-
Notifications
You must be signed in to change notification settings - Fork 257
/
Copy pathtorch_script_custom_classes.html
1123 lines (920 loc) Β· 83.4 KB
/
torch_script_custom_classes.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>Extending TorchScript with Custom C++ Classes — PyTorch Tutorials 1.6.0 documentation</title>
<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="../_static/gallery.css" type="text/css" />
<link rel="index" title="Index" href="../genindex.html" />
<link rel="search" title="Search" href="../search.html" />
<link rel="next" title="Dynamic Parallelism in TorchScript" href="torch-script-parallelism.html" />
<link rel="prev" title="Extending TorchScript with Custom C++ Operators" href="torch_script_custom_ops.html" />
<script src="../_static/js/modernizr.min.js"></script>
<!-- Preload the theme fonts -->
<link rel="preload" href="../_static/fonts/FreightSans/freight-sans-book.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="../_static/fonts/FreightSans/freight-sans-medium.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="../_static/fonts/IBMPlexMono/IBMPlexMono-Medium.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="../_static/fonts/FreightSans/freight-sans-bold.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="../_static/fonts/FreightSans/freight-sans-medium-italic.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="../_static/fonts/IBMPlexMono/IBMPlexMono-SemiBold.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<!-- Preload the katex fonts -->
<link rel="preload" href="https://cdn.jsdelivr.net/npm/katex@0.10.0/dist/fonts/KaTeX_Math-Italic.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/katex@0.10.0/dist/fonts/KaTeX_Main-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/katex@0.10.0/dist/fonts/KaTeX_Main-Bold.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/katex@0.10.0/dist/fonts/KaTeX_Size1-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/katex@0.10.0/dist/fonts/KaTeX_Size4-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/katex@0.10.0/dist/fonts/KaTeX_Size2-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/katex@0.10.0/dist/fonts/KaTeX_Size3-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/katex@0.10.0/dist/fonts/KaTeX_Caligraphic-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
</head>
<body class="pytorch-body">
<nav class="navbar sticky-top navbar-dark fixed-top navbar-expand-lg" style="background: rgba(55,55,55,.8)">
<div class="container-fluid">
<div class="navbar-brand">
<a href="https://pytorch.kr/" aria-label="PyTorch">
<img src="../_static/images/logo-kr.svg" width="260" height="28" fill="white" />
</a>
</div>
<button type="button" aria-label="Toggle navigation" class="navbar-toggler collapsed" aria-expanded="false" aria-controls="nav-collapse"><span class="navbar-toggler-icon"></span></button>
<div id="nav-collapse" class="navbar-collapse collapse">
<ul class="navbar-nav ml-auto">
<li class="nav-item">
<a href="//pytorch.kr/" target="_self" class="nav-link">ν</a></li>
<li class="nav-item">
<a href="//tutorials.pytorch.kr/" target="_self" class="nav-link">νν 리μΌ</a>
</li>
<li class="nav-item">
<a href="//pytorch.kr/about" target="_self" class="nav-link">
μκ°
</a></li>
</ul>
</div>
</div>
</nav>
<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">
1.6.0
</div>
<div role="search">
<form id="rtd-search-form" class="wy-form" action="../search.html" method="get">
<input type="text" name="q" placeholder="Search Tutorials" />
<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">νμ΄ν μΉ(PyTorch) λ μνΌ</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../recipes/recipes_index.html">λͺ¨λ λ μνΌ λ³΄κΈ°</a></li>
</ul>
<p class="caption"><span class="caption-text">νμ΄ν μΉ(PyTorch) λ°°μ°κΈ°</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../beginner/deep_learning_60min_blitz.html">νμ΄ν μΉ(PyTorch)λ‘ λ₯λ¬λνκΈ°: 60λΆλ§μ λμ₯λ΄κΈ°</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/pytorch_with_examples.html">μμ λ‘ λ°°μ°λ νμ΄ν μΉ(PyTorch)</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/nn_tutorial.html"><cite>torch.nn</cite> μ΄ <em>μ€μ λ‘</em> 무μμΈκ°μ?</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/tensorboard_tutorial.html">TensorBoardλ‘ λͺ¨λΈ, λ°μ΄ν°, νμ΅ μκ°ννκΈ°</a></li>
</ul>
<p class="caption"><span class="caption-text">μ΄λ―Έμ§/λΉλμ€</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/torchvision_tutorial.html">TorchVision κ°μ²΄ κ²μΆ λ―ΈμΈμ‘°μ (Finetuning) νν 리μΌ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/transfer_learning_tutorial.html">μ»΄ν¨ν° λΉμ (Vision)μ μν μ μ΄νμ΅(Transfer Learning)</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/fgsm_tutorial.html">μ λμ μμ μμ±(Adversarial Example Generation)</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/dcgan_faces_tutorial.html">DCGAN Tutorial</a></li>
</ul>
<p class="caption"><span class="caption-text">μ€λμ€</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../beginner/audio_preprocessing_tutorial.html">torchaudio Tutorial</a></li>
</ul>
<p class="caption"><span class="caption-text">ν
μ€νΈ</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../beginner/transformer_tutorial.html">nn.Transformer μ TorchText λ‘ μνμ€-ν¬-μνμ€(Sequence-to-Sequence) λͺ¨λΈλ§νκΈ°</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/char_rnn_classification_tutorial.html">κΈ°μ΄λΆν° μμνλ NLP: λ¬Έμ-λ¨μ RNNμΌλ‘ μ΄λ¦ λΆλ₯νκΈ°</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/char_rnn_generation_tutorial.html">κΈ°μ΄λΆν° μμνλ NLP: λ¬Έμ-λ¨μ RNNμΌλ‘ μ΄λ¦ μμ±νκΈ°</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/seq2seq_translation_tutorial.html">κΈ°μ΄λΆν° μμνλ NLP: Sequence to Sequence λ€νΈμν¬μ Attentionμ μ΄μ©ν λ²μ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/text_sentiment_ngrams_tutorial.html">SyntaxError</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/torchtext_translation_tutorial.html">TorchTextλ‘ μΈμ΄ λ²μνκΈ°</a></li>
</ul>
<p class="caption"><span class="caption-text">κ°ννμ΅</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/reinforcement_q_learning.html">κ°ν νμ΅ (DQN) νν 리μΌ</a></li>
</ul>
<p class="caption"><span class="caption-text">PyTorch λͺ¨λΈμ νλ‘λμ
νκ²½μ λ°°ν¬νκΈ°</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/flask_rest_api_tutorial.html">Flaskλ₯Ό μ΄μ©νμ¬ Pythonμμ PyTorchλ₯Ό REST APIλ‘ λ°°ν¬νκΈ°</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/Intro_to_TorchScript_tutorial.html">TorchScript μκ°</a></li>
<li class="toctree-l1"><a class="reference internal" href="cpp_export.html">C++μμ TorchScript λͺ¨λΈ λ‘λ©νκΈ°</a></li>
<li class="toctree-l1"><a class="reference internal" href="super_resolution_with_onnxruntime.html">(μ ν) PyTorch λͺ¨λΈμ ONNXμΌλ‘ λ³ννκ³ ONNX λ°νμμμ μ€ννκΈ°</a></li>
</ul>
<p class="caption"><span class="caption-text">νλ‘ νΈμλ API</span></p>
<ul class="current">
<li class="toctree-l1"><a class="reference internal" href="../intermediate/named_tensor_tutorial.html">(prototype) Introduction to Named Tensors in PyTorch</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/memory_format_tutorial.html">(beta) Channels Last Memory Format in PyTorch</a></li>
<li class="toctree-l1"><a class="reference internal" href="cpp_frontend.html">Using the PyTorch C++ Frontend</a></li>
<li class="toctree-l1"><a class="reference internal" href="cpp_extension.html">Custom C++ and CUDA Extensions</a></li>
<li class="toctree-l1"><a class="reference internal" href="torch_script_custom_ops.html">Extending TorchScript with Custom C++ Operators</a></li>
<li class="toctree-l1 current"><a class="current reference internal" href="#">Extending TorchScript with Custom C++ Classes</a></li>
<li class="toctree-l1"><a class="reference internal" href="torch-script-parallelism.html">Dynamic Parallelism in TorchScript</a></li>
<li class="toctree-l1"><a class="reference internal" href="cpp_autograd.html">Autograd in C++ Frontend</a></li>
<li class="toctree-l1"><a class="reference internal" href="dispatcher.html">Dispatcher in C++</a></li>
</ul>
<p class="caption"><span class="caption-text">λͺ¨λΈ μ΅μ ν</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/pruning_tutorial.html">Pruning Tutorial</a></li>
<li class="toctree-l1"><a class="reference internal" href="dynamic_quantization_tutorial.html">(beta) Dynamic Quantization on an LSTM Word Language Model</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/dynamic_quantization_bert_tutorial.html">(λ² ν) BERT λͺ¨λΈ λμ μμννκΈ°</a></li>
<li class="toctree-l1"><a class="reference internal" href="static_quantization_tutorial.html">(beta) Static Quantization with Eager Mode in PyTorch</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/quantized_transfer_learning_tutorial.html">(beta) μ»΄ν¨ν° λΉμ (Vision) νν 리μΌμ μν μμνλ μ μ΄νμ΅(Quantized Transfer Learning)</a></li>
</ul>
<p class="caption"><span class="caption-text">λ³λ ¬ λ° λΆμ° νμ΅</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../beginner/dist_overview.html">PyTorch Distributed Overview</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/model_parallel_tutorial.html">Single-Machine Model Parallel Best Practices</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/ddp_tutorial.html">Getting Started with Distributed Data Parallel</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/dist_tuto.html">PyTorchλ‘ λΆμ° μ΄ν리μΌμ΄μ
κ°λ°νκΈ°</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/rpc_tutorial.html">Getting Started with Distributed RPC Framework</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/aws_distributed_training_tutorial.html">(advanced) PyTorch 1.0 Distributed Trainer with Amazon AWS</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/rpc_param_server_tutorial.html">Implementing a Parameter Server Using Distributed RPC Framework</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/dist_pipeline_parallel_tutorial.html">Distributed Pipeline Parallelism Using RPC</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/rpc_async_execution.html">Implementing Batch RPC Processing Using Asynchronous Executions</a></li>
<li class="toctree-l1"><a class="reference internal" href="rpc_ddp_tutorial.html">Combining Distributed DataParallel with Distributed RPC Framework</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">
Tutorials
</a> >
</li>
<li>Extending TorchScript with Custom C++ Classes</li>
<li class="pytorch-breadcrumbs-aside">
<a href="../_sources/advanced/torch_script_custom_classes.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="pytorch-call-to-action-links">
<div id="tutorial-type">advanced/torch_script_custom_classes</div>
<div id="google-colab-link">
<img class="call-to-action-img" src="../_static/images/pytorch-colab.svg"/>
<div class="call-to-action-desktop-view">Run in Google Colab</div>
<div class="call-to-action-mobile-view">Colab</div>
</div>
<div id="download-notebook-link">
<img class="call-to-action-notebook-img" src="../_static/images/pytorch-download.svg"/>
<div class="call-to-action-desktop-view">Download Notebook</div>
<div class="call-to-action-mobile-view">Notebook</div>
</div>
<div id="github-view-link">
<img class="call-to-action-img" src="../_static/images/pytorch-github.svg"/>
<div class="call-to-action-desktop-view">View on GitHub</div>
<div class="call-to-action-mobile-view">GitHub</div>
</div>
</div>
<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="extending-torchscript-with-custom-c-classes">
<h1>Extending TorchScript with Custom C++ Classes<a class="headerlink" href="#extending-torchscript-with-custom-c-classes" title="Permalink to this headline">ΒΆ</a></h1>
<p>This tutorial is a follow-on to the
<a class="reference internal" href="torch_script_custom_ops.html"><span class="doc">custom operator</span></a>
tutorial, and introduces the API weβve built for binding C++ classes into TorchScript
and Python simultaneously. The API is very similar to
<a class="reference external" href="https://github.com/pybind/pybind11">pybind11</a>, and most of the concepts will transfer
over if youβre familiar with that system.</p>
<div class="section" id="implementing-and-binding-the-class-in-c">
<h2>Implementing and Binding the Class in C++<a class="headerlink" href="#implementing-and-binding-the-class-in-c" title="Permalink to this headline">ΒΆ</a></h2>
<p>For this tutorial, we are going to define a simple C++ class that maintains persistent
state in a member variable.</p>
<div class="highlight-cpp notranslate"><div class="highlight"><pre><span></span><span class="c1">// This header is all you need to do the C++ portions of this</span>
<span class="c1">// tutorial</span>
<span class="cp">#include</span> <span class="cpf"><torch/script.h></span><span class="cp"></span>
<span class="c1">// This header is what defines the custom class registration</span>
<span class="c1">// behavior specifically. script.h already includes this, but</span>
<span class="c1">// we include it here so you know it exists in case you want</span>
<span class="c1">// to look at the API or implementation.</span>
<span class="cp">#include</span> <span class="cpf"><torch/custom_class.h></span><span class="cp"></span>
<span class="cp">#include</span> <span class="cpf"><string></span><span class="cp"></span>
<span class="cp">#include</span> <span class="cpf"><vector></span><span class="cp"></span>
<span class="k">template</span> <span class="o"><</span><span class="k">class</span> <span class="nc">T</span><span class="o">></span>
<span class="k">struct</span> <span class="nl">MyStackClass</span> <span class="p">:</span> <span class="n">torch</span><span class="o">::</span><span class="n">CustomClassHolder</span> <span class="p">{</span>
<span class="n">std</span><span class="o">::</span><span class="n">vector</span><span class="o"><</span><span class="n">T</span><span class="o">></span> <span class="n">stack_</span><span class="p">;</span>
<span class="n">MyStackClass</span><span class="p">(</span><span class="n">std</span><span class="o">::</span><span class="n">vector</span><span class="o"><</span><span class="n">T</span><span class="o">></span> <span class="n">init</span><span class="p">)</span> <span class="o">:</span> <span class="n">stack_</span><span class="p">(</span><span class="n">init</span><span class="p">.</span><span class="n">begin</span><span class="p">(),</span> <span class="n">init</span><span class="p">.</span><span class="n">end</span><span class="p">())</span> <span class="p">{}</span>
<span class="kt">void</span> <span class="n">push</span><span class="p">(</span><span class="n">T</span> <span class="n">x</span><span class="p">)</span> <span class="p">{</span>
<span class="n">stack_</span><span class="p">.</span><span class="n">push_back</span><span class="p">(</span><span class="n">x</span><span class="p">);</span>
<span class="p">}</span>
<span class="n">T</span> <span class="n">pop</span><span class="p">()</span> <span class="p">{</span>
<span class="k">auto</span> <span class="n">val</span> <span class="o">=</span> <span class="n">stack_</span><span class="p">.</span><span class="n">back</span><span class="p">();</span>
<span class="n">stack_</span><span class="p">.</span><span class="n">pop_back</span><span class="p">();</span>
<span class="k">return</span> <span class="n">val</span><span class="p">;</span>
<span class="p">}</span>
<span class="n">c10</span><span class="o">::</span><span class="n">intrusive_ptr</span><span class="o"><</span><span class="n">MyStackClass</span><span class="o">></span> <span class="n">clone</span><span class="p">()</span> <span class="k">const</span> <span class="p">{</span>
<span class="k">return</span> <span class="n">c10</span><span class="o">::</span><span class="n">make_intrusive</span><span class="o"><</span><span class="n">MyStackClass</span><span class="o">></span><span class="p">(</span><span class="n">stack_</span><span class="p">);</span>
<span class="p">}</span>
<span class="kt">void</span> <span class="n">merge</span><span class="p">(</span><span class="k">const</span> <span class="n">c10</span><span class="o">::</span><span class="n">intrusive_ptr</span><span class="o"><</span><span class="n">MyStackClass</span><span class="o">>&</span> <span class="n">c</span><span class="p">)</span> <span class="p">{</span>
<span class="k">for</span> <span class="p">(</span><span class="k">auto</span><span class="o">&</span> <span class="nl">elem</span> <span class="p">:</span> <span class="n">c</span><span class="o">-></span><span class="n">stack_</span><span class="p">)</span> <span class="p">{</span>
<span class="n">push</span><span class="p">(</span><span class="n">elem</span><span class="p">);</span>
<span class="p">}</span>
<span class="p">}</span>
<span class="p">};</span>
</pre></div>
</div>
<p>There are several things to note:</p>
<ul class="simple">
<li><p><code class="docutils literal notranslate"><span class="pre">torch/custom_class.h</span></code> is the header you need to include to extend TorchScript
with your custom class.</p></li>
<li><p>Notice that whenever we are working with instances of the custom
class, we do it via instances of <code class="docutils literal notranslate"><span class="pre">c10::intrusive_ptr<></span></code>. Think of <code class="docutils literal notranslate"><span class="pre">intrusive_ptr</span></code>
as a smart pointer like <code class="docutils literal notranslate"><span class="pre">std::shared_ptr</span></code>, but the reference count is stored
directly in the object, as opposed to a separate metadata block (as is done in
<code class="docutils literal notranslate"><span class="pre">std::shared_ptr</span></code>. <code class="docutils literal notranslate"><span class="pre">torch::Tensor</span></code> internally uses the same pointer type;
and custom classes have to also use this pointer type so that we can
consistently manage different object types.</p></li>
<li><p>The second thing to notice is that the user-defined class must inherit from
<code class="docutils literal notranslate"><span class="pre">torch::CustomClassHolder</span></code>. This ensures that the custom class has space to
store the reference count.</p></li>
</ul>
<p>Now letβs take a look at how we will make this class visible to TorchScript, a process called
<em>binding</em> the class:</p>
<div class="highlight-cpp notranslate"><div class="highlight"><pre><span></span><span class="c1">// Notice a few things:</span>
<span class="c1">// - We pass the class to be registered as a template parameter to</span>
<span class="c1">// `torch::class_`. In this instance, we've passed the</span>
<span class="c1">// specialization of the MyStackClass class ``MyStackClass<std::string>``.</span>
<span class="c1">// In general, you cannot register a non-specialized template</span>
<span class="c1">// class. For non-templated classes, you can just pass the</span>
<span class="c1">// class name directly as the template parameter.</span>
<span class="c1">// - The arguments passed to the constructor make up the "qualified name"</span>
<span class="c1">// of the class. In this case, the registered class will appear in</span>
<span class="c1">// Python and C++ as `torch.classes.my_classes.MyStackClass`. We call</span>
<span class="c1">// the first argument the "namespace" and the second argument the</span>
<span class="c1">// actual class name.</span>
<span class="n">TORCH_LIBRARY</span><span class="p">(</span><span class="n">my_classes</span><span class="p">,</span> <span class="n">m</span><span class="p">)</span> <span class="p">{</span>
<span class="n">m</span><span class="p">.</span><span class="n">class_</span><span class="o"><</span><span class="n">MyStackClass</span><span class="o"><</span><span class="n">std</span><span class="o">::</span><span class="n">string</span><span class="o">>></span><span class="p">(</span><span class="s">"MyStackClass"</span><span class="p">)</span>
<span class="c1">// The following line registers the contructor of our MyStackClass</span>
<span class="c1">// class that takes a single `std::vector<std::string>` argument,</span>
<span class="c1">// i.e. it exposes the C++ method `MyStackClass(std::vector<T> init)`.</span>
<span class="c1">// Currently, we do not support registering overloaded</span>
<span class="c1">// constructors, so for now you can only `def()` one instance of</span>
<span class="c1">// `torch::init`.</span>
<span class="p">.</span><span class="n">def</span><span class="p">(</span><span class="n">torch</span><span class="o">::</span><span class="n">init</span><span class="o"><</span><span class="n">std</span><span class="o">::</span><span class="n">vector</span><span class="o"><</span><span class="n">std</span><span class="o">::</span><span class="n">string</span><span class="o">>></span><span class="p">())</span>
<span class="c1">// The next line registers a stateless (i.e. no captures) C++ lambda</span>
<span class="c1">// function as a method. Note that a lambda function must take a</span>
<span class="c1">// `c10::intrusive_ptr<YourClass>` (or some const/ref version of that)</span>
<span class="c1">// as the first argument. Other arguments can be whatever you want.</span>
<span class="p">.</span><span class="n">def</span><span class="p">(</span><span class="s">"top"</span><span class="p">,</span> <span class="p">[](</span><span class="k">const</span> <span class="n">c10</span><span class="o">::</span><span class="n">intrusive_ptr</span><span class="o"><</span><span class="n">MyStackClass</span><span class="o"><</span><span class="n">std</span><span class="o">::</span><span class="n">string</span><span class="o">>>&</span> <span class="n">self</span><span class="p">)</span> <span class="p">{</span>
<span class="k">return</span> <span class="n">self</span><span class="o">-></span><span class="n">stack_</span><span class="p">.</span><span class="n">back</span><span class="p">();</span>
<span class="p">})</span>
<span class="c1">// The following four lines expose methods of the MyStackClass<std::string></span>
<span class="c1">// class as-is. `torch::class_` will automatically examine the</span>
<span class="c1">// argument and return types of the passed-in method pointers and</span>
<span class="c1">// expose these to Python and TorchScript accordingly. Finally, notice</span>
<span class="c1">// that we must take the *address* of the fully-qualified method name,</span>
<span class="c1">// i.e. use the unary `&` operator, due to C++ typing rules.</span>
<span class="p">.</span><span class="n">def</span><span class="p">(</span><span class="s">"push"</span><span class="p">,</span> <span class="o">&</span><span class="n">MyStackClass</span><span class="o"><</span><span class="n">std</span><span class="o">::</span><span class="n">string</span><span class="o">>::</span><span class="n">push</span><span class="p">)</span>
<span class="p">.</span><span class="n">def</span><span class="p">(</span><span class="s">"pop"</span><span class="p">,</span> <span class="o">&</span><span class="n">MyStackClass</span><span class="o"><</span><span class="n">std</span><span class="o">::</span><span class="n">string</span><span class="o">>::</span><span class="n">pop</span><span class="p">)</span>
<span class="p">.</span><span class="n">def</span><span class="p">(</span><span class="s">"clone"</span><span class="p">,</span> <span class="o">&</span><span class="n">MyStackClass</span><span class="o"><</span><span class="n">std</span><span class="o">::</span><span class="n">string</span><span class="o">>::</span><span class="n">clone</span><span class="p">)</span>
<span class="p">.</span><span class="n">def</span><span class="p">(</span><span class="s">"merge"</span><span class="p">,</span> <span class="o">&</span><span class="n">MyStackClass</span><span class="o"><</span><span class="n">std</span><span class="o">::</span><span class="n">string</span><span class="o">>::</span><span class="n">merge</span><span class="p">)</span>
<span class="p">;</span>
<span class="p">}</span>
</pre></div>
</div>
</div>
<div class="section" id="building-the-example-as-a-c-project-with-cmake">
<h2>Building the Example as a C++ Project With CMake<a class="headerlink" href="#building-the-example-as-a-c-project-with-cmake" title="Permalink to this headline">ΒΆ</a></h2>
<p>Now, weβre going to build the above C++ code with the <a class="reference external" href="https://cmake.org">CMake</a> build system. First, take all the C++ code
weβve covered so far and place it in a file called <code class="docutils literal notranslate"><span class="pre">class.cpp</span></code>.
Then, write a simple <code class="docutils literal notranslate"><span class="pre">CMakeLists.txt</span></code> file and place it in the
same directory. Here is what <code class="docutils literal notranslate"><span class="pre">CMakeLists.txt</span></code> should look like:</p>
<div class="highlight-cmake notranslate"><div class="highlight"><pre><span></span><span class="nb">cmake_minimum_required</span><span class="p">(</span><span class="s">VERSION</span> <span class="s">3.1</span> <span class="s">FATAL_ERROR</span><span class="p">)</span>
<span class="nb">project</span><span class="p">(</span><span class="s">custom_class</span><span class="p">)</span>
<span class="nb">find_package</span><span class="p">(</span><span class="s">Torch</span> <span class="s">REQUIRED</span><span class="p">)</span>
<span class="c"># Define our library target</span>
<span class="nb">add_library</span><span class="p">(</span><span class="s">custom_class</span> <span class="s">SHARED</span> <span class="s">class.cpp</span><span class="p">)</span>
<span class="nb">set</span><span class="p">(</span><span class="s">CMAKE_CXX_STANDARD</span> <span class="s">14</span><span class="p">)</span>
<span class="c"># Link against LibTorch</span>
<span class="nb">target_link_libraries</span><span class="p">(</span><span class="s">custom_class</span> <span class="s2">"${TORCH_LIBRARIES}"</span><span class="p">)</span>
</pre></div>
</div>
<p>Also, create a <code class="docutils literal notranslate"><span class="pre">build</span></code> directory. Your file tree should look like this:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">custom_class_project</span><span class="o">/</span>
<span class="n">class</span><span class="o">.</span><span class="n">cpp</span>
<span class="n">CMakeLists</span><span class="o">.</span><span class="n">txt</span>
<span class="n">build</span><span class="o">/</span>
</pre></div>
</div>
<p>We assume youβve setup your environment in the same way as described in
the <a class="reference internal" href="torch_script_custom_ops.html"><span class="doc">previous tutorial</span></a>.
Go ahead and invoke cmake and then make to build the project:</p>
<div class="highlight-shell notranslate"><div class="highlight"><pre><span></span>$ <span class="nb">cd</span> build
$ cmake -DCMAKE_PREFIX_PATH<span class="o">=</span><span class="s2">"</span><span class="k">$(</span>python -c <span class="s1">'import torch.utils; print(torch.utils.cmake_prefix_path)'</span><span class="k">)</span><span class="s2">"</span> ..
-- The C compiler identification is GNU <span class="m">7</span>.3.1
-- The CXX compiler identification is GNU <span class="m">7</span>.3.1
-- Check <span class="k">for</span> working C compiler: /opt/rh/devtoolset-7/root/usr/bin/cc
-- Check <span class="k">for</span> working C compiler: /opt/rh/devtoolset-7/root/usr/bin/cc -- works
-- Detecting C compiler ABI info
-- Detecting C compiler ABI info - <span class="k">done</span>
-- Detecting C compile features
-- Detecting C compile features - <span class="k">done</span>
-- Check <span class="k">for</span> working CXX compiler: /opt/rh/devtoolset-7/root/usr/bin/c++
-- Check <span class="k">for</span> working CXX compiler: /opt/rh/devtoolset-7/root/usr/bin/c++ -- works
-- Detecting CXX compiler ABI info
-- Detecting CXX compiler ABI info - <span class="k">done</span>
-- Detecting CXX compile features
-- Detecting CXX compile features - <span class="k">done</span>
-- Looking <span class="k">for</span> pthread.h
-- Looking <span class="k">for</span> pthread.h - found
-- Looking <span class="k">for</span> pthread_create
-- Looking <span class="k">for</span> pthread_create - not found
-- Looking <span class="k">for</span> pthread_create in pthreads
-- Looking <span class="k">for</span> pthread_create in pthreads - not found
-- Looking <span class="k">for</span> pthread_create in pthread
-- Looking <span class="k">for</span> pthread_create in pthread - found
-- Found Threads: TRUE
-- Found torch: /torchbind_tutorial/libtorch/lib/libtorch.so
-- Configuring <span class="k">done</span>
-- Generating <span class="k">done</span>
-- Build files have been written to: /torchbind_tutorial/build
$ make -j
Scanning dependencies of target custom_class
<span class="o">[</span> <span class="m">50</span>%<span class="o">]</span> Building CXX object CMakeFiles/custom_class.dir/class.cpp.o
<span class="o">[</span><span class="m">100</span>%<span class="o">]</span> Linking CXX shared library libcustom_class.so
<span class="o">[</span><span class="m">100</span>%<span class="o">]</span> Built target custom_class
</pre></div>
</div>
<p>What youβll find is there is now (among other things) a dynamic library
file present in the build directory. On Linux, this is probably named
<code class="docutils literal notranslate"><span class="pre">libcustom_class.so</span></code>. So the file tree should look like:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">custom_class_project</span><span class="o">/</span>
<span class="n">class</span><span class="o">.</span><span class="n">cpp</span>
<span class="n">CMakeLists</span><span class="o">.</span><span class="n">txt</span>
<span class="n">build</span><span class="o">/</span>
<span class="n">libcustom_class</span><span class="o">.</span><span class="n">so</span>
</pre></div>
</div>
</div>
<div class="section" id="using-the-c-class-from-python-and-torchscript">
<h2>Using the C++ Class from Python and TorchScript<a class="headerlink" href="#using-the-c-class-from-python-and-torchscript" title="Permalink to this headline">ΒΆ</a></h2>
<p>Now that we have our class and its registration compiled into an <code class="docutils literal notranslate"><span class="pre">.so</span></code> file,
we can load that <cite>.so</cite> into Python and try it out. Hereβs a script that
demonstrates that:</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="kn">import</span> <span class="nn">torch</span>
<span class="c1"># `torch.classes.load_library()` allows you to pass the path to your .so file</span>
<span class="c1"># to load it in and make the custom C++ classes available to both Python and</span>
<span class="c1"># TorchScript</span>
<span class="n">torch</span><span class="o">.</span><span class="n">classes</span><span class="o">.</span><span class="n">load_library</span><span class="p">(</span><span class="s2">"build/libcustom_class.so"</span><span class="p">)</span>
<span class="c1"># You can query the loaded libraries like this:</span>
<span class="k">print</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">classes</span><span class="o">.</span><span class="n">loaded_libraries</span><span class="p">)</span>
<span class="c1"># prints {'/custom_class_project/build/libcustom_class.so'}</span>
<span class="c1"># We can find and instantiate our custom C++ class in python by using the</span>
<span class="c1"># `torch.classes` namespace:</span>
<span class="c1">#</span>
<span class="c1"># This instantiation will invoke the MyStackClass(std::vector<T> init)</span>
<span class="c1"># constructor we registered earlier</span>
<span class="n">s</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">classes</span><span class="o">.</span><span class="n">my_classes</span><span class="o">.</span><span class="n">MyStackClass</span><span class="p">([</span><span class="s2">"foo"</span><span class="p">,</span> <span class="s2">"bar"</span><span class="p">])</span>
<span class="c1"># We can call methods in Python</span>
<span class="n">s</span><span class="o">.</span><span class="n">push</span><span class="p">(</span><span class="s2">"pushed"</span><span class="p">)</span>
<span class="k">assert</span> <span class="n">s</span><span class="o">.</span><span class="n">pop</span><span class="p">()</span> <span class="o">==</span> <span class="s2">"pushed"</span>
<span class="c1"># Returning and passing instances of custom classes works as you'd expect</span>
<span class="n">s2</span> <span class="o">=</span> <span class="n">s</span><span class="o">.</span><span class="n">clone</span><span class="p">()</span>
<span class="n">s</span><span class="o">.</span><span class="n">merge</span><span class="p">(</span><span class="n">s2</span><span class="p">)</span>
<span class="k">for</span> <span class="n">expected</span> <span class="ow">in</span> <span class="p">[</span><span class="s2">"bar"</span><span class="p">,</span> <span class="s2">"foo"</span><span class="p">,</span> <span class="s2">"bar"</span><span class="p">,</span> <span class="s2">"foo"</span><span class="p">]:</span>
<span class="k">assert</span> <span class="n">s</span><span class="o">.</span><span class="n">pop</span><span class="p">()</span> <span class="o">==</span> <span class="n">expected</span>
<span class="c1"># We can also use the class in TorchScript</span>
<span class="c1"># For now, we need to assign the class's type to a local in order to</span>
<span class="c1"># annotate the type on the TorchScript function. This may change</span>
<span class="c1"># in the future.</span>
<span class="n">MyStackClass</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">classes</span><span class="o">.</span><span class="n">my_classes</span><span class="o">.</span><span class="n">MyStackClass</span>
<span class="nd">@torch.jit.script</span>
<span class="k">def</span> <span class="nf">do_stacks</span><span class="p">(</span><span class="n">s</span><span class="p">:</span> <span class="n">MyStackClass</span><span class="p">):</span> <span class="c1"># We can pass a custom class instance</span>
<span class="c1"># We can instantiate the class</span>
<span class="n">s2</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">classes</span><span class="o">.</span><span class="n">my_classes</span><span class="o">.</span><span class="n">MyStackClass</span><span class="p">([</span><span class="s2">"hi"</span><span class="p">,</span> <span class="s2">"mom"</span><span class="p">])</span>
<span class="n">s2</span><span class="o">.</span><span class="n">merge</span><span class="p">(</span><span class="n">s</span><span class="p">)</span> <span class="c1"># We can call a method on the class</span>
<span class="c1"># We can also return instances of the class</span>
<span class="c1"># from TorchScript function/methods</span>
<span class="k">return</span> <span class="n">s2</span><span class="o">.</span><span class="n">clone</span><span class="p">(),</span> <span class="n">s2</span><span class="o">.</span><span class="n">top</span><span class="p">()</span>
<span class="n">stack</span><span class="p">,</span> <span class="n">top</span> <span class="o">=</span> <span class="n">do_stacks</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">classes</span><span class="o">.</span><span class="n">my_classes</span><span class="o">.</span><span class="n">MyStackClass</span><span class="p">([</span><span class="s2">"wow"</span><span class="p">]))</span>
<span class="k">assert</span> <span class="n">top</span> <span class="o">==</span> <span class="s2">"wow"</span>
<span class="k">for</span> <span class="n">expected</span> <span class="ow">in</span> <span class="p">[</span><span class="s2">"wow"</span><span class="p">,</span> <span class="s2">"mom"</span><span class="p">,</span> <span class="s2">"hi"</span><span class="p">]:</span>
<span class="k">assert</span> <span class="n">stack</span><span class="o">.</span><span class="n">pop</span><span class="p">()</span> <span class="o">==</span> <span class="n">expected</span>
</pre></div>
</div>
</div>
<div class="section" id="saving-loading-and-running-torchscript-code-using-custom-classes">
<h2>Saving, Loading, and Running TorchScript Code Using Custom Classes<a class="headerlink" href="#saving-loading-and-running-torchscript-code-using-custom-classes" title="Permalink to this headline">ΒΆ</a></h2>
<p>We can also use custom-registered C++ classes in a C++ process using
libtorch. As an example, letβs define a simple <code class="docutils literal notranslate"><span class="pre">nn.Module</span></code> that
instantiates and calls a method on our MyStackClass class:</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="kn">import</span> <span class="nn">torch</span>
<span class="n">torch</span><span class="o">.</span><span class="n">classes</span><span class="o">.</span><span class="n">load_library</span><span class="p">(</span><span class="s1">'build/libcustom_class.so'</span><span class="p">)</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">nn</span><span class="o">.</span><span class="n">Module</span><span class="p">):</span>
<span class="k">def</span> <span class="fm">__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="o">.</span><span class="fm">__init__</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">s</span><span class="p">:</span> <span class="nb">str</span><span class="p">)</span> <span class="o">-></span> <span class="nb">str</span><span class="p">:</span>
<span class="n">stack</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">classes</span><span class="o">.</span><span class="n">my_classes</span><span class="o">.</span><span class="n">MyStackClass</span><span class="p">([</span><span class="s2">"hi"</span><span class="p">,</span> <span class="s2">"mom"</span><span class="p">])</span>
<span class="k">return</span> <span class="n">stack</span><span class="o">.</span><span class="n">pop</span><span class="p">()</span> <span class="o">+</span> <span class="n">s</span>
<span class="n">scripted_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">script</span><span class="p">(</span><span class="n">Foo</span><span class="p">())</span>
<span class="k">print</span><span class="p">(</span><span class="n">scripted_foo</span><span class="o">.</span><span class="n">graph</span><span class="p">)</span>
<span class="n">scripted_foo</span><span class="o">.</span><span class="n">save</span><span class="p">(</span><span class="s1">'foo.pt'</span><span class="p">)</span>
</pre></div>
</div>
<p><code class="docutils literal notranslate"><span class="pre">foo.pt</span></code> in our filesystem now contains the serialized TorchScript
program weβve just defined.</p>
<p>Now, weβre going to define a new CMake project to show how you can load
this model and its required .so file. For a full treatment of how to do this,
please have a look at the <a class="reference external" href="https://pytorch.org/tutorials/advanced/cpp_export.html">Loading a TorchScript Model in C++ Tutorial</a>.</p>
<p>Similarly to before, letβs create a file structure containing the following:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">cpp_inference_example</span><span class="o">/</span>
<span class="n">infer</span><span class="o">.</span><span class="n">cpp</span>
<span class="n">CMakeLists</span><span class="o">.</span><span class="n">txt</span>
<span class="n">foo</span><span class="o">.</span><span class="n">pt</span>
<span class="n">build</span><span class="o">/</span>
<span class="n">custom_class_project</span><span class="o">/</span>
<span class="n">class</span><span class="o">.</span><span class="n">cpp</span>
<span class="n">CMakeLists</span><span class="o">.</span><span class="n">txt</span>
<span class="n">build</span><span class="o">/</span>
</pre></div>
</div>
<p>Notice weβve copied over the serialized <code class="docutils literal notranslate"><span class="pre">foo.pt</span></code> file, as well as the source
tree from the <code class="docutils literal notranslate"><span class="pre">custom_class_project</span></code> above. We will be adding the
<code class="docutils literal notranslate"><span class="pre">custom_class_project</span></code> as a dependency to this C++ project so that we can
build the custom class into the binary.</p>
<p>Letβs populate <code class="docutils literal notranslate"><span class="pre">infer.cpp</span></code> with the following:</p>
<div class="highlight-cpp notranslate"><div class="highlight"><pre><span></span><span class="cp">#include</span> <span class="cpf"><torch/script.h></span><span class="cp"></span>
<span class="cp">#include</span> <span class="cpf"><iostream></span><span class="cp"></span>
<span class="cp">#include</span> <span class="cpf"><memory></span><span class="cp"></span>
<span class="kt">int</span> <span class="nf">main</span><span class="p">(</span><span class="kt">int</span> <span class="n">argc</span><span class="p">,</span> <span class="k">const</span> <span class="kt">char</span><span class="o">*</span> <span class="n">argv</span><span class="p">[])</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">Module</span> <span class="n">module</span><span class="p">;</span>
<span class="k">try</span> <span class="p">{</span>
<span class="c1">// Deserialize the ScriptModule from a file using torch::jit::load().</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">load</span><span class="p">(</span><span class="s">"foo.pt"</span><span class="p">);</span>
<span class="p">}</span>
<span class="k">catch</span> <span class="p">(</span><span class="k">const</span> <span class="n">c10</span><span class="o">::</span><span class="n">Error</span><span class="o">&</span> <span class="n">e</span><span class="p">)</span> <span class="p">{</span>
<span class="n">std</span><span class="o">::</span><span class="n">cerr</span> <span class="o"><<</span> <span class="s">"error loading the model</span><span class="se">\n</span><span class="s">"</span><span class="p">;</span>
<span class="k">return</span> <span class="o">-</span><span class="mi">1</span><span class="p">;</span>
<span class="p">}</span>
<span class="n">std</span><span class="o">::</span><span class="n">vector</span><span class="o"><</span><span class="n">c10</span><span class="o">::</span><span class="n">IValue</span><span class="o">></span> <span class="n">inputs</span> <span class="o">=</span> <span class="p">{</span><span class="s">"foobarbaz"</span><span class="p">};</span>
<span class="k">auto</span> <span class="n">output</span> <span class="o">=</span> <span class="n">module</span><span class="p">.</span><span class="n">forward</span><span class="p">(</span><span class="n">inputs</span><span class="p">).</span><span class="n">toString</span><span class="p">();</span>
<span class="n">std</span><span class="o">::</span><span class="n">cout</span> <span class="o"><<</span> <span class="n">output</span><span class="o">-></span><span class="n">string</span><span class="p">()</span> <span class="o"><<</span> <span class="n">std</span><span class="o">::</span><span class="n">endl</span><span class="p">;</span>
<span class="p">}</span>
</pre></div>
</div>
<p>And similarly letβs define our CMakeLists.txt file:</p>
<div class="highlight-cpp notranslate"><div class="highlight"><pre><span></span><span class="n">cmake_minimum_required</span><span class="p">(</span><span class="n">VERSION</span> <span class="mf">3.1</span> <span class="n">FATAL_ERROR</span><span class="p">)</span>
<span class="n">project</span><span class="p">(</span><span class="n">infer</span><span class="p">)</span>
<span class="n">find_package</span><span class="p">(</span><span class="n">Torch</span> <span class="n">REQUIRED</span><span class="p">)</span>
<span class="n">add_subdirectory</span><span class="p">(</span><span class="n">custom_class_project</span><span class="p">)</span>
<span class="cp"># Define our library target</span>
<span class="n">add_executable</span><span class="p">(</span><span class="n">infer</span> <span class="n">infer</span><span class="p">.</span><span class="n">cpp</span><span class="p">)</span>
<span class="n">set</span><span class="p">(</span><span class="n">CMAKE_CXX_STANDARD</span> <span class="mi">14</span><span class="p">)</span>
<span class="cp"># Link against LibTorch</span>
<span class="n">target_link_libraries</span><span class="p">(</span><span class="n">infer</span> <span class="s">"${TORCH_LIBRARIES}"</span><span class="p">)</span>
<span class="cp"># This is where we link in our libcustom_class code, making our</span>
<span class="cp"># custom class available in our binary.</span>
<span class="n">target_link_libraries</span><span class="p">(</span><span class="n">infer</span> <span class="o">-</span><span class="n">Wl</span><span class="p">,</span><span class="o">--</span><span class="n">no</span><span class="o">-</span><span class="n">as</span><span class="o">-</span><span class="n">needed</span> <span class="n">custom_class</span><span class="p">)</span>
</pre></div>
</div>
<p>You know the drill: <code class="docutils literal notranslate"><span class="pre">cd</span> <span class="pre">build</span></code>, <code class="docutils literal notranslate"><span class="pre">cmake</span></code>, and <code class="docutils literal notranslate"><span class="pre">make</span></code>:</p>
<div class="highlight-shell notranslate"><div class="highlight"><pre><span></span>$ <span class="nb">cd</span> build
$ cmake -DCMAKE_PREFIX_PATH<span class="o">=</span><span class="s2">"</span><span class="k">$(</span>python -c <span class="s1">'import torch.utils; print(torch.utils.cmake_prefix_path)'</span><span class="k">)</span><span class="s2">"</span> ..
-- The C compiler identification is GNU <span class="m">7</span>.3.1
-- The CXX compiler identification is GNU <span class="m">7</span>.3.1
-- Check <span class="k">for</span> working C compiler: /opt/rh/devtoolset-7/root/usr/bin/cc
-- Check <span class="k">for</span> working C compiler: /opt/rh/devtoolset-7/root/usr/bin/cc -- works
-- Detecting C compiler ABI info
-- Detecting C compiler ABI info - <span class="k">done</span>
-- Detecting C compile features
-- Detecting C compile features - <span class="k">done</span>
-- Check <span class="k">for</span> working CXX compiler: /opt/rh/devtoolset-7/root/usr/bin/c++
-- Check <span class="k">for</span> working CXX compiler: /opt/rh/devtoolset-7/root/usr/bin/c++ -- works
-- Detecting CXX compiler ABI info
-- Detecting CXX compiler ABI info - <span class="k">done</span>
-- Detecting CXX compile features
-- Detecting CXX compile features - <span class="k">done</span>
-- Looking <span class="k">for</span> pthread.h
-- Looking <span class="k">for</span> pthread.h - found
-- Looking <span class="k">for</span> pthread_create
-- Looking <span class="k">for</span> pthread_create - not found
-- Looking <span class="k">for</span> pthread_create in pthreads
-- Looking <span class="k">for</span> pthread_create in pthreads - not found
-- Looking <span class="k">for</span> pthread_create in pthread
-- Looking <span class="k">for</span> pthread_create in pthread - found
-- Found Threads: TRUE
-- Found torch: /local/miniconda3/lib/python3.7/site-packages/torch/lib/libtorch.so
-- Configuring <span class="k">done</span>
-- Generating <span class="k">done</span>
-- Build files have been written to: /cpp_inference_example/build
$ make -j
Scanning dependencies of target custom_class
<span class="o">[</span> <span class="m">25</span>%<span class="o">]</span> Building CXX object custom_class_project/CMakeFiles/custom_class.dir/class.cpp.o
<span class="o">[</span> <span class="m">50</span>%<span class="o">]</span> Linking CXX shared library libcustom_class.so
<span class="o">[</span> <span class="m">50</span>%<span class="o">]</span> Built target custom_class
Scanning dependencies of target infer
<span class="o">[</span> <span class="m">75</span>%<span class="o">]</span> Building CXX object CMakeFiles/infer.dir/infer.cpp.o
<span class="o">[</span><span class="m">100</span>%<span class="o">]</span> Linking CXX executable infer
<span class="o">[</span><span class="m">100</span>%<span class="o">]</span> Built target infer
</pre></div>
</div>
<p>And now we can run our exciting C++ binary:</p>
<div class="highlight-shell notranslate"><div class="highlight"><pre><span></span>$ ./infer
momfoobarbaz
</pre></div>
</div>
<p>Incredible!</p>
</div>
<div class="section" id="moving-custom-classes-to-from-ivalues">
<h2>Moving Custom Classes To/From IValues<a class="headerlink" href="#moving-custom-classes-to-from-ivalues" title="Permalink to this headline">ΒΆ</a></h2>
<p>Itβs also possible that you may need to move custom classes into or out of
<code class="docutils literal notranslate"><span class="pre">IValue``s,</span> <span class="pre">such</span> <span class="pre">as</span> <span class="pre">when</span> <span class="pre">you</span> <span class="pre">take</span> <span class="pre">or</span> <span class="pre">return</span> <span class="pre">``IValue``s</span> <span class="pre">from</span> <span class="pre">TorchScript</span> <span class="pre">methods</span>
<span class="pre">or</span> <span class="pre">you</span> <span class="pre">want</span> <span class="pre">to</span> <span class="pre">instantiate</span> <span class="pre">a</span> <span class="pre">custom</span> <span class="pre">class</span> <span class="pre">attribute</span> <span class="pre">in</span> <span class="pre">C++.</span> <span class="pre">For</span> <span class="pre">creating</span> <span class="pre">an</span>
<span class="pre">``IValue</span></code> from a custom C++ class instance:</p>
<ul class="simple">
<li><p><code class="docutils literal notranslate"><span class="pre">torch::make_custom_class<T>()</span></code> provides an API similar to c10::intrusive_ptr<T>
in that it will take whatever set of arguments you provide to it, call the constructor
of T that matches that set of arguments, and wrap that instance up and return it.
However, instead of returning just a pointer to a custom class object, it returns
an <code class="docutils literal notranslate"><span class="pre">IValue</span></code> wrapping the object. You can then pass this <code class="docutils literal notranslate"><span class="pre">IValue</span></code> directly to
TorchScript.</p></li>
<li><p>In the event that you already have an <code class="docutils literal notranslate"><span class="pre">intrusive_ptr</span></code> pointing to your class, you
can directly construct an IValue from it using the constructor <code class="docutils literal notranslate"><span class="pre">IValue(intrusive_ptr<T>)</span></code>.</p></li>
</ul>
<p>For converting <code class="docutils literal notranslate"><span class="pre">IValue</span></code> back to custom classes:</p>
<ul class="simple">
<li><p><code class="docutils literal notranslate"><span class="pre">IValue::toCustomClass<T>()</span></code> will return an <code class="docutils literal notranslate"><span class="pre">intrusive_ptr<T></span></code> pointing to the
custom class that the <code class="docutils literal notranslate"><span class="pre">IValue</span></code> contains. Internally, this function is checking
that <code class="docutils literal notranslate"><span class="pre">T</span></code> is registered as a custom class and that the <code class="docutils literal notranslate"><span class="pre">IValue</span></code> does in fact contain
a custom class. You can check whether the <code class="docutils literal notranslate"><span class="pre">IValue</span></code> contains a custom class manually by
calling <code class="docutils literal notranslate"><span class="pre">isCustomClass()</span></code>.</p></li>
</ul>
</div>
<div class="section" id="defining-serialization-deserialization-methods-for-custom-c-classes">
<h2>Defining Serialization/Deserialization Methods for Custom C++ Classes<a class="headerlink" href="#defining-serialization-deserialization-methods-for-custom-c-classes" title="Permalink to this headline">ΒΆ</a></h2>
<p>If you try to save a <code class="docutils literal notranslate"><span class="pre">ScriptModule</span></code> with a custom-bound C++ class as
an attribute, youβll get the following error:</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="c1"># export_attr.py</span>
<span class="kn">import</span> <span class="nn">torch</span>
<span class="n">torch</span><span class="o">.</span><span class="n">classes</span><span class="o">.</span><span class="n">load_library</span><span class="p">(</span><span class="s1">'build/libcustom_class.so'</span><span class="p">)</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">nn</span><span class="o">.</span><span class="n">Module</span><span class="p">):</span>
<span class="k">def</span> <span class="fm">__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="o">.</span><span class="fm">__init__</span><span class="p">()</span>
<span class="bp">self</span><span class="o">.</span><span class="n">stack</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">classes</span><span class="o">.</span><span class="n">my_classes</span><span class="o">.</span><span class="n">MyStackClass</span><span class="p">([</span><span class="s2">"just"</span><span class="p">,</span> <span class="s2">"testing"</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">s</span><span class="p">:</span> <span class="nb">str</span><span class="p">)</span> <span class="o">-></span> <span class="nb">str</span><span class="p">:</span>
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">stack</span><span class="o">.</span><span class="n">pop</span><span class="p">()</span> <span class="o">+</span> <span class="n">s</span>
<span class="n">scripted_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">script</span><span class="p">(</span><span class="n">Foo</span><span class="p">())</span>
<span class="n">scripted_foo</span><span class="o">.</span><span class="n">save</span><span class="p">(</span><span class="s1">'foo.pt'</span><span class="p">)</span>
<span class="n">loaded</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">load</span><span class="p">(</span><span class="s1">'foo.pt'</span><span class="p">)</span>
<span class="k">print</span><span class="p">(</span><span class="n">loaded</span><span class="o">.</span><span class="n">stack</span><span class="o">.</span><span class="n">pop</span><span class="p">())</span>
</pre></div>
</div>
<div class="highlight-shell notranslate"><div class="highlight"><pre><span></span>$ python export_attr.py
RuntimeError: Cannot serialize custom bound C++ class __torch__.torch.classes.my_classes.MyStackClass. Please define serialization methods via def_pickle <span class="k">for</span> this class. <span class="o">(</span>pushIValueImpl at ../torch/csrc/jit/pickler.cpp:128<span class="o">)</span>
</pre></div>
</div>
<p>This is because TorchScript cannot automatically figure out what information
save from your C++ class. You must specify that manually. The way to do that
is to define <code class="docutils literal notranslate"><span class="pre">__getstate__</span></code> and <code class="docutils literal notranslate"><span class="pre">__setstate__</span></code> methods on the class using
the special <code class="docutils literal notranslate"><span class="pre">def_pickle</span></code> method on <code class="docutils literal notranslate"><span class="pre">class_</span></code>.</p>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>The semantics of <code class="docutils literal notranslate"><span class="pre">__getstate__</span></code> and <code class="docutils literal notranslate"><span class="pre">__setstate__</span></code> in TorchScript are
equivalent to that of the Python pickle module. You can
<a class="reference external" href="https://github.com/pytorch/pytorch/blob/master/torch/csrc/jit/docs/serialization.md#getstate-and-setstate">read more</a>
about how we use these methods.</p>
</div>
<p>Here is an example of the <code class="docutils literal notranslate"><span class="pre">def_pickle</span></code> call we can add to the registration of
<code class="docutils literal notranslate"><span class="pre">MyStackClass</span></code> to include serialization methods:</p>
<div class="highlight-cpp notranslate"><div class="highlight"><pre><span></span> <span class="c1">// class_<>::def_pickle allows you to define the serialization</span>
<span class="c1">// and deserialization methods for your C++ class.</span>
<span class="c1">// Currently, we only support passing stateless lambda functions</span>
<span class="c1">// as arguments to def_pickle</span>
<span class="p">.</span><span class="n">def_pickle</span><span class="p">(</span>
<span class="c1">// __getstate__</span>
<span class="c1">// This function defines what data structure should be produced</span>
<span class="c1">// when we serialize an instance of this class. The function</span>
<span class="c1">// must take a single `self` argument, which is an intrusive_ptr</span>
<span class="c1">// to the instance of the object. The function can return</span>
<span class="c1">// any type that is supported as a return value of the TorchScript</span>
<span class="c1">// custom operator API. In this instance, we've chosen to return</span>
<span class="c1">// a std::vector<std::string> as the salient data to preserve</span>
<span class="c1">// from the class.</span>
<span class="p">[](</span><span class="k">const</span> <span class="n">c10</span><span class="o">::</span><span class="n">intrusive_ptr</span><span class="o"><</span><span class="n">MyStackClass</span><span class="o"><</span><span class="n">std</span><span class="o">::</span><span class="n">string</span><span class="o">>>&</span> <span class="n">self</span><span class="p">)</span>
<span class="o">-></span> <span class="n">std</span><span class="o">::</span><span class="n">vector</span><span class="o"><</span><span class="n">std</span><span class="o">::</span><span class="n">string</span><span class="o">></span> <span class="p">{</span>
<span class="k">return</span> <span class="n">self</span><span class="o">-></span><span class="n">stack_</span><span class="p">;</span>
<span class="p">},</span>
<span class="c1">// __setstate__</span>
<span class="c1">// This function defines how to create a new instance of the C++</span>
<span class="c1">// class when we are deserializing. The function must take a</span>
<span class="c1">// single argument of the same type as the return value of</span>
<span class="c1">// `__getstate__`. The function must return an intrusive_ptr</span>
<span class="c1">// to a new instance of the C++ class, initialized however</span>
<span class="c1">// you would like given the serialized state.</span>
<span class="p">[](</span><span class="n">std</span><span class="o">::</span><span class="n">vector</span><span class="o"><</span><span class="n">std</span><span class="o">::</span><span class="n">string</span><span class="o">></span> <span class="n">state</span><span class="p">)</span>
<span class="o">-></span> <span class="n">c10</span><span class="o">::</span><span class="n">intrusive_ptr</span><span class="o"><</span><span class="n">MyStackClass</span><span class="o"><</span><span class="n">std</span><span class="o">::</span><span class="n">string</span><span class="o">>></span> <span class="p">{</span>
<span class="c1">// A convenient way to instantiate an object and get an</span>
<span class="c1">// intrusive_ptr to it is via `make_intrusive`. We use</span>
<span class="c1">// that here to allocate an instance of MyStackClass<std::string></span>
<span class="c1">// and call the single-argument std::vector<std::string></span>
<span class="c1">// constructor with the serialized state.</span>
<span class="k">return</span> <span class="n">c10</span><span class="o">::</span><span class="n">make_intrusive</span><span class="o"><</span><span class="n">MyStackClass</span><span class="o"><</span><span class="n">std</span><span class="o">::</span><span class="n">string</span><span class="o">>></span><span class="p">(</span><span class="n">std</span><span class="o">::</span><span class="n">move</span><span class="p">(</span><span class="n">state</span><span class="p">));</span>
<span class="p">});</span>
</pre></div>
</div>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>We take a different approach from pybind11 in the pickle API. Whereas pybind11
as a special function <code class="docutils literal notranslate"><span class="pre">pybind11::pickle()</span></code> which you pass into <code class="docutils literal notranslate"><span class="pre">class_::def()</span></code>,
we have a separate method <code class="docutils literal notranslate"><span class="pre">def_pickle</span></code> for this purpose. This is because the
name <code class="docutils literal notranslate"><span class="pre">torch::jit::pickle</span></code> was already taken, and we didnβt want to cause confusion.</p>
</div>
<p>Once we have defined the (de)serialization behavior in this way, our script can
now run successfully:</p>
<div class="highlight-shell notranslate"><div class="highlight"><pre><span></span>$ python ../export_attr.py
testing
</pre></div>
</div>
</div>
<div class="section" id="defining-custom-operators-that-take-or-return-bound-c-classes">
<h2>Defining Custom Operators that Take or Return Bound C++ Classes<a class="headerlink" href="#defining-custom-operators-that-take-or-return-bound-c-classes" title="Permalink to this headline">ΒΆ</a></h2>
<p>Once youβve defined a custom C++ class, you can also use that class
as an argument or return from a custom operator (i.e. free functions). Suppose
you have the following free function:</p>
<div class="highlight-cpp notranslate"><div class="highlight"><pre><span></span><span class="n">c10</span><span class="o">::</span><span class="n">intrusive_ptr</span><span class="o"><</span><span class="n">MyStackClass</span><span class="o"><</span><span class="n">std</span><span class="o">::</span><span class="n">string</span><span class="o">>></span> <span class="n">manipulate_instance</span><span class="p">(</span><span class="k">const</span> <span class="n">c10</span><span class="o">::</span><span class="n">intrusive_ptr</span><span class="o"><</span><span class="n">MyStackClass</span><span class="o"><</span><span class="n">std</span><span class="o">::</span><span class="n">string</span><span class="o">>>&</span> <span class="n">instance</span><span class="p">)</span> <span class="p">{</span>
<span class="n">instance</span><span class="o">-></span><span class="n">pop</span><span class="p">();</span>
<span class="k">return</span> <span class="n">instance</span><span class="p">;</span>
<span class="p">}</span>
</pre></div>
</div>
<p>You can register it running the following code inside your <code class="docutils literal notranslate"><span class="pre">TORCH_LIBRARY</span></code>
block:</p>
<div class="highlight-cpp notranslate"><div class="highlight"><pre><span></span> <span class="n">m</span><span class="p">.</span><span class="n">def</span><span class="p">(</span>
<span class="s">"foo::manipulate_instance(__torch__.torch.classes.my_classes.MyStackClass x) -> __torch__.torch.classes.my_classes.MyStackClass Y"</span><span class="p">,</span>
<span class="n">manipulate_instance</span>
<span class="p">);</span>
</pre></div>
</div>
<p>Refer to the <a class="reference external" href="https://pytorch.org/tutorials/advanced/torch_script_custom_ops.html">custom op tutorial</a>
for more details on the registration API.</p>
<p>Once this is done, you can use the op like the following example:</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="k">class</span> <span class="nc">TryCustomOp</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="fm">__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">TryCustomOp</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">f</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">classes</span><span class="o">.</span><span class="n">my_classes</span><span class="o">.</span><span class="n">MyStackClass</span><span class="p">([</span><span class="s2">"foo"</span><span class="p">,</span> <span class="s2">"bar"</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="k">return</span> <span class="n">torch</span><span class="o">.</span><span class="n">ops</span><span class="o">.</span><span class="n">foo</span><span class="o">.</span><span class="n">manipulate_instance</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">f</span><span class="p">)</span>
</pre></div>
</div>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>Registration of an operator that takes a C++ class as an argument requires that
the custom class has already been registered. You can enforce this by
making sure the custom class registration and your free function definitions
are in the same <code class="docutils literal notranslate"><span class="pre">TORCH_LIBRARY</span></code> block, and that the custom class
registration comes first. In the future, we may relax this requirement,
so that these can be registered in any order.</p>
</div>
</div>
<div class="section" id="conclusion">
<h2>Conclusion<a class="headerlink" href="#conclusion" title="Permalink to this headline">ΒΆ</a></h2>
<p>This tutorial walked you through how to expose a C++ class to TorchScript
(and by extension Python), how to register its methods, how to use that
class from Python and TorchScript, and how to save and load code using
the class and run that code in a standalone C++ process. You are now ready
to extend your TorchScript models with C++ classes that interface with
third party C++ libraries or implement any other use case that requires the
lines between Python, TorchScript and C++ to blend smoothly.</p>
<p>As always, if you run into any problems or have questions, you can use our
<a class="reference external" href="https://discuss.pytorch.org/">forum</a> or <a class="reference external" href="https://github.com/pytorch/pytorch/issues">GitHub issues</a> to get in touch. Also, our
<a class="reference external" href="https://pytorch.org/cppdocs/notes/faq.html">frequently asked questions (FAQ) page</a> may have helpful information.</p>
</div>
</div>
</article>
</div>
<footer>
<div class="rst-footer-buttons" role="navigation" aria-label="footer navigation">
<a href="torch-script-parallelism.html" class="btn btn-neutral float-right" title="Dynamic Parallelism in TorchScript" accesskey="n" rel="next">Next <img src="../_static/images/chevron-right-orange.svg" class="next-page"></a>
<a href="torch_script_custom_ops.html" class="btn btn-neutral" title="Extending TorchScript with Custom C++ Operators" accesskey="p" rel="prev"><img src="../_static/images/chevron-right-orange.svg" class="previous-page"> Previous</a>
</div>
<hr class="helpful-hr hr-top">
<div class="helpful-container">
<div class="helpful-question">μ΄ λ¬Έμκ° λμμ΄ λμλμ?</div>
<div class="helpful-question yes-link" data-behavior="was-this-helpful-event" data-response="yes">λ€</div>
<div class="helpful-question no-link" data-behavior="was-this-helpful-event" data-response="no">μλμ€</div>
<div class="was-helpful-thank-you">νΌλλ°±μ μ£Όμ
μ κ°μ¬ν©λλ€.</div>
</div>
<hr class="helpful-hr hr-bottom"/>
<div role="contentinfo">
<p>
© Copyright 2019, PyTorch.
</p>
</div>
<div>
Built with <a href="http://sphinx-doc.org/">Sphinx</a> using a <a href="https://github.com/rtfd/sphinx_rtd_theme">theme</a> provided by <a href="https://readthedocs.org">Read the Docs</a>.
</div>
</footer>
</div>
</div>
<div class="pytorch-content-right" id="pytorch-content-right">
<div class="pytorch-right-menu" id="pytorch-right-menu">
<div class="pytorch-side-scroll" id="pytorch-side-scroll-right">
<ul>
<li><a class="reference internal" href="#">Extending TorchScript with Custom C++ Classes</a><ul>
<li><a class="reference internal" href="#implementing-and-binding-the-class-in-c">Implementing and Binding the Class in C++</a></li>
<li><a class="reference internal" href="#building-the-example-as-a-c-project-with-cmake">Building the Example as a C++ Project With CMake</a></li>
<li><a class="reference internal" href="#using-the-c-class-from-python-and-torchscript">Using the C++ Class from Python and TorchScript</a></li>
<li><a class="reference internal" href="#saving-loading-and-running-torchscript-code-using-custom-classes">Saving, Loading, and Running TorchScript Code Using Custom Classes</a></li>
<li><a class="reference internal" href="#moving-custom-classes-to-from-ivalues">Moving Custom Classes To/From IValues</a></li>
<li><a class="reference internal" href="#defining-serialization-deserialization-methods-for-custom-c-classes">Defining Serialization/Deserialization Methods for Custom C++ Classes</a></li>
<li><a class="reference internal" href="#defining-custom-operators-that-take-or-return-bound-c-classes">Defining Custom Operators that Take or Return Bound C++ Classes</a></li>
<li><a class="reference internal" href="#conclusion">Conclusion</a></li>
</ul>
</li>
</ul>
</div>
</div>
</div>
</section>
</div>
<script type="text/javascript" id="documentation_options" data-url_root="../" src="../_static/documentation_options.js"></script>
<script type="text/javascript" src="../_static/jquery.js"></script>
<script type="text/javascript" src="../_static/underscore.js"></script>
<script type="text/javascript" src="../_static/doctools.js"></script>
<script type="text/javascript" src="../_static/language_data.js"></script>
<script async="async" type="text/javascript" src="https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.5/latest.js?config=TeX-AMS-MML_HTMLorMML"></script>
<script type="text/javascript" src="../_static/js/vendor/popper.min.js"></script>
<script type="text/javascript" src="../_static/js/vendor/bootstrap.min.js"></script>
<script src="https://cdnjs.cloudflare.com/ajax/libs/list.js/1.5.0/list.min.js"></script>
<script type="text/javascript" src="../_static/js/theme.js"></script>
<script type="text/javascript">
jQuery(function () {
SphinxRtdTheme.Navigation.enable(true);
});
</script>
<script>
(function(i,s,o,g,r,a,m){i['GoogleAnalyticsObject']=r;i[r]=i[r]||function(){
(i[r].q=i[r].q||[]).push(arguments)},i[r].l=1*new Date();a=s.createElement(o),
m=s.getElementsByTagName(o)[0];a.async=1;a.src=g;m.parentNode.insertBefore(a,m)
})(window,document,'script','https://www.google-analytics.com/analytics.js','ga');
ga('create', 'UA-71919972-3', 'auto');
ga('send', 'pageview');
$("[data-behavior='call-to-action-event']").on('click', function(){
ga('send', {
hitType: 'event',
eventCategory: 'Download',
eventAction: 'click',
eventLabel: $(this).attr("data-response")
});
});
$("[data-behavior='was-this-helpful-event']").on('click', function(){
$(".helpful-question").hide();
$(".was-helpful-thank-you").show();
ga('send', {
hitType: 'event',
eventCategory: 'Was this Helpful?',
eventAction: 'click',
eventLabel: $(this).attr("data-response")
});
});
if (location.pathname == "/") {
$(".helpful-container").hide();
$(".hr-bottom").hide();
}
</script>
<!-- Begin Footer -->
<div class="container-fluid docs-tutorials-resources" id="docs-tutorials-resources">
<div class="container">
<div class="row">
<div class="col-md-4 text-center">
<h2>Docs</h2>
<p>Access comprehensive developer documentation for PyTorch</p>
<a class="with-right-arrow" href="https://pytorch.org/docs/stable/index.html">View Docs</a>
</div>
<div class="col-md-4 text-center">
<h2>Tutorials</h2>
<p>Get in-depth tutorials for beginners and advanced developers</p>
<a class="with-right-arrow" href="https://pytorch.org/tutorials">View Tutorials</a>
</div>
<div class="col-md-4 text-center">