File size: 61,649 Bytes
1112c3f
b0d1496
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
8c1d0f7
b0d1496
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
f5567a0
b0d1496
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
8c1d0f7
b0d1496
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
8c1d0f7
b0d1496
 
 
 
 
8c1d0f7
31cd561
 
 
 
 
 
 
 
 
 
 
 
 
b0d1496
 
31cd561
 
 
b0d1496
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
<!DOCTYPE html>
<html lang="en"><head>
<script src="Accelerate_files/libs/clipboard/clipboard.min.js"></script>
<script src="Accelerate_files/libs/quarto-html/tabby.min.js"></script>
<script src="Accelerate_files/libs/quarto-html/popper.min.js"></script>
<script src="Accelerate_files/libs/quarto-html/tippy.umd.min.js"></script>
<link href="Accelerate_files/libs/quarto-html/tippy.css" rel="stylesheet">
<link href="Accelerate_files/libs/quarto-html/quarto-html.min.css" rel="stylesheet" data-mode="light">
<link href="Accelerate_files/libs/quarto-html/quarto-syntax-highlighting-dark.css" rel="stylesheet" id="quarto-text-highlighting-styles"><meta charset="utf-8">
  <meta name="generator" content="quarto-1.2.237">

  <meta name="author" content="Zachary Mueller">
  <title>Accelerate, Three Powerful Sublibraries for PyTorch</title>
  <meta name="apple-mobile-web-app-capable" content="yes">
  <meta name="apple-mobile-web-app-status-bar-style" content="black-translucent">
  <meta name="viewport" content="width=device-width, initial-scale=1.0, maximum-scale=1.0, user-scalable=no, minimal-ui">
  <link rel="stylesheet" href="Accelerate_files/libs/revealjs/dist/reset.css">
  <link rel="stylesheet" href="Accelerate_files/libs/revealjs/dist/reveal.css">
  <style>
    code{white-space: pre-wrap;}
    span.smallcaps{font-variant: small-caps;}
    div.columns{display: flex; gap: min(4vw, 1.5em);}
    div.column{flex: auto; overflow-x: auto;}
    div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
    ul.task-list{list-style: none;}
    ul.task-list li input[type="checkbox"] {
      width: 0.8em;
      margin: 0 0.8em 0.2em -1.6em;
      vertical-align: middle;
    }
    pre > code.sourceCode { white-space: pre; position: relative; }
    pre > code.sourceCode > span { display: inline-block; line-height: 1.25; }
    pre > code.sourceCode > span:empty { height: 1.2em; }
    .sourceCode { overflow: visible; }
    code.sourceCode > span { color: inherit; text-decoration: inherit; }
    div.sourceCode { margin: 1em 0; }
    pre.sourceCode { margin: 0; }
    @media screen {
    div.sourceCode { overflow: auto; }
    }
    @media print {
    pre > code.sourceCode { white-space: pre-wrap; }
    pre > code.sourceCode > span { text-indent: -5em; padding-left: 5em; }
    }
    pre.numberSource code
      { counter-reset: source-line 0; }
    pre.numberSource code > span
      { position: relative; left: -4em; counter-increment: source-line; }
    pre.numberSource code > span > a:first-child::before
      { content: counter(source-line);
        position: relative; left: -1em; text-align: right; vertical-align: baseline;
        border: none; display: inline-block;
        -webkit-touch-callout: none; -webkit-user-select: none;
        -khtml-user-select: none; -moz-user-select: none;
        -ms-user-select: none; user-select: none;
        padding: 0 4px; width: 4em;
      }
    pre.numberSource { margin-left: 3em;  padding-left: 4px; }
    div.sourceCode
      { color: #f8f8f2;  }
    @media screen {
    pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
    }
    code span { color: #f8f8f2; } /* Normal */
    code span.al { color: #f07178; background-color: #2a0f15; font-weight: bold; } /* Alert */
    code span.an { color: #d4d0ab; } /* Annotation */
    code span.at { color: #00e0e0; } /* Attribute */
    code span.bn { color: #d4d0ab; } /* BaseN */
    code span.bu { color: #abe338; } /* BuiltIn */
    code span.cf { color: #ffa07a; font-weight: bold; } /* ControlFlow */
    code span.ch { color: #abe338; } /* Char */
    code span.cn { color: #ffd700; } /* Constant */
    code span.co { color: #f8f8f2; font-style: italic; } /* Comment */
    code span.cv { color: #ffd700; } /* CommentVar */
    code span.do { color: #f8f8f2; } /* Documentation */
    code span.dt { color: #ffa07a; } /* DataType */
    code span.dv { color: #d4d0ab; } /* DecVal */
    code span.er { color: #f07178; text-decoration: underline; } /* Error */
    code span.ex { color: #00e0e0; font-weight: bold; } /* Extension */
    code span.fl { color: #d4d0ab; } /* Float */
    code span.fu { color: #ffa07a; } /* Function */
    code span.im { color: #abe338; } /* Import */
    code span.in { color: #d4d0ab; } /* Information */
    code span.kw { color: #ffa07a; font-weight: bold; } /* Keyword */
    code span.op { color: #ffa07a; } /* Operator */
    code span.ot { color: #00e0e0; } /* Other */
    code span.pp { color: #dcc6e0; } /* Preprocessor */
    code span.re { color: #00e0e0; background-color: #f8f8f2; } /* RegionMarker */
    code span.sc { color: #abe338; } /* SpecialChar */
    code span.ss { color: #abe338; } /* SpecialString */
    code span.st { color: #abe338; } /* String */
    code span.va { color: #00e0e0; } /* Variable */
    code span.vs { color: #abe338; } /* VerbatimString */
    code span.wa { color: #dcc6e0; } /* Warning */
  </style>
  <link rel="stylesheet" href="Accelerate_files/libs/revealjs/dist/theme/quarto.css" id="theme">
  <link href="Accelerate_files/libs/revealjs/plugin/quarto-line-highlight/line-highlight.css" rel="stylesheet">
  <link href="Accelerate_files/libs/revealjs/plugin/reveal-menu/menu.css" rel="stylesheet">
  <link href="Accelerate_files/libs/revealjs/plugin/reveal-menu/quarto-menu.css" rel="stylesheet">
  <link href="Accelerate_files/libs/revealjs/plugin/quarto-support/footer.css" rel="stylesheet">
  <style type="text/css">

  .callout {
    margin-top: 1em;
    margin-bottom: 1em;  
    border-radius: .25rem;
  }

  .callout.callout-style-simple { 
    padding: 0em 0.5em;
    border-left: solid #acacac .3rem;
    border-right: solid 1px silver;
    border-top: solid 1px silver;
    border-bottom: solid 1px silver;
    display: flex;
  }

  .callout.callout-style-default {
    border-left: solid #acacac .3rem;
    border-right: solid 1px silver;
    border-top: solid 1px silver;
    border-bottom: solid 1px silver;
  }

  .callout .callout-body-container {
    flex-grow: 1;
  }

  .callout.callout-style-simple .callout-body {
    font-size: 1rem;
    font-weight: 400;
  }

  .callout.callout-style-default .callout-body {
    font-size: 0.9rem;
    font-weight: 400;
  }

  .callout.callout-captioned.callout-style-simple .callout-body {
    margin-top: 0.2em;
  }

  .callout:not(.callout-captioned) .callout-body {
      display: flex;
  }

  .callout:not(.no-icon).callout-captioned.callout-style-simple .callout-content {
    padding-left: 1.6em;
  }

  .callout.callout-captioned .callout-header {
    padding-top: 0.2em;
    margin-bottom: -0.2em;
  }

  .callout.callout-captioned .callout-caption  p {
    margin-top: 0.5em;
    margin-bottom: 0.5em;
  }
    
  .callout.callout-captioned.callout-style-simple .callout-content  p {
    margin-top: 0;
  }

  .callout.callout-captioned.callout-style-default .callout-content  p {
    margin-top: 0.7em;
  }

  .callout.callout-style-simple div.callout-caption {
    border-bottom: none;
    font-size: .9rem;
    font-weight: 600;
    opacity: 75%;
  }

  .callout.callout-style-default  div.callout-caption {
    border-bottom: none;
    font-weight: 600;
    opacity: 85%;
    font-size: 0.9rem;
    padding-left: 0.5em;
    padding-right: 0.5em;
  }

  .callout.callout-style-default div.callout-content {
    padding-left: 0.5em;
    padding-right: 0.5em;
  }

  .callout.callout-style-simple .callout-icon::before {
    height: 1rem;
    width: 1rem;
    display: inline-block;
    content: "";
    background-repeat: no-repeat;
    background-size: 1rem 1rem;
  }

  .callout.callout-style-default .callout-icon::before {
    height: 0.9rem;
    width: 0.9rem;
    display: inline-block;
    content: "";
    background-repeat: no-repeat;
    background-size: 0.9rem 0.9rem;
  }

  .callout-caption {
    display: flex
  }
    
  .callout-icon::before {
    margin-top: 1rem;
    padding-right: .5rem;
  }

  .callout.no-icon::before {
    display: none !important;
  }

  .callout.callout-captioned .callout-body > .callout-content > :last-child {
    margin-bottom: 0.5rem;
  }

  .callout.callout-captioned .callout-icon::before {
    margin-top: .5rem;
    padding-right: .5rem;
  }

  .callout:not(.callout-captioned) .callout-icon::before {
    margin-top: 1rem;
    padding-right: .5rem;
  }

  /* Callout Types */

  div.callout-note {
    border-left-color: #4582ec !important;
  }

  div.callout-note .callout-icon::before {
    background-image: url('data:image/png;base64,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');
  }

  div.callout-note.callout-style-default .callout-caption {
    background-color: #dae6fb
  }

  div.callout-important {
    border-left-color: #d9534f !important;
  }

  div.callout-important .callout-icon::before {
    background-image: url('data:image/png;base64,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');
  }

  div.callout-important.callout-style-default .callout-caption {
    background-color: #f7dddc
  }

  div.callout-warning {
    border-left-color: #f0ad4e !important;
  }

  div.callout-warning .callout-icon::before {
    background-image: url('data:image/png;base64,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');
  }

  div.callout-warning.callout-style-default .callout-caption {
    background-color: #fcefdc
  }

  div.callout-tip {
    border-left-color: #02b875 !important;
  }

  div.callout-tip .callout-icon::before {
    background-image: url('data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAACAAAAAgCAYAAABzenr0AAAAAXNSR0IArs4c6QAAAERlWElmTU0AKgAAAAgAAYdpAAQAAAABAAAAGgAAAAAAA6ABAAMAAAABAAEAAKACAAQAAAABAAAAIKADAAQAAAABAAAAIAAAAACshmLzAAADr0lEQVRYCe1XTWgTQRj9ZjZV8a9SPIkKgj8I1bMHsUWrqYLVg4Ue6v9BwZOxSYsIerFao7UiUryIqJcqgtpimhbBXoSCVxUFe9CTiogUrUp2Pt+3aUI2u5vdNh4dmMzOzHvvezuz8xNFM0mjnbXaNu1MvFWRXkXEyE6aYOYJpdW4IXuA4r0fo8qqSMDBU0v1HJUgVieAXxzCsdE/YJTdFcVIZQNMyhruOMJKXYFoLfIfIvVIMWdsrd+Rpd86ZmyzzjJmLStqRn0v8lzkb4rVIXvnpScOJuAn2ACC65FkPzEdEy4TPWRLJ2h7z4cArXzzaOdKlbOvKKX25Wl00jSnrwVxAg3o4dRxhO13RBSdNvH0xSARv3adTXbBdTf64IWO2vH0LT+cv4GR1DJt+DUItaQogeBX/chhbTBxEiZ6gftlDNXTrvT7co4ub5A6gp9HIcHvzTa46OS5fBeP87Qm0fQkr4FsYgVQ7Qg+ZayaDg9jhg1GkWj8RG6lkeSacrrHgDaxdoBiZPg+NXV/KifMuB6//JmYH4CntVEHy/keA6x4h4CU5oFy8GzrBS18cLJMXcljAKB6INjWsRcuZBWVaS3GDrqB7rdapVIeA+isQ57Eev9eCqzqOa81CY05VLd6SamW2wA2H3SiTbnbSxmzfp7WtKZkqy4mdyAlGx7ennghYf8voqp9cLSgKdqNfa6RdRsAAkPwRuJZNbpByn+RrJi1RXTwdi8RQF6ymDwGMAtZ6TVE+4uoKh+MYkcLsT0Hk8eAienbiGdjJHZTpmNjlbFJNKDVAp2fJlYju6IreQxQ08UJDNYdoLSl6AadO+fFuCQqVMB1NJwPm69T04Wv5WhfcWyfXQB+wXRs1pt+nCknRa0LVzSA/2B+a9+zQJadb7IyyV24YAxKp2Jqs3emZTuNnKxsah+uabKbMk7CbTgJx/zIgQYErIeTKRQ9yD9wxVof5YolPHqaWo7TD6tJlh7jQnK5z2n3+fGdggIOx2kaa2YI9QWarc5Ce1ipNWMKeSG4DysFF52KBmTNMmn5HqCFkwy34rDg05gDwgH3bBi+sgFhN/e8QvRn8kbamCOhgrZ9GJhFDgfcMHzFb6BAtjKpFhzTjwv1KCVuxHvCbsSiEz4CANnj84cwHdFXAbAOJ4LTSAawGWFn5tDhLMYz6nWeU2wJfIhmIJBefcd/A5FWQWGgrWzyORZ3Q6HuV+Jf0Bj+BTX69fm1zWgK7By1YTXchFDORywnfQ7GpzOo6S+qECrsx2ifVQAAAABJRU5ErkJggg==');
  }

  div.callout-tip.callout-style-default .callout-caption {
    background-color: #ccf1e3
  }

  div.callout-caution {
    border-left-color: #fd7e14 !important;
  }

  div.callout-caution .callout-icon::before {
    background-image: url('data:image/png;base64,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');
  }

  div.callout-caution.callout-style-default .callout-caption {
    background-color: #ffe5d0
  }

  </style>
  <style type="text/css">
    .reveal div.sourceCode {
      margin: 0;
      overflow: auto;
    }
    .reveal div.hanging-indent {
      margin-left: 1em;
      text-indent: -1em;
    }
    .reveal .slide:not(.center) {
      height: 100%;
    }
    .reveal .slide.scrollable {
      overflow-y: auto;
    }
    .reveal .footnotes {
      height: 100%;
      overflow-y: auto;
    }
    .reveal .slide .absolute {
      position: absolute;
      display: block;
    }
    .reveal .footnotes ol {
      counter-reset: ol;
      list-style-type: none; 
      margin-left: 0;
    }
    .reveal .footnotes ol li:before {
      counter-increment: ol;
      content: counter(ol) ". "; 
    }
    .reveal .footnotes ol li > p:first-child {
      display: inline-block;
    }
    .reveal .slide ul,
    .reveal .slide ol {
      margin-bottom: 0.5em;
    }
    .reveal .slide ul li,
    .reveal .slide ol li {
      margin-top: 0.4em;
      margin-bottom: 0.2em;
    }
    .reveal .slide ul[role="tablist"] li {
      margin-bottom: 0;
    }
    .reveal .slide ul li > *:first-child,
    .reveal .slide ol li > *:first-child {
      margin-block-start: 0;
    }
    .reveal .slide ul li > *:last-child,
    .reveal .slide ol li > *:last-child {
      margin-block-end: 0;
    }
    .reveal .slide .columns:nth-child(3) {
      margin-block-start: 0.8em;
    }
    .reveal blockquote {
      box-shadow: none;
    }
    .reveal .tippy-content>* {
      margin-top: 0.2em;
      margin-bottom: 0.7em;
    }
    .reveal .tippy-content>*:last-child {
      margin-bottom: 0.2em;
    }
    .reveal .slide > img.stretch.quarto-figure-center,
    .reveal .slide > img.r-stretch.quarto-figure-center {
      display: block;
      margin-left: auto;
      margin-right: auto; 
    }
    .reveal .slide > img.stretch.quarto-figure-left,
    .reveal .slide > img.r-stretch.quarto-figure-left  {
      display: block;
      margin-left: 0;
      margin-right: auto; 
    }
    .reveal .slide > img.stretch.quarto-figure-right,
    .reveal .slide > img.r-stretch.quarto-figure-right  {
      display: block;
      margin-left: auto;
      margin-right: 0; 
    }
  </style>
  <script src="Accelerate_files/libs/quarto-diagram/mermaid.min.js"></script>
  <script src="Accelerate_files/libs/quarto-diagram/mermaid-init.js"></script>
  <link href="Accelerate_files/libs/quarto-diagram/mermaid.css" rel="stylesheet">
</head>
<body class="quarto-dark">
  <div class="reveal">
    <div class="slides">

<section id="title-slide" class="quarto-title-block center">
  <h1 class="title">Accelerate, Three Powerful Sublibraries for PyTorch</h1>

<div class="quarto-title-authors">
<div class="quarto-title-author">
<div class="quarto-title-author-name">
Zachary Mueller 
</div>
</div>
</div>

</section>
<section id="who-am-i" class="slide level2">
<h2>Who am I?</h2>
<ul>
<li>Zachary Mueller</li>
<li>Deep Learning Software Engineer at 🤗</li>
<li>API design geek</li>
</ul>
</section>
<section id="what-is-accelerate" class="slide level2">
<h2>What is 🤗 Accelerate?</h2>
<div class="cell" data-reveal="true" data-fig-height="6">
<div class="cell-output-display">
<div>
<p>
</p><pre class="mermaid mermaid-js" data-tooltip-selector="#mermaid-tooltip-1">graph LR
    A{"🤗 Accelerate#32;"}
    A --&gt; B["Launching&lt;br&gt;Interface#32;"]
    A --&gt; C["Training Library#32;"]
    A --&gt; D["Big Model&lt;br&gt;Inference#32;"]
</pre>
<div id="mermaid-tooltip-1" class="mermaidTooltip">

</div>
<p></p>
</div>
</div>
</div>
</section>
<section>
<section id="a-launching-interface" class="title-slide slide level1 center">
<h1>A Launching Interface</h1>
<p>Can’t I just use <code>python do_the_thing.py</code>?</p>
</section>
<section id="a-launching-interface-1" class="slide level2">
<h2>A Launching Interface</h2>
<p>Launching scripts in different environments is complicated:</p>
<ul>
<li><div class="sourceCode" id="cb1"><pre class="sourceCode numberSource bash number-lines code-with-copy"><code class="sourceCode bash"><span id="cb1-1"><a href="#cb1-1"></a><span class="ex">python</span> script.py</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div></li>
<li><div class="sourceCode" id="cb2"><pre class="sourceCode numberSource bash number-lines code-with-copy"><code class="sourceCode bash"><span id="cb2-1"><a href="#cb2-1"></a><span class="ex">torchrun</span> <span class="at">--nnodes</span><span class="op">=</span>1 <span class="at">--nproc_per_node</span><span class="op">=</span>2 script.py</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div></li>
<li><div class="sourceCode" id="cb3"><pre class="sourceCode numberSource bash number-lines code-with-copy"><code class="sourceCode bash"><span id="cb3-1"><a href="#cb3-1"></a><span class="ex">deepspeed</span> <span class="at">--num_gpus</span><span class="op">=</span>2 script.py</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div></li>
</ul>
<p>And more!</p>
</section>
<section id="a-launching-interface-2" class="slide level2">
<h2>A Launching Interface</h2>
<p>But it doesn’t have to be:</p>
<div class="sourceCode" id="cb4"><pre class="sourceCode numberSource bash number-lines code-with-copy"><code class="sourceCode bash"><span id="cb4-1"><a href="#cb4-1"></a><span class="ex">accelerate</span> launch script.py</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
<p>A single command to launch with <code>DeepSpeed</code>, Fully Sharded Data Parallelism, across single and multi CPUs and GPUs, and to train on TPUs<sup>1</sup> too!</p>
<aside><ol class="aside-footnotes"><li id="fn1"><p>Without needing to modify your code and create a <code>_mp_fn</code></p></li></ol></aside></section>
<section id="a-launching-interface-3" class="slide level2">
<h2>A Launching Interface</h2>
<p>Generate a device-specific configuration through <code>accelerate config</code></p>

<img data-src="CLI.gif" class="r-stretch"></section>
<section id="a-launching-interface-4" class="slide level2">
<h2>A Launching Interface</h2>
<p>Or don’t. <code>accelerate config</code> doesn’t <em>have</em> to be done!</p>
<div class="sourceCode" id="cb5"><pre class="sourceCode numberSource bash number-lines code-with-copy"><code class="sourceCode bash"><span id="cb5-1"><a href="#cb5-1"></a><span class="ex">torchrun</span> <span class="at">--nnodes</span><span class="op">=</span>1 <span class="at">--nproc_per_node</span><span class="op">=</span>2 script.py</span>
<span id="cb5-2"><a href="#cb5-2"></a><span class="ex">accelerate</span> launch <span class="at">--multi_gpu</span> <span class="at">--nproc_per_node</span><span class="op">=</span>2 script.py</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
<p>A quick default configuration can be made too:</p>
<div class="sourceCode" id="cb6"><pre class="sourceCode numberSource bash number-lines code-with-copy"><code class="sourceCode bash"><span id="cb6-1"><a href="#cb6-1"></a><span class="ex">accelerate</span> config default</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
</section>
<section id="a-launching-interface-5" class="slide level2">
<h2>A Launching Interface</h2>
<p>With the <code>notebook_launcher</code> it’s also possible to launch code directly from your Jupyter environment too!</p>
<div class="sourceCode" id="cb7"><pre class="sourceCode numberSource python number-lines code-with-copy"><code class="sourceCode python"><span id="cb7-1"><a href="#cb7-1"></a><span class="im">from</span> accelerate <span class="im">import</span> notebook_launcher</span>
<span id="cb7-2"><a href="#cb7-2"></a>notebook_launcher(</span>
<span id="cb7-3"><a href="#cb7-3"></a>    training_loop_function, </span>
<span id="cb7-4"><a href="#cb7-4"></a>    args, </span>
<span id="cb7-5"><a href="#cb7-5"></a>    num_processes<span class="op">=</span><span class="dv">2</span></span>
<span id="cb7-6"><a href="#cb7-6"></a>)</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
<div class="sourceCode" id="cb8"><pre class="sourceCode numberSource python number-lines code-with-copy"><code class="sourceCode python"><span id="cb8-1"><a href="#cb8-1"></a>Launching training on <span class="dv">2</span> GPUs.</span>
<span id="cb8-2"><a href="#cb8-2"></a>epoch <span class="dv">0</span>: <span class="fl">88.12</span></span>
<span id="cb8-3"><a href="#cb8-3"></a>epoch <span class="dv">1</span>: <span class="fl">91.73</span></span>
<span id="cb8-4"><a href="#cb8-4"></a>epoch <span class="dv">2</span>: <span class="fl">92.58</span></span>
<span id="cb8-5"><a href="#cb8-5"></a>epoch <span class="dv">3</span>: <span class="fl">93.90</span></span>
<span id="cb8-6"><a href="#cb8-6"></a>epoch <span class="dv">4</span>: <span class="fl">94.71</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
</section></section>
<section>
<section id="a-training-library" class="title-slide slide level1 center">
<h1>A Training Library</h1>
<p>Okay, will <code>accelerate launch</code> make <code>do_the_thing.py</code> use all my GPUs magically?</p>
</section>
<section id="a-training-library-1" class="slide level2">
<h2>A Training Library</h2>
<ul>
<li>Just showed that its possible using <code>accelerate launch</code> to <em>launch</em> a python script in various distributed environments</li>
<li>This does <em>not</em> mean that the script will just “use” that code and still run on the new compute efficiently.</li>
<li>Training on different computes often means <em>many</em> lines of code changed for each specific compute.</li>
<li>🤗 <code>accelerate</code> solves this by ensuring the same code can be ran on a CPU or GPU, multiples, and on TPUs!</li>
</ul>
</section>
<section id="a-training-library-2" class="slide level2">
<h2>A Training Library</h2>
<div class="sourceCode" id="cb9"><pre class="sourceCode numberSource python number-lines code-with-copy"><code class="sourceCode python"><span id="cb9-1"><a href="#cb9-1"></a><span class="cf">for</span> batch <span class="kw">in</span> dataloader:</span>
<span id="cb9-2"><a href="#cb9-2"></a>    optimizer.zero_grad()</span>
<span id="cb9-3"><a href="#cb9-3"></a>    inputs, targets <span class="op">=</span> batch</span>
<span id="cb9-4"><a href="#cb9-4"></a>    inputs <span class="op">=</span> inputs.to(device)</span>
<span id="cb9-5"><a href="#cb9-5"></a>    targets <span class="op">=</span> targets.to(device)</span>
<span id="cb9-6"><a href="#cb9-6"></a>    outputs <span class="op">=</span> model(inputs)</span>
<span id="cb9-7"><a href="#cb9-7"></a>    loss <span class="op">=</span> loss_function(outputs, targets)</span>
<span id="cb9-8"><a href="#cb9-8"></a>    loss.backward()</span>
<span id="cb9-9"><a href="#cb9-9"></a>    optimizer.step()</span>
<span id="cb9-10"><a href="#cb9-10"></a>    scheduler.step()</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
</section>
<section id="a-training-library-3" class="slide level2 smaller">
<h2>A Training Library</h2>
<div class="columns">
<div class="column" style="width:43%;">
<p><br><br><br></p>
<div class="sourceCode" id="cb10" data-code-line-numbers="5-6,9"><pre class="sourceCode numberSource python number-lines code-with-copy"><code class="sourceCode python"><span id="cb10-1"><a href="#cb10-1"></a><span class="co"># For alignment purposes</span></span>
<span id="cb10-2"><a href="#cb10-2"></a><span class="cf">for</span> batch <span class="kw">in</span> dataloader:</span>
<span id="cb10-3"><a href="#cb10-3"></a>    optimizer.zero_grad()</span>
<span id="cb10-4"><a href="#cb10-4"></a>    inputs, targets <span class="op">=</span> batch</span>
<span id="cb10-5"><a href="#cb10-5"></a>    inputs <span class="op">=</span> inputs.to(device)</span>
<span id="cb10-6"><a href="#cb10-6"></a>    targets <span class="op">=</span> targets.to(device)</span>
<span id="cb10-7"><a href="#cb10-7"></a>    outputs <span class="op">=</span> model(inputs)</span>
<span id="cb10-8"><a href="#cb10-8"></a>    loss <span class="op">=</span> loss_function(outputs, targets)</span>
<span id="cb10-9"><a href="#cb10-9"></a>    loss.backward()</span>
<span id="cb10-10"><a href="#cb10-10"></a>    optimizer.step()</span>
<span id="cb10-11"><a href="#cb10-11"></a>    scheduler.step()</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
</div><div class="column" style="width:57%;">
<div class="sourceCode" id="cb11" data-code-line-numbers="1-7,12-13,16"><pre class="sourceCode numberSource python number-lines code-with-copy"><code class="sourceCode python"><span id="cb11-1"><a href="#cb11-1"></a><span class="im">from</span> accelerate <span class="im">import</span> Accelerator</span>
<span id="cb11-2"><a href="#cb11-2"></a>accelerator <span class="op">=</span> Accelerator()</span>
<span id="cb11-3"><a href="#cb11-3"></a>dataloader, model, optimizer scheduler <span class="op">=</span> (</span>
<span id="cb11-4"><a href="#cb11-4"></a>    accelerator.prepare(</span>
<span id="cb11-5"><a href="#cb11-5"></a>        dataloader, model, optimizer, scheduler</span>
<span id="cb11-6"><a href="#cb11-6"></a>    )</span>
<span id="cb11-7"><a href="#cb11-7"></a>)</span>
<span id="cb11-8"><a href="#cb11-8"></a></span>
<span id="cb11-9"><a href="#cb11-9"></a><span class="cf">for</span> batch <span class="kw">in</span> dataloader:</span>
<span id="cb11-10"><a href="#cb11-10"></a>    optimizer.zero_grad()</span>
<span id="cb11-11"><a href="#cb11-11"></a>    inputs, targets <span class="op">=</span> batch</span>
<span id="cb11-12"><a href="#cb11-12"></a>    <span class="co"># inputs = inputs.to(device)</span></span>
<span id="cb11-13"><a href="#cb11-13"></a>    <span class="co"># targets = targets.to(device)</span></span>
<span id="cb11-14"><a href="#cb11-14"></a>    outputs <span class="op">=</span> model(inputs)</span>
<span id="cb11-15"><a href="#cb11-15"></a>    loss <span class="op">=</span> loss_function(outputs, targets)</span>
<span id="cb11-16"><a href="#cb11-16"></a>    accelerator.backward(loss) <span class="co"># loss.backward()</span></span>
<span id="cb11-17"><a href="#cb11-17"></a>    optimizer.step()</span>
<span id="cb11-18"><a href="#cb11-18"></a>    scheduler.step()</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
</div>
</div>
</section>
<section id="a-training-library-4" class="slide level2">
<h2>A Training Library</h2>
<p>What all happened in <code>Accelerator.prepare</code>?</p>
<div>
<ol type="1">
<li class="fragment"><code>Accelerator</code> looked at the configuration</li>
<li class="fragment">The <code>dataloader</code> was converted into one that can dispatch each batch onto a seperate GPU</li>
<li class="fragment">The <code>model</code> was wrapped with the appropriate DDP wrapper from either <code>torch.distributed</code> or <code>torch_xla</code></li>
<li class="fragment">The <code>optimizer</code> and <code>scheduler</code> were both converted into an <code>AcceleratedOptimizer</code> and <code>AcceleratedScheduler</code> which knows how to handle any distributed scenario</li>
</ol>
</div>
</section>
<section id="a-training-library-mixed-precision" class="slide level2">
<h2>A Training Library, Mixed Precision</h2>
<p>🤗 <code>accelerate</code> also supports <em>automatic mixed precision</em>.</p>
<p>Through a single flag to the <code>Accelerator</code> object when calling <code>accelerator.backward()</code> the mixed precision of your choosing (such as <code>bf16</code> or <code>fp16</code>) will be applied:</p>
<div class="sourceCode" id="cb12" data-code-line-numbers="2,9"><pre class="sourceCode numberSource python number-lines code-with-copy"><code class="sourceCode python"><span id="cb12-1"><a href="#cb12-1"></a><span class="im">from</span> accelerate <span class="im">import</span> Accelerator</span>
<span id="cb12-2"><a href="#cb12-2"></a>accelerator <span class="op">=</span> Accelerator(mixed_precision<span class="op">=</span><span class="st">"fp16"</span>)</span>
<span id="cb12-3"><a href="#cb12-3"></a>...</span>
<span id="cb12-4"><a href="#cb12-4"></a><span class="cf">for</span> batch <span class="kw">in</span> dataloader:</span>
<span id="cb12-5"><a href="#cb12-5"></a>    optimizer.zero_grad()</span>
<span id="cb12-6"><a href="#cb12-6"></a>    inputs, targets <span class="op">=</span> batch</span>
<span id="cb12-7"><a href="#cb12-7"></a>    outputs <span class="op">=</span> model(inputs)</span>
<span id="cb12-8"><a href="#cb12-8"></a>    loss <span class="op">=</span> loss_function(outputs, targets)</span>
<span id="cb12-9"><a href="#cb12-9"></a>    accelerator.backward(loss)</span>
<span id="cb12-10"><a href="#cb12-10"></a>    optimizer.step()</span>
<span id="cb12-11"><a href="#cb12-11"></a>    scheduler.step()</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
</section>
<section id="a-training-library-gradient-accumulation" class="slide level2">
<h2>A Training Library, Gradient Accumulation</h2>
<p>Gradient accumulation in distributed setups often need extra care to ensure gradients are aligned when they need to be and the backward pass is computationally efficient.</p>
<p>🤗 <code>accelerate</code> can just easily handle this for you:</p>
<div class="sourceCode" id="cb13" data-code-line-numbers="2,5"><pre class="sourceCode numberSource python number-lines code-with-copy"><code class="sourceCode python"><span id="cb13-1"><a href="#cb13-1"></a><span class="im">from</span> accelerate <span class="im">import</span> Accelerator</span>
<span id="cb13-2"><a href="#cb13-2"></a>accelerator <span class="op">=</span> Accelerator(gradient_accumulation_steps<span class="op">=</span><span class="dv">4</span>)</span>
<span id="cb13-3"><a href="#cb13-3"></a>...</span>
<span id="cb13-4"><a href="#cb13-4"></a><span class="cf">for</span> batch <span class="kw">in</span> dataloader:</span>
<span id="cb13-5"><a href="#cb13-5"></a>    <span class="cf">with</span> accelerator.accumulate(model):</span>
<span id="cb13-6"><a href="#cb13-6"></a>        optimizer.zero_grad()</span>
<span id="cb13-7"><a href="#cb13-7"></a>        inputs, targets <span class="op">=</span> batch</span>
<span id="cb13-8"><a href="#cb13-8"></a>        outputs <span class="op">=</span> model(inputs)</span>
<span id="cb13-9"><a href="#cb13-9"></a>        loss <span class="op">=</span> loss_function(outputs, targets)</span>
<span id="cb13-10"><a href="#cb13-10"></a>        accelerator.backward(loss)</span>
<span id="cb13-11"><a href="#cb13-11"></a>        optimizer.step()</span>
<span id="cb13-12"><a href="#cb13-12"></a>        scheduler.step()</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
</section>
<section id="a-training-library-gradient-accumulation-1" class="slide level2">
<h2>A Training Library, Gradient Accumulation</h2>
<div class="sourceCode" id="cb14" data-code-line-numbers="5-7,10,11,12,15"><pre class="sourceCode numberSource python number-lines code-with-copy"><code class="sourceCode python"><span id="cb14-1"><a href="#cb14-1"></a>ddp_model, dataloader <span class="op">=</span> accelerator.prepare(model, dataloader)</span>
<span id="cb14-2"><a href="#cb14-2"></a></span>
<span id="cb14-3"><a href="#cb14-3"></a><span class="cf">for</span> index, batch <span class="kw">in</span> <span class="bu">enumerate</span>(dataloader):</span>
<span id="cb14-4"><a href="#cb14-4"></a>    inputs, targets <span class="op">=</span> batch</span>
<span id="cb14-5"><a href="#cb14-5"></a>    <span class="cf">if</span> index <span class="op">!=</span> (<span class="bu">len</span>(dataloader)<span class="op">-</span><span class="dv">1</span>) <span class="kw">or</span> (index <span class="op">%</span> <span class="dv">4</span>) <span class="op">!=</span> <span class="dv">0</span>:</span>
<span id="cb14-6"><a href="#cb14-6"></a>        <span class="co"># Gradients don't sync</span></span>
<span id="cb14-7"><a href="#cb14-7"></a>        <span class="cf">with</span> accelerator.no_sync(model):</span>
<span id="cb14-8"><a href="#cb14-8"></a>            outputs <span class="op">=</span> ddp_model(inputs)</span>
<span id="cb14-9"><a href="#cb14-9"></a>            loss <span class="op">=</span> loss_func(outputs, targets)</span>
<span id="cb14-10"><a href="#cb14-10"></a>            accelerator.backward(loss)</span>
<span id="cb14-11"><a href="#cb14-11"></a>    <span class="cf">else</span>:</span>
<span id="cb14-12"><a href="#cb14-12"></a>        <span class="co"># Gradients finally sync</span></span>
<span id="cb14-13"><a href="#cb14-13"></a>        outputs <span class="op">=</span> ddp_model(inputs)</span>
<span id="cb14-14"><a href="#cb14-14"></a>        loss <span class="op">=</span> loss_func(outputs)</span>
<span id="cb14-15"><a href="#cb14-15"></a>        accelerator.backward(loss)</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
</section></section>
<section>
<section id="big-model-inference" class="title-slide slide level1 center">
<h1>Big Model Inference</h1>
<p>Stable Diffusion taking the world by storm</p>
</section>
<section id="bigger-models-higher-compute" class="slide level2">
<h2>Bigger Models == Higher Compute</h2>
<p>As more large models were being released, Hugging Face quickly realized there must be a way to continue our decentralization of Machine Learning and have the day-to-day programmer be able to leverage these big models.</p>
<p>Born out of this effort by Sylvain Gugger:</p>
<p>🤗 Accelerate: Big Model Inference.</p>
</section>
<section id="the-basic-premise" class="slide level2">
<h2>The Basic Premise</h2>
<div>
<ul>
<li class="fragment"><p>In PyTorch, there exists the <code>meta</code> device.</p></li>
<li class="fragment"><p>Super small footprint to load in huge models quickly by not loading in their weights immediatly.</p></li>
<li class="fragment"><p>As an input gets passed through each layer, we can load and unload <em>parts</em> of the PyTorch model quickly so that only a small portion of the big model is loaded in at a single time.</p></li>
<li class="fragment"><p>The end result? Stable Diffusion v1 can be ran on &lt; 800mb of vRAM</p></li>
</ul>
</div>
</section>
<section id="the-code" class="slide level2">
<h2>The Code</h2>
<p>Generally you start with something like so:</p>
<div class="sourceCode" id="cb15"><pre class="sourceCode numberSource python number-lines code-with-copy"><code class="sourceCode python"><span id="cb15-1"><a href="#cb15-1"></a><span class="im">import</span> torch</span>
<span id="cb15-2"><a href="#cb15-2"></a></span>
<span id="cb15-3"><a href="#cb15-3"></a>my_model <span class="op">=</span> ModelClass(...)</span>
<span id="cb15-4"><a href="#cb15-4"></a>state_dict <span class="op">=</span> torch.load(checkpoint_file)</span>
<span id="cb15-5"><a href="#cb15-5"></a>my_model.load_state_dict(state_dict)</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
<p>But this has issues:</p>
<ol type="1">
<li>The full version of the model is loaded at <code>3</code></li>
<li>Another version of the model is loaded into memory at <code>4</code></li>
</ol>
<p>If a 6 <em>billion</em> parameter model is being loaded, each model class has a dictionary of 24GB so 48GB of vRAM is needed</p>
</section>
<section id="empty-model-weights" class="slide level2">
<h2>Empty Model Weights</h2>
<p>We can fix step 1 by loading in an empty model skeleton at first:</p>
<div class="sourceCode" id="cb16" data-code-line-numbers="1,3-4"><pre class="sourceCode numberSource python number-lines code-with-copy"><code class="sourceCode python"><span id="cb16-1"><a href="#cb16-1"></a><span class="im">from</span> accelerate <span class="im">import</span> init_empty_weights</span>
<span id="cb16-2"><a href="#cb16-2"></a></span>
<span id="cb16-3"><a href="#cb16-3"></a><span class="cf">with</span> init_empty_weights():</span>
<span id="cb16-4"><a href="#cb16-4"></a>    my_model <span class="op">=</span> ModelClass(...)</span>
<span id="cb16-5"><a href="#cb16-5"></a>state_dict <span class="op">=</span> torch.load(checkpoint_file)</span>
<span id="cb16-6"><a href="#cb16-6"></a>my_model.load_state_dict(state_dict)</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
<div class="callout callout-important callout-captioned callout-style-default">
<div class="callout-body">
<div class="callout-caption">
<div class="callout-icon-container">
<i class="callout-icon"></i>
</div>
<p><strong>This code will not run</strong></p>
</div>
<div class="callout-content">
<p>It is likely that just calling <code>my_model(x)</code> will fail as not all tensor operations are supported on the <code>meta</code> device.</p>
</div>
</div>
</div>
</section>
<section id="sharded-checkpoints---the-concept" class="slide level2">
<h2>Sharded Checkpoints - The Concept</h2>
<p>The next step is to have “Sharded Checkpoints” saved for your model.</p>
<p>Basically smaller chunks of your model weights stored that can be brought in at any particular time.</p>
<p>This reduces the amount of memory step 2 takes in since we can just load in a “chunk” of the model at a time, then swap it out for a new chunk through PyTorch hooks</p>
</section>
<section id="sharded-checkpoints---the-code" class="slide level2">
<h2>Sharded Checkpoints - The Code</h2>
<div class="sourceCode" id="cb17" data-code-line-numbers="1,6-8"><pre class="sourceCode numberSource python number-lines code-with-copy"><code class="sourceCode python"><span id="cb17-1"><a href="#cb17-1"></a><span class="im">from</span> accelerate <span class="im">import</span> init_empty_weights, load_checkpoint_and_dispatch</span>
<span id="cb17-2"><a href="#cb17-2"></a></span>
<span id="cb17-3"><a href="#cb17-3"></a><span class="cf">with</span> init_empty_weights():</span>
<span id="cb17-4"><a href="#cb17-4"></a>    my_model <span class="op">=</span> ModelClass(...)</span>
<span id="cb17-5"><a href="#cb17-5"></a></span>
<span id="cb17-6"><a href="#cb17-6"></a>my_model <span class="op">=</span> load_checkpoint_and_dispatch(</span>
<span id="cb17-7"><a href="#cb17-7"></a>    my_model, <span class="st">"sharded-weights"</span>, device_map<span class="op">=</span><span class="st">"auto"</span></span>
<span id="cb17-8"><a href="#cb17-8"></a>)</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
<p><code>device_map="auto"</code> will tell 🤗 Accelerate that it should determine where to put each layer of the model:</p>
<ol type="1">
<li>Maximum space on the GPU(s)</li>
<li>Maximum space on the CPU(s)</li>
<li>Utilize disk space through memory-mapped tensors</li>
</ol>
</section>
<section id="big-model-inference-put-together" class="slide level2">
<h2>Big Model Inference Put Together</h2>
<div class="sourceCode" id="cb18"><pre class="sourceCode numberSource python number-lines code-with-copy"><code class="sourceCode python"><span id="cb18-1"><a href="#cb18-1"></a><span class="im">from</span> accelerate <span class="im">import</span> init_empty_weights, load_checkpoint_and_dispatch</span>
<span id="cb18-2"><a href="#cb18-2"></a></span>
<span id="cb18-3"><a href="#cb18-3"></a><span class="cf">with</span> init_empty_weights():</span>
<span id="cb18-4"><a href="#cb18-4"></a>    my_model <span class="op">=</span> ModelClass(...)</span>
<span id="cb18-5"><a href="#cb18-5"></a></span>
<span id="cb18-6"><a href="#cb18-6"></a>my_model <span class="op">=</span> load_checkpoint_and_dispatch(</span>
<span id="cb18-7"><a href="#cb18-7"></a>    my_model, <span class="st">"sharded-weights"</span>, device_map<span class="op">=</span><span class="st">"auto"</span></span>
<span id="cb18-8"><a href="#cb18-8"></a>)</span>
<span id="cb18-9"><a href="#cb18-9"></a>my_model.<span class="bu">eval</span>()</span>
<span id="cb18-10"><a href="#cb18-10"></a></span>
<span id="cb18-11"><a href="#cb18-11"></a><span class="cf">for</span> batch <span class="kw">in</span> dataloader:</span>
<span id="cb18-12"><a href="#cb18-12"></a>    output <span class="op">=</span> my_model(batch)</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
</section>
<section id="is-there-an-easier-way" class="slide level2">
<h2>Is there an easier way?</h2>
<p>The <code>transformers</code> library combined with the Hub makes all this code wrapping much easier for you with the <code>pipeline</code></p>
<div class="sourceCode" id="cb19"><pre class="sourceCode numberSource python number-lines code-with-copy"><code class="sourceCode python"><span id="cb19-1"><a href="#cb19-1"></a><span class="im">import</span> torch</span>
<span id="cb19-2"><a href="#cb19-2"></a><span class="im">from</span> transformers <span class="im">import</span> pipeline</span>
<span id="cb19-3"><a href="#cb19-3"></a>pipe <span class="op">=</span> pipeline(</span>
<span id="cb19-4"><a href="#cb19-4"></a>    task<span class="op">=</span><span class="st">"text-generation"</span>,</span>
<span id="cb19-5"><a href="#cb19-5"></a>    model<span class="op">=</span><span class="st">"EleutherAI/gpt-j-6B"</span>,</span>
<span id="cb19-6"><a href="#cb19-6"></a>    device_map<span class="op">=</span><span class="st">"auto"</span>,</span>
<span id="cb19-7"><a href="#cb19-7"></a>    torch_dtype<span class="op">=</span>torch.float16</span>
<span id="cb19-8"><a href="#cb19-8"></a>)</span>
<span id="cb19-9"><a href="#cb19-9"></a></span>
<span id="cb19-10"><a href="#cb19-10"></a>text <span class="op">=</span> pipe(<span class="st">"This is some generated text, I think"</span>)</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
</section></section>
<section>
<section id="what-about-stable-diffusion" class="title-slide slide level1 center">
<h1>What about Stable Diffusion?</h1>
<p>A demo with <code>diffusers</code></p>
</section>
<section id="some-handy-resources" class="slide level2">
<h2>Some Handy Resources</h2>
<ul>
<li><a href="https://hf.co/docs/accelerate">🤗 Accelerate documentation</a></li>
<li><a href="https://huggingface.co/docs/accelerate/basic_tutorials/launch">Launching distributed code</a></li>
<li><a href="https://huggingface.co/docs/accelerate/basic_tutorials/notebook">Distributed code and Jupyter Notebooks</a></li>
<li><a href="https://huggingface.co/docs/accelerate/basic_tutorials/migration">Migrating to 🤗 Accelerate easily</a></li>
<li><a href="https://huggingface.co/docs/accelerate/usage_guides/big_modeling">Big Model Inference tutorial</a></li>
<li><a href="https://huggingface.co/docs/accelerate/usage_guides/deepspeed">DeepSpeed and 🤗 Accelerate</a></li>
<li><a href="https://huggingface.co/docs/accelerate/usage_guides/fsdp">Fully Sharded Data Parallelism and 🤗 Accelerate</a></li>
</ul>
<div class="footer footer-default">

</div>
</section></section>

    </div>
  </div>

  <script>window.backupDefine = window.define; window.define = undefined;</script>
  <script src="Accelerate_files/libs/revealjs/dist/reveal.js"></script>
  <!-- reveal.js plugins -->
  <script src="Accelerate_files/libs/revealjs/plugin/quarto-line-highlight/line-highlight.js"></script>
  <script src="Accelerate_files/libs/revealjs/plugin/pdf-export/pdfexport.js"></script>
  <script src="Accelerate_files/libs/revealjs/plugin/reveal-menu/menu.js"></script>
  <script src="Accelerate_files/libs/revealjs/plugin/reveal-menu/quarto-menu.js"></script>
  <script src="Accelerate_files/libs/revealjs/plugin/quarto-support/support.js"></script>
  

  <script src="Accelerate_files/libs/revealjs/plugin/notes/notes.js"></script>
  <script src="Accelerate_files/libs/revealjs/plugin/search/search.js"></script>
  <script src="Accelerate_files/libs/revealjs/plugin/zoom/zoom.js"></script>
  <script src="Accelerate_files/libs/revealjs/plugin/math/math.js"></script>
  <script>window.define = window.backupDefine; window.backupDefine = undefined;</script>

  <script>

      // Full list of configuration options available at:
      // https://revealjs.com/config/
      Reveal.initialize({
'controlsAuto': true,
'previewLinksAuto': false,
'smaller': false,
'pdfSeparateFragments': false,
'autoAnimateEasing': "ease",
'autoAnimateDuration': 1,
'autoAnimateUnmatched': true,
'menu': {"side":"left","useTextContentForMissingTitles":true,"markers":false,"loadIcons":false,"custom":[{"title":"Tools","icon":"<i class=\"fas fa-gear\"></i>","content":"<ul class=\"slide-menu-items\">\n<li class=\"slide-tool-item active\" data-item=\"0\"><a href=\"#\" onclick=\"RevealMenuToolHandlers.fullscreen(event)\"><kbd>f</kbd> Fullscreen</a></li>\n<li class=\"slide-tool-item\" data-item=\"1\"><a href=\"#\" onclick=\"RevealMenuToolHandlers.speakerMode(event)\"><kbd>s</kbd> Speaker View</a></li>\n<li class=\"slide-tool-item\" data-item=\"2\"><a href=\"#\" onclick=\"RevealMenuToolHandlers.overview(event)\"><kbd>o</kbd> Slide Overview</a></li>\n<li class=\"slide-tool-item\" data-item=\"3\"><a href=\"#\" onclick=\"RevealMenuToolHandlers.overview(event)\"><kbd>e</kbd> PDF Export Mode</a></li>\n<li class=\"slide-tool-item\" data-item=\"4\"><a href=\"#\" onclick=\"RevealMenuToolHandlers.keyboardHelp(event)\"><kbd>?</kbd> Keyboard Help</a></li>\n</ul>"}],"openButton":true},
'smaller': false,
 
        // Display controls in the bottom right corner
        controls: false,

        // Help the user learn the controls by providing hints, for example by
        // bouncing the down arrow when they first encounter a vertical slide
        controlsTutorial: false,

        // Determines where controls appear, "edges" or "bottom-right"
        controlsLayout: 'edges',

        // Visibility rule for backwards navigation arrows; "faded", "hidden"
        // or "visible"
        controlsBackArrows: 'faded',

        // Display a presentation progress bar
        progress: true,

        // Display the page number of the current slide
        slideNumber: false,

        // 'all', 'print', or 'speaker'
        showSlideNumber: 'all',

        // Add the current slide number to the URL hash so that reloading the
        // page/copying the URL will return you to the same slide
        hash: true,

        // Start with 1 for the hash rather than 0
        hashOneBasedIndex: false,

        // Flags if we should monitor the hash and change slides accordingly
        respondToHashChanges: true,

        // Push each slide change to the browser history
        history: true,

        // Enable keyboard shortcuts for navigation
        keyboard: true,

        // Enable the slide overview mode
        overview: true,

        // Disables the default reveal.js slide layout (scaling and centering)
        // so that you can use custom CSS layout
        disableLayout: false,

        // Vertical centering of slides
        center: false,

        // Enables touch navigation on devices with touch input
        touch: true,

        // Loop the presentation
        loop: false,

        // Change the presentation direction to be RTL
        rtl: false,

        // see https://revealjs.com/vertical-slides/#navigation-mode
        navigationMode: 'linear',

        // Randomizes the order of slides each time the presentation loads
        shuffle: false,

        // Turns fragments on and off globally
        fragments: true,

        // Flags whether to include the current fragment in the URL,
        // so that reloading brings you to the same fragment position
        fragmentInURL: false,

        // Flags if the presentation is running in an embedded mode,
        // i.e. contained within a limited portion of the screen
        embedded: false,

        // Flags if we should show a help overlay when the questionmark
        // key is pressed
        help: true,

        // Flags if it should be possible to pause the presentation (blackout)
        pause: true,

        // Flags if speaker notes should be visible to all viewers
        showNotes: false,

        // Global override for autoplaying embedded media (null/true/false)
        autoPlayMedia: null,

        // Global override for preloading lazy-loaded iframes (null/true/false)
        preloadIframes: null,

        // Number of milliseconds between automatically proceeding to the
        // next slide, disabled when set to 0, this value can be overwritten
        // by using a data-autoslide attribute on your slides
        autoSlide: 0,

        // Stop auto-sliding after user input
        autoSlideStoppable: true,

        // Use this method for navigation when auto-sliding
        autoSlideMethod: null,

        // Specify the average time in seconds that you think you will spend
        // presenting each slide. This is used to show a pacing timer in the
        // speaker view
        defaultTiming: null,

        // Enable slide navigation via mouse wheel
        mouseWheel: false,

        // The display mode that will be used to show slides
        display: 'block',

        // Hide cursor if inactive
        hideInactiveCursor: true,

        // Time before the cursor is hidden (in ms)
        hideCursorTime: 5000,

        // Opens links in an iframe preview overlay
        previewLinks: false,

        // Transition style (none/fade/slide/convex/concave/zoom)
        transition: 'none',

        // Transition speed (default/fast/slow)
        transitionSpeed: 'default',

        // Transition style for full page slide backgrounds
        // (none/fade/slide/convex/concave/zoom)
        backgroundTransition: 'none',

        // Number of slides away from the current that are visible
        viewDistance: 3,

        // Number of slides away from the current that are visible on mobile
        // devices. It is advisable to set this to a lower number than
        // viewDistance in order to save resources.
        mobileViewDistance: 2,

        // The "normal" size of the presentation, aspect ratio will be preserved
        // when the presentation is scaled to fit different resolutions. Can be
        // specified using percentage units.
        width: 1050,

        height: 700,

        // Factor of the display size that should remain empty around the content
        margin: 0.1,

        math: {
          mathjax: 'https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.0/MathJax.js',
          config: 'TeX-AMS_HTML-full',
          tex2jax: {
            inlineMath: [['\\(','\\)']],
            displayMath: [['\\[','\\]']],
            balanceBraces: true,
            processEscapes: false,
            processRefs: true,
            processEnvironments: true,
            preview: 'TeX',
            skipTags: ['script','noscript','style','textarea','pre','code'],
            ignoreClass: 'tex2jax_ignore',
            processClass: 'tex2jax_process'
          },
        },

        // reveal.js plugins
        plugins: [QuartoLineHighlight, PdfExport, RevealMenu, QuartoSupport,

          RevealMath,
          RevealNotes,
          RevealSearch,
          RevealZoom
        ]
      });
    </script>
    <script id="quarto-html-after-body" type="application/javascript">
    window.document.addEventListener("DOMContentLoaded", function (event) {
      const toggleBodyColorMode = (bsSheetEl) => {
        const mode = bsSheetEl.getAttribute("data-mode");
        const bodyEl = window.document.querySelector("body");
        if (mode === "dark") {
          bodyEl.classList.add("quarto-dark");
          bodyEl.classList.remove("quarto-light");
        } else {
          bodyEl.classList.add("quarto-light");
          bodyEl.classList.remove("quarto-dark");
        }
      }
      const toggleBodyColorPrimary = () => {
        const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
        if (bsSheetEl) {
          toggleBodyColorMode(bsSheetEl);
        }
      }
      toggleBodyColorPrimary();  
      const tabsets =  window.document.querySelectorAll(".panel-tabset-tabby")
      tabsets.forEach(function(tabset) {
        const tabby = new Tabby('#' + tabset.id);
      });
      const clipboard = new window.ClipboardJS('.code-copy-button', {
        target: function(trigger) {
          return trigger.previousElementSibling;
        }
      });
      clipboard.on('success', function(e) {
        // button target
        const button = e.trigger;
        // don't keep focus
        button.blur();
        // flash "checked"
        button.classList.add('code-copy-button-checked');
        var currentTitle = button.getAttribute("title");
        button.setAttribute("title", "Copied!");
        let tooltip;
        if (window.bootstrap) {
          button.setAttribute("data-bs-toggle", "tooltip");
          button.setAttribute("data-bs-placement", "left");
          button.setAttribute("data-bs-title", "Copied!");
          tooltip = new bootstrap.Tooltip(button, 
            { trigger: "manual", 
              customClass: "code-copy-button-tooltip",
              offset: [0, -8]});
          tooltip.show();    
        }
        setTimeout(function() {
          if (tooltip) {
            tooltip.hide();
            button.removeAttribute("data-bs-title");
            button.removeAttribute("data-bs-toggle");
            button.removeAttribute("data-bs-placement");
          }
          button.setAttribute("title", currentTitle);
          button.classList.remove('code-copy-button-checked');
        }, 1000);
        // clear code selection
        e.clearSelection();
      });
      function tippyHover(el, contentFn) {
        const config = {
          allowHTML: true,
          content: contentFn,
          maxWidth: 500,
          delay: 100,
          arrow: false,
          appendTo: function(el) {
              return el.closest('section.slide') || el.parentElement;
          },
          interactive: true,
          interactiveBorder: 10,
          theme: 'quarto-reveal',
          placement: 'bottom-start'
        };
          config['offset'] = [0,0];
          config['maxWidth'] = 700;
        window.tippy(el, config); 
      }
      const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
      for (var i=0; i<noterefs.length; i++) {
        const ref = noterefs[i];
        tippyHover(ref, function() {
          // use id or data attribute instead here
          let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
          try { href = new URL(href).hash; } catch {}
          const id = href.replace(/^#\/?/, "");
          const note = window.document.getElementById(id);
          return note.innerHTML;
        });
      }
      const findCites = (el) => {
        const parentEl = el.parentElement;
        if (parentEl) {
          const cites = parentEl.dataset.cites;
          if (cites) {
            return {
              el,
              cites: cites.split(' ')
            };
          } else {
            return findCites(el.parentElement)
          }
        } else {
          return undefined;
        }
      };
      var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
      for (var i=0; i<bibliorefs.length; i++) {
        const ref = bibliorefs[i];
        const citeInfo = findCites(ref);
        if (citeInfo) {
          tippyHover(citeInfo.el, function() {
            var popup = window.document.createElement('div');
            citeInfo.cites.forEach(function(cite) {
              var citeDiv = window.document.createElement('div');
              citeDiv.classList.add('hanging-indent');
              citeDiv.classList.add('csl-entry');
              var biblioDiv = window.document.getElementById('ref-' + cite);
              if (biblioDiv) {
                citeDiv.innerHTML = biblioDiv.innerHTML;
              }
              popup.appendChild(citeDiv);
            });
            return popup.innerHTML;
          });
        }
      }
    });
    </script>
    

</body></html>