-
Notifications
You must be signed in to change notification settings - Fork 257
/
Copy pathtorch_script_custom_classes.html
1207 lines (993 loc) · 93.9 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>커스텀 C++ 클래스로 TorchScript 확장하기 — PyTorch Tutorials 1.10.2+cu102 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/copybutton.css" type="text/css" />
<link rel="stylesheet" href="../_static/gallery.css" type="text/css" />
<link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/katex@0.10.0-beta/dist/katex.min.css" type="text/css" />
<link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/katex@0.13.11/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="Registering a Dispatched Operator in C++" href="dispatcher.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">
<link rel="stylesheet" href="https://use.fontawesome.com/releases/v5.15.2/css/all.css" integrity="sha384-vSIIfh2YWi9wW0r9iZe7RJPrKwp6bG+s9QZMoITbCckVJqGCCRhc+ccxNcdpHuYu" crossorigin="anonymous">
</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.kr/" aria-label="PyTorch"></a>
<div class="main-menu">
<ul>
<li>
<a href="https://pytorch.kr/get-started">시작하기</a>
</li>
<li class="active">
<a href="https://tutorials.pytorch.kr">튜토리얼</a>
</li>
<li>
<a href="https://pytorch.kr/hub">허브</a>
</li>
<li>
<a href="https://discuss.pytorch.kr">커뮤니티</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">
1.10.2+cu102
</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">파이토치(PyTorch) 레시피</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../recipes/recipes_index.html">모든 레시피 보기</a></li>
<li class="toctree-l1"><a class="reference internal" href="../prototype/prototype_index.html">모든 프로토타입 레시피 보기</a></li>
</ul>
<p class="caption"><span class="caption-text">파이토치(PyTorch) 시작하기</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../beginner/basics/intro.html">파이토치(PyTorch) 기본 익히기</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/basics/quickstart_tutorial.html">빠른 시작(Quickstart)</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/basics/tensorqs_tutorial.html">텐서(Tensor)</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/basics/data_tutorial.html">Dataset과 DataLoader</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/basics/transforms_tutorial.html">변형(Transform)</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/basics/buildmodel_tutorial.html">신경망 모델 구성하기</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/basics/autogradqs_tutorial.html"><code class="docutils literal notranslate"><span class="pre">torch.autograd</span></code>를 사용한 자동 미분</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/basics/optimization_tutorial.html">모델 매개변수 최적화하기</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/basics/saveloadrun_tutorial.html">모델 저장하고 불러오기</a></li>
</ul>
<p class="caption"><span class="caption-text">Introduction to PyTorch on YouTube</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../beginner/introyt.html">Introduction to PyTorch - YouTube Series</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/introyt/introyt1_tutorial.html">Introduction to PyTorch</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/introyt/tensors_deeper_tutorial.html">Introduction to PyTorch Tensors</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/introyt/autogradyt_tutorial.html">The Fundamentals of Autograd</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/introyt/modelsyt_tutorial.html">Building Models with PyTorch</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/introyt/tensorboardyt_tutorial.html">PyTorch TensorBoard Support</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/introyt/trainingyt.html">Training with PyTorch</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/introyt/captumyt.html">Model Understanding with Captum</a></li>
</ul>
<p class="caption"><span class="caption-text">파이토치(PyTorch) 배우기</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../beginner/deep_learning_60min_blitz.html">PyTorch로 딥러닝하기: 60분만에 끝장내기</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/pytorch_with_examples.html">예제로 배우는 파이토치(PyTorch)</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/nn_tutorial.html"><cite>torch.nn</cite> 이 <em>실제로</em> 무엇인가요?</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/tensorboard_tutorial.html">TensorBoard로 모델, 데이터, 학습 시각화하기</a></li>
</ul>
<p class="caption"><span class="caption-text">이미지/비디오</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/torchvision_tutorial.html">TorchVision 객체 검출 미세조정(Finetuning) 튜토리얼</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/transfer_learning_tutorial.html">컴퓨터 비전(Vision)을 위한 전이학습(Transfer Learning)</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/fgsm_tutorial.html">적대적 예제 생성(Adversarial Example Generation)</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/dcgan_faces_tutorial.html">DCGAN 튜토리얼</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/vt_tutorial.html">배포를 위한 비전 트랜스포머(Vision Transformer) 모델 최적화하기</a></li>
</ul>
<p class="caption"><span class="caption-text">오디오</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../beginner/audio_io_tutorial.html">Audio I/O</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/audio_resampling_tutorial.html">Audio Resampling</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/audio_data_augmentation_tutorial.html">Audio Data Augmentation</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/audio_feature_extractions_tutorial.html">Audio Feature Extractions</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/audio_feature_augmentation_tutorial.html">Audio Feature Augmentation</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/audio_datasets_tutorial.html">Audio Datasets</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/speech_recognition_pipeline_tutorial.html">Speech Recognition with Wav2Vec2</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/speech_command_classification_with_torchaudio_tutorial.html">Speech Command Classification with torchaudio</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/text_to_speech_with_torchaudio.html">Text-to-speech with torchaudio</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/forced_alignment_with_torchaudio_tutorial.html">Forced Alignment with Wav2Vec2</a></li>
</ul>
<p class="caption"><span class="caption-text">텍스트</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../beginner/transformer_tutorial.html">nn.Transformer 와 TorchText 로 시퀀스-투-시퀀스(Sequence-to-Sequence) 모델링하기</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/char_rnn_classification_tutorial.html">기초부터 시작하는 NLP: 문자-단위 RNN으로 이름 분류하기</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/char_rnn_generation_tutorial.html">기초부터 시작하는 NLP: 문자-단위 RNN으로 이름 생성하기</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/seq2seq_translation_tutorial.html">기초부터 시작하는 NLP: Sequence to Sequence 네트워크와 Attention을 이용한 번역</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/text_sentiment_ngrams_tutorial.html">torchtext 라이브러리로 텍스트 분류하기</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/translation_transformer.html">nn.Transformer와 torchtext로 언어 번역하기</a></li>
</ul>
<p class="caption"><span class="caption-text">강화학습</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/reinforcement_q_learning.html">강화 학습 (DQN) 튜토리얼</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/mario_rl_tutorial.html">Train a Mario-playing RL Agent</a></li>
</ul>
<p class="caption"><span class="caption-text">PyTorch 모델을 프로덕션 환경에 배포하기</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/flask_rest_api_tutorial.html">Flask를 사용하여 Python에서 PyTorch를 REST API로 배포하기</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/Intro_to_TorchScript_tutorial.html">TorchScript 소개</a></li>
<li class="toctree-l1"><a class="reference internal" href="cpp_export.html">C++에서 TorchScript 모델 로딩하기</a></li>
<li class="toctree-l1"><a class="reference internal" href="super_resolution_with_onnxruntime.html">(선택) PyTorch 모델을 ONNX으로 변환하고 ONNX 런타임에서 실행하기</a></li>
</ul>
<p class="caption"><span class="caption-text">Code Transforms with FX</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/fx_conv_bn_fuser.html">(베타) FX에서 합성곱/배치 정규화(Convolution/Batch Norm) 결합기(Fuser) 만들기</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/fx_profiling_tutorial.html">(beta) Building a Simple CPU Performance Profiler with FX</a></li>
</ul>
<p class="caption"><span class="caption-text">프론트엔드 API</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/memory_format_tutorial.html">(베타) PyTorch를 사용한 Channels Last 메모리 형식</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/forward_ad_usage.html">Forward-mode Automatic Differentiation (Beta)</a></li>
<li class="toctree-l1"><a class="reference internal" href="cpp_frontend.html">PyTorch C++ 프론트엔드 사용하기</a></li>
<li class="toctree-l1"><a class="reference internal" href="torch-script-parallelism.html">TorchScript의 동적 병렬 처리(Dynamic Parallelism)</a></li>
<li class="toctree-l1"><a class="reference internal" href="cpp_autograd.html">C++ 프론트엔드의 자동 미분 (autograd)</a></li>
</ul>
<p class="caption"><span class="caption-text">PyTorch 확장하기</span></p>
<ul class="current">
<li class="toctree-l1"><a class="reference internal" href="../intermediate/custom_function_double_backward_tutorial.html">Double Backward with Custom Functions</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/custom_function_conv_bn_tutorial.html">Fusing Convolution and Batch Norm using Custom Function</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="#">커스텀 C++ 클래스로 TorchScript 확장하기</a></li>
<li class="toctree-l1"><a class="reference internal" href="dispatcher.html">Registering a Dispatched Operator in C++</a></li>
<li class="toctree-l1"><a class="reference internal" href="extend_dispatcher.html">Extending dispatcher for a new backend in C++</a></li>
</ul>
<p class="caption"><span class="caption-text">모델 최적화</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../beginner/profiler.html">PyTorch 모듈 프로파일링 하기</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/tensorboard_profiler_tutorial.html">PyTorch Profiler With TensorBoard</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/hyperparameter_tuning_tutorial.html">Hyperparameter tuning with Ray Tune</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/vt_tutorial.html">배포를 위한 비전 트랜스포머(Vision Transformer) 모델 최적화하기</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/parametrizations.html">Parametrizations Tutorial</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/pruning_tutorial.html">가지치기 기법(Pruning) 튜토리얼</a></li>
<li class="toctree-l1"><a class="reference internal" href="dynamic_quantization_tutorial.html">(베타) LSTM 기반 단어 단위 언어 모델의 동적 양자화</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/dynamic_quantization_bert_tutorial.html">(베타) BERT 모델 동적 양자화하기</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/quantized_transfer_learning_tutorial.html">(베타) 컴퓨터 비전 튜토리얼을 위한 양자화된 전이학습(Quantized Transfer Learning)</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>
</ul>
<p class="caption"><span class="caption-text">병렬 및 분산 학습</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">단일 머신을 사용한 모델 병렬화 모범 사례</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/ddp_tutorial.html">분산 데이터 병렬 처리 시작하기</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/dist_tuto.html">PyTorch로 분산 어플리케이션 개발하기</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="../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">분산 데이터 병렬(DDP)과 분산 RPC 프레임워크 결합</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/pipeline_tutorial.html">파이프라인 병렬화로 트랜스포머 모델 학습시키기</a></li>
<li class="toctree-l1"><a class="reference internal" href="ddp_pipeline.html">분산 데이터 병렬 처리와 병렬 처리 파이프라인을 사용한 트랜스포머 모델 학습</a></li>
<li class="toctree-l1"><a class="reference internal" href="generic_join.html">Distributed Training with Uneven Inputs Using the Join Context Manager</a></li>
</ul>
<p class="caption"><span class="caption-text">Mobile</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../beginner/deeplabv3_on_ios.html">iOS에서의 이미지 분할 DeepLapV3</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/deeplabv3_on_android.html">안드로이드에서의 이미지 분할 DeepLapV3</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>커스텀 C++ 클래스로 TorchScript 확장하기</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="c-torchscript">
<h1>커스텀 C++ 클래스로 TorchScript 확장하기<a class="headerlink" href="#c-torchscript" title="Permalink to this headline">¶</a></h1>
<p>이 튜토리얼은 <a class="reference internal" href="torch_script_custom_ops.html"><span class="doc">커스텀 오퍼레이터</span></a> 튜토리얼의 후속이며
C++ 클래스를 TorchScript와 Python에 동시에 바인딩하기 위해 구축한 API를 소개합니다.
API는 <a class="reference external" href="https://github.com/pybind/pybind11">pybind11</a> 과
매우 유사하며 해당 시스템에 익숙하다면 대부분의 개념이 이전됩니다.</p>
<div class="section" id="c">
<h2>C++에서 클래스 구현 및 바인딩<a class="headerlink" href="#c" title="Permalink to this headline">¶</a></h2>
<p>이 튜토리얼에서는 멤버 변수에서 지속 상태를 유지하는 간단한 C++ 클래스를 정의할 것입니다.</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="nc">MyStackClass</span> <span class="o">:</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>몇 가지 주의할 사항이 있습니다:</p>
<ul class="simple">
<li><code class="docutils literal notranslate"><span class="pre">torch/custom_class.h</span></code> 는 커스텀 클래스로 TorchScript를 확장하기 위해 포함해야하는 헤더입니다.</li>
<li>커스텀 클래스의 인스턴스로 작업할 때마다 <code class="docutils literal notranslate"><span class="pre">c10::intrusive_ptr<></span></code> 의 인스턴스를 통해 작업을 수행합니다.
<code class="docutils literal notranslate"><span class="pre">intrusive_ptr</span></code> 를 <code class="docutils literal notranslate"><span class="pre">std::shared_ptr</span></code> 과 같은 스마트 포인터로 생각하세요. 그러나 참조 계수는
<code class="docutils literal notranslate"><span class="pre">std::shared_ptr</span></code> 같이 별도의 메타데이터 블록과 달리 객체에 직접 저장됩니다.
<code class="docutils literal notranslate"><span class="pre">torch::Tensor</span></code> 는 내부적으로 동일한 포인터 유형을 사용합니다.
커스텀 클래스도 <code class="docutils literal notranslate"><span class="pre">torch::Tensor</span></code> 포인터 유형을 사용해야 다양한 객체 유형을 일관되게 관리할 수 있습니다.</li>
<li>두 번째로 주목해야 할 점은 커스텀 클래스가 <code class="docutils literal notranslate"><span class="pre">torch::CustomClassHolder</span></code> 에서 상속되어야 한다는 것입니다.
이렇게 하면 커스텀 클래스에 참조 계수를 저장할 공간이 있습니다.</li>
</ul>
<p>이제 이 클래스를 어떻게 TorchScript에서 사용가능하게 하는지 살펴보겠습니다.
이런 과정은 클래스를 <em>바인딩</em> 한다고 합니다:</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="cmake-c">
<h2>CMake를 사용하여 C++ 프로젝트로 예제 빌드<a class="headerlink" href="#cmake-c" title="Permalink to this headline">¶</a></h2>
<p>이제 <a class="reference external" href="https://cmake.org">CMake</a> 빌드 시스템을 사용하여 위의 C++ 코드를 빌드합니다.
먼저, 지금까지 다룬 모든 C++ code를 <code class="docutils literal notranslate"><span class="pre">class.cpp</span></code> 라는 파일에 넣습니다.
그런 다음 간단한 <code class="docutils literal notranslate"><span class="pre">CMakeLists.txt</span></code> 파일을 작성하여 동일한 디렉토리에 배치합니다.
<code class="docutils literal notranslate"><span class="pre">CMakeLists.txt</span></code> 는 다음과 같아야 합니다:</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">"${TORCH_LIBRARIES}"</span><span class="p">)</span>
</pre></div>
</div>
<p>또한 <code class="docutils literal notranslate"><span class="pre">build</span></code> 디렉토리를 만듭니다. 파일 트리는 다음과 같아야 합니다:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">custom_class_project</span><span class="o">/</span>
<span class="n">class</span><span class="o">.</span><span class="n">cpp</span>
<span class="n">CMakeLists</span><span class="o">.</span><span class="n">txt</span>
<span class="n">build</span><span class="o">/</span>
</pre></div>
</div>
<p><a class="reference internal" href="torch_script_custom_ops.html"><span class="doc">이전 튜토리얼</span></a> 에서 설명한 것과 동일한 방식으로 환경을 설정했다고 가정합니다.
계속해서 cmake를 호출한 다음 make를 호출하여 프로젝트를 빌드합니다:</p>
<div class="highlight-shell notranslate"><div class="highlight"><pre><span></span>$ <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 <span class="k">in</span> pthreads
-- Looking <span class="k">for</span> pthread_create <span class="k">in</span> pthreads - not found
-- Looking <span class="k">for</span> pthread_create <span class="k">in</span> pthread
-- Looking <span class="k">for</span> pthread_create <span class="k">in</span> 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>이제 무엇보다도 빌드 디렉토리에 동적 라이브러리 파일이 있다는 것을 알게 될 것입니다.
리눅스에서는 아마도 <code class="docutils literal notranslate"><span class="pre">libcustom_class.so</span></code> 로 이름이 지정될 것입니다.
따라서 파일 트리는 다음과 같아야 합니다:</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="python-torchscript-c">
<h2>Python 및 TorchScript의 C++ 클래스 사용<a class="headerlink" href="#python-torchscript-c" title="Permalink to this headline">¶</a></h2>
<p>이제 클래스와 등록이 <code class="docutils literal notranslate"><span class="pre">.so</span></code> 파일로 컴파일되었으므로 해당 <cite>.so</cite> 를 Python에 읽어들이고 사용해 볼 수 있습니다.
다음은 이를 보여주는 스크립트입니다:</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="nb">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"># Test custom operator</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="n">torch</span><span class="o">.</span><span class="n">ops</span><span class="o">.</span><span class="n">my_classes</span><span class="o">.</span><span class="n">manipulate_instance</span><span class="p">(</span><span class="n">s</span><span class="p">)</span> <span class="c1"># acting as s.pop()</span>
<span class="k">assert</span> <span class="n">s</span><span class="o">.</span><span class="n">top</span><span class="p">()</span> <span class="o">==</span> <span class="s2">"bar"</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</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">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="torchscript">
<h2>커스텀 클래스를 사용하여 TorchScript 코드 저장, 읽기 및 실행<a class="headerlink" href="#torchscript" title="Permalink to this headline">¶</a></h2>
<p>libtorch를 사용하여 C++ 프로세스에서 커스텀 등록 C++ 클래스를 사용할 수도 있습니다.
예를 들어 MyStackClass 클래스에서 메소드를 인스턴스화하고 호출하는 간단한 <code class="docutils literal notranslate"><span class="pre">nn.Module</span></code> 을 정의해 보겠습니다:</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="nb">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>파일 시스템의 foo.pt는 방금 정의한 직렬화된 TorchScript 프로그램을 포함합니다.</p>
<p>이제 이 모델과 필요한 .so 파일을 읽어들이는 방법을 보여주기 위해 새 CMake 프로젝트를 정의하겠습니다.
이 작업을 수행하는 방법에 대한 자세한 내용은 <a class="reference external" href="https://tutorials.pytorch.kr/advanced/cpp_export.html">C++에서 TorchScript 모델 로딩하기</a> 를
참조하세요.</p>
<p>이전과 유사하게 다음을 포함하는 파일 구조를 생성해 보겠습니다:</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>직렬화된 foo.pt 파일과 위의 <code class="docutils literal notranslate"><span class="pre">custom_class_project</span></code> 소스 트리를 복사했음을 주목하세요.
커스텀 클래스를 바이너리로 빌드할 수 있도록 <code class="docutils literal notranslate"><span class="pre">custom_class_project</span></code> 를 이 C++ 프로젝트에 의존성으로 추가할 것입니다.</p>
<p><code class="docutils literal notranslate"><span class="pre">infer.cpp</span></code> 를 다음으로 채우겠습니다:</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="k">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="k">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="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="k">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>마찬가지로 CMakeLists.txt 파일을 정의해 보겠습니다:</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">"${TORCH_LIBRARIES}"</span><span class="p">)</span>
<span class="cp"># This is where we link in our libcustom_class code, making our</span>
<span class="cp"># custom class available in our binary.</span>
<span class="n">target_link_libraries</span><span class="p">(</span><span class="n">infer</span> <span class="o">-</span><span class="n">Wl</span><span class="p">,</span><span class="o">--</span><span class="n">no</span><span class="o">-</span><span class="n">as</span><span class="o">-</span><span class="n">needed</span> <span class="n">custom_class</span><span class="p">)</span>
</pre></div>
</div>
<p><code class="docutils literal notranslate"><span class="pre">cd</span> <span class="pre">build</span></code>, <code class="docutils literal notranslate"><span class="pre">cmake</span></code>, 및 <code class="docutils literal notranslate"><span class="pre">make</span></code> 에 대한 사용 방법을 알고 있습니다:</p>
<div class="highlight-shell notranslate"><div class="highlight"><pre><span></span>$ <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 <span class="k">in</span> pthreads
-- Looking <span class="k">for</span> pthread_create <span class="k">in</span> pthreads - not found
-- Looking <span class="k">for</span> pthread_create <span class="k">in</span> pthread
-- Looking <span class="k">for</span> pthread_create <span class="k">in</span> 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>이제 흥미로운 C++ 바이너리를 실행할 수 있습니다:</p>
<div class="highlight-shell notranslate"><div class="highlight"><pre><span></span>$ ./infer
momfoobarbaz
</pre></div>
</div>
<p>대단합니다!</p>
</div>
<div class="section" id="ivalues">
<h2>커스텀 클래스를 IValues로/에서 이동<a class="headerlink" href="#ivalues" title="Permalink to this headline">¶</a></h2>
<p>TorchScript 메소드에서 <code class="docutils literal notranslate"><span class="pre">IValue</span></code> 를 가져오거나 반환하기, 또는 C++에서 커스텀 클래스 속성을 인스턴스화하려는
경우와 같이 커스텀 클래스를 <code class="docutils literal notranslate"><span class="pre">IValue</span></code> 안팎으로 이동해야 할 수도 있습니다.
커스텀 C++ 클래스 인스턴스에서 <code class="docutils literal notranslate"><span class="pre">IValue</span></code> 를 생성하려면:</p>
<ul class="simple">
<li><code class="docutils literal notranslate"><span class="pre">torch::make_custom_class<T>()</span></code> 는 제공하는 인수 집합을 사용하고 해당 인수 집합과 일치하는
T의 생성자를 호출하며 해당 인스턴스를 래핑하고 반환하는 c10::intrusive_ptr<T>와 유사한 API를 제공합니다.
그러나 커스텀 클래스 객체에 대한 포인터만 반환하는 대신 객체를 래핑하는 <code class="docutils literal notranslate"><span class="pre">IValue</span></code> 를 반환합니다.
그런 다음 이 <code class="docutils literal notranslate"><span class="pre">IValue</span></code> 를 TorchScript에 직접 전달할 수 있습니다.</li>
<li>이미 클래스를 가리키는 <code class="docutils literal notranslate"><span class="pre">intrusive_ptr</span></code> 이 있는 경우 생성자 <code class="docutils literal notranslate"><span class="pre">IValue(intrusive_ptr<T>)</span></code> 를 사용하여
해당 클래스에서 IValue를 직접 생성할 수 있습니다.</li>
</ul>
<p><code class="docutils literal notranslate"><span class="pre">IValue</span></code> 를 커스텀 클래스로 다시 변환하려면:</p>
<ul class="simple">
<li><code class="docutils literal notranslate"><span class="pre">IValue::toCustomClass<T>()</span></code> 는 <code class="docutils literal notranslate"><span class="pre">IValue</span></code> 에 포함된 커스텀 클래스를 가리키는 <code class="docutils literal notranslate"><span class="pre">intrusive_ptr<T></span></code> 를
반환합니다. 내부적으로 이 함수는 <code class="docutils literal notranslate"><span class="pre">T</span></code> 가 커스텀 클래스로 등록되어 있고 <code class="docutils literal notranslate"><span class="pre">IValue</span></code> 에 실제로 커스텀 클래스가
포함되어 있는지 확인합니다. <code class="docutils literal notranslate"><span class="pre">isCustomClass()</span></code> 를 호출하여 <code class="docutils literal notranslate"><span class="pre">IValue</span></code> 에 커스텀 클래스가 포함되어 있는지
수동으로 확인할 수 있습니다.</li>
</ul>
</div>
<div class="section" id="id2">
<h2>커스텀 C++ 클래스에 대한 직렬화/역직렬화 방법 정의<a class="headerlink" href="#id2" title="Permalink to this headline">¶</a></h2>
<p>커스텀 바인딩 된 C++ 클래스를 속성으로 사용하여 <code class="docutils literal notranslate"><span class="pre">ScriptModule</span></code> 을 저장하려고 하면
다음 오류가 발생합니다:</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="nb">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>TorchScript가 C++ 클래스에서 저장한 정보를 자동으로 파악할 수 없기 때문입니다.
수동으로 지정해야 합니다. 그렇게 하는 방법은 <code class="docutils literal notranslate"><span class="pre">class_</span></code> 에서 특별한 <code class="docutils literal notranslate"><span class="pre">def_pickle</span></code> 메소드를
사용하여 클래스에서 <code class="docutils literal notranslate"><span class="pre">__getstate__</span></code> 및 <code class="docutils literal notranslate"><span class="pre">__setstate__</span></code> 메소드를 정의하는 것입니다.</p>
<div class="admonition note">
<p class="first admonition-title">Note</p>
<p class="last">TorchScript에서 <code class="docutils literal notranslate"><span class="pre">__getstate__</span></code> 및 <code class="docutils literal notranslate"><span class="pre">__setstate__</span></code> 의 의미는 Python pickle 모듈의 의미와 동일합니다.
이러한 방법을 어떻게 사용하는지에 대하여 <a class="reference external" href="https://github.com/pytorch/pytorch/blob/master/torch/csrc/jit/docs/serialization.md#getstate-and-setstate">자세한 내용</a> 을
참조하세요.</p>
</div>
<p>다음은 직렬화 메소드를 포함하기 위해 <code class="docutils literal notranslate"><span class="pre">MyStackClass</span></code> 등록에 추가할 수 있는 <code class="docutils literal notranslate"><span class="pre">def_pickle</span></code> 호출의 예시입니다:</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="first admonition-title">Note</p>
<p class="last">pickle API에서 pybind11과 다른 접근 방식을 취합니다. pybind11이 <code class="docutils literal notranslate"><span class="pre">class_::def()</span></code> 로
전달되는 특수 함수 <code class="docutils literal notranslate"><span class="pre">pybind11::pickle()</span></code> 인 반면, 이 목적을 위해 별도의 메소드
<code class="docutils literal notranslate"><span class="pre">def_pickle</span></code> 를 가지고 있습니다. 이미 <code class="docutils literal notranslate"><span class="pre">torch::jit::pickle</span></code> 라는 이름이 사용되었고
혼동을 일으키고 싶지 않았기 때문입니다.</p>
</div>
<p>이러한 방식으로 (역)직렬화 동작을 정의하면 이제 스크립트를 성공적으로 실행할 수 있습니다:</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="id4">
<h2>바인딩된 C++ 클래스를 사용하거나 반환하는 커스텀 연산자 정의<a class="headerlink" href="#id4" title="Permalink to this headline">¶</a></h2>
<p>커스텀 C++ 클래스를 정의한 후에는 해당 클래스를 인수로 사용하거나 커스텀 연산자(예를 들어 free 함수)에서
반환할 수도 있습니다. 다음과 같은 free 함수가 있다고 가정합니다:</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><code class="docutils literal notranslate"><span class="pre">TORCH_LIBRARY</span></code> 블록 내에서 다음 코드를 실행하여 등록할 수 있습니다:</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">"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>등록 API에 대한 자세한 내용은 <a class="reference external" href="https://tutorials.pytorch.kr/advanced/torch_script_custom_ops.html">커스텀 C++ 연산자로 TorchScript 확장</a> 을
참조하세요.</p>
<p>이 작업이 완료되면 다음 예제와 같이 연산자를 사용할 수 있습니다:</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">my_classes</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="first admonition-title">Note</p>
<p class="last">C++ 클래스를 인수로 사용하는 연산자를 등록하려면 커스텀 클래스가 이미 등록되어
있어야 합니다. 커스텀 클래스 등록과 free 함수 정의가 동일한 <code class="docutils literal notranslate"><span class="pre">TORCH_LIBRARY</span></code>
블록에 있고 커스텀 클래스 등록이 먼저 오게 하여 이를 시행할 수 있습니다.
향후 어떤 순서로든 등록할 수 있도록 이 요구사항을 완화할 수 있습니다.</p>
</div>
</div>
<div class="section" id="id6">
<h2>결론<a class="headerlink" href="#id6" title="Permalink to this headline">¶</a></h2>
<p>이 튜토리얼에서는 독립된 C++ 프로세스에서 C++ 클래스를 TorchScript 및
확장 Python에 나타내는 방법, 해당 메소드를 등록하는 방법, Python 및 TorchScript에서
해당 클래스를 사용하는 방법, 클래스를 사용하여 코드를 저장 및 읽어들이고 해당 코드를 실행하는 방법을 안내했습니다.
이제 타사 C++ 라이브러리와 인터페이스가 있는 C++ 클래스로 TorchScript 모델을 확장하거나,
Python, TorchScript 및 C++ 간의 라인이 원활하게 혼합되어야 하는 다른 사용 사례를 구현할 준비가 되었습니다.</p>
<p>언제나처럼 문제를 마주치거나 질문이 있으면 저희 <a class="reference external" href="https://discuss.pytorch.org/">forum</a> 또는
<a class="reference external" href="https://github.com/pytorch/pytorch/issues">GitHub issues</a> 에 올려주시면 되겠습니다.
또한 <a class="reference external" href="https://pytorch.org/cppdocs/notes/faq.html">자주 묻는 질문(FAQ) 페이지</a> 에
유용한 정보가 있을 수 있습니다.</p>
</div>
</div>
</article>
</div>
<footer>
<div class="rst-footer-buttons" role="navigation" aria-label="footer navigation">
<a href="dispatcher.html" class="btn btn-neutral float-right" title="Registering a Dispatched Operator in C++" 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="rating-hr hr-top">
<div class="rating-container">
<div class="rating-prompt">Rate this Tutorial</div>
<div class="stars-outer">
<i class="far fa-star" title="1 Star" data-behavior="tutorial-rating" data-count="1"></i>
<i class="far fa-star" title="2 Stars" data-behavior="tutorial-rating" data-count="2"></i>
<i class="far fa-star" title="3 Stars" data-behavior="tutorial-rating" data-count="3"></i>
<i class="far fa-star" title="4 Stars" data-behavior="tutorial-rating" data-count="4"></i>
<i class="far fa-star" title="5 Stars" data-behavior="tutorial-rating" data-count="5"></i>
</div>
</div>
<hr class="rating-hr hr-bottom"/>
<div role="contentinfo">
<p>
© Copyright 2021, PyTorch & PyTorch Korea Community.
</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="#">커스텀 C++ 클래스로 TorchScript 확장하기</a><ul>
<li><a class="reference internal" href="#c">C++에서 클래스 구현 및 바인딩</a></li>
<li><a class="reference internal" href="#cmake-c">CMake를 사용하여 C++ 프로젝트로 예제 빌드</a></li>
<li><a class="reference internal" href="#python-torchscript-c">Python 및 TorchScript의 C++ 클래스 사용</a></li>
<li><a class="reference internal" href="#torchscript">커스텀 클래스를 사용하여 TorchScript 코드 저장, 읽기 및 실행</a></li>
<li><a class="reference internal" href="#ivalues">커스텀 클래스를 IValues로/에서 이동</a></li>
<li><a class="reference internal" href="#id2">커스텀 C++ 클래스에 대한 직렬화/역직렬화 방법 정의</a></li>
<li><a class="reference internal" href="#id4">바인딩된 C++ 클래스를 사용하거나 반환하는 커스텀 연산자 정의</a></li>
<li><a class="reference internal" href="#id6">결론</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/clipboard.min.js"></script>
<script type="text/javascript" src="../_static/copybutton.js"></script>
<script type="text/javascript" src="https://cdn.jsdelivr.net/npm/katex@0.13.11/dist/katex.min.js"></script>
<script type="text/javascript" src="https://cdn.jsdelivr.net/npm/katex@0.13.11/dist/contrib/auto-render.min.js"></script>
<script type="text/javascript" src="../_static/katex_autorenderer.js"></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);