-
Notifications
You must be signed in to change notification settings - Fork 257
/
Copy pathrpc_ddp_tutorial.html
936 lines (738 loc) ยท 69.6 KB
/
rpc_ddp_tutorial.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
<!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>๋ถ์ฐ ๋ฐ์ดํฐ ๋ณ๋ ฌ(DDP)๊ณผ ๋ถ์ฐ RPC ํ๋ ์์ํฌ ๊ฒฐํฉ — 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="ํ์ดํ๋ผ์ธ ๋ณ๋ ฌํ๋ก ํธ๋์คํฌ๋จธ ๋ชจ๋ธ ํ์ต์ํค๊ธฐ" href="../intermediate/pipeline_tutorial.html" />
<link rel="prev" title="Implementing Batch RPC Processing Using Asynchronous Executions" href="../intermediate/rpc_async_execution.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">์์ํ๊ธฐ</a>
</li>
<li class="active">
<a href="https://tutorials.pytorch.kr">ํํ ๋ฆฌ์ผ</a>
</li>
<li>
<a href="https://pytorch.kr/hub">ํ๋ธ</a>
</li>
<li>
<a href="https://discuss.pytorch.kr">์ปค๋ฎค๋ํฐ</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">ํ์ดํ ์น(PyTorch) ๋ ์ํผ</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../recipes/recipes_index.html">๋ชจ๋ ๋ ์ํผ ๋ณด๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../prototype/prototype_index.html">๋ชจ๋ ํ๋กํ ํ์
๋ ์ํผ ๋ณด๊ธฐ</a></li>
</ul>
<p class="caption"><span class="caption-text">ํ์ดํ ์น(PyTorch) ์์ํ๊ธฐ</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../beginner/basics/intro.html">ํ์ดํ ์น(PyTorch) ๊ธฐ๋ณธ ์ตํ๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/basics/quickstart_tutorial.html">๋น ๋ฅธ ์์(Quickstart)</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/basics/tensorqs_tutorial.html">ํ
์(Tensor)</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/basics/data_tutorial.html">Dataset๊ณผ DataLoader</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/basics/transforms_tutorial.html">๋ณํ(Transform)</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/basics/buildmodel_tutorial.html">์ ๊ฒฝ๋ง ๋ชจ๋ธ ๊ตฌ์ฑํ๊ธฐ</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>๋ฅผ ์ฌ์ฉํ ์๋ ๋ฏธ๋ถ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/basics/optimization_tutorial.html">๋ชจ๋ธ ๋งค๊ฐ๋ณ์ ์ต์ ํํ๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/basics/saveloadrun_tutorial.html">๋ชจ๋ธ ์ ์ฅํ๊ณ ๋ถ๋ฌ์ค๊ธฐ</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">ํ์ดํ ์น(PyTorch) ๋ฐฐ์ฐ๊ธฐ</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../beginner/deep_learning_60min_blitz.html">PyTorch๋ก ๋ฅ๋ฌ๋ํ๊ธฐ: 60๋ถ๋ง์ ๋์ฅ๋ด๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/pytorch_with_examples.html">์์ ๋ก ๋ฐฐ์ฐ๋ ํ์ดํ ์น(PyTorch)</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/nn_tutorial.html"><cite>torch.nn</cite> ์ด <em>์ค์ ๋ก</em> ๋ฌด์์ธ๊ฐ์?</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/tensorboard_tutorial.html">TensorBoard๋ก ๋ชจ๋ธ, ๋ฐ์ดํฐ, ํ์ต ์๊ฐํํ๊ธฐ</a></li>
</ul>
<p class="caption"><span class="caption-text">์ด๋ฏธ์ง/๋น๋์ค</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/torchvision_tutorial.html">TorchVision ๊ฐ์ฒด ๊ฒ์ถ ๋ฏธ์ธ์กฐ์ (Finetuning) ํํ ๋ฆฌ์ผ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/transfer_learning_tutorial.html">์ปดํจํฐ ๋น์ (Vision)์ ์ํ ์ ์ดํ์ต(Transfer Learning)</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/fgsm_tutorial.html">์ ๋์ ์์ ์์ฑ(Adversarial Example Generation)</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/dcgan_faces_tutorial.html">DCGAN ํํ ๋ฆฌ์ผ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/vt_tutorial.html">๋ฐฐํฌ๋ฅผ ์ํ ๋น์ ํธ๋์คํฌ๋จธ(Vision Transformer) ๋ชจ๋ธ ์ต์ ํํ๊ธฐ</a></li>
</ul>
<p class="caption"><span class="caption-text">์ค๋์ค</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">ํ
์คํธ</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../beginner/transformer_tutorial.html">nn.Transformer ์ TorchText ๋ก ์ํ์ค-ํฌ-์ํ์ค(Sequence-to-Sequence) ๋ชจ๋ธ๋งํ๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/char_rnn_classification_tutorial.html">๊ธฐ์ด๋ถํฐ ์์ํ๋ NLP: ๋ฌธ์-๋จ์ RNN์ผ๋ก ์ด๋ฆ ๋ถ๋ฅํ๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/char_rnn_generation_tutorial.html">๊ธฐ์ด๋ถํฐ ์์ํ๋ NLP: ๋ฌธ์-๋จ์ RNN์ผ๋ก ์ด๋ฆ ์์ฑํ๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/seq2seq_translation_tutorial.html">๊ธฐ์ด๋ถํฐ ์์ํ๋ NLP: Sequence to Sequence ๋คํธ์ํฌ์ Attention์ ์ด์ฉํ ๋ฒ์ญ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/text_sentiment_ngrams_tutorial.html">torchtext ๋ผ์ด๋ธ๋ฌ๋ฆฌ๋ก ํ
์คํธ ๋ถ๋ฅํ๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/translation_transformer.html">nn.Transformer์ torchtext๋ก ์ธ์ด ๋ฒ์ญํ๊ธฐ</a></li>
</ul>
<p class="caption"><span class="caption-text">๊ฐํํ์ต</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/reinforcement_q_learning.html">๊ฐํ ํ์ต (DQN) ํํ ๋ฆฌ์ผ</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 ๋ชจ๋ธ์ ํ๋ก๋์
ํ๊ฒฝ์ ๋ฐฐํฌํ๊ธฐ</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/flask_rest_api_tutorial.html">Flask๋ฅผ ์ฌ์ฉํ์ฌ Python์์ PyTorch๋ฅผ REST API๋ก ๋ฐฐํฌํ๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/Intro_to_TorchScript_tutorial.html">TorchScript ์๊ฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="cpp_export.html">C++์์ TorchScript ๋ชจ๋ธ ๋ก๋ฉํ๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="super_resolution_with_onnxruntime.html">(์ ํ) PyTorch ๋ชจ๋ธ์ ONNX์ผ๋ก ๋ณํํ๊ณ ONNX ๋ฐํ์์์ ์คํํ๊ธฐ</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">(๋ฒ ํ) FX์์ ํฉ์ฑ๊ณฑ/๋ฐฐ์น ์ ๊ทํ(Convolution/Batch Norm) ๊ฒฐํฉ๊ธฐ(Fuser) ๋ง๋ค๊ธฐ</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">ํ๋ก ํธ์๋ API</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/memory_format_tutorial.html">(๋ฒ ํ) PyTorch๋ฅผ ์ฌ์ฉํ Channels Last ๋ฉ๋ชจ๋ฆฌ ํ์</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++ ํ๋ก ํธ์๋ ์ฌ์ฉํ๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="torch-script-parallelism.html">TorchScript์ ๋์ ๋ณ๋ ฌ ์ฒ๋ฆฌ(Dynamic Parallelism)</a></li>
<li class="toctree-l1"><a class="reference internal" href="cpp_autograd.html">C++ ํ๋ก ํธ์๋์ ์๋ ๋ฏธ๋ถ (autograd)</a></li>
</ul>
<p class="caption"><span class="caption-text">PyTorch ํ์ฅํ๊ธฐ</span></p>
<ul>
<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"><a class="reference internal" href="torch_script_custom_classes.html">์ปค์คํ
C++ ํด๋์ค๋ก TorchScript ํ์ฅํ๊ธฐ</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">๋ชจ๋ธ ์ต์ ํ</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../beginner/profiler.html">PyTorch ๋ชจ๋ ํ๋กํ์ผ๋ง ํ๊ธฐ</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">๋ฐฐํฌ๋ฅผ ์ํ ๋น์ ํธ๋์คํฌ๋จธ(Vision Transformer) ๋ชจ๋ธ ์ต์ ํํ๊ธฐ</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">๊ฐ์ง์น๊ธฐ ๊ธฐ๋ฒ(Pruning) ํํ ๋ฆฌ์ผ</a></li>
<li class="toctree-l1"><a class="reference internal" href="dynamic_quantization_tutorial.html">(๋ฒ ํ) LSTM ๊ธฐ๋ฐ ๋จ์ด ๋จ์ ์ธ์ด ๋ชจ๋ธ์ ๋์ ์์ํ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/dynamic_quantization_bert_tutorial.html">(๋ฒ ํ) BERT ๋ชจ๋ธ ๋์ ์์ํํ๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/quantized_transfer_learning_tutorial.html">(๋ฒ ํ) ์ปดํจํฐ ๋น์ ํํ ๋ฆฌ์ผ์ ์ํ ์์ํ๋ ์ ์ดํ์ต(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">๋ณ๋ ฌ ๋ฐ ๋ถ์ฐ ํ์ต</span></p>
<ul class="current">
<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">๋จ์ผ ๋จธ์ ์ ์ฌ์ฉํ ๋ชจ๋ธ ๋ณ๋ ฌํ ๋ชจ๋ฒ ์ฌ๋ก</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/ddp_tutorial.html">๋ถ์ฐ ๋ฐ์ดํฐ ๋ณ๋ ฌ ์ฒ๋ฆฌ ์์ํ๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/dist_tuto.html">PyTorch๋ก ๋ถ์ฐ ์ดํ๋ฆฌ์ผ์ด์
๊ฐ๋ฐํ๊ธฐ</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 current"><a class="current reference internal" href="#">๋ถ์ฐ ๋ฐ์ดํฐ ๋ณ๋ ฌ(DDP)๊ณผ ๋ถ์ฐ RPC ํ๋ ์์ํฌ ๊ฒฐํฉ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/pipeline_tutorial.html">ํ์ดํ๋ผ์ธ ๋ณ๋ ฌํ๋ก ํธ๋์คํฌ๋จธ ๋ชจ๋ธ ํ์ต์ํค๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="ddp_pipeline.html">๋ถ์ฐ ๋ฐ์ดํฐ ๋ณ๋ ฌ ์ฒ๋ฆฌ์ ๋ณ๋ ฌ ์ฒ๋ฆฌ ํ์ดํ๋ผ์ธ์ ์ฌ์ฉํ ํธ๋์คํฌ๋จธ ๋ชจ๋ธ ํ์ต</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">iOS์์์ ์ด๋ฏธ์ง ๋ถํ DeepLapV3</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/deeplabv3_on_android.html">์๋๋ก์ด๋์์์ ์ด๋ฏธ์ง ๋ถํ 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>๋ถ์ฐ ๋ฐ์ดํฐ ๋ณ๋ ฌ(DDP)๊ณผ ๋ถ์ฐ RPC ํ๋ ์์ํฌ ๊ฒฐํฉ</li>
<li class="pytorch-breadcrumbs-aside">
<a href="../_sources/advanced/rpc_ddp_tutorial.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/rpc_ddp_tutorial</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="ddp-rpc">
<h1>๋ถ์ฐ ๋ฐ์ดํฐ ๋ณ๋ ฌ(DDP)๊ณผ ๋ถ์ฐ RPC ํ๋ ์์ํฌ ๊ฒฐํฉ<a class="headerlink" href="#ddp-rpc" title="Permalink to this headline">ยถ</a></h1>
<p><strong>์ ์</strong>: <a class="reference external" href="https://github.com/pritamdamania87">Pritam Damania</a> and <a class="reference external" href="https://github.com/SciPioneer">Yi Wang</a></p>
<p><strong>๋ฒ์ญ</strong>: <a class="reference external" href="https://github.com/dajeongPark-dev">๋ฐ๋ค์ </a></p>
<p>์ด ํํ ๋ฆฌ์ผ์ ๊ฐ๋จํ ์์ ๋ฅผ ์ฌ์ฉํ์ฌ ๋ถ์ฐ ๋ฐ์ดํฐ ๋ณ๋ ฌ ์ฒ๋ฆฌ(distributed data parallelism)์
๋ถ์ฐ ๋ชจ๋ธ ๋ณ๋ ฌ ์ฒ๋ฆฌ(distributed model parallelism)๋ฅผ ๊ฒฐํฉํ์ฌ ๊ฐ๋จํ ๋ชจ๋ธ ํ์ต์ํฌ ๋
<a class="reference external" href="https://pytorch.org/docs/stable/nn.html#torch.nn.parallel.DistributedDataParallel">๋ถ์ฐ ๋ฐ์ดํฐ ๋ณ๋ ฌ(DistributedDataParallel)</a> (DDP)๊ณผ
<a href="#id2"><span class="problematic" id="id3">`</span></a>๋ถ์ฐ RPC ํ๋ ์์ํฌ(Distributed RPC framework) <<a class="reference external" href="https://pytorch.org/docs/master/rpc.html">https://pytorch.org/docs/master/rpc.html</a>>`__๋ฅผ ๊ฒฐํฉํ๋ ๋ฐฉ๋ฒ์ ๋ํด ์ค๋ช
ํฉ๋๋ค.
์์ ์ ์์ค ์ฝ๋๋ <a href="#id4"><span class="problematic" id="id5">`</span></a>์ฌ๊ธฐ <<a class="reference external" href="https://github.com/pytorch/examples/tree/master/distributed/rpc/ddp_rpc">https://github.com/pytorch/examples/tree/master/distributed/rpc/ddp_rpc</a>>`__์์ ํ์ธํ ์ ์์ต๋๋ค.</p>
<p>์ด์ ํํ ๋ฆฌ์ผ ๋ด์ฉ์ด์๋
<a href="#id6"><span class="problematic" id="id7">`</span></a>๋ถ์ฐ ๋ฐ์ดํฐ ๋ณ๋ ฌ ์์ํ๊ธฐ <<a class="reference external" href="https://tutorials.pytorch.kr/intermediate/ddp_tutorial.html">https://tutorials.pytorch.kr/intermediate/ddp_tutorial.html</a>>`__์
<a href="#id8"><span class="problematic" id="id9">`</span></a>๋ถ์ฐ RPC ํ๋ ์์ํฌ ์์ํ๊ธฐ <<a class="reference external" href="https://tutorials.pytorch.kr/intermediate/rpc_tutorial.html">https://tutorials.pytorch.kr/intermediate/rpc_tutorial.html</a>>`__๋
๋ถ์ฐ ๋ฐ์ดํฐ ๋ณ๋ ฌ ๋ฐ ๋ถ์ฐ ๋ชจ๋ธ ๋ณ๋ ฌ ํ์ต์ ๊ฐ๊ฐ ์ํํ๋ ๋ฐฉ๋ฒ์ ๋ํด ์ค๋ช
ํฉ๋๋ค.
๊ทธ๋ฌ๋ ์ด ๋ ๊ฐ์ง ๊ธฐ์ ์ ๊ฒฐํฉํ ์ ์๋ ๋ช ๊ฐ์ง ํ์ต ํจ๋ฌ๋ค์์ด ์์ต๋๋ค. ์๋ฅผ ๋ค์ด:</p>
<ol class="arabic simple">
<li>ํฌ์ ๋ถ๋ถ(ํฐ ์๋ฒ ๋ฉ ํ
์ด๋ธ)๊ณผ ๋ฐ์ง ๋ถ๋ถ(FC ๋ ์ด์ด)์ด ์๋ ๋ชจ๋ธ์ด ์๋ ๊ฒฝ์ฐ,
๋งค๊ฐ๋ณ์ ์๋ฒ(parameter server)์ ์๋ฒ ๋ฉ ํ
์ด๋ธ(embedding table)์ ๋๊ณ <a href="#id10"><span class="problematic" id="id11">`</span></a>๋ถ์ฐ ๋ฐ์ดํฐ ๋ณ๋ ฌ <<a class="reference external" href="https://pytorch.org/docs/stable/nn.html#torch.nn.parallel.DistributedDataParallel">https://pytorch.org/docs/stable/nn.html#torch.nn.parallel.DistributedDataParallel</a>>`__์ ์ฌ์ฉํ์ฌ
์ฌ๋ฌ ํธ๋ ์ด๋์ ๊ฑธ์ณ FC ๋ ์ด์ด๋ฅผ ๋ณต์ ํ๋ ๊ฒ์ ์ํ ์๋ ์์ต๋๋ค.
์ด๋ <a href="#id12"><span class="problematic" id="id13">`</span></a>๋ถ์ฐ RPC ํ๋ ์์ํฌ <<a class="reference external" href="https://pytorch.org/docs/master/rpc.html">https://pytorch.org/docs/master/rpc.html</a>>`__๋
๋งค๊ฐ๋ณ์ ์๋ฒ์์ ์๋ฒ ๋ฉ ์ฐพ๊ธฐ ์์
(embedding lookup)์ ์ํํ๋ ๋ฐ ์ฌ์ฉํ ์ ์์ต๋๋ค.</li>
<li>๋ค์์ <a class="reference external" href="https://arxiv.org/abs/1806.03377">PipeDream</a> ๋ฌธ์์์ ์ค๋ช
๋ ํ์ด๋ธ๋ฆฌ๋ ๋ณ๋ ฌ ์ฒ๋ฆฌ ํ์ฑํํ๊ธฐ ์
๋๋ค.
<a href="#id14"><span class="problematic" id="id15">`</span></a>๋ถ์ฐ RPC ํ๋ ์์ํฌ <<a class="reference external" href="https://pytorch.org/docs/master/rpc.html">https://pytorch.org/docs/master/rpc.html</a>>`__๋ฅผ ์ฌ์ฉํ์ฌ
์ฌ๋ฌ worker์ ๊ฑธ์ณ ๋ชจ๋ธ์ ๋จ๊ณ๋ฅผ ํ์ดํ๋ผ์ธ(pipeline)ํ ์ ์๊ณ
(ํ์์ ๋ฐ๋ผ) <a href="#id16"><span class="problematic" id="id17">`</span></a>๋ถ์ฐ ๋ฐ์ดํฐ ๋ณ๋ ฌ <<a class="reference external" href="https://pytorch.org/docs/stable/nn.html#torch.nn.parallel.DistributedDataParallel">https://pytorch.org/docs/stable/nn.html#torch.nn.parallel.DistributedDataParallel</a>>`__์ ์ด์ฉํด์
๊ฐ ๋จ๊ณ๋ฅผ ๋ณต์ ํ ์ ์์ต๋๋ค.</li>
</ol>
<div class="line-block">
<div class="line"><br /></div>
</div>
<p>์ด ํํ ๋ฆฌ์ผ์์๋ ์์์ ์ธ๊ธํ ์ฒซ ๋ฒ์งธ ๊ฒฝ์ฐ๋ฅผ ๋ค๋ฃฐ ๊ฒ์
๋๋ค.
๋ค์๊ณผ ๊ฐ์ด ์ด 4๊ฐ์ worker๊ฐ ์์ต๋๋ค:</p>
<ol class="arabic simple">
<li>1๊ฐ์ ๋ง์คํฐ๋ ๋งค๊ฐ๋ณ์ ์๋ฒ์ ์๋ฒ ๋ฉ ํ
์ด๋ธ(nn.EmbeddingBag) ์์ฑ์ ๋ด๋นํฉ๋๋ค.
๋ํ ๋ง์คํฐ๋ ๋ ํธ๋ ์ด๋์ ํ์ต ๋ฃจํ๋ฅผ ์ํํฉ๋๋ค.</li>
<li>1๊ฐ์ ๋งค๊ฐ๋ณ์ ์๋ฒ๋ ๊ธฐ๋ณธ์ ์ผ๋ก ๋ฉ๋ชจ๋ฆฌ์ ์๋ฒ ๋ฉ ํ
์ด๋ธ์ ๋ณด์ ํ๊ณ ๋ง์คํฐ ๋ฐ ํธ๋ ์ด๋์ RPC์ ์๋ตํฉ๋๋ค.</li>
<li>2๊ฐ์ ํธ๋ ์ด๋๋ <a href="#id18"><span class="problematic" id="id19">`</span></a>๋ถ์ฐ ๋ฐ์ดํฐ ๋ณ๋ ฌ <<a class="reference external" href="https://pytorch.org/docs/stable/nn.html#torch.nn.parallel.DistributedDataParallel">https://pytorch.org/docs/stable/nn.html#torch.nn.parallel.DistributedDataParallel</a>>`__์
์ฌ์ฉํ์ฌ ์์ฒด์ ์ผ๋ก ๋ณต์ ๋๋ FC ๋ ์ด์ด(nn.Linear)๋ฅผ ์ ์ฅํฉ๋๋ค.
ํธ๋ ์ด๋๋ ๋ํ ์๋ฐฉํฅ ์ ๋ฌ(forward pass), ์ญ๋ฐฉํฅ ์ ๋ฌ(backward pass) ๋ฐ ์ต์ ํ ๋จ๊ณ๋ฅผ ์คํํด์ผ ํฉ๋๋ค.</li>
</ol>
<div class="line-block">
<div class="line"><br /></div>
</div>
<p>์ ์ฒด์ ์ธ ํ์ต๊ณผ์ ์ ๋ค์๊ณผ ๊ฐ์ด ์คํ๋ฉ๋๋ค:</p>
<ol class="arabic simple">
<li>๋ง์คํฐ๋ ๋งค๊ฐ๋ณ์ ์๋ฒ์ ์๋ฒ ๋ฉ ํ
์ด๋ธ์ ๋ด๊ณ ์๋
<a href="#id20"><span class="problematic" id="id21">`</span></a>์๊ฒฉ ๋ชจ๋(RemoteModule) <<a class="reference external" href="https://pytorch.org/docs/master/rpc.html#remotemodule">https://pytorch.org/docs/master/rpc.html#remotemodule</a>>`__์ ์์ฑํฉ๋๋ค.</li>
<li>๊ทธ๋ฐ ๋ค์ ๋ง์คํฐ๋ ํธ๋ ์ด๋์ ํ์ต ๋ฃจํ๋ฅผ ์์ํ๊ณ ์๊ฒฉ ๋ชจ๋(remote module)์ ํธ๋ ์ด๋์๊ฒ ์ ๋ฌํฉ๋๋ค.</li>
<li>ํธ๋ ์ด๋๋ ๋จผ์ ๋ง์คํฐ์์ ์ ๊ณตํ๋ ์๊ฒฉ ๋ชจ๋์ ์ฌ์ฉํ์ฌ
์๋ฒ ๋ฉ ์ฐพ๊ธฐ ์์
(embedding lookup)์ ์ํํ ๋ค์ DDP ๋ด๋ถ์ ๊ฐ์ธ์ง FC ๋ ์ด์ด๋ฅผ ์คํํ๋ <a href="#id22"><span class="problematic" id="id23">``</span></a>HybridModel``์ ์์ฑํฉ๋๋ค.</li>
<li>ํธ๋ ์ด๋๋ ๋ชจ๋ธ์ ์๋ฐฉํฅ ์ ๋ฌ์ ์คํํ๊ณ ์์ค์ ์ฌ์ฉํ์ฌ <a href="#id24"><span class="problematic" id="id25">`</span></a>๋ถ์ฐ Autograd <<a class="reference external" href="https://pytorch.org/docs/master/rpc.html#distributed-autograd-framework">https://pytorch.org/docs/master/rpc.html#distributed-autograd-framework</a>>`__๋ฅผ
์ฌ์ฉํ์ฌ ์ญ๋ฐฉํฅ ์ ๋ฌ์ ์คํํฉ๋๋ค.</li>
<li>์ญ๋ฐฉํฅ ์ ๋ฌ์ ์ผ๋ถ๋ก FC ๋ ์ด์ด์ ๋ณํ๋๊ฐ ๋จผ์ ๊ณ์ฐ๋๊ณ DDP์ allreduce๋ฅผ ํตํด ๋ชจ๋ ํธ๋ ์ด๋์ ๋๊ธฐํ๋ฉ๋๋ค.</li>
<li>๋ค์์ผ๋ก, ๋ถ์ฐ Autograd๋ ๋งค๊ฐ๋ณ์ ์๋ฒ๋ก ๋ณํ๋๋ฅผ ์ ํํ๊ณ ๊ทธ๊ณณ์์ ์๋ฒ ๋ฉ ํ
์ด๋ธ์ ๋ณํ๋๊ฐ ์
๋ฐ์ดํธ๋ฉ๋๋ค.</li>
<li>๋ง์ง๋ง์ผ๋ก, <a href="#id26"><span class="problematic" id="id27">`</span></a>๋ถ์ฐ ์ตํฐ๋ง์ด์ (DistributedOptimizer) <<a class="reference external" href="https://pytorch.org/docs/master/rpc.html#module-torch.distributed.optim">https://pytorch.org/docs/master/rpc.html#module-torch.distributed.optim</a>>`__๋ ๋ชจ๋ ๋งค๊ฐ๋ณ์๋ฅผ ์
๋ฐ์ดํธํ๋ ๋ฐ ์ฌ์ฉ๋ฉ๋๋ค.</li>
</ol>
<p>์ด์ ๊ฐ ๋ถ๋ถ์ ์์ธํ ์ดํด๋ณด๊ฒ ์ต๋๋ค.
๋จผ์ ํ์ต์ ์ํํ๊ธฐ ์ ์ ๋ชจ๋ ์์
์๋ฅผ ์ค์ ํด์ผ ํฉ๋๋ค.
์์ 0๊ณผ 1์ ํธ๋ ์ด๋, ์์ 2๋ ๋ง์คํฐ, ์์ 3์ ๋งค๊ฐ๋ณ์ ์๋ฒ์ธ 4๊ฐ์ ํ๋ก์ธ์ค๋ฅผ ๋ง๋ญ๋๋ค.</p>
<p>TCP init_method๋ฅผ ์ฌ์ฉํ์ฌ 4๊ฐ์ ๋ชจ๋ worker์์ RPC ํ๋ ์์ํฌ๋ฅผ ์ด๊ธฐํํฉ๋๋ค.
RPC ์ด๊ธฐํ๊ฐ ๋๋๋ฉด, ๋ง์คํฐ๋ <a class="reference external" href="https://pytorch.org/docs/master/generated/torch.nn.EmbeddingBag.html">EmbeddingBag</a> ๋ ์ด์ด๋ฅผ
<a href="#id28"><span class="problematic" id="id29">`</span></a>์๊ฒฉ ๋ชจ๋(RemoteModule) <<a class="reference external" href="https://pytorch.org/docs/master/rpc.html#remotemodule">https://pytorch.org/docs/master/rpc.html#remotemodule</a>>`__์ ์ฌ์ฉํ์ฌ
๋งค๊ฐ๋ณ์ ์๋ฒ์ ๋ด๊ณ ์๋ ์๊ฒฉ ๋ชจ๋ ํ๋๋ฅผ ์์ฑํฉ๋๋ค.
๊ทธ๋ฐ ๋ค์ ๋ง์คํฐ๋ ๊ฐ ํธ๋ ์ด๋๋ฅผ ๋ฐ๋ณตํ๊ณ <a href="#id30"><span class="problematic" id="id31">`</span></a>rpc_async <<a class="reference external" href="https://pytorch.org/docs/master/rpc.html#torch.distributed.rpc.rpc_async">https://pytorch.org/docs/master/rpc.html#torch.distributed.rpc.rpc_async</a>>`__๋ฅผ
์ฌ์ฉํ์ฌ ๊ฐ ํธ๋ ์ด๋์์ <a href="#id32"><span class="problematic" id="id33">``</span></a>_run_trainer``๋ฅผ ํธ์ถํ์ฌ ๋ฐ๋ณต ํ์ต์ ์์ํฉ๋๋ค.
๋ง์ง๋ง์ผ๋ก ๋ง์คํฐ๋ ์ข
๋ฃํ๊ธฐ ์ ์ ๋ชจ๋ ํ์ต์ด ์๋ฃ๋ ๋๊น์ง ๊ธฐ๋ค๋ฆฝ๋๋ค.</p>
<p>ํธ๋ ์ด๋๋ <cite>init_process_group <https://pytorch.org/docs/stable/distributed.html#torch.distributed.init_process_group>`__์ ์ฌ์ฉํ์ฌ
(2๊ฐ์ ํธ๋ ์ด๋) world_size=2๋ก DDP๋ฅผ ์ํด ``ProcessGroup``์ ์ด๊ธฐํํฉ๋๋ค.
๋ค์์ผ๋ก TCP init_method๋ฅผ ์ฌ์ฉํ์ฌ RPC ํ๋ ์์ํฌ๋ฅผ ์ด๊ธฐํํฉ๋๋ค.
์ฌ๊ธฐ์ ์ฃผ์ ํ ์ ์ RPC ์ด๊ธฐํ์ ProgressGroup ์ด๊ธฐํ์์ ์ฐ์ด๋ ํฌํธ(port)๊ฐ ๋ค๋ฅด๋ค๋ ๊ฒ์
๋๋ค.
์ด๋ ๋ ํ๋ ์์ํฌ์ ์ด๊ธฐํ ๊ฐ์ ํฌํธ ์ถฉ๋์ ํผํ๊ธฐ ์ํด์ ์
๋๋ค.
์ด๊ธฐํ๊ฐ ์๋ฃ๋๋ฉด ํธ๋ ์ด๋๋ ๋ง์คํฐ์ ``_run_trainer</cite> RPC๋ฅผ ๊ธฐ๋ค๋ฆฌ๊ธฐ๋ง ํ๋ฉด ๋ฉ๋๋ค.</p>
<p>ํ๋ผํผํฐ ์๋ฒ๋ RPC ํ๋ ์์ํฌ๋ฅผ ์ด๊ธฐํํ๊ณ ํธ๋ ์ด๋์ ๋ง์คํฐ์ RPC๋ฅผ ๊ธฐ๋ค๋ฆฝ๋๋ค.</p>
<div class="highlight-py notranslate"><div class="highlight"><pre><span></span><span class="k">def</span> <span class="nf">run_worker</span><span class="p">(</span><span class="n">rank</span><span class="p">,</span> <span class="n">world_size</span><span class="p">):</span>
<span class="sa">r</span><span class="sd">"""</span>
<span class="sd"> ์ด ํจ์๋ RPC๋ฅผ ์ด๊ธฐํํ๊ณ , ํจ์๋ฅผ ํธ์ถํ๊ณ ,</span>
<span class="sd"> RPC๋ฅผ ์ข
๋ฃํ๋ ๋ํผ ํจ์(wrapper function)์
๋๋ค.</span>
<span class="sd"> """</span>
<span class="c1"># ํฌํธ ์ถฉ๋์ ํผํ๊ธฐ ์ํด init_rpc ๋ฐ init_process_group์ ๋ํด</span>
<span class="c1"># TCP init_method์์ ๋ค๋ฅธ ํฌํธ ๋ฒํธ๋ฅผ ์ฌ์ฉํด์ผ ํฉ๋๋ค.</span>
<span class="n">rpc_backend_options</span> <span class="o">=</span> <span class="n">TensorPipeRpcBackendOptions</span><span class="p">()</span>
<span class="n">rpc_backend_options</span><span class="o">.</span><span class="n">init_method</span> <span class="o">=</span> <span class="s2">"tcp://localhost:29501"</span>
<span class="c1"># ์์(rank) 2๋ master๋ฅผ ์๋ฏธํ๊ณ , 3์ ๋งค๊ฐ๋ณ์ ์๋ฒ, ๊ทธ๋ฆฌ๊ณ 0๊ณผ 1์ ํธ๋ ์ด๋๋ฅผ ๋ปํฉ๋๋ค.</span>
<span class="k">if</span> <span class="n">rank</span> <span class="o">==</span> <span class="mi">2</span><span class="p">:</span>
<span class="n">rpc</span><span class="o">.</span><span class="n">init_rpc</span><span class="p">(</span>
<span class="s2">"master"</span><span class="p">,</span>
<span class="n">rank</span><span class="o">=</span><span class="n">rank</span><span class="p">,</span>
<span class="n">world_size</span><span class="o">=</span><span class="n">world_size</span><span class="p">,</span>
<span class="n">rpc_backend_options</span><span class="o">=</span><span class="n">rpc_backend_options</span><span class="p">,</span>
<span class="p">)</span>
<span class="n">remote_emb_module</span> <span class="o">=</span> <span class="n">RemoteModule</span><span class="p">(</span>
<span class="s2">"ps"</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">EmbeddingBag</span><span class="p">,</span>
<span class="n">args</span><span class="o">=</span><span class="p">(</span><span class="n">NUM_EMBEDDINGS</span><span class="p">,</span> <span class="n">EMBEDDING_DIM</span><span class="p">),</span>
<span class="n">kwargs</span><span class="o">=</span><span class="p">{</span><span class="s2">"mode"</span><span class="p">:</span> <span class="s2">"sum"</span><span class="p">},</span>
<span class="p">)</span>
<span class="c1"># ํธ๋ ์ด๋์์ ํ์ต ๋ฃจํ๋ฅผ ์คํํฉ๋๋ค.</span>
<span class="n">futs</span> <span class="o">=</span> <span class="p">[]</span>
<span class="k">for</span> <span class="n">trainer_rank</span> <span class="ow">in</span> <span class="p">[</span><span class="mi">0</span><span class="p">,</span> <span class="mi">1</span><span class="p">]:</span>
<span class="n">trainer_name</span> <span class="o">=</span> <span class="s2">"trainer</span><span class="si">{}</span><span class="s2">"</span><span class="o">.</span><span class="n">format</span><span class="p">(</span><span class="n">trainer_rank</span><span class="p">)</span>
<span class="n">fut</span> <span class="o">=</span> <span class="n">rpc</span><span class="o">.</span><span class="n">rpc_async</span><span class="p">(</span>
<span class="n">trainer_name</span><span class="p">,</span> <span class="n">_run_trainer</span><span class="p">,</span> <span class="n">args</span><span class="o">=</span><span class="p">(</span><span class="n">remote_emb_module</span><span class="p">,</span> <span class="n">trainer_rank</span><span class="p">)</span>
<span class="p">)</span>
<span class="n">futs</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="n">fut</span><span class="p">)</span>
<span class="c1"># ๋ชจ๋ ํ์ต์ด ๋๋ ๋๊น์ง ๊ธฐ๋ค๋ฆฝ๋๋ค.</span>
<span class="k">for</span> <span class="n">fut</span> <span class="ow">in</span> <span class="n">futs</span><span class="p">:</span>
<span class="n">fut</span><span class="o">.</span><span class="n">wait</span><span class="p">()</span>
<span class="k">elif</span> <span class="n">rank</span> <span class="o"><=</span> <span class="mi">1</span><span class="p">:</span>
<span class="c1"># ํธ๋ ์ด๋์์ ๋ถ์ฐ ๋ฐ์ดํฐ ๋ณ๋ ฌ์ ๋ํ ํ๋ก์ธ์ค ๊ทธ๋ฃน์ ์ด๊ธฐํํฉ๋๋ค.</span>
<span class="n">dist</span><span class="o">.</span><span class="n">init_process_group</span><span class="p">(</span>
<span class="n">backend</span><span class="o">=</span><span class="s2">"gloo"</span><span class="p">,</span> <span class="n">rank</span><span class="o">=</span><span class="n">rank</span><span class="p">,</span> <span class="n">world_size</span><span class="o">=</span><span class="mi">2</span><span class="p">,</span> <span class="n">init_method</span><span class="o">=</span><span class="s2">"tcp://localhost:29500"</span>
<span class="p">)</span>
<span class="c1"># RPC๋ฅผ ์ด๊ธฐํํฉ๋๋ค.</span>
<span class="n">trainer_name</span> <span class="o">=</span> <span class="s2">"trainer</span><span class="si">{}</span><span class="s2">"</span><span class="o">.</span><span class="n">format</span><span class="p">(</span><span class="n">rank</span><span class="p">)</span>
<span class="n">rpc</span><span class="o">.</span><span class="n">init_rpc</span><span class="p">(</span>
<span class="n">trainer_name</span><span class="p">,</span>
<span class="n">rank</span><span class="o">=</span><span class="n">rank</span><span class="p">,</span>
<span class="n">world_size</span><span class="o">=</span><span class="n">world_size</span><span class="p">,</span>
<span class="n">rpc_backend_options</span><span class="o">=</span><span class="n">rpc_backend_options</span><span class="p">,</span>
<span class="p">)</span>
<span class="c1"># ํธ๋ ์ด๋๋ ๋ง์คํฐ์ RPC๋ฅผ ๊ธฐ๋ค๋ฆฌ๊ธฐ๋ง ํฉ๋๋ค.</span>
<span class="k">else</span><span class="p">:</span>
<span class="n">rpc</span><span class="o">.</span><span class="n">init_rpc</span><span class="p">(</span>
<span class="s2">"ps"</span><span class="p">,</span>
<span class="n">rank</span><span class="o">=</span><span class="n">rank</span><span class="p">,</span>
<span class="n">world_size</span><span class="o">=</span><span class="n">world_size</span><span class="p">,</span>
<span class="n">rpc_backend_options</span><span class="o">=</span><span class="n">rpc_backend_options</span><span class="p">,</span>
<span class="p">)</span>
<span class="c1"># ๋งค๊ฐ๋ณ์ ์๋ฒ๋ ์๋ฌด๊ฒ๋ ํ์ง ์์ต๋๋ค.</span>
<span class="k">pass</span>
<span class="c1"># ๋ชจ๋ rpc๊ฐ ๋๋ ๋๊น์ง ์ฐจ๋จํฉ๋๋ค.</span>
<span class="n">rpc</span><span class="o">.</span><span class="n">shutdown</span><span class="p">()</span>
<span class="k">if</span> <span class="vm">__name__</span> <span class="o">==</span> <span class="s2">"__main__"</span><span class="p">:</span>
<span class="c1"># 2๊ฐ์ ํธ๋ ์ด๋, 1๊ฐ์ ๋งค๊ฐ๋ณ์ ์๋ฒ, 1๊ฐ์ ๋ง์คํฐ</span>
<span class="n">world_size</span> <span class="o">=</span> <span class="mi">4</span>
<span class="n">mp</span><span class="o">.</span><span class="n">spawn</span><span class="p">(</span><span class="n">run_worker</span><span class="p">,</span> <span class="n">args</span><span class="o">=</span><span class="p">(</span><span class="n">world_size</span><span class="p">,),</span> <span class="n">nprocs</span><span class="o">=</span><span class="n">world_size</span><span class="p">,</span> <span class="n">join</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span>
</pre></div>
</div>
<p>ํธ๋ ์ด๋์ ๋ํ ์์ธํ ์ค๋ช
์ ์์, ํธ๋ ์ด๋๊ฐ ์ฌ์ฉํ๋ <code class="docutils literal notranslate"><span class="pre">HybridModel``์</span> <span class="pre">๋ํด</span> <span class="pre">์ค๋ช
๋๋ฆฌ๊ฒ ์ต๋๋ค.</span>
<span class="pre">์๋์</span> <span class="pre">์ค๋ช
๋</span> <span class="pre">๋๋ก</span> <span class="pre">``HybridModel``์</span> <span class="pre">๋งค๊ฐ๋ณ์</span> <span class="pre">์๋ฒ์</span> <span class="pre">์๋ฒ ๋ฉ</span> <span class="pre">ํ
์ด๋ธ(``remote_emb_module</span></code>)๊ณผ DDP์ ์ฌ์ฉํ <a href="#id34"><span class="problematic" id="id35">``</span></a>device``๋ฅผ ๋ณด์ ํ๋ ์๊ฒฉ ๋ชจ๋์ ์ฌ์ฉํ์ฌ ์ด๊ธฐํ๋ฉ๋๋ค.
๋ชจ๋ธ ์ด๊ธฐํ๋ DDP ๋ด๋ถ์ <a class="reference external" href="https://pytorch.org/docs/master/generated/torch.nn.Linear.html">nn.Linear</a> ๋ ์ด์ด๋ฅผ
๊ฐ์ธ ๋ชจ๋ ํธ๋ ์ด๋์์ ์ด ๋ ์ด์ด๋ฅผ ๋ณต์ ํ๊ณ ๋๊ธฐํํฉ๋๋ค.</p>
<p>๋ชจ๋ธ์ ์๋ฐฉํฅ(forward) ํจ์๋ ๊ฝค ๊ฐ๋จํฉ๋๋ค.
RemoteModule์ <a href="#id36"><span class="problematic" id="id37">``</span></a>forward``๋ฅผ ์ฌ์ฉํ์ฌ ๋งค๊ฐ๋ณ์ ์๋ฒ์์ ์๋ฒ ๋ฉ ์ฐพ๊ธฐ ์์
(embedding lookup)์ ์ํํ๊ณ ๊ทธ ์ถ๋ ฅ์ FC ๋ ์ด์ด์ ์ ๋ฌํฉ๋๋ค.</p>
<div class="highlight-py notranslate"><div class="highlight"><pre><span></span><span class="k">class</span> <span class="nc">HybridModel</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="sa">r</span><span class="sd">"""</span>
<span class="sd"> ํ์ด๋ธ๋ฆฌ๋ ๋ชจ๋ธ์ ํฌ์ ๋ถ๋ถ๊ณผ ๋ฐ์ง ๋ถ๋ถ์ด ์๋ ๋ชจ๋ธ์
๋๋ค.</span>
<span class="sd"> 1) ๋ฐ์ง ๋ถ๋ถ์ ๋ถ์ฐ ๋ฐ์ดํฐ ๋ณ๋ ฌ์ ์ฌ์ฉํ์ฌ ๋ชจ๋ ํธ๋ ์ด๋์ ๊ฑธ์ณ ๋ณต์ ๋๋ nn.Linear ๋ชจ๋์
๋๋ค.</span>
<span class="sd"> 2) ํฌ์ ๋ถ๋ถ์ ๋งค๊ฐ๋ณ์ ์๋ฒ์์ nn.EmbeddingBag๋ฅผ ๊ฐ์ง๊ณ ์๋ ์๊ฒฉ ๋ชจ๋์
๋๋ค.</span>
<span class="sd"> ์ด ์๊ฒฉ ๋ชจ๋ธ์ ๋งค๊ฐ๋ณ์ ์๋ฒ์ ์๋ฒ ๋ฉ ํ
์ด๋ธ์ ๋ํ ์๊ฒฉ ์ฐธ์กฐ(Remote Reference)๋ฅผ ์ป์ ์ ์์ต๋๋ค.</span>
<span class="sd"> """</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="n">remote_emb_module</span><span class="p">,</span> <span class="n">device</span><span class="p">):</span>
<span class="nb">super</span><span class="p">(</span><span class="n">HybridModel</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">remote_emb_module</span> <span class="o">=</span> <span class="n">remote_emb_module</span>
<span class="bp">self</span><span class="o">.</span><span class="n">fc</span> <span class="o">=</span> <span class="n">DDP</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">Linear</span><span class="p">(</span><span class="mi">16</span><span class="p">,</span> <span class="mi">8</span><span class="p">)</span><span class="o">.</span><span class="n">cuda</span><span class="p">(</span><span class="n">device</span><span class="p">),</span> <span class="n">device_ids</span><span class="o">=</span><span class="p">[</span><span class="n">device</span><span class="p">])</span>
<span class="bp">self</span><span class="o">.</span><span class="n">device</span> <span class="o">=</span> <span class="n">device</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">indices</span><span class="p">,</span> <span class="n">offsets</span><span class="p">):</span>
<span class="n">emb_lookup</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">remote_emb_module</span><span class="o">.</span><span class="n">forward</span><span class="p">(</span><span class="n">indices</span><span class="p">,</span> <span class="n">offsets</span><span class="p">)</span>
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">fc</span><span class="p">(</span><span class="n">emb_lookup</span><span class="o">.</span><span class="n">cuda</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">device</span><span class="p">))</span>
</pre></div>
</div>
<p>๋ค์์ผ๋ก ํธ๋ ์ด๋์ ์ค์ ์ ์ดํด๋ณด๊ฒ ์ต๋๋ค.
ํธ๋ ์ด๋๋ ๋จผ์ ๋งค๊ฐ๋ณ์ ์๋ฒ์ ์๋ฒ ๋ฉ ํ
์ด๋ธ๊ณผ ์์ฒด ์์๋ฅผ ๋ณด์ ํ๋ ์๊ฒฉ ๋ชจ๋์ ์ฌ์ฉํ์ฌ
์์์ ์ค๋ช
ํ <a href="#id38"><span class="problematic" id="id39">``</span></a>HybridModel``์ ์์ฑํฉ๋๋ค.</p>
<p>์ด์ <a href="#id40"><span class="problematic" id="id41">`</span></a>๋ถ์ฐ ์ตํฐ๋ง์ด์ (DistributedOptimizer) <<a class="reference external" href="https://pytorch.org/docs/master/rpc.html#module-torch.distributed.optim">https://pytorch.org/docs/master/rpc.html#module-torch.distributed.optim</a>>`__๋ก
์ต์ ํํ๋ ค๋ ๋ชจ๋ ๋งค๊ฐ๋ณ์์ ๋ํ RRef ๋ชฉ๋ก์ ๊ฒ์ํด์ผ ํฉ๋๋ค.
๋งค๊ฐ๋ณ์ ์๋ฒ์์ ์๋ฒ ๋ฉ ํ
์ด๋ธ์ ๋งค๊ฐ๋ณ์๋ฅผ ๊ฒ์ํ๊ธฐ ์ํด
RemoteModule์ <a href="#id42"><span class="problematic" id="id43">`</span></a>remote_parameters <<a class="reference external" href="https://pytorch.org/docs/master/rpc.html#torch.distributed.nn.api.remote_module.RemoteModule.remote_parameters">https://pytorch.org/docs/master/rpc.html#torch.distributed.nn.api.remote_module.RemoteModule.remote_parameters</a>>`__๋ฅผ ํธ์ถํ ์ ์์ต๋๋ค.
๊ทธ๋ฆฌ๊ณ ์ด๊ฒ์ ๊ธฐ๋ณธ์ ์ผ๋ก ์๋ฒ ๋ฉ ํ
์ด๋ธ์ ๋ชจ๋ ๋งค๊ฐ๋ณ์๋ฅผ ์ดํด๋ณด๊ณ RRef ๋ชฉ๋ก์ ๋ฐํํฉ๋๋ค.
ํธ๋ ์ด๋๋ RPC๋ฅผ ํตํด ๋งค๊ฐ๋ณ์ ์๋ฒ์์ ์ด ๋ฉ์๋๋ฅผ ํธ์ถํ์ฌ ์ํ๋ ๋งค๊ฐ๋ณ์์ ๋ํ RRef ๋ชฉ๋ก์ ์์ ํฉ๋๋ค.
DistributedOptimizer๋ ํญ์ ์ต์ ํํด์ผ ํ๋ ๋งค๊ฐ๋ณ์์ ๋ํ RRef ๋ชฉ๋ก์ ๊ฐ์ ธ์ค๊ธฐ ๋๋ฌธ์ FC ๋ ์ด์ด์ ์ ์ญ ๋งค๊ฐ๋ณ์์ ๋ํด์๋ RRef๋ฅผ ์์ฑํด์ผ ํฉ๋๋ค.
์ด๊ฒ์ <a href="#id44"><span class="problematic" id="id45">``</span></a>model.fc.parameters()``๋ฅผ ํ์ํ๊ณ ๊ฐ ๋งค๊ฐ๋ณ์์ ๋ํ RRef๋ฅผ ์์ฑํ๊ณ
<a href="#id46"><span class="problematic" id="id47">``</span></a>remote_parameters()``์์ ๋ฐํ๋ ๋ชฉ๋ก์ ์ถ๊ฐํจ์ผ๋ก์จ ์ํ๋ฉ๋๋ค.
์ฐธ๊ณ ๋ก <a href="#id48"><span class="problematic" id="id49">``</span></a>model.parameters()``๋ ์ฌ์ฉํ ์ ์์ต๋๋ค. <a href="#id50"><span class="problematic" id="id51">``</span></a>RemoteModule``์์ ์ง์ํ์ง ์๋ <a href="#id52"><span class="problematic" id="id53">``</span></a>model.remote_emb_module.parameters()``๋ฅผ ์ฌ๊ท์ ์ผ๋ก ํธ์ถํ๊ธฐ ๋๋ฌธ์
๋๋ค.</p>
<p>๋ง์ง๋ง์ผ๋ก ๋ชจ๋ RRef๋ฅผ ์ฌ์ฉํ์ฌ DistributedOptimizer๋ฅผ ๋ง๋ค๊ณ CrossEntropyLoss ํจ์๋ฅผ ์ ์ํฉ๋๋ค.</p>
<div class="highlight-py notranslate"><div class="highlight"><pre><span></span><span class="k">def</span> <span class="nf">_run_trainer</span><span class="p">(</span><span class="n">remote_emb_module</span><span class="p">,</span> <span class="n">rank</span><span class="p">):</span>
<span class="sa">r</span><span class="sd">"""</span>
<span class="sd"> ๊ฐ ํธ๋ ์ด๋๋ ๋งค๊ฐ๋ณ์ ์๋ฒ์์ ์๋ฒ ๋ฉ ์ฐพ๊ธฐ ์์
(embedding lookup)์ ํฌํจํ๊ณ </span>
<span class="sd"> ์ ์ญ์์ nn.Linear๋ฅผ ์คํํ๋ ์๋ฐฉํฅ ์ ๋ฌ(forward pass)๋ฅผ ์คํํฉ๋๋ค.</span>
<span class="sd"> ์ญ๋ฐฉํฅ ์ ๋ฌ(backward pass) ๋์์, ๋ถ์ฐ ๋ฐ์ดํฐ ๋ณ๋ ฌ์ ๋ฐ์ง ๋ถ๋ถ(nn.Linear)์ ๋ํ ๋ณํ๋ ์ง๊ณ๋ฅผ ๋ด๋นํ๊ณ </span>
<span class="sd"> ๋ถ์ฐ autograd๋ ๋ณํ๋ ์
๋ฐ์ดํธ๊ฐ ๋งค๊ฐ๋ณ์ ์๋ฒ๋ก ์ ํ๋๋๋ก ํฉ๋๋ค.</span>
<span class="sd"> """</span>
<span class="c1"># ๋ชจ๋ธ์ ์์ฑํฉ๋๋ค.</span>
<span class="n">model</span> <span class="o">=</span> <span class="n">HybridModel</span><span class="p">(</span><span class="n">remote_emb_module</span><span class="p">,</span> <span class="n">rank</span><span class="p">)</span>
<span class="c1"># ๋ชจ๋ ๋ชจ๋ธ ๋งค๊ฐ๋ณ์๋ฅผ ๋ถ์ฐ ์ตํฐ๋ง์ด์ ์ ๋ํ rrefs๋ก ๊ฒ์ํฉ๋๋ค.</span>
<span class="c1"># ์๋ฒ ๋ฉ ํ
์ด๋ธ์ ๋ํ ๋งค๊ฐ๋ณ์๋ฅผ ๊ฒ์ํฉ๋๋ค.</span>
<span class="n">model_parameter_rrefs</span> <span class="o">=</span> <span class="n">model</span><span class="o">.</span><span class="n">remote_emb_module</span><span class="o">.</span><span class="n">remote_parameters</span><span class="p">()</span>
<span class="c1"># model.fc.parameters()์ ์ค์ง ์ง์ญ ๋งค๊ฐ๋ณ์๋ง ํฌํจํฉ๋๋ค.</span>
<span class="c1"># ์ฐธ๊ณ : ์ฌ๊ธฐ์์๋ model.parameters()๋ฅผ ํธ์ถํ ์ ์์ต๋๋ค.</span>
<span class="c1"># ์๋ํ๋ฉด parameters()๊ฐ ์๋๋ผ remote_parameters()๋ฅผ ์ง์ํ๋</span>
<span class="c1"># remote_emb_module.parameters()๋ฅผ ํธ์ถํ๊ธฐ ๋๋ฌธ์
๋๋ค.</span>
<span class="k">for</span> <span class="n">param</span> <span class="ow">in</span> <span class="n">model</span><span class="o">.</span><span class="n">fc</span><span class="o">.</span><span class="n">parameters</span><span class="p">():</span>
<span class="n">model_parameter_rrefs</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="n">RRef</span><span class="p">(</span><span class="n">param</span><span class="p">))</span>
<span class="c1"># ๋ถ์ฐ ์ตํฐ๋ง์ด์ ๋ฅผ ์ค์ ํฉ๋๋ค.</span>
<span class="n">opt</span> <span class="o">=</span> <span class="n">DistributedOptimizer</span><span class="p">(</span>
<span class="n">optim</span><span class="o">.</span><span class="n">SGD</span><span class="p">,</span>
<span class="n">model_parameter_rrefs</span><span class="p">,</span>
<span class="n">lr</span><span class="o">=</span><span class="mf">0.05</span><span class="p">,</span>
<span class="p">)</span>
<span class="n">criterion</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">nn</span><span class="o">.</span><span class="n">CrossEntropyLoss</span><span class="p">()</span>
</pre></div>
</div>
<p>์ด์ ๊ฐ ํธ๋ ์ด๋์์ ์คํ๋๋ ๊ธฐ๋ณธ ํ์ต ๋ฃจํ๋ฅผ ์๊ฐํ๊ฒ ์ต๋๋ค.
<a href="#id54"><span class="problematic" id="id55">``</span></a>get_next_batch``๋ ํ์ต์ ์ํ ์์์ ์
๋ ฅ๊ณผ ๋์์ ์์ฑํ๋ ๊ฒ์ ๋์์ฃผ๋ ํจ์์ผ ๋ฟ์
๋๋ค.
์ฌ๋ฌ ์ํญ(epoch)๊ณผ ๊ฐ ๋ฐฐ์น(batch)์ ๋ํด ํ์ต ๋ฃจํ๋ฅผ ์คํํฉ๋๋ค:</p>
<ol class="arabic simple">
<li>๋จผ์ ๋ถ์ฐ Autograd์ ๋ํด
<a href="#id56"><span class="problematic" id="id57">`</span></a>๋ถ์ฐ Autograd Context <<a class="reference external" href="https://pytorch.org/docs/master/rpc.html#torch.distributed.autograd.context">https://pytorch.org/docs/master/rpc.html#torch.distributed.autograd.context</a>>`__๋ฅผ ์ค์ ํฉ๋๋ค.</li>
<li>๋ชจ๋ธ์ ์๋ฐฉํฅ ์ ๋ฌ์ ์คํํ๊ณ ํด๋น ์ถ๋ ฅ์ ๊ฒ์(retrieve)ํฉ๋๋ค.</li>
<li>์์ค ํจ์๋ฅผ ์ฌ์ฉํ์ฌ ์ถ๋ ฅ๊ณผ ๋ชฉํ๋ฅผ ๊ธฐ๋ฐ์ผ๋ก ์์ค์ ๊ณ์ฐํฉ๋๋ค.</li>
<li>๋ถ์ฐ Autograd๋ฅผ ์ฌ์ฉํ์ฌ ์์ค์ ์ฌ์ฉํ์ฌ ๋ถ์ฐ ์ญ๋ฐฉํฅ ์ ๋ฌ์ ์คํํฉ๋๋ค.</li>
<li>๋ง์ง๋ง์ผ๋ก ๋ถ์ฐ ์ตํฐ๋ง์ด์ ๋จ๊ณ๋ฅผ ์คํํ์ฌ ๋ชจ๋ ๋งค๊ฐ๋ณ์๋ฅผ ์ต์ ํํฉ๋๋ค.</li>
</ol>
<div class="highlight-py notranslate"><div class="highlight"><pre><span></span> <span class="k">def</span> <span class="nf">get_next_batch</span><span class="p">(</span><span class="n">rank</span><span class="p">):</span>
<span class="k">for</span> <span class="n">_</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="mi">10</span><span class="p">):</span>
<span class="n">num_indices</span> <span class="o">=</span> <span class="n">random</span><span class="o">.</span><span class="n">randint</span><span class="p">(</span><span class="mi">20</span><span class="p">,</span> <span class="mi">50</span><span class="p">)</span>
<span class="n">indices</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">LongTensor</span><span class="p">(</span><span class="n">num_indices</span><span class="p">)</span><span class="o">.</span><span class="n">random_</span><span class="p">(</span><span class="mi">0</span><span class="p">,</span> <span class="n">NUM_EMBEDDINGS</span><span class="p">)</span>
<span class="c1"># ์คํ์
(offset)์ ์์ฑํฉ๋๋ค.</span>
<span class="n">offsets</span> <span class="o">=</span> <span class="p">[]</span>
<span class="n">start</span> <span class="o">=</span> <span class="mi">0</span>
<span class="n">batch_size</span> <span class="o">=</span> <span class="mi">0</span>
<span class="k">while</span> <span class="n">start</span> <span class="o"><</span> <span class="n">num_indices</span><span class="p">:</span>
<span class="n">offsets</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="n">start</span><span class="p">)</span>
<span class="n">start</span> <span class="o">+=</span> <span class="n">random</span><span class="o">.</span><span class="n">randint</span><span class="p">(</span><span class="mi">1</span><span class="p">,</span> <span class="mi">10</span><span class="p">)</span>
<span class="n">batch_size</span> <span class="o">+=</span> <span class="mi">1</span>
<span class="n">offsets_tensor</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">LongTensor</span><span class="p">(</span><span class="n">offsets</span><span class="p">)</span>
<span class="n">target</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">LongTensor</span><span class="p">(</span><span class="n">batch_size</span><span class="p">)</span><span class="o">.</span><span class="n">random_</span><span class="p">(</span><span class="mi">8</span><span class="p">)</span><span class="o">.</span><span class="n">cuda</span><span class="p">(</span><span class="n">rank</span><span class="p">)</span>
<span class="k">yield</span> <span class="n">indices</span><span class="p">,</span> <span class="n">offsets_tensor</span><span class="p">,</span> <span class="n">target</span>
<span class="c1"># 100๊ฐ์ ์ํญ(epoch)์ ๋ํด ํ์ตํฉ๋๋ค.</span>
<span class="k">for</span> <span class="n">epoch</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="mi">100</span><span class="p">):</span>
<span class="c1"># ๋ถ์ฐ autograd context๋ฅผ ์์ฑํฉ๋๋ค.</span>
<span class="k">for</span> <span class="n">indices</span><span class="p">,</span> <span class="n">offsets</span><span class="p">,</span> <span class="n">target</span> <span class="ow">in</span> <span class="n">get_next_batch</span><span class="p">(</span><span class="n">rank</span><span class="p">):</span>
<span class="k">with</span> <span class="n">dist_autograd</span><span class="o">.</span><span class="n">context</span><span class="p">()</span> <span class="k">as</span> <span class="n">context_id</span><span class="p">:</span>
<span class="n">output</span> <span class="o">=</span> <span class="n">model</span><span class="p">(</span><span class="n">indices</span><span class="p">,</span> <span class="n">offsets</span><span class="p">)</span>
<span class="n">loss</span> <span class="o">=</span> <span class="n">criterion</span><span class="p">(</span><span class="n">output</span><span class="p">,</span> <span class="n">target</span><span class="p">)</span>
<span class="c1"># ๋ถ์ฐ ์ญ๋ฐฉํฅ ์ ๋ฌ(distributed backward pass)์ ์คํํฉ๋๋ค.</span>
<span class="n">dist_autograd</span><span class="o">.</span><span class="n">backward</span><span class="p">(</span><span class="n">context_id</span><span class="p">,</span> <span class="p">[</span><span class="n">loss</span><span class="p">])</span>
<span class="c1"># ๋ถ์ฐ ์ตํฐ๋ง์ด์ ๋ฅผ ๊ฐฑ์ ํฉ๋๋ค.</span>
<span class="n">opt</span><span class="o">.</span><span class="n">step</span><span class="p">(</span><span class="n">context_id</span><span class="p">)</span>
<span class="c1"># ๋ฐ๋ณต๋ ๋๋ง๋ค ๋ค๋ฅธ ๋ณํ๋๋ฅผ ํธ์คํ
ํ๋ ํ๋์ ๋ค๋ฅธ ๋ถ์ฐ autograd context๋ฅผ ์์ฑํ๋ฏ๋ก</span>
<span class="c1"># ๋ณํ๋๋ฅผ 0์ผ๋ก ๋ง๋ค ํ์๊ฐ ์์ต๋๋ค.</span>
<span class="nb">print</span><span class="p">(</span><span class="s2">"Training done for epoch </span><span class="si">{}</span><span class="s2">"</span><span class="o">.</span><span class="n">format</span><span class="p">(</span><span class="n">epoch</span><span class="p">))</span>
</pre></div>
</div>
<p>์ ์ฒด ์์ ์ ์์ค ์ฝ๋๋ <a href="#id58"><span class="problematic" id="id59">`</span></a>์ฌ๊ธฐ <<a class="reference external" href="https://github.com/pytorch/examples/tree/master/distributed/rpc/ddp_rpc">https://github.com/pytorch/examples/tree/master/distributed/rpc/ddp_rpc</a>>`__์์ ์ฐพ์ ์ ์์ต๋๋ค.</p>
</div>
</article>
</div>
<footer>
<div class="rst-footer-buttons" role="navigation" aria-label="footer navigation">
<a href="../intermediate/pipeline_tutorial.html" class="btn btn-neutral float-right" title="ํ์ดํ๋ผ์ธ ๋ณ๋ ฌํ๋ก ํธ๋์คํฌ๋จธ ๋ชจ๋ธ ํ์ต์ํค๊ธฐ" accesskey="n" rel="next">Next <img src="../_static/images/chevron-right-orange.svg" class="next-page"></a>
<a href="../intermediate/rpc_async_execution.html" class="btn btn-neutral" title="Implementing Batch RPC Processing Using Asynchronous Executions" 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="#">๋ถ์ฐ ๋ฐ์ดํฐ ๋ณ๋ ฌ(DDP)๊ณผ ๋ถ์ฐ RPC ํ๋ ์์ํฌ ๊ฒฐํฉ</a></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);
});
</script>
<script>
//add microsoft link
if(window.location.href.indexOf("/beginner/basics/")!= -1)
{
var url="https://docs.microsoft.com/learn/paths/pytorch-fundamentals/?wt.mc_id=aiml-7486-cxa";
switch(window.location.pathname.split("/").pop().replace('.html',''))
{
case"quickstart_tutorial":
url="https://docs.microsoft.com/learn/modules/intro-machine-learning-pytorch/9-quickstart?WT.mc_id=aiml-7486-cxa";
break;
case"tensorqs_tutorial":
url="https://docs.microsoft.com/learn/modules/intro-machine-learning-pytorch/2-tensors?WT.mc_id=aiml-7486-cxa";
break;
case"data_tutorial":
url="https://docs.microsoft.com/learn/modules/intro-machine-learning-pytorch/3-data?WT.mc_id=aiml-7486-cxa";
break;
case"transforms_tutorial":
url="https://docs.microsoft.com/learn/modules/intro-machine-learning-pytorch/4-transforms?WT.mc_id=aiml-7486-cxa";
break;
case"buildmodel_tutorial":
url="https://docs.microsoft.com/learn/modules/intro-machine-learning-pytorch/5-model?WT.mc_id=aiml-7486-cxa";
break;
case"autogradqs_tutorial":
url="https://docs.microsoft.com/learn/modules/intro-machine-learning-pytorch/6-autograd?WT.mc_id=aiml-7486-cxa";
break;
case"optimization_tutorial":
url="https://docs.microsoft.com/learn/modules/intro-machine-learning-pytorch/7-optimization?WT.mc_id=aiml-7486-cxa";
break;
case"saveloadrun_tutorial":
url="https://docs.microsoft.com/learn/modules/intro-machine-learning-pytorch/8-inference?WT.mc_id=aiml-7486-cxa";
}
$(".pytorch-call-to-action-links").children().first().before("<a href="+url+' data-behavior="call-to-action-event" data-response="Run in Microsoft Learn" target="_blank"><div id="microsoft-learn-link" style="padding-bottom: 0.625rem;border-bottom: 1px solid #f3f4f7;padding-right: 2.5rem;display: -webkit-box; display: -ms-flexbox; isplay: flex; -webkit-box-align: center;-ms-flex-align: center;align-items: center;"><img class="call-to-action-img" src="../../_static/images/microsoft-logo.svg"/><div class="call-to-action-desktop-view">Run in Microsoft Learn</div><div class="call-to-action-mobile-view">Learn</div></div></a>')
}
</script>
<script async src="https://www.googletagmanager.com/gtag/js?id=UA-71919972-3"></script>
<script>
window.dataLayer = window.dataLayer || [];
function gtag(){dataLayer.push(arguments);}
gtag('js', new Date());
gtag('config', 'UA-71919972-3');
</script>
<script>
$("[data-behavior='call-to-action-event']").on('click', function(){
ga('send', {
hitType: 'event',
eventCategory: $(this).attr("data-response"),
eventAction: 'click',
eventLabel: window.location.href
});
gtag('event', 'click', {
'event_category': $(this).attr("data-response"),
'event_label': $("h1").first().text(),
'tutorial_link': window.location.href
});
});
$("[data-behavior='tutorial-rating']").on('click', function(){
gtag('event', 'click', {
'event_category': 'Tutorial Rating',
'event_label': $("h1").first().text(),
'value': $(this).attr("data-count")
});
});
if (location.pathname == "/") {
$(".rating-container").hide();
$(".hr-bottom").hide();
}
</script>
<script type="text/javascript">
var collapsedSections = ['ํ์ดํ ์น(PyTorch) ๋ ์ํผ', 'ํ์ดํ ์น(PyTorch) ๋ฐฐ์ฐ๊ธฐ', '์ด๋ฏธ์ง/๋น๋์ค', '์ค๋์ค', 'ํ
์คํธ', '๊ฐํํ์ต', 'PyTorch ๋ชจ๋ธ์ ํ๋ก๋์
ํ๊ฒฝ์ ๋ฐฐํฌํ๊ธฐ', 'Code Transforms with FX', 'ํ๋ก ํธ์๋ API', 'PyTorch ํ์ฅํ๊ธฐ', '๋ชจ๋ธ ์ต์ ํ', '๋ณ๋ ฌ ๋ฐ ๋ถ์ฐ ํ์ต', 'Mobile'];
</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>๊ณต์ ๋ฌธ์ (์์ด)</h2>
<p>PyTorch ๊ณต์ ๋ฌธ์์
๋๋ค.</p>
<a id="orgTutorialLink" class="with-right-arrow" href="https://pytorch.org/docs/stable/index.html" target="_blank">๊ณต์ ๋ฌธ์๋ก ์ด๋</a>
</div>
<div class="col-md-4 text-center">
<h2>ํ๊ตญ์ด ํํ ๋ฆฌ์ผ</h2>
<p>ํ๊ตญ์ด๋ก ๋ฒ์ญ ์ค์ธ PyTorch ํํ ๋ฆฌ์ผ์
๋๋ค.</p>
<a class="with-right-arrow" href="https://tutorials.pytorch.kr">ํํ ๋ฆฌ์ผ๋ก ์ด๋</a>
</div>
<div class="col-md-4 text-center">
<h2>์ปค๋ฎค๋ํฐ</h2>
<p>๋ค๋ฅธ ์ฌ์ฉ์๋ค๊ณผ ์๊ฒฌ์ ๋๋ ๋ณด์ธ์!</p>
<a class="with-right-arrow" href="https://discuss.pytorch.kr">์ปค๋ฎค๋ํฐ๋ก ์ด๋</a>
</div>
</div>
</div>
</div>
<footer class="site-footer">
<div class="container footer-container">
<div class="footer-logo-wrapper">
<a href="https://pytorch.kr/" class="footer-logo"></a>
</div>
<div class="footer-links-wrapper">
<div class="footer-links-col">
<ul>
<li class="list-title"><a href="https://pytorch.org">PyTorch ํํ์ด์ง (๊ณต์)</a></li>
<li><a href="https://pytorch.org" target="_blank">๊ณต์ ํํ์ด์ง</a></li>
<li><a href="https://pytorch.org/tutorials" target="_blank">๊ณต์ ํํ ๋ฆฌ์ผ</a></li>
<li><a href="https://pytorch.org/docs" target="_blank">๊ณต์ ๋ฌธ์</a></li>
</ul>
</div>
<div class="footer-links-col">
<ul>
<li class="list-title"><a href="">ํ๊ตญ ์ฌ์ฉ์ ๋ชจ์</a></li>
<li><a href="https://pytorch.kr/about">์ฌ์ดํธ ์๊ฐ</a></li>
<li><a href="https://tutorials.pytorch.kr/">ํ๊ตญ์ด ํํ ๋ฆฌ์ผ</a></li>
<li><a href="https://github.com/9bow/PyTorch-tutorials-kr" target="_blank">ํ๊ตญ์ด ํํ ๋ฆฌ์ผ ์ ์ฅ์</a></li>
</ul>
</div>
</div>
<div class="trademark-disclaimer">
<ul>
<li>์ด ์ฌ์ดํธ๋ PyTorch ํ๊ตญ ์ฌ์ฉ์ ์ปค๋ฎค๋ํฐ๋ก Facebook, Inc์์ ์ด์ํ๋ ์ฌ์ดํธ๊ฐ ์๋๋๋ค. PyTorch, PyTorch ๋ก๊ณ ๋ฐ ๋ชจ๋ ๊ด๋ จ ํ๊ธฐ๋ Facebook, Inc์ ์ํ์
๋๋ค.</li>
<li>This site is a user community and is not operated by Facebook, Inc. PyTorch, the PyTorch logo and any related marks are trademarks of Facebook, Inc.</li>
</ul>
</div>
</div>
</footer>
<!-- End Footer -->
<!-- Begin Mobile Menu -->
<div class="mobile-main-menu">
<div class="container-fluid">
<div class="container">
<div class="mobile-main-menu-header-container">
<a class="header-logo" href="https://pytorch.kr/" aria-label="PyTorch"></a>
<a class="main-menu-close-button" href="#" data-behavior="close-mobile-menu"></a>
</div>
</div>
</div>
<div class="mobile-main-menu-links-container">
<div class="main-menu">
<ul>
<li>
<a href="https://pytorch.kr/get-started">์์ํ๊ธฐ</a>
</li>
<li class="active">
<a href="https://tutorials.pytorch.kr">ํํ ๋ฆฌ์ผ</a>
</li>
<li>
<a href="https://pytorch.kr/hub">ํ๋ธ</a>
</li>
<li>
<a href="https://discuss.pytorch.kr">์ปค๋ฎค๋ํฐ</a>
</li>
</ul>
</div>
</div>
</div>
<!-- End Mobile Menu -->
<script type="text/javascript" src="../_static/js/vendor/anchor.min.js"></script>
<script type="text/javascript">
$(document).ready(function() {
mobileMenu.bind();
mobileTOC.bind();
pytorchAnchors.bind();
sideMenus.bind();
scrollToAnchor.bind();
highlightNavigation.bind();
mainMenuDropdown.bind();
filterTags.bind();
// Add class to links that have code blocks, since we cannot create links in code blocks
$("article.pytorch-article a span.pre").each(function(e) {
$(this).closest("a").addClass("has-code");
});
})
</script>
</body>
</html>