-
Notifications
You must be signed in to change notification settings - Fork 257
/
Copy pathquantized_transfer_learning_tutorial.html
1204 lines (990 loc) ยท 101 KB
/
quantized_transfer_learning_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
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="ko" > <![endif]-->
<!--[if gt IE 8]><!--> <html class="no-js" lang="ko" > <!--<![endif]-->
<head>
<meta charset="utf-8">
<meta property="og:title" content="(๋ฒ ํ) ์ปดํจํฐ ๋น์ ํํ ๋ฆฌ์ผ์ ์ํ ์์ํ๋ ์ ์ดํ์ต(Quantized Transfer Learning)" />
<meta property="og:type" content="article" />
<meta property="og:url" content="https://tutorials.pytorch.kr/intermediate/quantized_transfer_learning_tutorial.html" />
<meta property="og:site_name" content="PyTorch Tutorials KR" />
<meta property="og:description" content="Author: Zafar Takhirov Reviewed by: Raghuraman Krishnamoorthi Edited by: Jessica Lin ๋ฒ์ญ: ์ ์ฌ๋ฏผ ์ด ํํ ๋ฆฌ์ผ์ Sasank Chilamkurthy ๊ฐ ์์ฑํ ์ปดํจํฐ ๋น์ (Vision)์ ์ํ ์ ์ดํ์ต(Transfer Learning) ์ ๊ธฐ๋ฐ์ผ๋ก ํฉ๋๋ค. ์ ์ดํ์ต(Transfer learning)์ ๋ค๋ฅธ ๋ฐ์ดํฐ์
์ ์ ์ฉํ๊ธฐ ์ํด์ ๋ฏธ๋ฆฌ ํ์ต๋ ๋ชจ๋ธ์ ์ฌ์ฉํ๋ ๊ธฐ์ ์ ๋งํฉ๋๋ค. ์ ์ดํ์ต์ ์ฌ์ฉํ๋ 2๊ฐ์ง ์ฃผ์ ๋ฐฉ๋ฒ์ด ์์ต๋๋ค. ๊ณ ์ ๋ ํน์ง ์ถ์ถ๊ธฐ๋ก์จ Conv..." />
<meta property="og:image" content="https://tutorials.pytorch.kr/_static/logos/logo-kr-sm-dark.png" />
<meta property="og:image:alt" content="PyTorch Tutorials KR" />
<meta name="description" content="Author: Zafar Takhirov Reviewed by: Raghuraman Krishnamoorthi Edited by: Jessica Lin ๋ฒ์ญ: ์ ์ฌ๋ฏผ ์ด ํํ ๋ฆฌ์ผ์ Sasank Chilamkurthy ๊ฐ ์์ฑํ ์ปดํจํฐ ๋น์ (Vision)์ ์ํ ์ ์ดํ์ต(Transfer Learning) ์ ๊ธฐ๋ฐ์ผ๋ก ํฉ๋๋ค. ์ ์ดํ์ต(Transfer learning)์ ๋ค๋ฅธ ๋ฐ์ดํฐ์
์ ์ ์ฉํ๊ธฐ ์ํด์ ๋ฏธ๋ฆฌ ํ์ต๋ ๋ชจ๋ธ์ ์ฌ์ฉํ๋ ๊ธฐ์ ์ ๋งํฉ๋๋ค. ์ ์ดํ์ต์ ์ฌ์ฉํ๋ 2๊ฐ์ง ์ฃผ์ ๋ฐฉ๋ฒ์ด ์์ต๋๋ค. ๊ณ ์ ๋ ํน์ง ์ถ์ถ๊ธฐ๋ก์จ Conv..." />
<meta property="og:ignore_canonical" content="true" />
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>(๋ฒ ํ) ์ปดํจํฐ ๋น์ ํํ ๋ฆฌ์ผ์ ์ํ ์์ํ๋ ์ ์ดํ์ต(Quantized Transfer Learning) — ํ์ดํ ์น ํ๊ตญ์ด ํํ ๋ฆฌ์ผ (PyTorch tutorials in Korean)</title>
<link rel="shortcut icon" href="../_static/favicon.ico"/>
<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/pygments.css" type="text/css" />
<link rel="stylesheet" href="../_static/css/theme.css" type="text/css" />
<link rel="stylesheet" href="../_static/copybutton.css" type="text/css" />
<link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/katex@0.16.10/dist/katex.min.css" type="text/css" />
<link rel="stylesheet" href="../_static/katex-math.css" type="text/css" />
<link rel="stylesheet" href="../_static/sg_gallery.css" type="text/css" />
<link rel="stylesheet" href="../_static/sg_gallery-binder.css" type="text/css" />
<link rel="stylesheet" href="../_static/sg_gallery-dataframe.css" type="text/css" />
<link rel="stylesheet" href="../_static/sg_gallery-rendered-html.css" type="text/css" />
<link rel="stylesheet" href="../_static/sphinx-design.5ea377869091fd0449014c60fc090103.min.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="../_static/css/custom.css" type="text/css" />
<link rel="stylesheet" href="../_static/css/custom2.css" type="text/css" />
<link rel="index" title="์์ธ" href="../genindex.html" />
<link rel="search" title="๊ฒ์" href="../search.html" />
<link rel="next" title="(๋ฒ ํ) PyTorch์์ Eager Mode๋ฅผ ์ด์ฉํ ์ ์ ์์ํ" href="../advanced/static_quantization_tutorial.html" />
<link rel="prev" title="(๋ฒ ํ) BERT ๋ชจ๋ธ ๋์ ์์ํํ๊ธฐ" href="dynamic_quantization_bert_tutorial.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">
2.3.1+cu121
</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" role="heading"><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" role="heading"><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" role="heading"><span class="caption-text">Introduction to PyTorch on YouTube</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../beginner/introyt.html">PyTorch ์๊ฐ - YouTube ์๋ฆฌ์ฆ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/introyt/introyt1_tutorial.html">PyTorch ์๊ฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/introyt/tensors_deeper_tutorial.html">Pytorch Tensor ์๊ฐ</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" role="heading"><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="tensorboard_tutorial.html">TensorBoard๋ก ๋ชจ๋ธ, ๋ฐ์ดํฐ, ํ์ต ์๊ฐํํ๊ธฐ</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">์ด๋ฏธ์ง/๋น๋์ค</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="torchvision_tutorial.html">TorchVision Object Detection Finetuning Tutorial</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 ์ํด ๋น์ ํธ๋์คํฌ๋จธ(Vision Transformer) ๋ชจ๋ธ ์ต์ ํํ๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="tiatoolbox_tutorial.html">Whole Slide Image Classification Using PyTorch and TIAToolbox</a></li>
</ul>
<p class="caption" role="heading"><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="speech_recognition_pipeline_tutorial.html">Speech Recognition with Wav2Vec2</a></li>
<li class="toctree-l1"><a class="reference internal" href="text_to_speech_with_torchaudio.html">Text-to-speech with Tacotron2</a></li>
<li class="toctree-l1"><a class="reference internal" href="forced_alignment_with_torchaudio_tutorial.html">wav2vec2์ ์ด์ฉํ ๊ฐ์ ์ ๋ ฌ</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">ํ
์คํธ</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../beginner/bettertransformer_tutorial.html">Fast Transformer Inference with Better Transformer</a></li>
<li class="toctree-l1"><a class="reference internal" href="char_rnn_classification_tutorial.html">๊ธฐ์ด๋ถํฐ ์์ํ๋ NLP: ๋ฌธ์-๋จ์ RNN์ผ๋ก ์ด๋ฆ ๋ถ๋ฅํ๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="char_rnn_generation_tutorial.html">๊ธฐ์ด๋ถํฐ ์์ํ๋ NLP: ๋ฌธ์-๋จ์ RNN์ผ๋ก ์ด๋ฆ ์์ฑํ๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="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"><code class="docutils literal notranslate"><span class="pre">nn.Transformer</span></code> ์ torchtext๋ก ์ธ์ด ๋ฒ์ญํ๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/torchtext_custom_dataset_tutorial.html">Preprocess custom text dataset using Torchtext</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">๋ฐฑ์๋</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../beginner/onnx/intro_onnx.html">Introduction to ONNX</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">๊ฐํํ์ต</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="reinforcement_q_learning.html">๊ฐํ ํ์ต (DQN) ํํ ๋ฆฌ์ผ</a></li>
<li class="toctree-l1"><a class="reference internal" href="reinforcement_ppo.html">Reinforcement Learning (PPO) with TorchRL Tutorial</a></li>
<li class="toctree-l1"><a class="reference internal" href="mario_rl_tutorial.html">๋ง๋ฆฌ์ค ๊ฒ์ RL ์์ด์ ํธ๋ก ํ์ตํ๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../advanced/pendulum.html">Pendulum: Writing your environment and transforms with TorchRL</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">PyTorch ๋ชจ๋ธ์ ํ๋ก๋์
ํ๊ฒฝ์ ๋ฐฐํฌํ๊ธฐ</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../beginner/onnx/intro_onnx.html">Introduction to ONNX</a></li>
<li class="toctree-l1"><a class="reference internal" href="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="../advanced/cpp_export.html">C++์์ TorchScript ๋ชจ๋ธ ๋ก๋ฉํ๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../advanced/super_resolution_with_onnxruntime.html">(์ ํ) PyTorch ๋ชจ๋ธ์ ONNX์ผ๋ก ๋ณํํ๊ณ ONNX ๋ฐํ์์์ ์คํํ๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="realtime_rpi.html">Raspberry Pi 4 ์์ ์ค์๊ฐ ์ถ๋ก (Inference) (30fps!)</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">PyTorch ํ๋กํ์ผ๋ง</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../beginner/profiler.html">PyTorch ๋ชจ๋ ํ๋กํ์ผ๋งํ๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/hta_intro_tutorial.html">Introduction to Holistic Trace Analysis</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/hta_trace_diff_tutorial.html">Trace Diff using Holistic Trace Analysis</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">Code Transforms with FX</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="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="fx_profiling_tutorial.html">(beta) Building a Simple CPU Performance Profiler with FX</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">ํ๋ก ํธ์๋ API</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="memory_format_tutorial.html">(๋ฒ ํ) PyTorch๋ฅผ ์ฌ์ฉํ Channels Last ๋ฉ๋ชจ๋ฆฌ ํ์</a></li>
<li class="toctree-l1"><a class="reference internal" href="forward_ad_usage.html">Forward-mode Automatic Differentiation (Beta)</a></li>
<li class="toctree-l1"><a class="reference internal" href="jacobians_hessians.html">Jacobians, Hessians, hvp, vhp, and more: composing function transforms</a></li>
<li class="toctree-l1"><a class="reference internal" href="ensembling.html">๋ชจ๋ธ ์์๋ธ</a></li>
<li class="toctree-l1"><a class="reference internal" href="per_sample_grads.html">Per-sample-gradients</a></li>
<li class="toctree-l1"><a class="reference internal" href="../advanced/cpp_frontend.html">PyTorch C++ ํ๋ก ํธ์๋ ์ฌ์ฉํ๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../advanced/torch-script-parallelism.html">TorchScript์ ๋์ ๋ณ๋ ฌ ์ฒ๋ฆฌ(Dynamic Parallelism)</a></li>
<li class="toctree-l1"><a class="reference internal" href="../advanced/cpp_autograd.html">C++ ํ๋ก ํธ์๋์ ์๋ ๋ฏธ๋ถ (autograd)</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">PyTorch ํ์ฅํ๊ธฐ</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="custom_function_double_backward_tutorial.html">Double Backward with Custom Functions</a></li>
<li class="toctree-l1"><a class="reference internal" href="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="../advanced/cpp_extension.html">Custom C++ and CUDA Extensions</a></li>
<li class="toctree-l1"><a class="reference internal" href="../advanced/torch_script_custom_ops.html">Extending TorchScript with Custom C++ Operators</a></li>
<li class="toctree-l1"><a class="reference internal" href="../advanced/torch_script_custom_classes.html">์ปค์คํ
C++ ํด๋์ค๋ก TorchScript ํ์ฅํ๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../advanced/dispatcher.html">Registering a Dispatched Operator in C++</a></li>
<li class="toctree-l1"><a class="reference internal" href="../advanced/extend_dispatcher.html">Extending dispatcher for a new backend in C++</a></li>
<li class="toctree-l1"><a class="reference internal" href="../advanced/privateuseone.html">Facilitating New Backend Integration by PrivateUse1</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">๋ชจ๋ธ ์ต์ ํ</span></p>
<ul class="current">
<li class="toctree-l1"><a class="reference internal" href="../beginner/profiler.html">PyTorch ๋ชจ๋ ํ๋กํ์ผ๋งํ๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="tensorboard_profiler_tutorial.html">ํ
์๋ณด๋๋ฅผ ์ด์ฉํ ํ์ดํ ์น ํ๋กํ์ผ๋ฌ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/hyperparameter_tuning_tutorial.html">Ray Tune์ ์ฌ์ฉํ ํ์ดํผํ๋ผ๋ฏธํฐ ํ๋</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/vt_tutorial.html">๋ฐฐํฌ๋ฅผ ์ํด ๋น์ ํธ๋์คํฌ๋จธ(Vision Transformer) ๋ชจ๋ธ ์ต์ ํํ๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="parametrizations.html">Parametrizations Tutorial</a></li>
<li class="toctree-l1"><a class="reference internal" href="pruning_tutorial.html">๊ฐ์ง์น๊ธฐ ๊ธฐ๋ฒ(Pruning) ํํ ๋ฆฌ์ผ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../advanced/dynamic_quantization_tutorial.html">(๋ฒ ํ) LSTM ๊ธฐ๋ฐ ๋จ์ด ๋จ์ ์ธ์ด ๋ชจ๋ธ์ ๋์ ์์ํ</a></li>
<li class="toctree-l1"><a class="reference internal" href="dynamic_quantization_bert_tutorial.html">(๋ฒ ํ) BERT ๋ชจ๋ธ ๋์ ์์ํํ๊ธฐ</a></li>
<li class="toctree-l1 current"><a class="current reference internal" href="#">(๋ฒ ํ) ์ปดํจํฐ ๋น์ ํํ ๋ฆฌ์ผ์ ์ํ ์์ํ๋ ์ ์ดํ์ต(Quantized Transfer Learning)</a></li>
<li class="toctree-l1"><a class="reference internal" href="../advanced/static_quantization_tutorial.html">(๋ฒ ํ) PyTorch์์ Eager Mode๋ฅผ ์ด์ฉํ ์ ์ ์์ํ</a></li>
<li class="toctree-l1"><a class="reference internal" href="torchserve_with_ipex.html">Grokking PyTorch Intel CPU performance from first principles</a></li>
<li class="toctree-l1"><a class="reference internal" href="torchserve_with_ipex_2.html">Grokking PyTorch Intel CPU performance from first principles (Part 2)</a></li>
<li class="toctree-l1"><a class="reference internal" href="nvfuser_intro_tutorial.html">Getting Started - Accelerate Your Scripts with nvFuser</a></li>
<li class="toctree-l1"><a class="reference internal" href="ax_multiobjective_nas_tutorial.html">Multi-Objective NAS with Ax</a></li>
<li class="toctree-l1"><a class="reference internal" href="torch_compile_tutorial.html">Introduction to <code class="docutils literal notranslate"><span class="pre">torch.compile</span></code></a></li>
<li class="toctree-l1"><a class="reference internal" href="inductor_debug_cpu.html">Inductor CPU backend debugging and profiling</a></li>
<li class="toctree-l1"><a class="reference internal" href="scaled_dot_product_attention_tutorial.html">(Beta) Scaled Dot Product Attention (SDPA)๋ก ๊ณ ์ฑ๋ฅ ํธ๋์คํฌ๋จธ(Transformers) ๊ตฌํํ๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="scaled_dot_product_attention_tutorial.html#torch-compile-sdpa"><code class="docutils literal notranslate"><span class="pre">torch.compile</span></code> ๊ณผ ํจ๊ป SDPA ์ฌ์ฉํ๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="scaled_dot_product_attention_tutorial.html#sdpa-atteition-bias">SDPA๋ฅผ <code class="docutils literal notranslate"><span class="pre">atteition.bias</span></code> ํ์ ํด๋์ค์ ์ฌ์ฉํ๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="scaled_dot_product_attention_tutorial.html#id8">๊ฒฐ๋ก </a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/knowledge_distillation_tutorial.html">Knowledge Distillation Tutorial</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">๋ณ๋ ฌ ๋ฐ ๋ถ์ฐ ํ์ต</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../distributed/home.html">Distributed and Parallel Training Tutorials</a></li>
<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="../beginner/ddp_series_intro.html">Distributed Data Parallel in PyTorch - Video Tutorials</a></li>
<li class="toctree-l1"><a class="reference internal" href="model_parallel_tutorial.html">๋จ์ผ ๋จธ์ ์ ์ฌ์ฉํ ๋ชจ๋ธ ๋ณ๋ ฌํ ๋ชจ๋ฒ ์ฌ๋ก</a></li>
<li class="toctree-l1"><a class="reference internal" href="ddp_tutorial.html">๋ถ์ฐ ๋ฐ์ดํฐ ๋ณ๋ ฌ ์ฒ๋ฆฌ ์์ํ๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="dist_tuto.html">PyTorch๋ก ๋ถ์ฐ ์ดํ๋ฆฌ์ผ์ด์
๊ฐ๋ฐํ๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="FSDP_tutorial.html">Getting Started with Fully Sharded Data Parallel(FSDP)</a></li>
<li class="toctree-l1"><a class="reference internal" href="FSDP_adavnced_tutorial.html">Advanced Model Training with Fully Sharded Data Parallel (FSDP)</a></li>
<li class="toctree-l1"><a class="reference internal" href="TP_tutorial.html">Large Scale Transformer model training with Tensor Parallel (TP)</a></li>
<li class="toctree-l1"><a class="reference internal" href="process_group_cpp_extension_tutorial.html">Cpp ํ์ฅ์ ์ฌ์ฉํ ํ๋ก์ธ์ค ๊ทธ๋ฃน ๋ฐฑ์๋ ์ฌ์ฉ์ ์ ์</a></li>
<li class="toctree-l1"><a class="reference internal" href="rpc_tutorial.html">Getting Started with Distributed RPC Framework</a></li>
<li class="toctree-l1"><a class="reference internal" href="rpc_param_server_tutorial.html">Implementing a Parameter Server Using Distributed RPC Framework</a></li>
<li class="toctree-l1"><a class="reference internal" href="dist_pipeline_parallel_tutorial.html">Distributed Pipeline Parallelism Using RPC</a></li>
<li class="toctree-l1"><a class="reference internal" href="rpc_async_execution.html">Implementing Batch RPC Processing Using Asynchronous Executions</a></li>
<li class="toctree-l1"><a class="reference internal" href="../advanced/rpc_ddp_tutorial.html">๋ถ์ฐ ๋ฐ์ดํฐ ๋ณ๋ ฌ(DDP)๊ณผ ๋ถ์ฐ RPC ํ๋ ์์ํฌ ๊ฒฐํฉ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../advanced/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="../advanced/generic_join.html">Distributed Training with Uneven Inputs Using the Join Context Manager</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">Edge with ExecuTorch</span></p>
<ul>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/executorch/stable/tutorials/export-to-executorch-tutorial.html">Exporting to ExecuTorch Tutorial</a></li>
<li class="toctree-l1"><a class="reference external" href=" https://pytorch.org/executorch/stable/running-a-model-cpp-tutorial.html">Running an ExecuTorch Model in C++ Tutorial</a></li>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/executorch/stable/tutorials/sdk-integration-tutorial.html">Using the ExecuTorch SDK to Profile a Model</a></li>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/executorch/stable/demo-apps-ios.html">Building an ExecuTorch iOS Demo App</a></li>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/executorch/stable/demo-apps-android.html">Building an ExecuTorch Android Demo App</a></li>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/executorch/stable/examples-end-to-end-to-lower-model-to-delegate.html">Lowering a Model as a Delegate</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">์ถ์ฒ ์์คํ
</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="torchrec_tutorial.html">TorchRec ์๊ฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../advanced/sharding.html">Exploring TorchRec sharding</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">Multimodality</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../beginner/flava_finetuning_tutorial.html">TorchMultimodal ํํ ๋ฆฌ์ผ: FLAVA ๋ฏธ์ธ์กฐ์ </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>(๋ฒ ํ) ์ปดํจํฐ ๋น์ ํํ ๋ฆฌ์ผ์ ์ํ ์์ํ๋ ์ ์ดํ์ต(Quantized Transfer Learning)</li>
<li class="pytorch-breadcrumbs-aside">
<a href="../_sources/intermediate/quantized_transfer_learning_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">intermediate/quantized_transfer_learning_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="quantized-transfer-learning">
<h1>(๋ฒ ํ) ์ปดํจํฐ ๋น์ ํํ ๋ฆฌ์ผ์ ์ํ ์์ํ๋ ์ ์ดํ์ต(Quantized Transfer Learning)<a class="headerlink" href="#quantized-transfer-learning" title="์ด ์ ๋ชฉ์ ๋ํ ํผ๋จธ๋งํฌ">ยถ</a></h1>
<div class="admonition tip">
<p class="admonition-title">ํ</p>
<p>์ด ํํ ๋ฆฌ์ผ์ ์ต๋ํ ํ์ฉํ์๋ ค๋ฉด, ๋ค์์ ๋งํฌ๋ฅผ ์ด์ฉํ์๊ธธ ์ถ์ฒํฉ๋๋ค.
<a class="reference external" href="https://colab.research.google.com/github/pytorch/tutorials/blob/gh-pages/_downloads/quantized_transfer_learning_tutorial.ipynb">Colab ๋ฒ์ </a>.
์ด๋ฅผ ํตํด ์๋์ ์ ์๋ ์ ๋ณด๋ก ์คํ์ ํด ๋ณผ ์ ์์ต๋๋ค.</p>
</div>
<p><strong>Author</strong>: <a class="reference external" href="https://github.com/z-a-f">Zafar Takhirov</a>
<strong>Reviewed by</strong>: <a class="reference external" href="https://github.com/raghuramank100">Raghuraman Krishnamoorthi</a>
<strong>Edited by</strong>: <a class="reference external" href="https://github.com/jlin27">Jessica Lin</a>
<strong>๋ฒ์ญ</strong>: <a class="reference external" href="https://github.com/jjeamin">์ ์ฌ๋ฏผ</a></p>
<p>์ด ํํ ๋ฆฌ์ผ์ <a class="reference external" href="https://chsasank.github.io/">Sasank Chilamkurthy</a> ๊ฐ ์์ฑํ
<a class="reference internal" href="../beginner/transfer_learning_tutorial.html"><span class="doc">์ปดํจํฐ ๋น์ (Vision)์ ์ํ ์ ์ดํ์ต(Transfer Learning)</span></a> ์ ๊ธฐ๋ฐ์ผ๋ก ํฉ๋๋ค.</p>
<p>์ ์ดํ์ต(Transfer learning)์ ๋ค๋ฅธ ๋ฐ์ดํฐ์
์ ์ ์ฉํ๊ธฐ ์ํด์ ๋ฏธ๋ฆฌ ํ์ต๋ ๋ชจ๋ธ์ ์ฌ์ฉํ๋ ๊ธฐ์ ์ ๋งํฉ๋๋ค.
์ ์ดํ์ต์ ์ฌ์ฉํ๋ 2๊ฐ์ง ์ฃผ์ ๋ฐฉ๋ฒ์ด ์์ต๋๋ค.</p>
<ol class="arabic">
<li><p><strong>๊ณ ์ ๋ ํน์ง ์ถ์ถ๊ธฐ๋ก์จ ConvNet</strong>: ์ฌ๊ธฐ์๋ ๋ง์ง๋ง ๋ช๊ฐ์ ๊ณ์ธต(์ผ๋ช
โํค๋(the head)โ, ์ผ๋ฐ์ ์ผ๋ก ์์ ํ ์ฐ๊ฒฐ๋ ๊ณ์ธต)
์ ์ ์ธํ๊ณ ๋คํธ์ํฌ์ ๋ชจ๋ ๋งค๊ฐ ๋ณ์ ๊ฐ์ค์น๋ฅผ <a class="reference external" href="https://arxiv.org/abs/1706.04983">โ๊ณ ์ (freeze)โ</a> ํฉ๋๋ค.
๋ง์ง๋ง ๊ณ์ธต์ ์์์ ๊ฐ์ค์น๋ก ์ด๊ธฐํ๋ ์๋ก์ด ๊ณ์ธต์ผ๋ก ๋์ฒด๋๋ฉฐ ์ค์ง ์ด ๊ณ์ธต๋ง ํ์ต๋ฉ๋๋ค.</p></li>
<li><p><strong>ConvNet ๋ฏธ์ธ์กฐ์ (Finetuning)</strong>: ๋๋ค ์ด๊ธฐํ ๋์ , ๋ฏธ๋ฆฌ ํ์ต๋ ๋คํธ์ํฌ๋ฅผ ์ด์ฉํ์ฌ ๋ชจ๋ธ์ ์ด๊ธฐํํฉ๋๋ค.
์ดํ ํ์์ฒ๋ผ ํ์ต์ ์งํํ์ง๋ง ๋ค๋ฅธ ๋ฐ์ดํฐ์
์ ์ฌ์ฉํฉ๋๋ค.
ํ์์ฒ๋ผ ํ์ต์ด ์งํ๋์ง๋ง ๋ค๋ฅธ ๋ฐ์ดํฐ์
์ ์ฌ์ฉํฉ๋๋ค.</p>
<p>์ถ๋ ฅ์ ์๊ฐ ๋ค๋ฅผ ์ ์๊ธฐ ๋๋ฌธ์, ์ผ๋ฐ์ ์ผ๋ก ์ ๊ฒฝ๋ง์์ ํค๋(๋๋ ๊ทธ ์ผ๋ถ)๋ ๊ต์ฒด๋ฉ๋๋ค.
์ด ๋ฐฉ๋ฒ์์๋ ํ์ต๋ฅ ์ ๋ ์์ ์๋ก ์ค์ ํ๋ ๊ฒ์ด ์ผ๋ฐ์ ์
๋๋ค.
์ด๋ ๋คํธ์ํฌ๊ฐ ์ด๋ฏธ ํ์ต๋์๊ธฐ ๋๋ฌธ์ด๋ฉฐ ์๋ก์ด ๋ฐ์ดํฐ์
์ผ๋ก ใ๋ฏธ์ธ์กฐ์ (finetuning)ใํ๋ ค๋ฉด ์ฝ๊ฐ์ ๋ณ๊ฒฝ๋ง์ด ํ์ํฉ๋๋ค.</p>
</li>
</ol>
<p>๋ํ ์์ ๋ ๋ฐฉ๋ฒ์ ๊ฒฐํฉํ ์๋ ์์ต๋๋ค.
๋จผ์ ํน์ง ์ถ์ถ๊ธฐ๋ฅผ ๊ณ ์ (freeze)ํ๊ณ ํค๋(the head)๋ฅผ ํ์ต์ํต๋๋ค.
๊ทธ๋ฐ ๋ค์ ํน์ง ์ถ์ถ๊ธฐ(๋๋ ๊ทธ ์ผ๋ถ)๋ฅผ ๊ณ ์ ํด์ (unfreeze)ํ๊ณ ํ์ต๋ฅ ์
๋ ์์ ์๋ก ์ค์ ํ ๋ค์ ํ์ต์ ๊ณ์ํ ์ ์์ต๋๋ค.</p>
<p>์ด๋ฒ ํํธ์์๋ ์ฒซ๋ฒ์งธ ๋ฐฉ๋ฒ์ ์ฌ์ฉํด ์์ํ๋ ๋ชจ๋ธ๋ก ํน์ง์ ์ถ์ถํด๋ด
์๋ค.</p>
<div class="section" id="id2">
<h2>ํํธ 0. ์๊ตฌ์ฌํญ<a class="headerlink" href="#id2" title="์ด ์ ๋ชฉ์ ๋ํ ํผ๋จธ๋งํฌ">ยถ</a></h2>
<p>์ ์ด ํ์ต(transfer learning)์ ์์ํ๊ธฐ ์ ์,
๋ฐ์ดํฐ ๋ถ๋ฌ์ค๊ธฐ / ์๊ฐํ์ ๊ฐ์ ใ์๊ตฌ์ฌํญ(prerequisites)ใ์ ๊ฒํ ํ๊ฒ ์ต๋๋ค.</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="c1"># Imports</span>
<span class="kn">import</span> <span class="nn">copy</span>
<span class="kn">import</span> <span class="nn">matplotlib.pyplot</span> <span class="k">as</span> <span class="nn">plt</span>
<span class="kn">import</span> <span class="nn">numpy</span> <span class="k">as</span> <span class="nn">np</span>
<span class="kn">import</span> <span class="nn">os</span>
<span class="kn">import</span> <span class="nn">time</span>
<span class="n">plt</span><span class="o">.</span><span class="n">ion</span><span class="p">()</span>
</pre></div>
</div>
<div class="section" id="nightly-build">
<h3>Nightly Build ์ค์นํ๊ธฐ<a class="headerlink" href="#nightly-build" title="์ด ์ ๋ชฉ์ ๋ํ ํผ๋จธ๋งํฌ">ยถ</a></h3>
<p>PyTorch์ ๋ฒ ํ(beta)๋ฅผ ์ฌ์ฉํ ๊ฒ์ด๋ฏ๋ก ์ต์ ๋ฒ์ ์ <code class="docutils literal notranslate"><span class="pre">torch</span></code> ์ <code class="docutils literal notranslate"><span class="pre">torchvision</span></code> ์ ์ค์นํ๋ ๊ฒ์ ๊ถ์ฅํฉ๋๋ค.
๋ก์ปฌ(local) ์ค์น์ ๋ํ ์ต์ ์ง์นจ์ <a class="reference external" href="https://pytorch.org/get-started/locally/">์ฌ๊ธฐ</a> ์์ ์ฐพ์ ์ ์์ต๋๋ค.
์๋ฅผ ๋ค์ด GPU ์ง์ ์์ด ์ค์นํ๋ ค๋ฉด :</p>
<div class="highlight-shell notranslate"><div class="highlight"><pre><span></span>pip<span class="w"> </span>install<span class="w"> </span>numpy
pip<span class="w"> </span>install<span class="w"> </span>--pre<span class="w"> </span>torch<span class="w"> </span>torchvision<span class="w"> </span>-f<span class="w"> </span>https://download.pytorch.org/whl/nightly/cpu/torch_nightly.html
<span class="c1"># CUDA ์ง์์ https://download.pytorch.org/whl/nightly/cu101/torch_nightly.html๋ฅผ ์ฌ์ฉํ์ธ์.</span>
</pre></div>
</div>
</div>
<div class="section" id="id4">
<h3>๋ฐ์ดํฐ ๋ถ๋ฌ์ค๊ธฐ<a class="headerlink" href="#id4" title="์ด ์ ๋ชฉ์ ๋ํ ํผ๋จธ๋งํฌ">ยถ</a></h3>
<div class="admonition note">
<p class="admonition-title">์ฐธ๊ณ </p>
<p>์ด๋ฒ ์น์
์ ์๋ณธ ์ ์ดํ์ต(Transfer Learning) ํํ ๋ฆฌ์ผ๊ณผ ๋์ผํฉ๋๋ค.</p>
</div>
<p><code class="docutils literal notranslate"><span class="pre">torchvision</span></code> ๊ณผ <code class="docutils literal notranslate"><span class="pre">torch.utils.data</span></code> ํจํค์ง๋ฅผ ์ฌ์ฉํ์ฌ ๋ฐ์ดํฐ๋ฅผ ๋ถ๋ฌ์ต๋๋ค.</p>
<p>์ฌ๊ธฐ์ ํ๊ณ ์ ํ๋ ๋ฌธ์ ๋ ์ด๋ฏธ์ง๋ก๋ถํฐ <strong>๊ฐ๋ฏธ</strong> ์ <strong>๋ฒ</strong> ์ ๋ถ๋ฅํ๋ ๊ฒ์
๋๋ค.
์ด ๋ฐ์ดํฐ์
์ ๊ฐ๋ฏธ์ ๋ฒ์ ๋ํด ๊ฐ๊ฐ 120์ฅ์ ํ์ต์ฉ ์ด๋ฏธ์ง, 75๊ฐ์ ๊ฒ์ฆ์ฉ ์ด๋ฏธ์ง๋ฅผ ํฌํจํฉ๋๋ค.
์ด๋ ์ผ๋ฐํํ๊ธฐ์๋ ์์ฃผ ์์ ๋ฐ์ดํฐ์
์
๋๋ค.
ํ์ง๋ง ์ฐ๋ฆฌ๋ ์ ์ดํ์ต(Transfer Learning)์ ์ฌ์ฉํ๊ธฐ ๋๋ฌธ์, ์ผ๋ฐํ๋ฅผ ๊ฝค ์ ํ ์ ์์ ๊ฒ์
๋๋ค.</p>
<p>์ด ๋ฐ์ดํฐ์
์ imagenet์ ์์ฃผ ์์ ์ผ๋ถ์
๋๋ค.</p>
<div class="admonition note">
<p class="admonition-title">์ฐธ๊ณ </p>
<p><a class="reference external" href="https://download.pytorch.org/tutorial/hymenoptera_data.zip">์ฌ๊ธฐ</a> ์์ ๋ฐ์ดํฐ๋ฅผ ๋ค์ด๋ก๋ ๋ฐ์ <code class="docutils literal notranslate"><span class="pre">data</span></code> ๋๋ ํ ๋ฆฌ์ ์์ถ์ ํธ์ธ์.</p>
</div>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="kn">import</span> <span class="nn">torch</span>
<span class="kn">from</span> <span class="nn">torchvision</span> <span class="kn">import</span> <span class="n">transforms</span><span class="p">,</span> <span class="n">datasets</span>
<span class="c1"># ํ์ต์ ์ํ ๋ฐ์ดํฐ ๋ณด๊ฐ(Data augmentation)๊ณผ ์ ๊ทํ</span>
<span class="c1"># ๊ฒ์ฆ์ ์ํ ์ ๊ทํ</span>
<span class="n">data_transforms</span> <span class="o">=</span> <span class="p">{</span>
<span class="s1">'train'</span><span class="p">:</span> <span class="n">transforms</span><span class="o">.</span><span class="n">Compose</span><span class="p">([</span>
<span class="n">transforms</span><span class="o">.</span><span class="n">Resize</span><span class="p">(</span><span class="mi">224</span><span class="p">),</span>
<span class="n">transforms</span><span class="o">.</span><span class="n">RandomCrop</span><span class="p">(</span><span class="mi">224</span><span class="p">),</span>
<span class="n">transforms</span><span class="o">.</span><span class="n">RandomHorizontalFlip</span><span class="p">(),</span>
<span class="n">transforms</span><span class="o">.</span><span class="n">ToTensor</span><span class="p">(),</span>
<span class="n">transforms</span><span class="o">.</span><span class="n">Normalize</span><span class="p">([</span><span class="mf">0.485</span><span class="p">,</span> <span class="mf">0.456</span><span class="p">,</span> <span class="mf">0.406</span><span class="p">],</span> <span class="p">[</span><span class="mf">0.229</span><span class="p">,</span> <span class="mf">0.224</span><span class="p">,</span> <span class="mf">0.225</span><span class="p">])</span>
<span class="p">]),</span>
<span class="s1">'val'</span><span class="p">:</span> <span class="n">transforms</span><span class="o">.</span><span class="n">Compose</span><span class="p">([</span>
<span class="n">transforms</span><span class="o">.</span><span class="n">Resize</span><span class="p">(</span><span class="mi">224</span><span class="p">),</span>
<span class="n">transforms</span><span class="o">.</span><span class="n">CenterCrop</span><span class="p">(</span><span class="mi">224</span><span class="p">),</span>
<span class="n">transforms</span><span class="o">.</span><span class="n">ToTensor</span><span class="p">(),</span>
<span class="n">transforms</span><span class="o">.</span><span class="n">Normalize</span><span class="p">([</span><span class="mf">0.485</span><span class="p">,</span> <span class="mf">0.456</span><span class="p">,</span> <span class="mf">0.406</span><span class="p">],</span> <span class="p">[</span><span class="mf">0.229</span><span class="p">,</span> <span class="mf">0.224</span><span class="p">,</span> <span class="mf">0.225</span><span class="p">])</span>
<span class="p">]),</span>
<span class="p">}</span>
<span class="n">data_dir</span> <span class="o">=</span> <span class="s1">'data/hymenoptera_data'</span>
<span class="n">image_datasets</span> <span class="o">=</span> <span class="p">{</span><span class="n">x</span><span class="p">:</span> <span class="n">datasets</span><span class="o">.</span><span class="n">ImageFolder</span><span class="p">(</span><span class="n">os</span><span class="o">.</span><span class="n">path</span><span class="o">.</span><span class="n">join</span><span class="p">(</span><span class="n">data_dir</span><span class="p">,</span> <span class="n">x</span><span class="p">),</span>
<span class="n">data_transforms</span><span class="p">[</span><span class="n">x</span><span class="p">])</span>
<span class="k">for</span> <span class="n">x</span> <span class="ow">in</span> <span class="p">[</span><span class="s1">'train'</span><span class="p">,</span> <span class="s1">'val'</span><span class="p">]}</span>
<span class="n">dataloaders</span> <span class="o">=</span> <span class="p">{</span><span class="n">x</span><span class="p">:</span> <span class="n">torch</span><span class="o">.</span><span class="n">utils</span><span class="o">.</span><span class="n">data</span><span class="o">.</span><span class="n">DataLoader</span><span class="p">(</span><span class="n">image_datasets</span><span class="p">[</span><span class="n">x</span><span class="p">],</span> <span class="n">batch_size</span><span class="o">=</span><span class="mi">16</span><span class="p">,</span>
<span class="n">shuffle</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span> <span class="n">num_workers</span><span class="o">=</span><span class="mi">8</span><span class="p">)</span>
<span class="k">for</span> <span class="n">x</span> <span class="ow">in</span> <span class="p">[</span><span class="s1">'train'</span><span class="p">,</span> <span class="s1">'val'</span><span class="p">]}</span>
<span class="n">dataset_sizes</span> <span class="o">=</span> <span class="p">{</span><span class="n">x</span><span class="p">:</span> <span class="nb">len</span><span class="p">(</span><span class="n">image_datasets</span><span class="p">[</span><span class="n">x</span><span class="p">])</span> <span class="k">for</span> <span class="n">x</span> <span class="ow">in</span> <span class="p">[</span><span class="s1">'train'</span><span class="p">,</span> <span class="s1">'val'</span><span class="p">]}</span>
<span class="n">class_names</span> <span class="o">=</span> <span class="n">image_datasets</span><span class="p">[</span><span class="s1">'train'</span><span class="p">]</span><span class="o">.</span><span class="n">classes</span>
<span class="n">device</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">device</span><span class="p">(</span><span class="s2">"cuda:0"</span> <span class="k">if</span> <span class="n">torch</span><span class="o">.</span><span class="n">cuda</span><span class="o">.</span><span class="n">is_available</span><span class="p">()</span> <span class="k">else</span> <span class="s2">"cpu"</span><span class="p">)</span>
</pre></div>
</div>
</div>
<div class="section" id="id6">
<h3>์ผ๋ถ ์ด๋ฏธ์ง ์๊ฐํํ๊ธฐ<a class="headerlink" href="#id6" title="์ด ์ ๋ชฉ์ ๋ํ ํผ๋จธ๋งํฌ">ยถ</a></h3>
<p>๋ฐ์ดํฐ ๋ณด๊ฐ์ ์ดํดํ๊ธฐ ์ํด ์ผ๋ถ ํ์ต์ฉ ์ด๋ฏธ์ง๋ฅผ ์๊ฐํ ํด๋ณด๊ฒ ์ต๋๋ค.</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="kn">import</span> <span class="nn">torchvision</span>
<span class="k">def</span> <span class="nf">imshow</span><span class="p">(</span><span class="n">inp</span><span class="p">,</span> <span class="n">title</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span> <span class="n">ax</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span> <span class="n">figsize</span><span class="o">=</span><span class="p">(</span><span class="mi">5</span><span class="p">,</span> <span class="mi">5</span><span class="p">)):</span>
<span class="w"> </span><span class="sd">"""Imshow for Tensor."""</span>
<span class="n">inp</span> <span class="o">=</span> <span class="n">inp</span><span class="o">.</span><span class="n">numpy</span><span class="p">()</span><span class="o">.</span><span class="n">transpose</span><span class="p">((</span><span class="mi">1</span><span class="p">,</span> <span class="mi">2</span><span class="p">,</span> <span class="mi">0</span><span class="p">))</span>
<span class="n">mean</span> <span class="o">=</span> <span class="n">np</span><span class="o">.</span><span class="n">array</span><span class="p">([</span><span class="mf">0.485</span><span class="p">,</span> <span class="mf">0.456</span><span class="p">,</span> <span class="mf">0.406</span><span class="p">])</span>
<span class="n">std</span> <span class="o">=</span> <span class="n">np</span><span class="o">.</span><span class="n">array</span><span class="p">([</span><span class="mf">0.229</span><span class="p">,</span> <span class="mf">0.224</span><span class="p">,</span> <span class="mf">0.225</span><span class="p">])</span>
<span class="n">inp</span> <span class="o">=</span> <span class="n">std</span> <span class="o">*</span> <span class="n">inp</span> <span class="o">+</span> <span class="n">mean</span>
<span class="n">inp</span> <span class="o">=</span> <span class="n">np</span><span class="o">.</span><span class="n">clip</span><span class="p">(</span><span class="n">inp</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="k">if</span> <span class="n">ax</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
<span class="n">fig</span><span class="p">,</span> <span class="n">ax</span> <span class="o">=</span> <span class="n">plt</span><span class="o">.</span><span class="n">subplots</span><span class="p">(</span><span class="mi">1</span><span class="p">,</span> <span class="n">figsize</span><span class="o">=</span><span class="n">figsize</span><span class="p">)</span>
<span class="n">ax</span><span class="o">.</span><span class="n">imshow</span><span class="p">(</span><span class="n">inp</span><span class="p">)</span>
<span class="n">ax</span><span class="o">.</span><span class="n">set_xticks</span><span class="p">([])</span>
<span class="n">ax</span><span class="o">.</span><span class="n">set_yticks</span><span class="p">([])</span>
<span class="k">if</span> <span class="n">title</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
<span class="n">ax</span><span class="o">.</span><span class="n">set_title</span><span class="p">(</span><span class="n">title</span><span class="p">)</span>
<span class="c1"># ํ์ต ๋ฐ์ดํฐ์ ๋ฐฐ์น๋ฅผ ์ป์ต๋๋ค.</span>
<span class="n">inputs</span><span class="p">,</span> <span class="n">classes</span> <span class="o">=</span> <span class="nb">next</span><span class="p">(</span><span class="nb">iter</span><span class="p">(</span><span class="n">dataloaders</span><span class="p">[</span><span class="s1">'train'</span><span class="p">]))</span>
<span class="c1"># ๋ฐฐ์น๋ก๋ถํฐ ๊ฒฉ์ ํํ์ ์ด๋ฏธ์ง๋ฅผ ๋ง๋ญ๋๋ค.</span>
<span class="n">out</span> <span class="o">=</span> <span class="n">torchvision</span><span class="o">.</span><span class="n">utils</span><span class="o">.</span><span class="n">make_grid</span><span class="p">(</span><span class="n">inputs</span><span class="p">,</span> <span class="n">nrow</span><span class="o">=</span><span class="mi">4</span><span class="p">)</span>
<span class="n">fig</span><span class="p">,</span> <span class="n">ax</span> <span class="o">=</span> <span class="n">plt</span><span class="o">.</span><span class="n">subplots</span><span class="p">(</span><span class="mi">1</span><span class="p">,</span> <span class="n">figsize</span><span class="o">=</span><span class="p">(</span><span class="mi">10</span><span class="p">,</span> <span class="mi">10</span><span class="p">))</span>
<span class="n">imshow</span><span class="p">(</span><span class="n">out</span><span class="p">,</span> <span class="n">title</span><span class="o">=</span><span class="p">[</span><span class="n">class_names</span><span class="p">[</span><span class="n">x</span><span class="p">]</span> <span class="k">for</span> <span class="n">x</span> <span class="ow">in</span> <span class="n">classes</span><span class="p">],</span> <span class="n">ax</span><span class="o">=</span><span class="n">ax</span><span class="p">)</span>
</pre></div>
</div>
</div>
<div class="section" id="id7">
<h3>๋ชจ๋ธ ํ์ต์ ์ํ ์ง์ ํจ์<a class="headerlink" href="#id7" title="์ด ์ ๋ชฉ์ ๋ํ ํผ๋จธ๋งํฌ">ยถ</a></h3>
<p>๋ค์์ ๋ชจ๋ธ์ ํ์ตํ๊ธฐ ์ํ ์ผ๋ฐ ํจ์ ์
๋๋ค.</p>
<ul class="simple">
<li><p>ํ์ต๋ฅ (learning rate)์ ๊ด๋ฆฌํฉ๋๋ค(schedules).</p></li>
<li><p>์ต์ ์ ๋ชจ๋ธ์ ์ ์ฅํฉ๋๋ค.</p></li>
</ul>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="k">def</span> <span class="nf">train_model</span><span class="p">(</span><span class="n">model</span><span class="p">,</span> <span class="n">criterion</span><span class="p">,</span> <span class="n">optimizer</span><span class="p">,</span> <span class="n">scheduler</span><span class="p">,</span> <span class="n">num_epochs</span><span class="o">=</span><span class="mi">25</span><span class="p">,</span> <span class="n">device</span><span class="o">=</span><span class="s1">'cpu'</span><span class="p">):</span>
<span class="w"> </span><span class="sd">"""</span>
<span class="sd"> Support function for model training.</span>
<span class="sd"> ๋ชจ๋ธ ํ์ต์ ์ํ ์ง์ ํจ์</span>
<span class="sd"> ๋งค๊ฐ๋ณ์:</span>
<span class="sd"> model: ํ์ตํ ๋ชจ๋ธ</span>
<span class="sd"> criterion: ์ต์ ํ ๊ธฐ์ค(์์ค)</span>
<span class="sd"> optimizer: ํ์ต์ ์ฌ์ฉํ ์ตํฐ๋ง์ด์ </span>
<span class="sd"> scheduler: ``torch.optim.lr_scheduler`` ์ ์ธ์คํด์ค</span>
<span class="sd"> num_epochs: ์ํญ์ ์</span>
<span class="sd"> device: ํ์ต์ ๋์์ํฌ ์ฅ์น. 'cpu' ๋๋ 'cuda'์ฌ์ผ ํฉ๋๋ค.</span>
<span class="sd"> """</span>
<span class="n">since</span> <span class="o">=</span> <span class="n">time</span><span class="o">.</span><span class="n">time</span><span class="p">()</span>
<span class="n">best_model_wts</span> <span class="o">=</span> <span class="n">copy</span><span class="o">.</span><span class="n">deepcopy</span><span class="p">(</span><span class="n">model</span><span class="o">.</span><span class="n">state_dict</span><span class="p">())</span>
<span class="n">best_acc</span> <span class="o">=</span> <span class="mf">0.0</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="n">num_epochs</span><span class="p">):</span>
<span class="nb">print</span><span class="p">(</span><span class="s1">'Epoch </span><span class="si">{}</span><span class="s1">/</span><span class="si">{}</span><span class="s1">'</span><span class="o">.</span><span class="n">format</span><span class="p">(</span><span class="n">epoch</span><span class="p">,</span> <span class="n">num_epochs</span> <span class="o">-</span> <span class="mi">1</span><span class="p">))</span>
<span class="nb">print</span><span class="p">(</span><span class="s1">'-'</span> <span class="o">*</span> <span class="mi">10</span><span class="p">)</span>
<span class="c1"># ๊ฐ ์ํญ์๋ ํ์ต ๋ฐ ๊ฒ์ฆ ๋จ๊ณ๊ฐ ์์ต๋๋ค.</span>
<span class="k">for</span> <span class="n">phase</span> <span class="ow">in</span> <span class="p">[</span><span class="s1">'train'</span><span class="p">,</span> <span class="s1">'val'</span><span class="p">]:</span>
<span class="k">if</span> <span class="n">phase</span> <span class="o">==</span> <span class="s1">'train'</span><span class="p">:</span>
<span class="n">model</span><span class="o">.</span><span class="n">train</span><span class="p">()</span> <span class="c1"># ๋ชจ๋ธ์ ํ์ต ๋ชจ๋๋ก ์ค์ ํ๊ธฐ</span>
<span class="k">else</span><span class="p">:</span>
<span class="n">model</span><span class="o">.</span><span class="n">eval</span><span class="p">()</span> <span class="c1"># ๋ชจ๋ธ์ ํ๊ฐ ๋ชจ๋๋ก ์ค์ ํ๊ธฐ</span>
<span class="n">running_loss</span> <span class="o">=</span> <span class="mf">0.0</span>
<span class="n">running_corrects</span> <span class="o">=</span> <span class="mi">0</span>
<span class="c1"># ๋ฐ์ดํฐ ๋ฐ๋ณตํ๊ธฐ</span>
<span class="k">for</span> <span class="n">inputs</span><span class="p">,</span> <span class="n">labels</span> <span class="ow">in</span> <span class="n">dataloaders</span><span class="p">[</span><span class="n">phase</span><span class="p">]:</span>
<span class="n">inputs</span> <span class="o">=</span> <span class="n">inputs</span><span class="o">.</span><span class="n">to</span><span class="p">(</span><span class="n">device</span><span class="p">)</span>
<span class="n">labels</span> <span class="o">=</span> <span class="n">labels</span><span class="o">.</span><span class="n">to</span><span class="p">(</span><span class="n">device</span><span class="p">)</span>
<span class="c1"># ๋งค๊ฐ ๋ณ์ ๊ธฐ์ธ๊ธฐ๋ฅผ 0์ผ๋ก ์ค์ ํ๊ธฐ</span>
<span class="n">optimizer</span><span class="o">.</span><span class="n">zero_grad</span><span class="p">()</span>
<span class="c1"># ์์ ํ</span>
<span class="c1"># ํ์ต ๋์๋ง ์ฐ์ฐ ๊ธฐ๋ก์ ์ถ์ ํ๊ธฐ</span>
<span class="k">with</span> <span class="n">torch</span><span class="o">.</span><span class="n">set_grad_enabled</span><span class="p">(</span><span class="n">phase</span> <span class="o">==</span> <span class="s1">'train'</span><span class="p">):</span>
<span class="n">outputs</span> <span class="o">=</span> <span class="n">model</span><span class="p">(</span><span class="n">inputs</span><span class="p">)</span>
<span class="n">_</span><span class="p">,</span> <span class="n">preds</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">max</span><span class="p">(</span><span class="n">outputs</span><span class="p">,</span> <span class="mi">1</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">outputs</span><span class="p">,</span> <span class="n">labels</span><span class="p">)</span>
<span class="c1"># ์ญ์ ํ + ํ์ต ๋จ๊ณ์์๋ง ์ต์ ํ</span>
<span class="k">if</span> <span class="n">phase</span> <span class="o">==</span> <span class="s1">'train'</span><span class="p">:</span>
<span class="n">loss</span><span class="o">.</span><span class="n">backward</span><span class="p">()</span>
<span class="n">optimizer</span><span class="o">.</span><span class="n">step</span><span class="p">()</span>
<span class="c1"># ํต๊ณ ๋ณด๊ธฐ</span>
<span class="n">running_loss</span> <span class="o">+=</span> <span class="n">loss</span><span class="o">.</span><span class="n">item</span><span class="p">()</span> <span class="o">*</span> <span class="n">inputs</span><span class="o">.</span><span class="n">size</span><span class="p">(</span><span class="mi">0</span><span class="p">)</span>
<span class="n">running_corrects</span> <span class="o">+=</span> <span class="n">torch</span><span class="o">.</span><span class="n">sum</span><span class="p">(</span><span class="n">preds</span> <span class="o">==</span> <span class="n">labels</span><span class="o">.</span><span class="n">data</span><span class="p">)</span>
<span class="k">if</span> <span class="n">phase</span> <span class="o">==</span> <span class="s1">'train'</span><span class="p">:</span>
<span class="n">scheduler</span><span class="o">.</span><span class="n">step</span><span class="p">()</span>
<span class="n">epoch_loss</span> <span class="o">=</span> <span class="n">running_loss</span> <span class="o">/</span> <span class="n">dataset_sizes</span><span class="p">[</span><span class="n">phase</span><span class="p">]</span>
<span class="n">epoch_acc</span> <span class="o">=</span> <span class="n">running_corrects</span><span class="o">.</span><span class="n">double</span><span class="p">()</span> <span class="o">/</span> <span class="n">dataset_sizes</span><span class="p">[</span><span class="n">phase</span><span class="p">]</span>
<span class="nb">print</span><span class="p">(</span><span class="s1">'</span><span class="si">{}</span><span class="s1"> Loss: </span><span class="si">{:.4f}</span><span class="s1"> Acc: </span><span class="si">{:.4f}</span><span class="s1">'</span><span class="o">.</span><span class="n">format</span><span class="p">(</span>
<span class="n">phase</span><span class="p">,</span> <span class="n">epoch_loss</span><span class="p">,</span> <span class="n">epoch_acc</span><span class="p">))</span>
<span class="c1"># ๋ชจ๋ธ ๋ณต์ฌํ๊ธฐ</span>
<span class="k">if</span> <span class="n">phase</span> <span class="o">==</span> <span class="s1">'val'</span> <span class="ow">and</span> <span class="n">epoch_acc</span> <span class="o">></span> <span class="n">best_acc</span><span class="p">:</span>
<span class="n">best_acc</span> <span class="o">=</span> <span class="n">epoch_acc</span>
<span class="n">best_model_wts</span> <span class="o">=</span> <span class="n">copy</span><span class="o">.</span><span class="n">deepcopy</span><span class="p">(</span><span class="n">model</span><span class="o">.</span><span class="n">state_dict</span><span class="p">())</span>
<span class="nb">print</span><span class="p">()</span>
<span class="n">time_elapsed</span> <span class="o">=</span> <span class="n">time</span><span class="o">.</span><span class="n">time</span><span class="p">()</span> <span class="o">-</span> <span class="n">since</span>
<span class="nb">print</span><span class="p">(</span><span class="s1">'Training complete in </span><span class="si">{:.0f}</span><span class="s1">m </span><span class="si">{:.0f}</span><span class="s1">s'</span><span class="o">.</span><span class="n">format</span><span class="p">(</span>
<span class="n">time_elapsed</span> <span class="o">//</span> <span class="mi">60</span><span class="p">,</span> <span class="n">time_elapsed</span> <span class="o">%</span> <span class="mi">60</span><span class="p">))</span>
<span class="nb">print</span><span class="p">(</span><span class="s1">'Best val Acc: </span><span class="si">{:4f}</span><span class="s1">'</span><span class="o">.</span><span class="n">format</span><span class="p">(</span><span class="n">best_acc</span><span class="p">))</span>
<span class="c1"># ์ต์ ์ ๋ชจ๋ธ ๊ฐ์ค์น ๋ถ๋ฌ์ค๊ธฐ</span>
<span class="n">model</span><span class="o">.</span><span class="n">load_state_dict</span><span class="p">(</span><span class="n">best_model_wts</span><span class="p">)</span>
<span class="k">return</span> <span class="n">model</span>
</pre></div>
</div>
</div>
<div class="section" id="id8">
<h3>๋ชจ๋ธ ์์ธก์ ์๊ฐํํ๊ธฐ ์ํ ์ง์ ํจ์<a class="headerlink" href="#id8" title="์ด ์ ๋ชฉ์ ๋ํ ํผ๋จธ๋งํฌ">ยถ</a></h3>
<p>์ผ๋ถ ์ด๋ฏธ์ง์ ๋ํ ์์ธก์ ์ถ๋ ฅํ๋ ์ผ๋ฐ ํจ์</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="k">def</span> <span class="nf">visualize_model</span><span class="p">(</span><span class="n">model</span><span class="p">,</span> <span class="n">rows</span><span class="o">=</span><span class="mi">3</span><span class="p">,</span> <span class="n">cols</span><span class="o">=</span><span class="mi">3</span><span class="p">):</span>
<span class="n">was_training</span> <span class="o">=</span> <span class="n">model</span><span class="o">.</span><span class="n">training</span>
<span class="n">model</span><span class="o">.</span><span class="n">eval</span><span class="p">()</span>
<span class="n">current_row</span> <span class="o">=</span> <span class="n">current_col</span> <span class="o">=</span> <span class="mi">0</span>
<span class="n">fig</span><span class="p">,</span> <span class="n">ax</span> <span class="o">=</span> <span class="n">plt</span><span class="o">.</span><span class="n">subplots</span><span class="p">(</span><span class="n">rows</span><span class="p">,</span> <span class="n">cols</span><span class="p">,</span> <span class="n">figsize</span><span class="o">=</span><span class="p">(</span><span class="n">cols</span><span class="o">*</span><span class="mi">2</span><span class="p">,</span> <span class="n">rows</span><span class="o">*</span><span class="mi">2</span><span class="p">))</span>
<span class="k">with</span> <span class="n">torch</span><span class="o">.</span><span class="n">no_grad</span><span class="p">():</span>
<span class="k">for</span> <span class="n">idx</span><span class="p">,</span> <span class="p">(</span><span class="n">imgs</span><span class="p">,</span> <span class="n">lbls</span><span class="p">)</span> <span class="ow">in</span> <span class="nb">enumerate</span><span class="p">(</span><span class="n">dataloaders</span><span class="p">[</span><span class="s1">'val'</span><span class="p">]):</span>
<span class="n">imgs</span> <span class="o">=</span> <span class="n">imgs</span><span class="o">.</span><span class="n">cpu</span><span class="p">()</span>
<span class="n">lbls</span> <span class="o">=</span> <span class="n">lbls</span><span class="o">.</span><span class="n">cpu</span><span class="p">()</span>
<span class="n">outputs</span> <span class="o">=</span> <span class="n">model</span><span class="p">(</span><span class="n">imgs</span><span class="p">)</span>
<span class="n">_</span><span class="p">,</span> <span class="n">preds</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">max</span><span class="p">(</span><span class="n">outputs</span><span class="p">,</span> <span class="mi">1</span><span class="p">)</span>
<span class="k">for</span> <span class="n">jdx</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="n">imgs</span><span class="o">.</span><span class="n">size</span><span class="p">()[</span><span class="mi">0</span><span class="p">]):</span>
<span class="n">imshow</span><span class="p">(</span><span class="n">imgs</span><span class="o">.</span><span class="n">data</span><span class="p">[</span><span class="n">jdx</span><span class="p">],</span> <span class="n">ax</span><span class="o">=</span><span class="n">ax</span><span class="p">[</span><span class="n">current_row</span><span class="p">,</span> <span class="n">current_col</span><span class="p">])</span>
<span class="n">ax</span><span class="p">[</span><span class="n">current_row</span><span class="p">,</span> <span class="n">current_col</span><span class="p">]</span><span class="o">.</span><span class="n">axis</span><span class="p">(</span><span class="s1">'off'</span><span class="p">)</span>
<span class="n">ax</span><span class="p">[</span><span class="n">current_row</span><span class="p">,</span> <span class="n">current_col</span><span class="p">]</span><span class="o">.</span><span class="n">set_title</span><span class="p">(</span><span class="s1">'predicted: </span><span class="si">{}</span><span class="s1">'</span><span class="o">.</span><span class="n">format</span><span class="p">(</span><span class="n">class_names</span><span class="p">[</span><span class="n">preds</span><span class="p">[</span><span class="n">jdx</span><span class="p">]]))</span>
<span class="n">current_col</span> <span class="o">+=</span> <span class="mi">1</span>
<span class="k">if</span> <span class="n">current_col</span> <span class="o">>=</span> <span class="n">cols</span><span class="p">:</span>
<span class="n">current_row</span> <span class="o">+=</span> <span class="mi">1</span>
<span class="n">current_col</span> <span class="o">=</span> <span class="mi">0</span>
<span class="k">if</span> <span class="n">current_row</span> <span class="o">>=</span> <span class="n">rows</span><span class="p">:</span>
<span class="n">model</span><span class="o">.</span><span class="n">train</span><span class="p">(</span><span class="n">mode</span><span class="o">=</span><span class="n">was_training</span><span class="p">)</span>
<span class="k">return</span>
<span class="n">model</span><span class="o">.</span><span class="n">train</span><span class="p">(</span><span class="n">mode</span><span class="o">=</span><span class="n">was_training</span><span class="p">)</span>
</pre></div>
</div>
</div>
</div>
<div class="section" id="quantized-feature-extractor">
<h2>ํํธ 1. ์์ํ๋ ํน์ง ์ถ์ถ๊ธฐ(Quantized Feature Extractor)๋ฅผ ๊ธฐ๋ฐ์ผ๋ก ์ฌ์ฉ์ ์ง์ ๋ถ๋ฅ๊ธฐ ํ๋ จํ๊ธฐ<a class="headerlink" href="#quantized-feature-extractor" title="์ด ์ ๋ชฉ์ ๋ํ ํผ๋จธ๋งํฌ">ยถ</a></h2>
<p>์ด๋ฒ ์น์
์์๋ โ๊ณ ์ ๋(frozen)โ ์์ํ ํน์ง ์ถ์ถ๊ธฐ๋ฅผ ์ฌ์ฉํ๊ณ ๊ทธ ์์ ์ฌ์ฉ์ ์ง์ ๋ถ๋ฅ๊ธฐ ํค๋๋ฅผ
ํ์ตํฉ๋๋ค. ๋ถ๋ ์์์ ๋ชจ๋ธ๊ณผ ๋ค๋ฅด๊ฒ ์์ํ๋ ๋ชจ๋ธ์๋ ํ์ต ๊ฐ๋ฅํ ๋งค๊ฐ ๋ณ์๊ฐ ์์ผ๋ฏ๋ก
requires_grad = False๋ฅผ ์ค์ ํ ํ์๊ฐ ์์ต๋๋ค. ์์ธํ ๋ด์ฉ์ <a class="reference external" href="https://pytorch.org/docs/stable/quantization.html">์ค๋ช
์</a> ๋ฅผ ์ฐธ์กฐํ์ธ์.</p>
<p>๋ฏธ๋ฆฌ ํ์ต๋ ๋ชจ๋ธ์ ๋ถ๋ฌ์ต๋๋ค: ์ด๋ฒ ์์ ์์๋ <a class="reference external" href="https://pytorch.org/hub/pytorch_vision_resnet/">ResNet-18</a> ์ ์ฌ์ฉํ ๊ฒ์
๋๋ค.</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="kn">import</span> <span class="nn">torchvision.models.quantization</span> <span class="k">as</span> <span class="nn">models</span>
<span class="c1"># ๋์ค์ ์ฌ์ฉํ ์ ์๊ฒ `fc` ์ ํํฐ์ ์๊ฐ ํ์ํฉ๋๋ค.</span>
<span class="c1"># ์ฌ๊ธฐ์ ๊ฐ ์ถ๋ ฅ ์ํ์ ํฌ๊ธฐ๋ 2๋ก ์ค์ ํฉ๋๋ค.</span>
<span class="c1"># ๋ํ, nn.Linear(num_ftrs, len(class_names))๋ก ์ผ๋ฐํ ํ ์ ์์ต๋๋ค.</span>
<span class="n">model_fe</span> <span class="o">=</span> <span class="n">models</span><span class="o">.</span><span class="n">resnet18</span><span class="p">(</span><span class="n">pretrained</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span> <span class="n">progress</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span> <span class="n">quantize</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span>
<span class="n">num_ftrs</span> <span class="o">=</span> <span class="n">model_fe</span><span class="o">.</span><span class="n">fc</span><span class="o">.</span><span class="n">in_features</span>
</pre></div>
</div>
<p>์ด ์์ ์์ ๋ฏธ๋ฆฌ ํ์ต๋ ๋ชจ๋ธ์ ์์ ํด์ผ ํฉ๋๋ค. ๋ชจ๋ธ์ ์์๊ณผ ๋์๋ ์์ํ/์ญ์์ํ ๋ธ๋ก์ด ์์ต๋๋ค.
๊ทธ๋ฌ๋ ํน์ง ์ถ์ถ๊ธฐ๋ง ์ฌ์ฉํ๊ธฐ ๋๋ฌธ์ ์ญ์์ํ(dequantization) ๊ณ์ธต์ ์ ํ ๊ณ์ธต(ํค๋) ๋ฐ๋ก ์ ์ผ๋ก ์ด๋์์ผ์ผ ํฉ๋๋ค.
๊ฐ์ฅ ์ฌ์ด ๋ฐฉ๋ฒ์ ๋ชจ๋ธ์ <code class="docutils literal notranslate"><span class="pre">nn.Sequential</span></code> ๋ชจ๋๋ก ๊ฐ์ธ๋ ๊ฒ์
๋๋ค.</p>
<p>์ฒซ๋ฒ์งธ ๋จ๊ณ๋ ResNet ๋ชจ๋ธ์์ ํน์ง ์ถ์ถ๊ธฐ๋ฅผ ๋ถ๋ฆฌํ๋ ๊ฒ์
๋๋ค.
์ด ์์ ์์๋ <code class="docutils literal notranslate"><span class="pre">fc</span></code> ๋ฅผ ์ ์ธํ ๋ชจ๋ ๊ณ์ธต์ ํน์ง ์ถ์ถ๊ธฐ๋ก ์ฌ์ฉํด์ผ ํ์ง๋ง, ์ค์ ๋ก๋ ํ์ํ ๋งํผ ๋ง์ ๋ถ๋ถ์ ์ฌ์ฉํ ์ ์์ต๋๋ค.
์ด๊ฒ์ ํฉ์ฑ๊ณฑ ๊ณ์ธต ์ค ์ผ๋ถ๋ฅผ ๊ต์ฒดํ๋ ค๋ ๊ฒฝ์ฐ์๋ ์ ์ฉํฉ๋๋ค.</p>
<div class="admonition note">
<p class="admonition-title">์ฐธ๊ณ </p>
<p>์์ํ ๋ชจ๋ธ์์ ํน์ง ์ถ์ถ๊ธฐ๋ฅผ ๋ถ๋ฆฌํ ๋ ์์ํ๋ฅผ ์ ์งํ๋ ค๋ ๋ถ๋ถ์ ์์๊ณผ ๋์ ์๋์ผ๋ก ์์ํ/์ญ์์ํ๋ฅผ ๋ฐฐ์นํด์ผ ํฉ๋๋ค.</p>
</div>
<p>์๋ ํจ์๋ ์ฌ์ฉ์ ์ง์ ํค๋๋ก ๋ชจ๋ธ์ ์์ฑํ๋ ํจ์์
๋๋ค.</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="kn">from</span> <span class="nn">torch</span> <span class="kn">import</span> <span class="n">nn</span>
<span class="k">def</span> <span class="nf">create_combined_model</span><span class="p">(</span><span class="n">model_fe</span><span class="p">):</span>
<span class="c1"># 1 ๋จ๊ณ. ํน์ง ์ถ์ถ๊ธฐ๋ฅผ ๋ถ๋ฆฌํฉ๋๋ค.</span>
<span class="n">model_fe_features</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">Sequential</span><span class="p">(</span>
<span class="n">model_fe</span><span class="o">.</span><span class="n">quant</span><span class="p">,</span> <span class="c1"># Quantize the input</span>
<span class="n">model_fe</span><span class="o">.</span><span class="n">conv1</span><span class="p">,</span>
<span class="n">model_fe</span><span class="o">.</span><span class="n">bn1</span><span class="p">,</span>
<span class="n">model_fe</span><span class="o">.</span><span class="n">relu</span><span class="p">,</span>
<span class="n">model_fe</span><span class="o">.</span><span class="n">maxpool</span><span class="p">,</span>
<span class="n">model_fe</span><span class="o">.</span><span class="n">layer1</span><span class="p">,</span>
<span class="n">model_fe</span><span class="o">.</span><span class="n">layer2</span><span class="p">,</span>
<span class="n">model_fe</span><span class="o">.</span><span class="n">layer3</span><span class="p">,</span>
<span class="n">model_fe</span><span class="o">.</span><span class="n">layer4</span><span class="p">,</span>
<span class="n">model_fe</span><span class="o">.</span><span class="n">avgpool</span><span class="p">,</span>
<span class="n">model_fe</span><span class="o">.</span><span class="n">dequant</span><span class="p">,</span> <span class="c1"># ์ถ๋ ฅ์ ์ญ์์ํํ๊ธฐ</span>
<span class="p">)</span>
<span class="c1"># 2 ๋จ๊ณ. ์๋ก์ด "ํค๋(head)"๋ฅผ ๋ง๋ญ๋๋ค.</span>
<span class="n">new_head</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">Sequential</span><span class="p">(</span>
<span class="n">nn</span><span class="o">.</span><span class="n">Dropout</span><span class="p">(</span><span class="n">p</span><span class="o">=</span><span class="mf">0.5</span><span class="p">),</span>
<span class="n">nn</span><span class="o">.</span><span class="n">Linear</span><span class="p">(</span><span class="n">num_ftrs</span><span class="p">,</span> <span class="mi">2</span><span class="p">),</span>
<span class="p">)</span>
<span class="c1"># 3 ๋จ๊ณ. ๊ฒฐํฉํ๊ณ ์์ ์คํ
(stubs)์ ์์ผ๋ฉด ์๋ฉ๋๋ค.</span>
<span class="n">new_model</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">Sequential</span><span class="p">(</span>
<span class="n">model_fe_features</span><span class="p">,</span>
<span class="n">nn</span><span class="o">.</span><span class="n">Flatten</span><span class="p">(</span><span class="mi">1</span><span class="p">),</span>
<span class="n">new_head</span><span class="p">,</span>
<span class="p">)</span>
<span class="k">return</span> <span class="n">new_model</span>
</pre></div>
</div>
<div class="admonition warning">
<p class="admonition-title">๊ฒฝ๊ณ </p>
<p>ํ์ฌ ์์ํ๋ ๋ชจ๋ธ์ CPU์์๋ง ์คํํ ์ ์์ต๋๋ค.
๊ทธ๋ฌ๋ ๋ชจ๋ธ์ ์์ํ ๋์ง ์์ ๋ถ๋ถ์ GPU๋ก ๋ณด๋ผ ์ ์์ต๋๋ค.</p>
</div>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="kn">import</span> <span class="nn">torch.optim</span> <span class="k">as</span> <span class="nn">optim</span>
<span class="n">new_model</span> <span class="o">=</span> <span class="n">create_combined_model</span><span class="p">(</span><span class="n">model_fe</span><span class="p">)</span>
<span class="n">new_model</span> <span class="o">=</span> <span class="n">new_model</span><span class="o">.</span><span class="n">to</span><span class="p">(</span><span class="s1">'cpu'</span><span class="p">)</span>
<span class="n">criterion</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">CrossEntropyLoss</span><span class="p">()</span>
<span class="c1"># ํค๋(the head)๋ง ํ๋ จ ํ๋ค๋ ์ ์ ์ ์ํ์ธ์</span>
<span class="n">optimizer_ft</span> <span class="o">=</span> <span class="n">optim</span><span class="o">.</span><span class="n">SGD</span><span class="p">(</span><span class="n">new_model</span><span class="o">.</span><span class="n">parameters</span><span class="p">(),</span> <span class="n">lr</span><span class="o">=</span><span class="mf">0.01</span><span class="p">,</span> <span class="n">momentum</span><span class="o">=</span><span class="mf">0.9</span><span class="p">)</span>
<span class="c1"># 7 ์ํญ๋ง๋ค 0.1๋ฐฐ์ฉ ํ์ต๋ฅ ์ด ๊ฐ์</span>
<span class="n">exp_lr_scheduler</span> <span class="o">=</span> <span class="n">optim</span><span class="o">.</span><span class="n">lr_scheduler</span><span class="o">.</span><span class="n">StepLR</span><span class="p">(</span><span class="n">optimizer_ft</span><span class="p">,</span> <span class="n">step_size</span><span class="o">=</span><span class="mi">7</span><span class="p">,</span> <span class="n">gamma</span><span class="o">=</span><span class="mf">0.1</span><span class="p">)</span>
</pre></div>
</div>
<div class="section" id="id10">
<h3>ํ์ต๊ณผ ํ๊ฐ<a class="headerlink" href="#id10" title="์ด ์ ๋ชฉ์ ๋ํ ํผ๋จธ๋งํฌ">ยถ</a></h3>
<p>์ด ๋จ๊ณ๋ CPU์์ ์ฝ 15 ~ 25๋ถ ๊ฑธ๋ฆฝ๋๋ค. ์์ํ๋ ๋ชจ๋ธ์ CPU์์๋ง ์คํ๋๊ธฐ ๋๋ฌธ์
GPU์์๋ ํ๋ จ์ ์คํํ ์ ์์ต๋๋ค.</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="n">new_model</span> <span class="o">=</span> <span class="n">train_model</span><span class="p">(</span><span class="n">new_model</span><span class="p">,</span> <span class="n">criterion</span><span class="p">,</span> <span class="n">optimizer_ft</span><span class="p">,</span> <span class="n">exp_lr_scheduler</span><span class="p">,</span>
<span class="n">num_epochs</span><span class="o">=</span><span class="mi">25</span><span class="p">,</span> <span class="n">device</span><span class="o">=</span><span class="s1">'cpu'</span><span class="p">)</span>
<span class="n">visualize_model</span><span class="p">(</span><span class="n">new_model</span><span class="p">)</span>
<span class="n">plt</span><span class="o">.</span><span class="n">tight_layout</span><span class="p">()</span>
</pre></div>
</div>
</div>
</div>
<div class="section" id="finetuning">
<h2>ํํธ 2. ์์ํ ๊ฐ๋ฅํ ๋ชจ๋ธ ๋ฏธ์ธ์กฐ์ (Finetuning)<a class="headerlink" href="#finetuning" title="์ด ์ ๋ชฉ์ ๋ํ ํผ๋จธ๋งํฌ">ยถ</a></h2>
<p>์ด๋ฒ ํํธ์์๋ ์ ์ดํ์ต(Transfer Learning)์ ์ฌ์ฉํ์ฌ ํน์ง ์ถ์ถ๊ธฐ(Feature Extractor)๋ฅผ
๋ฏธ์ธ์กฐ์ (Finetuning) ํฉ๋๋ค. ํํธ 1๊ณผ 2 ๋ชจ๋์์ ํน์ง ์ถ์ถ๊ธฐ๋ ์์ํ๋ฉ๋๋ค. ์ฐจ์ด์ ์ ํํธ 1์์
๋ฏธ๋ฆฌ ํ์ต ๋ ์์ํ ๋ชจ๋ธ์ ์ฌ์ฉํฉ๋๋ค. ์ด๋ฒ ํํธ์์, ์ฐ๋ฆฌ๋ ๊ด์ฌ์๋ ๋ฐ์ดํฐ์
์ผ๋ก ๋ฏธ์ธ์กฐ์ (Finetuning)ํ ํ
์์ํ๋ ํน์ง ์ถ์ถ๊ธฐ๋ฅผ ์์ฑํ๋ฏ๋ก, ์์ํ์ ์ฅ์ ์ ๊ฐ์ง๋ฉด์ ์ ์ด ํ์ต(Transfer Learning)์ผ๋ก ๋ ๋์ ์ ํ๋๋ฅผ
์ป์ ์ ์์ต๋๋ค. ํน์ ํ ์์ ์์๋ ํ์ต์ฉ ์
์ ๋งค์ฐ ์๊ธฐ ๋๋ฌธ์(120๊ฐ์ ์ด๋ฏธ์ง) ์ ์ฒด ๋ชจ๋ธ์
๋ฏธ์ธ์กฐ์ (Finetuning)ํ๋ ์ฅ์ ์ด ๋ถ๋ถ๋ช
ํฉ๋๋ค. ๊ทธ๋ฌ๋ ์ฌ๊ธฐ์ ํ์๋ ์ ์ฐจ๋ ๋์ฑ ๋ ํฐ ๋ฐ์ดํฐ์
์ ์ฌ์ฉํ ์ ์ด ํ์ต(Transfer Learning)์
์ ํ๋๋ฅผ ํฅ์์ํต๋๋ค.</p>
<p>๋ฏธ๋ฆฌ ํ์ต๋ ํน์ง ์ถ์ถ๊ธฐ๋ ์์ํ๊ฐ ๊ฐ๋ฅํด์ผ ํฉ๋๋ค.
์์ํ๊ฐ ๊ฐ๋ฅํ์ง ํ์ธํ๊ธฐ ์ํด์ ๋ค์ ๋จ๊ณ๋ฅผ ์ํํ์ธ์:</p>
<blockquote>
<div><ol class="arabic simple">
<li><p><code class="docutils literal notranslate"><span class="pre">torch.quantization.fuse_modules</span></code> ๋ฅผ ์ฌ์ฉํ์ฌ <code class="docutils literal notranslate"><span class="pre">(Conv,</span> <span class="pre">BN,</span> <span class="pre">ReLU)</span></code> ,
<code class="docutils literal notranslate"><span class="pre">(Conv,</span> <span class="pre">BN)</span></code>, ๊ทธ๋ฆฌ๊ณ <code class="docutils literal notranslate"><span class="pre">(Conv,</span> <span class="pre">ReLU)</span></code> ๋ฅผ ์ตํฉํฉ๋๋ค.</p></li>
<li><p>ํน์ง ์ถ์ถ๊ธฐ๋ฅผ ์ฌ์ฉ์ ์ง์ ํค๋์ ์ฐ๊ฒฐํฉ๋๋ค. ์ด๋ฅผ ์ํด์ ํน์ง ์ถ์ถ๊ธฐ์ ์ถ๋ ฅ์ ์ญ์ผ๋ก ์์ํ ํด์ผํฉ๋๋ค.</p></li>
<li><p>ํน์ง ์ถ์ถ๊ธฐ์ ์ ํฉํ ์์น์ ๊ฐ์ง ์์ํ ๋ชจ๋์ ์ฝ์
ํ์ฌ ํ์ตํ๋ ๋์์ ์์ํ๋ฅผ ๋ชจ๋ฐฉํฉ๋๋ค.</p></li>
</ol>
</div></blockquote>
<p>(1) ๋จ๊ณ์ ๊ฒฝ์ฐ ๋ฉค๋ฒ ๋ฉ์๋(member method) <code class="docutils literal notranslate"><span class="pre">fuse_model</span></code> ์ด ์๋
<code class="docutils literal notranslate"><span class="pre">torchvision/models/quantization</span></code> ์ ๋ชจ๋ธ์ ์ฌ์ฉํฉ๋๋ค.
์ด ํจ์๋ ๋ชจ๋ <code class="docutils literal notranslate"><span class="pre">conv</span></code> , <code class="docutils literal notranslate"><span class="pre">bn</span></code> , ๊ทธ๋ฆฌ๊ณ <code class="docutils literal notranslate"><span class="pre">relu</span></code> ๋ชจ๋์ ํตํฉํฉ๋๋ค.
์ฌ์ฉ์ ์ง์ ๋ชจ๋ธ์ ๊ฒฝ์ฐ, ์๋์ผ๋ก ํตํฉํ ๋ชจ๋์ ๋ชฉ๋ก๊ณผ ํจ๊ป <code class="docutils literal notranslate"><span class="pre">torch.quantization.fuse_modules</span></code> API๋ฅผ ํธ์ถํด์ผํฉ๋๋ค.</p>
<ol class="arabic simple" start="2">
<li><p>๋จ๊ณ๋ ์ด์ ์น์
์์ ์ฌ์ฉํ <code class="docutils literal notranslate"><span class="pre">create_combined_model</span></code> ํจ์์ ์ํด์ ์ํ๋ฉ๋๋ค.</p></li>
<li><p>๋จ๊ณ๋ ๊ฐ์ง ์์ํ ๋ชจ๋์ ์ฝ์
ํ๋ <code class="docutils literal notranslate"><span class="pre">torch.quantization.prepare_qat</span></code> ๋ฅผ ์ฌ์ฉํ์ฌ ์ํ๋ฉ๋๋ค.</p></li>
<li><p>๋จ๊ณ๋ก ๋ชจ๋ธ์ ใ๋ฏธ์ธ์กฐ์ (Finetuning)ใํ ํ, ์์ ํ๊ฒ ์์ํ๋ ๋ฒ์ ์ผ๋ก ๋ณํ(5๋จ๊ณ) ํ ์ ์์ต๋๋ค.</p></li>
</ol>
<p>๋ฏธ์ธ์กฐ์ (Finetuning) ๋ชจ๋ธ์ ์์ํ๋ ๋ชจ๋ธ๋ก ๋ณํํ๋ ค๋ฉด <code class="docutils literal notranslate"><span class="pre">torch.quantization.convert</span></code> ํจ์๋ฅผ
ํธ์ถ ํ ์ ์์ต๋๋ค. (์ด ๊ฒฝ์ฐ ํน์ง ์ถ์ถ๊ธฐ๋ง ์์ํ ๋ฉ๋๋ค.)</p>
<div class="admonition note">
<p class="admonition-title">์ฐธ๊ณ </p>
<p>๋๋ค ์ด๊ธฐํ ๋๋ฌธ์ ์ฌ๋ฌ๋ถ์ ๊ฒฐ๊ณผ๊ฐ ํํ ๋ฆฌ์ผ์ ํ์๋ ๊ฒฐ๊ณผ์ ๋ค๋ฅผ ์ ์์ต๋๋ค.</p>
</div>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="c1"># `quantize=False` ๋ฅผ ์ฃผ๋ชฉํ์ธ์</span>
<span class="n">model</span> <span class="o">=</span> <span class="n">models</span><span class="o">.</span><span class="n">resnet18</span><span class="p">(</span><span class="n">pretrained</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span> <span class="n">progress</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span> <span class="n">quantize</span><span class="o">=</span><span class="kc">False</span><span class="p">)</span>
<span class="n">num_ftrs</span> <span class="o">=</span> <span class="n">model</span><span class="o">.</span><span class="n">fc</span><span class="o">.</span><span class="n">in_features</span>
<span class="c1"># 1 ๋จ๊ณ</span>
<span class="n">model</span><span class="o">.</span><span class="n">train</span><span class="p">()</span>
<span class="n">model</span><span class="o">.</span><span class="n">fuse_model</span><span class="p">()</span>
<span class="c1"># 2 ๋จ๊ณ</span>
<span class="n">model_ft</span> <span class="o">=</span> <span class="n">create_combined_model</span><span class="p">(</span><span class="n">model</span><span class="p">)</span>
<span class="n">model_ft</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span><span class="o">.</span><span class="n">qconfig</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">quantization</span><span class="o">.</span><span class="n">default_qat_qconfig</span> <span class="c1"># Use default QAT configuration</span>
<span class="c1"># 3 ๋จ๊ณ</span>
<span class="n">model_ft</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">quantization</span><span class="o">.</span><span class="n">prepare_qat</span><span class="p">(</span><span class="n">model_ft</span><span class="p">,</span> <span class="n">inplace</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span>
</pre></div>
</div>
<div class="section" id="id11">
<h3>๋ชจ๋ธ ๋ฏธ์ธ์กฐ์ <a class="headerlink" href="#id11" title="์ด ์ ๋ชฉ์ ๋ํ ํผ๋จธ๋งํฌ">ยถ</a></h3>
<p>ํ์ฌ ํํ ๋ฆฌ์ผ์์๋ ์ ์ฒด ๋ชจ๋ธ์ด ๋ฏธ์ธ์กฐ์ ๋์์ต๋๋ค.
์ผ๋ฐ์ ์ผ๋ก ์ด๊ฒ์ ๋ ๋์ ์ ํ๋๋ก ์ด์ด์ง ๊ฒ์
๋๋ค.
๊ทธ๋ฌ๋ ์ฌ๊ธฐ์๋ ํฌ๊ธฐ๊ฐ ์์ ํ์ต์ฉ ๋ฐ์ดํฐ์
์ ์ฌ์ฉํ๊ธฐ ๋๋ฌธ์ ๊ฒฐ๊ตญ ๊ณผ์ ํฉํ๊ฒ ๋ฉ๋๋ค.</p>
<p>4 ๋จ๊ณ. ๋ชจ๋ธ ๋ฏธ์ธ์กฐ์ ํ๊ธฐ</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="k">for</span> <span class="n">param</span> <span class="ow">in</span> <span class="n">model_ft</span><span class="o">.</span><span class="n">parameters</span><span class="p">():</span>
<span class="n">param</span><span class="o">.</span><span class="n">requires_grad</span> <span class="o">=</span> <span class="kc">True</span>
<span class="n">model_ft</span><span class="o">.</span><span class="n">to</span><span class="p">(</span><span class="n">device</span><span class="p">)</span> <span class="c1"># GPU์์ ๋ฏธ์ธ์กฐ์ (Finetuning) ํ ์ ์์ต๋๋ค.</span>
<span class="n">criterion</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">CrossEntropyLoss</span><span class="p">()</span>
<span class="c1"># ์ด๋ฏธ ๋ชจ๋ ๊ฒ์ด ํ์ต๋ ์ํ์ด๋ฏ๋ก ํ์ต๋ฅ ์ด ๋ฎ์ต๋๋ค.</span>
<span class="c1"># ๋ ์์ Learning rate์ ์ฃผ๋ชฉํ์ธ์</span>
<span class="n">optimizer_ft</span> <span class="o">=</span> <span class="n">optim</span><span class="o">.</span><span class="n">SGD</span><span class="p">(</span><span class="n">model_ft</span><span class="o">.</span><span class="n">parameters</span><span class="p">(),</span> <span class="n">lr</span><span class="o">=</span><span class="mf">1e-3</span><span class="p">,</span> <span class="n">momentum</span><span class="o">=</span><span class="mf">0.9</span><span class="p">,</span> <span class="n">weight_decay</span><span class="o">=</span><span class="mf">0.1</span><span class="p">)</span>
<span class="c1"># ํ์ต๋ฅ ์ ๋ช ์ํญ๋ง๋ค 0.3๋ฐฐ ๊ฐ์์ํค๊ธฐ</span>
<span class="n">exp_lr_scheduler</span> <span class="o">=</span> <span class="n">optim</span><span class="o">.</span><span class="n">lr_scheduler</span><span class="o">.</span><span class="n">StepLR</span><span class="p">(</span><span class="n">optimizer_ft</span><span class="p">,</span> <span class="n">step_size</span><span class="o">=</span><span class="mi">5</span><span class="p">,</span> <span class="n">gamma</span><span class="o">=</span><span class="mf">0.3</span><span class="p">)</span>
<span class="n">model_ft_tuned</span> <span class="o">=</span> <span class="n">train_model</span><span class="p">(</span><span class="n">model_ft</span><span class="p">,</span> <span class="n">criterion</span><span class="p">,</span> <span class="n">optimizer_ft</span><span class="p">,</span> <span class="n">exp_lr_scheduler</span><span class="p">,</span>
<span class="n">num_epochs</span><span class="o">=</span><span class="mi">25</span><span class="p">,</span> <span class="n">device</span><span class="o">=</span><span class="n">device</span><span class="p">)</span>
</pre></div>
</div>
<p>5 ๋จ๊ณ. ์์ํ๋ ๋ชจ๋ธ๋ก ๋ณํํ๊ธฐ</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="kn">from</span> <span class="nn">torch.quantization</span> <span class="kn">import</span> <span class="n">convert</span>
<span class="n">model_ft_tuned</span><span class="o">.</span><span class="n">cpu</span><span class="p">()</span>
<span class="n">model_quantized_and_trained</span> <span class="o">=</span> <span class="n">convert</span><span class="p">(</span><span class="n">model_ft_tuned</span><span class="p">,</span> <span class="n">inplace</span><span class="o">=</span><span class="kc">False</span><span class="p">)</span>
</pre></div>
</div>
<p>์์ํ๋ ๋ชจ๋ธ์ด ์ผ๋ถ ์ด๋ฏธ์ง์์ ์ด๋ป๊ฒ ๋์ํ๋์ง ์ดํด๋ณด๊ฒ ์ต๋๋ค.</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="n">visualize_model</span><span class="p">(</span><span class="n">model_quantized_and_trained</span><span class="p">)</span>
<span class="n">plt</span><span class="o">.</span><span class="n">ioff</span><span class="p">()</span>
<span class="n">plt</span><span class="o">.</span><span class="n">tight_layout</span><span class="p">()</span>
<span class="n">plt</span><span class="o">.</span><span class="n">show</span><span class="p">()</span>
</pre></div>
</div>
</div>
</div>
</div>
</article>
</div>
<footer>
<div class="rst-footer-buttons" role="navigation" aria-label="footer navigation">
<a href="../advanced/static_quantization_tutorial.html" class="btn btn-neutral float-right" title="(๋ฒ ํ) PyTorch์์ Eager Mode๋ฅผ ์ด์ฉํ ์ ์ ์์ํ" accesskey="n" rel="next">Next <img src="../_static/images/chevron-right-orange.svg" class="next-page"></a>
<a href="dynamic_quantization_bert_tutorial.html" class="btn btn-neutral" title="(๋ฒ ํ) BERT ๋ชจ๋ธ ๋์ ์์ํํ๊ธฐ" accesskey="p" rel="prev"><img src="../_static/images/chevron-right-orange.svg" class="previous-page"> Previous</a>
</div>
<hr class="community-hr hr-top" />
<div class="community-container">
<div class="community-prompt">๋ ๊ถ๊ธํ์๊ฑฐ๋ ๊ฐ์ ํ ๋ด์ฉ์ด ์์ผ์ ๊ฐ์? ์ปค๋ฎค๋ํฐ์ ์ฐธ์ฌํด๋ณด์ธ์!</div>
<div class="community-link"><a href="https://discuss.pytorch.kr/" aria-label="PyTorchKoreaCommunity">ํ๊ตญ์ด ์ปค๋ฎค๋ํฐ ๋ฐ๋ก๊ฐ๊ธฐ</a></div>
</div>
<hr class="community-hr hr-bottom"/>
<hr class="rating-hr hr-top" />
<div class="rating-container">
<div class="rating-prompt">์ด ํํ ๋ฆฌ์ผ์ด ์ด๋ ์
จ๋์? ํ๊ฐํด์ฃผ์๋ฉด ์ดํ ๊ฐ์ ์ ์ฐธ๊ณ ํ๊ฒ ์ต๋๋ค! :)</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 2018-2024, PyTorch & ํ์ดํ ์น ํ๊ตญ ์ฌ์ฉ์ ๋ชจ์(PyTorch Korea User Group).
</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>
<script>
if((window.location.href.indexOf("/prototype/")!= -1) && (window.location.href.indexOf("/prototype/prototype_index")< 1))
{
var div = '<div class="admonition note"><p class="admonition-title">Note</p><p><i class="fa fa-flask" aria-hidden="true"> </i> ์ด ํํ ๋ฆฌ์ผ์ ํ๋กํ ํ์
(prototype) ๊ธฐ๋ฅ๋ค์ ๋ํด์ ์ค๋ช
ํ๊ณ ์์ต๋๋ค. ํ๋กํ ํ์
๊ธฐ๋ฅ์ ์ผ๋ฐ์ ์ผ๋ก ํผ๋๋ฐฑ ๋ฐ ํ
์คํธ์ฉ์ผ๋ก, ๋ฐํ์ ํ๋๊ทธ ์์ด๋ PyPI๋ Conda๋ก ๋ฐฐํฌ๋๋ ๋ฐ์ด๋๋ฆฌ์์๋ ์ฌ์ฉํ ์ ์์ต๋๋ค.</p></div>'
document.getElementById("pytorch-article").insertAdjacentHTML('afterBegin', div)
}
</script>
</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="#">(๋ฒ ํ) ์ปดํจํฐ ๋น์ ํํ ๋ฆฌ์ผ์ ์ํ ์์ํ๋ ์ ์ดํ์ต(Quantized Transfer Learning)</a><ul>
<li><a class="reference internal" href="#id2">ํํธ 0. ์๊ตฌ์ฌํญ</a><ul>
<li><a class="reference internal" href="#nightly-build">Nightly Build ์ค์นํ๊ธฐ</a></li>
<li><a class="reference internal" href="#id4">๋ฐ์ดํฐ ๋ถ๋ฌ์ค๊ธฐ</a></li>
<li><a class="reference internal" href="#id6">์ผ๋ถ ์ด๋ฏธ์ง ์๊ฐํํ๊ธฐ</a></li>
<li><a class="reference internal" href="#id7">๋ชจ๋ธ ํ์ต์ ์ํ ์ง์ ํจ์</a></li>
<li><a class="reference internal" href="#id8">๋ชจ๋ธ ์์ธก์ ์๊ฐํํ๊ธฐ ์ํ ์ง์ ํจ์</a></li>
</ul>
</li>
<li><a class="reference internal" href="#quantized-feature-extractor">ํํธ 1. ์์ํ๋ ํน์ง ์ถ์ถ๊ธฐ(Quantized Feature Extractor)๋ฅผ ๊ธฐ๋ฐ์ผ๋ก ์ฌ์ฉ์ ์ง์ ๋ถ๋ฅ๊ธฐ ํ๋ จํ๊ธฐ</a><ul>
<li><a class="reference internal" href="#id10">ํ์ต๊ณผ ํ๊ฐ</a></li>
</ul>
</li>
<li><a class="reference internal" href="#finetuning">ํํธ 2. ์์ํ ๊ฐ๋ฅํ ๋ชจ๋ธ ๋ฏธ์ธ์กฐ์ (Finetuning)</a><ul>
<li><a class="reference internal" href="#id11">๋ชจ๋ธ ๋ฏธ์ธ์กฐ์ </a></li>
</ul>
</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 data-url_root="../" id="documentation_options" src="../_static/documentation_options.js"></script>
<script src="../_static/jquery.js"></script>
<script src="../_static/underscore.js"></script>
<script src="../_static/_sphinx_javascript_frameworks_compat.js"></script>
<script src="../_static/doctools.js"></script>
<script src="../_static/clipboard.min.js"></script>
<script src="../_static/copybutton.js"></script>
<script src="../_static/translations.js"></script>
<script src="../_static/katex.min.js"></script>
<script src="../_static/auto-render.min.js"></script>
<script src="../_static/katex_autorenderer.js"></script>
<script src="../_static/design-tabs.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>