into Torch Script values using
the rules described in <a class="reference internal" href="#use-of-python-values">Use of Python Values</a>.</p>
</div>
<div class="section" id="use-of-python-values">
<h3><a class="toc-backref" href="#id8">Use of Python Values</a><a class="headerlink" href="#use-of-python-values" title="Permalink to this headline">¶</a></h3>
<p>To make writing Torch Script more convenient, we allow script code to refer
to Python values in the surrounding scope. For instance, any time there is a
reference to <code class="docutils literal notranslate"><span class="pre">torch</span></code>, the Torch Script compiler is actually resolving it to the
<code class="docutils literal notranslate"><span class="pre">torch</span></code> Python module when the function is declared. These Python values are
not a first class part of Torch Script. Instead they are desugared at compile-time
into the primitive types that Torch Script supports. This section describes the
rules that are used when accessing Python values in Torch Script. They depend
on the dynamic type of the python valued referenced.</p>
<dl class="docutils">
<dt>Functions</dt>
<dd><p class="first">Torch Script can call python functions. This functionality is very useful when
incrementally converting a model into script. The model can be moved function-by-function
to script, leaving calls to Python functions in place. This way you can incrementally
check the correctness of the model as you go.</p>
<p>Example:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="k">def</span> <span class="nf">foo</span><span class="p">(</span><span class="n">x</span><span class="p">):</span>
<span class="nb">print</span><span class="p">(</span><span class="s2">"I am called with </span><span class="si">{}</span><span class="s2">"</span><span class="o">.</span><span class="n">format</span><span class="p">(</span><span class="n">x</span><span class="p">))</span>
<span class="kn">import</span> <span class="nn">pdb</span><span class="p">;</span> <span class="n">pdb</span><span class="o">.</span><span class="n">set_trace</span><span class="p">()</span>
<span class="k">return</span> <span class="n">x</span>
<span class="nd">@torch</span><span class="o">.</span><span class="n">jit</span><span class="o">.</span><span class="n">script</span>
<span class="k">def</span> <span class="nf">bar</span><span class="p">(</span><span class="n">x</span><span class="p">)</span>
<span class="k">return</span> <span class="n">foo</span><span class="p">(</span><span class="n">x</span> <span class="o">+</span> <span class="mi">1</span><span class="p">)</span>
</pre></div>
</div>
<div class="last admonition note">
<p class="first admonition-title">Note</p>
<p class="last">Attempting to call <code class="docutils literal notranslate"><span class="pre">save</span></code> on a ScriptModule that contains calls to Python
functions will fail. The intention is that this pathway is used for debugging
and the calls removed or turned into script functions before saving.</p>
</div>
</dd>
<dt>Attribute Lookup On Python Modules</dt>
<dd>Torch Script can lookup attributes on modules. Builtin functions like <code class="docutils literal notranslate"><span class="pre">torch.add</span></code>
are accessed this way. This allows Torch Script to call functions defined in
other modules.</dd>
<dt>Python-defined Constants</dt>
<dd><p class="first">Torch Script also provides a way to use constants that are defined in Python.
These can be used to hard-code hyper-parameters into the function, or to
define universal constants. There are two ways of specifying that a Python
value should be treated as a constant.</p>
<ol class="arabic">
<li><p class="first">Values looked up as attributes of a module are assumed to be constant.
Example: <code class="docutils literal notranslate"><span class="pre">math.pi</span></code></p>
</li>
<li><p class="first">Attributes of a ScriptModule can be marked constant by listing them
as a member of the <code class="docutils literal notranslate"><span class="pre">__constants__</span></code> property of the class:</p>
<p>Example:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="k">class</span> <span class="nc">Foo</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">jit</span><span class="o">.</span><span class="n">ScriptModule</span><span class="p">):</span>
<span class="n">__constants__</span> <span class="o">=</span> <span class="p">[</span><span class="s1">'a'</span><span class="p">]</span>
<span class="k">def</span> <span class="nf">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="nb">super</span><span class="p">(</span><span class="n">Foo</span><span class="p">,</span> <span class="bp">self</span><span class="p">)</span><span class="o">.</span><span class="fm">__init__</span><span class="p">(</span><span class="kc">False</span><span class="p">)</span>
<span class="bp">self</span><span class="o">.</span><span class="n">a</span> <span class="o">=</span> <span class="mi">1</span> <span class="o">+</span> <span class="mi">4</span>
<span class="nd">@torch</span><span class="o">.</span><span class="n">jit</span><span class="o">.</span><span class="n">ScriptModule</span>
<span class="k">def</span> <span class="nf">forward</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="nb">input</span><span class="p">):</span>
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">a</span> <span class="o">+</span> <span class="nb">input</span>
</pre></div>
</div>
</li>
</ol>
<p>Supported constant Python Values are</p>
<ul class="last simple">
<li><code class="docutils literal notranslate"><span class="pre">int</span></code></li>
<li><code class="docutils literal notranslate"><span class="pre">bool</span></code></li>
<li><code class="docutils literal notranslate"><span class="pre">torch.device</span></code></li>
<li><code class="docutils literal notranslate"><span class="pre">torch.layout</span></code></li>
<li><code class="docutils literal notranslate"><span class="pre">torch.dtype</span></code></li>
<li>tuples containing supported types</li>