-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathjit.html
2158 lines (1897 loc) · 194 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 name="robots" content="noindex">
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>TorchScript — 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="torch.hub" href="hub.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'>1.2.0 ▼</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/cpu_threading_torchscript_inference.html">CPU threading and TorchScript inference</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/large_scale_deployments.html">Features for large-scale deployments</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">Community</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="community/contribution_guide.html">PyTorch Contribution Guide</a></li>
<li class="toctree-l1"><a class="reference internal" href="community/governance.html">PyTorch Governance</a></li>
<li class="toctree-l1"><a class="reference internal" href="community/persons_of_interest.html">PyTorch Governance | Persons of Interest</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.functional.html">torch.nn.functional</a></li>
<li class="toctree-l1"><a class="reference internal" href="nn.init.html">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"><a class="reference internal" href="hub.html">torch.hub</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="random.html">torch.random</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="model_zoo.html">torch.utils.model_zoo</a></li>
<li class="toctree-l1"><a class="reference internal" href="tensorboard.html">torch.utils.tensorboard</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="__config__.html">torch.__config__</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>TorchScript</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="torchscript">
<h1>TorchScript<a class="headerlink" href="#torchscript" title="Permalink to this headline">¶</a></h1>
<div class="contents local topic" id="contents">
<ul class="simple">
<li><p><a class="reference internal" href="#creating-torchscript-code" id="id1">Creating TorchScript Code</a></p></li>
<li><p><a class="reference internal" href="#mixing-tracing-and-scripting" id="id2">Mixing Tracing and Scripting</a></p></li>
<li><p><a class="reference internal" href="#torchscript-language-reference" id="id3">TorchScript Language Reference</a></p>
<ul>
<li><p><a class="reference internal" href="#types" id="id4">Types</a></p>
<ul>
<li><p><a class="reference internal" href="#default-types" id="id5">Default Types</a></p></li>
<li><p><a class="reference internal" href="#optional-type-refinement" id="id6">Optional Type Refinement</a></p></li>
<li><p><a class="reference internal" href="#user-defined-types" id="id7">User Defined Types</a></p></li>
</ul>
</li>
<li><p><a class="reference internal" href="#expressions" id="id8">Expressions</a></p>
<ul>
<li><p><a class="reference internal" href="#literals" id="id9">Literals</a></p>
<ul>
<li><p><a class="reference internal" href="#list-construction" id="id10">List Construction</a></p></li>
<li><p><a class="reference internal" href="#tuple-construction" id="id11">Tuple Construction</a></p></li>
<li><p><a class="reference internal" href="#dict-construction" id="id12">Dict Construction</a></p></li>
</ul>
</li>
<li><p><a class="reference internal" href="#variables" id="id13">Variables</a></p></li>
<li><p><a class="reference internal" href="#arithmetic-operators" id="id14">Arithmetic Operators</a></p></li>
<li><p><a class="reference internal" href="#comparison-operators" id="id15">Comparison Operators</a></p></li>
<li><p><a class="reference internal" href="#logical-operators" id="id16">Logical Operators</a></p></li>
<li><p><a class="reference internal" href="#subscripts" id="id17">Subscripts</a></p></li>
<li><p><a class="reference internal" href="#function-calls" id="id18">Function Calls</a></p></li>
<li><p><a class="reference internal" href="#method-calls" id="id19">Method Calls</a></p></li>
<li><p><a class="reference internal" href="#ternary-expressions" id="id20">Ternary Expressions</a></p></li>
<li><p><a class="reference internal" href="#casts" id="id21">Casts</a></p></li>
<li><p><a class="reference internal" href="#accessing-module-parameters" id="id22">Accessing Module Parameters</a></p></li>
</ul>
</li>
<li><p><a class="reference internal" href="#statements" id="id23">Statements</a></p></li>
<li><p><a class="reference internal" href="#variable-resolution" id="id24">Variable Resolution</a></p></li>
<li><p><a class="reference internal" href="#use-of-python-values" id="id25">Use of Python Values</a></p>
<ul>
<li><p><a class="reference internal" href="#functions" id="id26">Functions</a></p></li>
<li><p><a class="reference internal" href="#attribute-lookup-on-python-modules" id="id27">Attribute Lookup On Python Modules</a></p></li>
<li><p><a class="reference internal" href="#python-defined-constants" id="id28">Python-defined Constants</a></p></li>
<li><p><a class="reference internal" href="#module-attributes" id="id29">Module Attributes</a></p></li>
</ul>
</li>
<li><p><a class="reference internal" href="#debugging" id="id30">Debugging</a></p>
<ul>
<li><p><a class="reference internal" href="#disable-jit-for-debugging" id="id31">Disable JIT for Debugging</a></p></li>
<li><p><a class="reference internal" href="#inspecting-code" id="id32">Inspecting Code</a></p></li>
<li><p><a class="reference internal" href="#interpreting-graphs" id="id33">Interpreting Graphs</a></p></li>
<li><p><a class="reference internal" href="#tracing-edge-cases" id="id34">Tracing Edge Cases</a></p></li>
<li><p><a class="reference internal" href="#automatic-trace-checking" id="id35">Automatic Trace Checking</a></p></li>
<li><p><a class="reference internal" href="#tracer-warnings" id="id36">Tracer Warnings</a></p></li>
</ul>
</li>
</ul>
</li>
<li><p><a class="reference internal" href="#frequently-asked-questions" id="id37">Frequently Asked Questions</a></p>
<ul>
<li><p><a class="reference internal" href="#builtin-functions" id="id38">Builtin Functions</a></p></li>
</ul>
</li>
</ul>
</div>
<span class="target" id="module-torch.jit"></span><p>TorchScript is a way to create serializable and optimizable models from PyTorch code.
Any TorchScript program can be saved from a Python
process and loaded in a process where there is no Python dependency.</p>
<p>We provide tools to incrementally transition a model from a pure Python program
to a TorchScript program that can be run independently from Python, such as in a standalone C++ program.
This makes it possible to train models in PyTorch using familiar tools in Python and then export
the model via TorchScript to a production environment where Python programs may be disadvantageous.
for performance and multi-threading reasons.</p>
<div class="section" id="creating-torchscript-code">
<h2><a class="toc-backref" href="#id1">Creating TorchScript Code</a><a class="headerlink" href="#creating-torchscript-code" title="Permalink to this headline">¶</a></h2>
<dl class="function">
<dt id="torch.jit.script">
<code class="sig-prename descclassname">torch.jit.</code><code class="sig-name descname">script</code><span class="sig-paren">(</span><em class="sig-param">obj</em>, <em class="sig-param">optimize=None</em>, <em class="sig-param">_frames_up=0</em>, <em class="sig-param">_rcb=None</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/jit.html#script"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.jit.script" title="Permalink to this definition">¶</a></dt>
<dd><p>Scripting a function or <code class="docutils literal notranslate"><span class="pre">nn.Module</span></code> will inspect the source code, compile
it as TorchScript code using the TorchScript compiler, and return a <code class="docutils literal notranslate"><span class="pre">ScriptModule</span></code> or
<code class="docutils literal notranslate"><span class="pre">torch._C.Function</span></code>.</p>
<dl>
<dt><strong>Scripting a function</strong></dt><dd><p>The <code class="docutils literal notranslate"><span class="pre">@torch.jit.script</span></code> decorator will construct a <code class="docutils literal notranslate"><span class="pre">torch._C.Function</span></code>.</p>
<p>Example (scripting a function):</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>
</dd>
<dt><strong>Scripting an nn.Module</strong></dt><dd><p>Scripting an <code class="docutils literal notranslate"><span class="pre">nn.Module</span></code> by default will compile the <code class="docutils literal notranslate"><span class="pre">forward</span></code> method and recursively
compile any methods, submodules, and functions called by <code class="docutils literal notranslate"><span class="pre">forward</span></code>. If a <code class="docutils literal notranslate"><span class="pre">nn.Module</span></code> only uses
features supported in TorchScript, no changes to the original module code should be necessary.</p>
<p>Example (scripting a simple module with a Parameter):</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">nn</span><span class="o">.</span><span class="n">Module</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="c1"># This parameter will be copied to the new ScriptModule</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="c1"># When this submodule is used, it will be compiled</span>
<span class="bp">self</span><span class="o">.</span><span class="n">linear</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">Linear</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="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="n">output</span> <span class="o">=</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>
<span class="c1"># This calls the `forward` method of the `nn.Linear` module, which will</span>
<span class="c1"># cause the `self.linear` submodule to be compiled to a `ScriptModule` here</span>
<span class="n">output</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">linear</span><span class="p">(</span><span class="n">output</span><span class="p">)</span>
<span class="k">return</span> <span class="n">output</span>
<span class="n">scripted_module</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">script</span><span class="p">(</span><span class="n">MyModule</span><span class="p">())</span>
</pre></div>
</div>
<p>Example (scripting a module with traced submodules):</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="k">class</span> <span class="nc">MyModule</span><span class="p">(</span><span class="n">nn</span><span class="o">.</span><span class="n">Module</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="c1"># torch.jit.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">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">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">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">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="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>
<span class="n">scripted_module</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">script</span><span class="p">(</span><span class="n">MyModule</span><span class="p">())</span>
</pre></div>
</div>
<p>To compile a method other than <code class="docutils literal notranslate"><span class="pre">forward</span></code> (and recursively compile anything it calls), add
the <code class="docutils literal notranslate"><span class="pre">@torch.jit.export</span></code> decorator to the method.</p>
</dd>
</dl>
</dd></dl>
<dl class="function">
<dt id="torch.jit.trace">
<code class="sig-prename descclassname">torch.jit.</code><code class="sig-name descname">trace</code><span class="sig-paren">(</span><em class="sig-param">func</em>, <em class="sig-param">example_inputs</em>, <em class="sig-param">optimize=None</em>, <em class="sig-param">check_trace=True</em>, <em class="sig-param">check_inputs=None</em>, <em class="sig-param">check_tolerance=1e-05</em>, <em class="sig-param">_force_outplace=False</em>, <em class="sig-param">_module_class=None</em>, <em class="sig-param">_compilation_unit=<torch._C.CompilationUnit object></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 <code class="docutils literal notranslate"><span class="pre">ScriptModule</span></code> or <code class="docutils literal notranslate"><span class="pre">torch.jit._C.Function</span></code>
that will be optimized using just-in-time compilation.</p>
<div class="admonition warning">
<p class="admonition-title">Warning</p>
<p>Tracing only correctly records functions and modules which are not data
dependent (e.g., do not 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>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><ul class="simple">
<li><p><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 <code class="docutils literal notranslate"><span class="pre">torch.nn.Module</span></code>
that will be run with <code class="docutils literal notranslate"><span class="pre">example_inputs</span></code>.
arguments and returns to <code class="docutils literal notranslate"><span class="pre">func</span></code> must be tensors
or (possibly nested) tuples that
contain tensors.</p></li>
<li><p><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. <code class="docutils literal notranslate"><span class="pre">example_inputs</span></code> may also be a single
Tensor in which case it is automatically wrapped in a tuple</p></li>
</ul>
</dd>
<dt class="field-even">Keyword Arguments</dt>
<dd class="field-even"><ul class="simple">
<li><p><strong>check_trace</strong> (<a class="reference internal" href="storage.html#torch.FloatStorage.bool" title="torch.FloatStorage.bool"><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.</p></li>
<li><p><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 set of input arguments that would
be specified in <code class="docutils literal notranslate"><span class="pre">example_inputs</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">example_inputs</span></code> are used for checking</p></li>
<li><p><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.</p></li>
</ul>
</dd>
<dt class="field-odd">Returns</dt>
<dd class="field-odd"><p>if <code class="docutils literal notranslate"><span class="pre">callable</span></code> is <code class="docutils literal notranslate"><span class="pre">nn.Module</span></code> or <code class="docutils literal notranslate"><span class="pre">forward()</span></code> of <code class="docutils literal notranslate"><span class="pre">nn.Module</span></code>, <code class="docutils literal notranslate"><span class="pre">trace</span></code> returns
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.
The returned <code class="docutils literal notranslate"><span class="pre">ScriptModule</span></code> will have the same set of sub-modules and parameters as the
original <code class="docutils literal notranslate"><span class="pre">nn.Module</span></code>.
If <code class="docutils literal notranslate"><span class="pre">callable</span></code> is a standalone function, <code class="docutils literal notranslate"><span class="pre">trace</span></code> returns <code class="docutils literal notranslate"><span class="pre">torch.jit._C.Function</span></code></p>
</dd>
</dl>
<p>Example:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="k">class</span> <span class="nc">Net</span><span class="p">(</span><span class="n">nn</span><span class="o">.</span><span class="n">Module</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">Net</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">conv</span> <span class="o">=</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">1</span><span class="p">,</span> <span class="mi">3</span><span class="p">)</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">x</span><span class="p">):</span>
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">conv</span><span class="p">(</span><span class="n">x</span><span class="p">)</span>
<span class="k">def</span> <span class="nf">weighted_kernel_sum</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">weight</span><span class="p">):</span>
<span class="k">return</span> <span class="n">weight</span> <span class="o">*</span> <span class="bp">self</span><span class="o">.</span><span class="n">conv</span><span class="o">.</span><span class="n">weight</span>
<span class="n">example_weight</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="mi">1</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="n">example_forward_input</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="mi">1</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="n">n</span> <span class="o">=</span> <span class="n">Net</span><span class="p">()</span>
<span class="c1"># the following two calls are equivalent</span>
<span class="n">module</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_module</span><span class="p">(</span><span class="n">n</span><span class="p">,</span> <span class="n">example_forward_input</span><span class="p">)</span>
<span class="n">module</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_module</span><span class="p">(</span><span class="n">n</span><span class="o">.</span><span class="n">forward</span><span class="p">,</span> <span class="n">example_forward_input</span><span class="p">)</span>
</pre></div>
</div>
</dd></dl>
<dl class="class">
<dt id="torch.jit.ScriptModule">
<em class="property">class </em><code class="sig-prename descclassname">torch.jit.</code><code class="sig-name descname">ScriptModule</code><span class="sig-paren">(</span><em class="sig-param">optimize=None</em>, <em class="sig-param">_qualified_name=None</em>, <em class="sig-param">_compilation_unit=None</em>, <em class="sig-param">_cpp_module=None</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 TorchScript is the <code class="docutils literal notranslate"><span class="pre">ScriptModule</span></code>. It is an
analogue of torch’s <code class="docutils literal notranslate"><span class="pre">nn.Module</span></code> and represents an entire model as a tree of
submodules. Like normal modules, each individual module in a <code class="docutils literal notranslate"><span class="pre">ScriptModule</span></code> can
have submodules, parameters, and methods. In <code class="docutils literal notranslate"><span class="pre">nn.Module</span></code>s methods are implemented
as Python functions, but in <code class="docutils literal notranslate"><span class="pre">ScriptModule</span></code>s methods are implemented as
TorchScript 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><code class="docutils literal notranslate"><span class="pre">ScriptModule</span></code>s 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> and <code class="docutils literal notranslate"><span class="pre">torch.jit.trace_module</span></code>, you can turn an existing module or Python
function into a TorchScript <code class="docutils literal notranslate"><span class="pre">torch._C.Function</span></code> or <code class="docutils literal notranslate"><span class="pre">ScriptModule</span></code>. You must provide example inputs,
and we run the function, recording the operations performed on all the tensors.
* The resulting recording of a standalone function produces <code class="docutils literal notranslate"><span class="pre">torch._C.Function</span></code>.
* The resulting recording of <code class="docutils literal notranslate"><span class="pre">forward</span></code> function of <code class="docutils literal notranslate"><span class="pre">nn.Module</span></code> or <code class="docutils literal notranslate"><span class="pre">nn.Module</span></code> produces <code class="docutils literal notranslate"><span class="pre">ScriptModule</span></code>.
This module also contains any parameters that the original
module had as well.</p>
<p>Example (tracing a function):</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="admonition-title">Note</p>
<p>Tracing a standalone function will construct a <code class="docutils literal notranslate"><span class="pre">torch._C.Function</span></code>
Tracing <code class="docutils literal notranslate"><span class="pre">nn.Module``s</span> <span class="pre">``forward</span></code> will construct a <code class="docutils literal notranslate"><span class="pre">ScriptModule</span></code></p>
</div>
<p>Example (tracing an existing module):</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">Net</span><span class="p">(</span><span class="n">nn</span><span class="o">.</span><span class="n">Module</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">Net</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">conv</span> <span class="o">=</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">1</span><span class="p">,</span> <span class="mi">3</span><span class="p">)</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">x</span><span class="p">):</span>
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">conv</span><span class="p">(</span><span class="n">x</span><span class="p">)</span>
<span class="k">def</span> <span class="nf">weighted_kernel_sum</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">weight</span><span class="p">):</span>
<span class="k">return</span> <span class="n">weight</span> <span class="o">*</span> <span class="bp">self</span><span class="o">.</span><span class="n">conv</span><span class="o">.</span><span class="n">weight</span>
<span class="n">n</span> <span class="o">=</span> <span class="n">Net</span><span class="p">()</span>
<span class="n">example_weight</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="mi">1</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="n">example_forward_input</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="mi">1</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="c1"># all three trace calls below are equivalent</span>
<span class="c1"># and construct `ScriptModule` with a single `forward` method</span>
<span class="n">module</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">n</span><span class="o">.</span><span class="n">forward</span><span class="p">,</span> <span class="n">example_forward_input</span><span class="p">)</span> <span class="c1"># produces ScriptModule with `forward`</span>
<span class="n">module</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">n</span><span class="p">,</span> <span class="n">example_forward_input</span><span class="p">)</span> <span class="c1"># produces ScriptModule with `forward`</span>
<span class="n">module</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_module</span><span class="p">(</span><span class="n">n</span><span class="p">,</span> <span class="n">inputs</span><span class="p">)</span> <span class="c1"># produces ScriptModule with `forward`</span>
<span class="n">inputs</span> <span class="o">=</span> <span class="p">{</span><span class="s1">'forward'</span> <span class="p">:</span> <span class="n">example_forward_input</span><span class="p">,</span> <span class="s1">'weighted_kernel_sum'</span> <span class="p">:</span> <span class="n">example_weight</span><span class="p">}</span>
<span class="c1"># trace_module produces `ScriptModule` with two methods:</span>
<span class="c1"># `forward` and `weighted_kernel_sum`</span>
<span class="n">module</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_module</span><span class="p">(</span><span class="n">n</span><span class="p">,</span> <span class="n">inputs</span><span class="p">,</span> <span class="kc">True</span><span class="p">,</span> <span class="kc">True</span><span class="p">)</span>
</pre></div>
</div>
<div class="admonition note">
<p class="admonition-title">Note</p>
<ul class="simple">
<li><p>The first three trace/trace_module calls are equivalent and return <code class="docutils literal notranslate"><span class="pre">ScriptModule</span></code></p></li>
</ul>
<p>with a single <code class="docutils literal notranslate"><span class="pre">forward</span></code> method.
* The last <code class="docutils literal notranslate"><span class="pre">trace_module</span></code> call produces a <code class="docutils literal notranslate"><span class="pre">ScriptModule</span></code> with two methods.
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><p>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
inlines the control-flow decisions. But sometimes the control-flow is
actually part of the model itself. For instance, a recurrent network is
a loop over the (possibly dynamic) length of an input sequence.</p></li>
<li><p>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.</p></li>
</ul>
</div></blockquote>
<p>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 TorchScript code directly using Python syntax. You do this
using the <code class="docutils literal notranslate"><span class="pre">@torch.jit.script</span></code> decorator for functions and modules. You can
also call <code class="docutils literal notranslate"><span class="pre">torch.jit.script</span></code> directly with the function or module you wish to
compile. On functions, the body of the function is compiled to TorchScript. If
applied to an <code class="docutils literal notranslate"><span class="pre">nn.Module</span></code>, by default the <code class="docutils literal notranslate"><span class="pre">forward</span></code> method and any methods it
calls are compiled, and all buffer and Parameters of the original module are copied
to a new <code class="docutils literal notranslate"><span class="pre">ScriptModule</span></code>. You should not need to construct a <code class="docutils literal notranslate"><span class="pre">ScriptModule</span></code> manually.
TorchScript 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>
</div></blockquote>
</dd></dl>
<dl class="function">
<dt id="torch.jit.save">
<code class="sig-prename descclassname">torch.jit.</code><code class="sig-name descname">save</code><span class="sig-paren">(</span><em class="sig-param">m</em>, <em class="sig-param">f</em>, <em class="sig-param">_extra_files=ExtraFilesMap{}</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/jit.html#save"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.jit.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, submodules, parameters, and attributes 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 <a class="reference internal" href="#torch.jit.load" title="torch.jit.load"><code class="xref py py-func docutils literal notranslate"><span class="pre">load</span></code></a>.</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 <code class="docutils literal notranslate"><span class="pre">torch.jit.ScriptModule</span></code> as well.</p>
<div class="admonition danger">
<p class="admonition-title">Danger</p>
<p>All modules, no matter their device, are always loaded onto the CPU during loading.
This is different from <a class="reference internal" href="#torch.jit.load" title="torch.jit.load"><code class="xref py py-func docutils literal notranslate"><span class="pre">load</span></code></a>’s semantics and may change in the future.</p>
</div>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><ul class="simple">
<li><p><strong>m</strong> – a ScriptModule to save</p></li>
<li><p><strong>f</strong> – a file-like object (has to implement write and flush) or a string
containing a file name</p></li>
<li><p><strong>_extra_files</strong> – Map from filename to contents which will be stored as part of ‘f’</p></li>
</ul>
</dd>
</dl>
<div class="admonition warning">
<p class="admonition-title">Warning</p>
<p>If you are using Python 2, <code class="docutils literal notranslate"><span class="pre">torch.save</span></code> does NOT support <code class="docutils literal notranslate"><span class="pre">StringIO.StringIO</span></code>
as a valid file-like object. This is because the write method should return
the number of bytes written; <code class="docutils literal notranslate"><span class="pre">StringIO.write()</span></code> does not do this.</p>
<p>Please use something like <code class="docutils literal notranslate"><span class="pre">io.BytesIO</span></code> instead.</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">io</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">nn</span><span class="o">.</span><span class="n">Module</span><span class="p">):</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">x</span><span class="p">):</span>
<span class="k">return</span> <span class="n">x</span> <span class="o">+</span> <span class="mi">10</span>
<span class="n">m</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">script</span><span class="p">(</span><span class="n">MyModule</span><span class="p">())</span>
<span class="c1"># Save to file</span>
<span class="n">torch</span><span class="o">.</span><span class="n">jit</span><span class="o">.</span><span class="n">save</span><span class="p">(</span><span class="n">m</span><span class="p">,</span> <span class="s1">'scriptmodule.pt'</span><span class="p">)</span>
<span class="c1"># Save to io.BytesIO buffer</span>
<span class="n">buffer</span> <span class="o">=</span> <span class="n">io</span><span class="o">.</span><span class="n">BytesIO</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">save</span><span class="p">(</span><span class="n">m</span><span class="p">,</span> <span class="n">buffer</span><span class="p">)</span>
<span class="c1"># Save with extra files</span>
<span class="n">extra_files</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">_C</span><span class="o">.</span><span class="n">ExtraFilesMap</span><span class="p">()</span>
<span class="n">extra_files</span><span class="p">[</span><span class="s1">'foo.txt'</span><span class="p">]</span> <span class="o">=</span> <span class="s1">'bar'</span>
<span class="n">torch</span><span class="o">.</span><span class="n">jit</span><span class="o">.</span><span class="n">save</span><span class="p">(</span><span class="n">m</span><span class="p">,</span> <span class="s1">'scriptmodule.pt'</span><span class="p">,</span> <span class="n">_extra_files</span><span class="o">=</span><span class="n">extra_files</span><span class="p">)</span>
</pre></div>
</div>
</dd></dl>
<dl class="function">
<dt id="torch.jit.load">
<code class="sig-prename descclassname">torch.jit.</code><code class="sig-name descname">load</code><span class="sig-paren">(</span><em class="sig-param">f</em>, <em class="sig-param">map_location=None</em>, <em class="sig-param">_extra_files=ExtraFilesMap{}</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 <a class="reference internal" href="#torch.jit.save" title="torch.jit.save"><code class="xref py py-func docutils literal notranslate"><span class="pre">save</span></code></a></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>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><ul class="simple">
<li><p><strong>f</strong> – a file-like object (has to implement read, readline, tell, and seek),
or a string containing a file name</p></li>
<li><p><strong>map_location</strong> – can a string (e.g., ‘cpu’, ‘cuda:0’), a device (e.g.,
torch.device(‘cpu’))</p></li>
<li><p><strong>_extra_files</strong> – map from filename to content. The extra
filenames given in the map would be loaded and their content
would be stored in the provided map.</p></li>
</ul>
</dd>
<dt class="field-even">Returns</dt>
<dd class="field-even"><p>A <code class="docutils literal notranslate"><span class="pre">ScriptModule</span></code> object.</p>
</dd>
</dl>
<p>Example:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></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="c1"># Load ScriptModule from io.BytesIO object</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="n">buffer</span> <span class="o">=</span> <span class="n">io</span><span class="o">.</span><span class="n">BytesIO</span><span class="p">(</span><span class="n">f</span><span class="o">.</span><span class="n">read</span><span class="p">())</span>
<span class="c1"># Load all tensors to the original device</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="c1"># Load all tensors onto CPU, using a device</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="c1"># Load all tensors onto CPU, using a string</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>
<span class="c1"># Load with extra files.</span>
<span class="n">files</span> <span class="o">=</span> <span class="p">{</span><span class="s1">'metadata.json'</span> <span class="p">:</span> <span class="s1">''</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">load</span><span class="p">(</span><span class="s1">'scriptmodule.pt'</span><span class="p">,</span> <span class="n">_extra_files</span> <span class="o">=</span> <span class="n">files</span><span class="p">)</span>
<span class="nb">print</span> <span class="p">(</span><span class="n">files</span><span class="p">[</span><span class="s1">'metadata.json'</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 scripting is an easier approach for converting a model to TorchScript.
Tracing and scripting can be composed to suit the particular requirements
of a part of a model.</p>
<p>Scripted functions can call traced functions. 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 <code class="docutils literal notranslate"><span class="pre">nn.Module</span></code>s 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">nn</span><span class="o">.</span><span class="n">Module</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="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>
<span class="n">my_script_module</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">script</span><span class="p">(</span><span class="n">MyScriptModule</span><span class="p">())</span>
</pre></div>
</div>
</div>
<div class="section" id="torchscript-language-reference">
<h2><a class="toc-backref" href="#id3">TorchScript Language Reference</a><a class="headerlink" href="#torchscript-language-reference" title="Permalink to this headline">¶</a></h2>
<p>TorchScript is a statically typed subset of Python that can either be written directly (using
the <code class="docutils literal notranslate"><span class="pre">@torch.jit.script</span></code> decorator) 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 TorchScript directly using <code class="docutils literal notranslate"><span class="pre">@torch.jit.script</span></code> decorator, the programmer must
only use the subset of Python supported in TorchScript. This section documents
what is supported in TorchScript 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 TorchScript.</p>
<p>As a subset of Python any valid TorchScript function is also a valid Python
function. This makes it possible to remove the <code class="docutils literal notranslate"><span class="pre">@torch.jit.script</span></code> decorator and debug the
function using standard Python tools like <code class="docutils literal notranslate"><span class="pre">pdb</span></code>. The reverse is not true: there
are many valid python programs that are not valid TorchScript programs.
Instead, TorchScript 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="sig-name 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 TorchScript and the full Python language is that
TorchScript only supports a small set of types that are needed to express neural
net models. In particular, TorchScript supports:</p>
<table class="docutils align-default">
<colgroup>
<col style="width: 50%" />
<col style="width: 50%" />
</colgroup>
<thead>
<tr class="row-odd"><th class="head"><p>Type</p></th>
<th class="head"><p>Description</p></th>
</tr>
</thead>
<tbody>
<tr class="row-even"><td><p><code class="docutils literal notranslate"><span class="pre">Tensor</span></code></p></td>
<td><p>A PyTorch tensor of any dtype, dimension, or backend</p></td>
</tr>
<tr class="row-odd"><td><p><code class="docutils literal notranslate"><span class="pre">Tuple[T0,</span> <span class="pre">T1,</span> <span class="pre">...]</span></code></p></td>
<td><p>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>)</p></td>
</tr>
<tr class="row-even"><td><p><code class="docutils literal notranslate"><span class="pre">bool</span></code></p></td>
<td><p>A boolean value</p></td>
</tr>
<tr class="row-odd"><td><p><code class="docutils literal notranslate"><span class="pre">int</span></code></p></td>
<td><p>A scalar integer</p></td>
</tr>
<tr class="row-even"><td><p><code class="docutils literal notranslate"><span class="pre">float</span></code></p></td>
<td><p>A scalar floating point number</p></td>
</tr>
<tr class="row-odd"><td><p><code class="docutils literal notranslate"><span class="pre">List[T]</span></code></p></td>
<td><p>A list of which all members are type <code class="docutils literal notranslate"><span class="pre">T</span></code></p></td>
</tr>
<tr class="row-even"><td><p><code class="docutils literal notranslate"><span class="pre">Optional[T]</span></code></p></td>
<td><p>A value which is either None or type <code class="docutils literal notranslate"><span class="pre">T</span></code></p></td>
</tr>
<tr class="row-odd"><td><p><code class="docutils literal notranslate"><span class="pre">Dict[K,</span> <span class="pre">V]</span></code></p></td>
<td><p>A dict with key type <code class="docutils literal notranslate"><span class="pre">K</span></code> and value type <code class="docutils literal notranslate"><span class="pre">V</span></code>. Only <code class="docutils literal notranslate"><span class="pre">str</span></code>, <code class="docutils literal notranslate"><span class="pre">int</span></code>, and <code class="docutils literal notranslate"><span class="pre">float</span></code> are allowed as key types.</p></td>
</tr>
</tbody>
</table>
<p>Unlike Python, each variable in TorchScript function must have a single static type.
This makes it easier to optimize TorchScript functions.</p>
<p>Example (a type mismatch):</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>
<div class="section" id="default-types">
<h4><a class="toc-backref" href="#id5">Default Types</a><a class="headerlink" href="#default-types" title="Permalink to this headline">¶</a></h4>
<p>By default, all parameters to a TorchScript function are assumed to be Tensor.
To specify that an argument to a TorchScript 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="admonition-title">Note</p>
<p>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>
<p>An empty list is assumed to be <code class="docutils literal notranslate"><span class="pre">List[Tensor]</span></code> and empty dicts
<code class="docutils literal notranslate"><span class="pre">Dict[str,</span> <span class="pre">Tensor]</span></code>. To instantiate an empty list or dict of other types,
use <code class="docutils literal notranslate"><span class="pre">torch.jit.annotate</span></code>.</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">from</span> <span class="nn">torch.jit</span> <span class="k">import</span> <span class="n">Tensor</span>
<span class="kn">from</span> <span class="nn">typing</span> <span class="k">import</span> <span class="n">List</span><span class="p">,</span> <span class="n">Tuple</span>
<span class="k">class</span> <span class="nc">EmptyDataStructures</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">EmptyDataStructures</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="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">x</span><span class="p">):</span>
<span class="c1"># type: (Tensor) -> Tuple[List[Tuple[int, float]], Dict[str, int]]</span>
<span class="c1"># This annotates the list to be a `List[Tuple[int, float]]`</span>
<span class="n">my_list</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">annotate</span><span class="p">(</span><span class="n">List</span><span class="p">[</span><span class="n">Tuple</span><span class="p">[</span><span class="nb">int</span><span class="p">,</span> <span class="nb">float</span><span class="p">]],</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="n">my_list</span><span class="o">.</span><span class="n">append</span><span class="p">((</span><span class="n">x</span><span class="p">,</span> <span class="n">x</span><span class="p">))</span>
<span class="n">my_dict</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">annotate</span><span class="p">(</span><span class="n">Dict</span><span class="p">[</span><span class="nb">str</span><span class="p">,</span> <span class="nb">int</span><span class="p">],</span> <span class="p">{})</span>
<span class="k">return</span> <span class="n">my_list</span><span class="p">,</span> <span class="n">my_dict</span>
</pre></div>
</div>
</div>
<div class="section" id="optional-type-refinement">
<h4><a class="toc-backref" href="#id6">Optional Type Refinement</a><a class="headerlink" href="#optional-type-refinement" title="Permalink to this headline">¶</a></h4>
<p>TorchScript will refine the type of a variable of type <code class="docutils literal notranslate"><span class="pre">Optional[T]</span></code> when
a comparison to <code class="docutils literal notranslate"><span class="pre">None</span></code> is made inside the conditional of an if-statement.
The compiler can reason about multiple <code class="docutils literal notranslate"><span class="pre">None</span></code> checks that are combined with
<code class="docutils literal notranslate"><span class="pre">and</span></code>, <code class="docutils literal notranslate"><span class="pre">or</span></code>, and <code class="docutils literal notranslate"><span class="pre">not</span></code>. Refinement will also occur for else blocks of if-statements
that are not explicitly written.</p>
<p>The expression must be emitted within the conditional; assigning
a <code class="docutils literal notranslate"><span class="pre">None</span></code> check to a variable and using it in the conditional will not refine types.
An attribute like <cite>self.x</cite> will not be refined, but assigning <cite>self.x</cite> to a local
variable first will work.</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_method</span>
<span class="k">def</span> <span class="nf">optional_unwrap</span><span class="p">(</span><span class="bp">self</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="c1"># type: (Optional[int], Optional[int]) -> int</span>
<span class="k">if</span> <span class="n">x</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
<span class="n">x</span> <span class="o">=</span> <span class="mi">1</span>
<span class="n">x</span> <span class="o">=</span> <span class="n">x</span> <span class="o">+</span> <span class="mi">1</span>
<span class="n">z</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">z</span>
<span class="k">if</span> <span class="n">y</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span> <span class="ow">and</span> <span class="n">z</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
<span class="n">x</span> <span class="o">=</span> <span class="n">y</span> <span class="o">+</span> <span class="n">z</span>
<span class="k">return</span> <span class="n">x</span>
</pre></div>
</div>
</div>
<div class="section" id="user-defined-types">
<h4><a class="toc-backref" href="#id7">User Defined Types</a><a class="headerlink" href="#user-defined-types" title="Permalink to this headline">¶</a></h4>
<p>Python classes can be used in TorchScript if they are annotated with <code class="docutils literal notranslate"><span class="pre">@torch.jit.script</span></code>,
similar to how you would declare a TorchScript function:</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">class</span> <span class="nc">Foo</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">x</span><span class="p">,</span> <span class="n">y</span><span class="p">):</span>
<span class="bp">self</span><span class="o">.</span><span class="n">x</span> <span class="o">=</span> <span class="n">x</span>
<span class="k">def</span> <span class="nf">aug_add_x</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">inc</span><span class="p">):</span>
<span class="bp">self</span><span class="o">.</span><span class="n">x</span> <span class="o">+=</span> <span class="n">inc</span>
</pre></div>
</div>
<p>This subset is restricted:</p>
<ul>
<li><p>All functions must be valid TorchScript functions (including <code class="docutils literal notranslate"><span class="pre">__init__()</span></code>)</p></li>
<li><p>Classes must be new-style classes, as we use <code class="docutils literal notranslate"><span class="pre">__new__()</span></code> to construct them with pybind11</p></li>
<li><p>TorchScript classes are statically typed. Members are declared by assigning to
self in the <code class="docutils literal notranslate"><span class="pre">__init__()</span></code> method</p>
<blockquote>
<div><p>For example, assigning outside of the <code class="docutils literal notranslate"><span class="pre">__init__()</span></code> method:</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">class</span> <span class="nc">Foo</span><span class="p">:</span>
<span class="k">def</span> <span class="nf">assign_x</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="bp">self</span><span class="o">.</span><span class="n">x</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">2</span><span class="p">,</span> <span class="mi">3</span><span class="p">)</span>
</pre></div>
</div>
<p>Will result in:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span>RuntimeError:
Tried to set nonexistent attribute: x. Did you forget to initialize it in __init__()?:
def assign_x(self):
self.x = torch.rand(2, 3)
~~~~~~~~~~~~~~~~~~~~~~~~ <--- HERE
</pre></div>
</div>
</div></blockquote>
</li>
<li><p>No expressions except method definitions are allowed in the body of the class</p></li>
<li><p>No support for inheritance or any other polymorphism strategy, except for inheriting
from object to specify a new-style class</p></li>
</ul>
<p>After a class is defined, it can be used in both TorchScript and Python interchangeably
like any other TorchScript type:</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">class</span> <span class="nc">Pair</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">first</span><span class="p">,</span> <span class="n">second</span><span class="p">):</span>
<span class="bp">self</span><span class="o">.</span><span class="n">first</span> <span class="o">=</span> <span class="n">first</span>
<span class="bp">self</span><span class="o">.</span><span class="n">second</span> <span class="o">=</span> <span class="n">second</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">sum_pair</span><span class="p">(</span><span class="n">p</span><span class="p">):</span>
<span class="c1"># type: (Pair) -> Tensor</span>
<span class="k">return</span> <span class="n">p</span><span class="o">.</span><span class="n">first</span> <span class="o">+</span> <span class="n">p</span><span class="o">.</span><span class="n">second</span>
<span class="n">p</span> <span class="o">=</span> <span class="n">Pair</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">2</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">2</span><span class="p">,</span> <span class="mi">3</span><span class="p">))</span>
<span class="nb">print</span><span class="p">(</span><span class="n">sum_pair</span><span class="p">(</span><span class="n">p</span><span class="p">))</span>
</pre></div>
</div>
</div>
</div>
<div class="section" id="expressions">
<h3><a class="toc-backref" href="#id8">Expressions</a><a class="headerlink" href="#expressions" title="Permalink to this headline">¶</a></h3>
<p>The following Python Expressions are supported</p>
<div class="section" id="literals">
<h4><a class="toc-backref" href="#id9">Literals</a><a class="headerlink" href="#literals" title="Permalink to this headline">¶</a></h4>
<blockquote>
<div><p><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> (interpreted as a float)</p>
</div></blockquote>
<div class="section" id="list-construction">
<h5><a class="toc-backref" href="#id10">List Construction</a><a class="headerlink" href="#list-construction" title="Permalink to this headline">¶</a></h5>
<blockquote>
<div><p><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>