-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathcheckpoint.html
1195 lines (1009 loc) · 77.4 KB
/
checkpoint.html
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
<!DOCTYPE html>
<!--[if IE 8]><html class="no-js lt-ie9" lang="en" > <![endif]-->
<!--[if gt IE 8]><!--> <html class="no-js" lang="en" > <!--<![endif]-->
<head>
<meta name="robots" content="noindex">
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>torch.utils.checkpoint — PyTorch 2.5 documentation</title>
<link rel="canonical" href="https://pytorch.org/docs/stable/checkpoint.html"/>
<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/[email protected]/dist/katex.min.css" type="text/css" />
<link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/katex.min.css" type="text/css" />
<link rel="stylesheet" href="_static/katex-math.css" type="text/css" />
<link rel="stylesheet" href="_static/sphinx-dropdown.css" type="text/css" />
<link rel="stylesheet" href="_static/panels-bootstrap.min.css" type="text/css" />
<link rel="stylesheet" href="_static/css/jit.css" type="text/css" />
<link rel="index" title="Index" href="genindex.html" />
<link rel="search" title="Search" href="search.html" />
<link rel="next" title="torch.utils.cpp_extension" href="cpp_extension.html" />
<link rel="prev" title="torch.utils.bottleneck" href="bottleneck.html" />
<!-- Google Tag Manager -->
<script>(function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':
new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],
j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src=
'https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);
})(window,document,'script','dataLayer','GTM-T8XT4PS');</script>
<!-- End Google Tag Manager -->
<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/[email protected]/dist/fonts/KaTeX_Math-Italic.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/fonts/KaTeX_Main-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/fonts/KaTeX_Main-Bold.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/fonts/KaTeX_Size1-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/fonts/KaTeX_Size4-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/fonts/KaTeX_Size2-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/fonts/KaTeX_Size3-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/[email protected]/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.org/" aria-label="PyTorch"></a>
<div class="main-menu">
<ul>
<li class="main-menu-item">
<div id="resourcesDropdownButton" data-toggle="resources-dropdown" class="resources-dropdown">
<a class="with-down-arrow">
Learn
</a>
<div class="resources-dropdown-menu">
<a class="nav-dropdown-item" href="https://pytorch.org/get-started">
<span class=dropdown-title>Get Started</span>
<p>Run PyTorch locally or get started quickly with one of the supported cloud platforms</p>
</a>
<a class="nav-dropdown-item" href="https://pytorch.org/tutorials">
<span class="dropdown-title">Tutorials</span>
<p>Whats new in PyTorch tutorials</p>
</a>
<a class="nav-dropdown-item" href="https://pytorch.org/tutorials/beginner/basics/intro.html">
<span class="dropdown-title">Learn the Basics</span>
<p>Familiarize yourself with PyTorch concepts and modules</p>
</a>
<a class="nav-dropdown-item" href="https://pytorch.org/tutorials/recipes/recipes_index.html">
<span class="dropdown-title">PyTorch Recipes</span>
<p>Bite-size, ready-to-deploy PyTorch code examples</p>
</a>
<a class="nav-dropdown-item" href="https://pytorch.org/tutorials/beginner/introyt.html">
<span class="dropdown-title">Intro to PyTorch - YouTube Series</span>
<p>Master PyTorch basics with our engaging YouTube tutorial series</p>
</a>
</div>
</div>
</li>
<li>
<div id="resourcesDropdownButton" data-toggle="resources-dropdown" class="resources-dropdown">
<a class="with-down-arrow">
Ecosystem
</a>
<div class="resources-dropdown-menu">
<a class="nav-dropdown-item" href="https://pytorch.org/ecosystem">
<span class="dropdown-title">Tools</span>
<p>Learn about the tools and frameworks in the PyTorch Ecosystem</p>
</a>
<a class="nav-dropdown-item" href="https://pytorch.org/#community-module">
<span class=dropdown-title>Community</span>
<p>Join the PyTorch developer community to contribute, learn, and get your questions answered</p>
</a>
<a class="nav-dropdown-item" href="https://discuss.pytorch.org/" target="_blank">
<span class=dropdown-title>Forums</span>
<p>A place to discuss PyTorch code, issues, install, research</p>
</a>
<a class="nav-dropdown-item" href="https://pytorch.org/resources">
<span class=dropdown-title>Developer Resources</span>
<p>Find resources and get questions answered</p>
</a>
<a class="nav-dropdown-item" href="https://pytorch.org/ecosystem/contributor-awards-2023">
<span class="dropdown-title">Contributor Awards - 2023</span>
<p>Award winners announced at this year's PyTorch Conference</p>
</a>
</div>
</div>
</li>
<li>
<div id="resourcesDropdownButton" data-toggle="resources-dropdown" class="resources-dropdown">
<a class="with-down-arrow">
Edge
</a>
<div class="resources-dropdown-menu">
<a class="nav-dropdown-item" href="https://pytorch.org/edge">
<span class="dropdown-title">About PyTorch Edge</span>
<p>Build innovative and privacy-aware AI experiences for edge devices</p>
</a>
<a class="nav-dropdown-item" href="https://pytorch.org/executorch-overview">
<span class="dropdown-title">ExecuTorch</span>
<p>End-to-end solution for enabling on-device inference capabilities across mobile and edge devices</p>
</a>
</div>
</div>
</li>
<li class="main-menu-item">
<div id="resourcesDropdownButton" data-toggle="resources-dropdown" class="resources-dropdown">
<a class="with-down-arrow">
Docs
</a>
<div class="resources-dropdown-menu">
<a class="nav-dropdown-item" href="https://pytorch.org/docs/stable/index.html">
<span class="dropdown-title">PyTorch</span>
<p>Explore the documentation for comprehensive guidance on how to use PyTorch</p>
</a>
<a class="nav-dropdown-item" href="https://pytorch.org/pytorch-domains">
<span class="dropdown-title">PyTorch Domains</span>
<p>Read the PyTorch Domains documentation to learn more about domain-specific libraries</p>
</a>
</div>
</div>
</li>
<li>
<div id="resourcesDropdownButton" data-toggle="resources-dropdown" class="resources-dropdown">
<a class="with-down-arrow">
Blogs & News
</a>
<div class="resources-dropdown-menu">
<a class="nav-dropdown-item" href="https://pytorch.org/blog/">
<span class="dropdown-title">PyTorch Blog</span>
<p>Catch up on the latest technical news and happenings</p>
</a>
<a class="nav-dropdown-item" href="https://pytorch.org/community-blog">
<span class="dropdown-title">Community Blog</span>
<p>Stories from the PyTorch ecosystem</p>
</a>
<a class="nav-dropdown-item" href="https://pytorch.org/videos">
<span class="dropdown-title">Videos</span>
<p>Learn about the latest PyTorch tutorials, new, and more </p>
<a class="nav-dropdown-item" href="https://pytorch.org/community-stories">
<span class="dropdown-title">Community Stories</span>
<p>Learn how our community solves real, everyday machine learning problems with PyTorch</p>
</a>
<a class="nav-dropdown-item" href="https://pytorch.org/events">
<span class="dropdown-title">Events</span>
<p>Find events, webinars, and podcasts</p>
</a>
</div>
</li>
<li>
<div id="resourcesDropdownButton" data-toggle="resources-dropdown" class="resources-dropdown">
<a class="with-down-arrow">
About
</a>
<div class="resources-dropdown-menu">
<a class="nav-dropdown-item" href="https://pytorch.org/foundation">
<span class="dropdown-title">PyTorch Foundation</span>
<p>Learn more about the PyTorch Foundation</p>
</a>
<a class="nav-dropdown-item" href="https://pytorch.org/governing-board">
<span class="dropdown-title">Governing Board</span>
<p></p>
</a>
</div>
</div>
</li>
<li class="main-menu-item">
<div class="no-dropdown">
<a href="https://pytorch.org/join" data-cta="join">
Become a Member
</a>
</div>
</li>
<li>
<div class="main-menu-item">
<a href="https://github.com/pytorch/pytorch" class="github-icon">
</a>
</div>
</li>
<!--- TODO: This block adds the search icon to the nav bar. We will enable it later.
<li>
<div class="main-menu-item">
<a href="https://github.com/pytorch/pytorch" class="search-icon">
</a>
</div>
</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">
<a href='https://pytorch.org/docs/versions.html'>2.5 ▼</a>
</div>
<div role="search">
<form id="rtd-search-form" class="wy-form" action="search.html" method="get">
<input type="text" name="q" placeholder="Search Docs" />
<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">Community</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="community/build_ci_governance.html">PyTorch Governance | Build + CI</a></li>
<li class="toctree-l1"><a class="reference internal" href="community/contribution_guide.html">PyTorch Contribution Guide</a></li>
<li class="toctree-l1"><a class="reference internal" href="community/design.html">PyTorch Design Philosophy</a></li>
<li class="toctree-l1"><a class="reference internal" href="community/governance.html">PyTorch Governance | Mechanics</a></li>
<li class="toctree-l1"><a class="reference internal" href="community/persons_of_interest.html">PyTorch Governance | Maintainers</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">Developer Notes</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="notes/amp_examples.html">Automatic Mixed Precision examples</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/autograd.html">Autograd mechanics</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/broadcasting.html">Broadcasting semantics</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/cpu_threading_torchscript_inference.html">CPU threading and TorchScript inference</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/cuda.html">CUDA semantics</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/custom_operators.html">PyTorch Custom Operators Landing Page</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/ddp.html">Distributed Data Parallel</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/extending.html">Extending PyTorch</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/extending.func.html">Extending torch.func with autograd.Function</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/faq.html">Frequently Asked Questions</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/fsdp.html">FSDP Notes</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/get_start_xpu.html">Getting Started on Intel GPU</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/gradcheck.html">Gradcheck mechanics</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/hip.html">HIP (ROCm) semantics</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/large_scale_deployments.html">Features for large-scale deployments</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/modules.html">Modules</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/mps.html">MPS backend</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/multiprocessing.html">Multiprocessing best practices</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/numerical_accuracy.html">Numerical accuracy</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/randomness.html">Reproducibility</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/serialization.html">Serialization semantics</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/windows.html">Windows FAQ</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">Language Bindings</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="cpp_index.html">C++</a></li>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/javadoc/">Javadoc</a></li>
<li class="toctree-l1"><a class="reference internal" href="deploy.html">torch::deploy</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">Python API</span></p>
<ul class="current">
<li class="toctree-l1"><a class="reference internal" href="torch.html">torch</a></li>
<li class="toctree-l1"><a class="reference internal" href="nn.html">torch.nn</a></li>
<li class="toctree-l1"><a class="reference internal" href="nn.functional.html">torch.nn.functional</a></li>
<li class="toctree-l1"><a class="reference internal" href="tensors.html">torch.Tensor</a></li>
<li class="toctree-l1"><a class="reference internal" href="tensor_attributes.html">Tensor Attributes</a></li>
<li class="toctree-l1"><a class="reference internal" href="tensor_view.html">Tensor Views</a></li>
<li class="toctree-l1"><a class="reference internal" href="amp.html">torch.amp</a></li>
<li class="toctree-l1"><a class="reference internal" href="autograd.html">torch.autograd</a></li>
<li class="toctree-l1"><a class="reference internal" href="library.html">torch.library</a></li>
<li class="toctree-l1"><a class="reference internal" href="cpu.html">torch.cpu</a></li>
<li class="toctree-l1"><a class="reference internal" href="cuda.html">torch.cuda</a></li>
<li class="toctree-l1"><a class="reference internal" href="torch_cuda_memory.html">Understanding CUDA Memory Usage</a></li>
<li class="toctree-l1"><a class="reference internal" href="torch_cuda_memory.html#generating-a-snapshot">Generating a Snapshot</a></li>
<li class="toctree-l1"><a class="reference internal" href="torch_cuda_memory.html#using-the-visualizer">Using the visualizer</a></li>
<li class="toctree-l1"><a class="reference internal" href="torch_cuda_memory.html#snapshot-api-reference">Snapshot API Reference</a></li>
<li class="toctree-l1"><a class="reference internal" href="mps.html">torch.mps</a></li>
<li class="toctree-l1"><a class="reference internal" href="xpu.html">torch.xpu</a></li>
<li class="toctree-l1"><a class="reference internal" href="mtia.html">torch.mtia</a></li>
<li class="toctree-l1"><a class="reference internal" href="meta.html">Meta device</a></li>
<li class="toctree-l1"><a class="reference internal" href="backends.html">torch.backends</a></li>
<li class="toctree-l1"><a class="reference internal" href="export.html">torch.export</a></li>
<li class="toctree-l1"><a class="reference internal" href="distributed.html">torch.distributed</a></li>
<li class="toctree-l1"><a class="reference internal" href="distributed.tensor.html">torch.distributed.tensor</a></li>
<li class="toctree-l1"><a class="reference internal" href="distributed.algorithms.join.html">torch.distributed.algorithms.join</a></li>
<li class="toctree-l1"><a class="reference internal" href="distributed.elastic.html">torch.distributed.elastic</a></li>
<li class="toctree-l1"><a class="reference internal" href="fsdp.html">torch.distributed.fsdp</a></li>
<li class="toctree-l1"><a class="reference internal" href="distributed.tensor.parallel.html">torch.distributed.tensor.parallel</a></li>
<li class="toctree-l1"><a class="reference internal" href="distributed.optim.html">torch.distributed.optim</a></li>
<li class="toctree-l1"><a class="reference internal" href="distributed.pipelining.html">torch.distributed.pipelining</a></li>
<li class="toctree-l1"><a class="reference internal" href="distributed.checkpoint.html">torch.distributed.checkpoint</a></li>
<li class="toctree-l1"><a class="reference internal" href="distributions.html">torch.distributions</a></li>
<li class="toctree-l1"><a class="reference internal" href="torch.compiler.html">torch.compiler</a></li>
<li class="toctree-l1"><a class="reference internal" href="fft.html">torch.fft</a></li>
<li class="toctree-l1"><a class="reference internal" href="func.html">torch.func</a></li>
<li class="toctree-l1"><a class="reference internal" href="futures.html">torch.futures</a></li>
<li class="toctree-l1"><a class="reference internal" href="fx.html">torch.fx</a></li>
<li class="toctree-l1"><a class="reference internal" href="fx.experimental.html">torch.fx.experimental</a></li>
<li class="toctree-l1"><a class="reference internal" href="hub.html">torch.hub</a></li>
<li class="toctree-l1"><a class="reference internal" href="jit.html">torch.jit</a></li>
<li class="toctree-l1"><a class="reference internal" href="linalg.html">torch.linalg</a></li>
<li class="toctree-l1"><a class="reference internal" href="monitor.html">torch.monitor</a></li>
<li class="toctree-l1"><a class="reference internal" href="signal.html">torch.signal</a></li>
<li class="toctree-l1"><a class="reference internal" href="special.html">torch.special</a></li>
<li class="toctree-l1"><a class="reference internal" href="torch.overrides.html">torch.overrides</a></li>
<li class="toctree-l1"><a class="reference internal" href="package.html">torch.package</a></li>
<li class="toctree-l1"><a class="reference internal" href="profiler.html">torch.profiler</a></li>
<li class="toctree-l1"><a class="reference internal" href="nn.init.html">torch.nn.init</a></li>
<li class="toctree-l1"><a class="reference internal" href="nn.attention.html">torch.nn.attention</a></li>
<li class="toctree-l1"><a class="reference internal" href="onnx.html">torch.onnx</a></li>
<li class="toctree-l1"><a class="reference internal" href="optim.html">torch.optim</a></li>
<li class="toctree-l1"><a class="reference internal" href="complex_numbers.html">Complex Numbers</a></li>
<li class="toctree-l1"><a class="reference internal" href="ddp_comm_hooks.html">DDP Communication Hooks</a></li>
<li class="toctree-l1"><a class="reference internal" href="quantization.html">Quantization</a></li>
<li class="toctree-l1"><a class="reference internal" href="rpc.html">Distributed RPC Framework</a></li>
<li class="toctree-l1"><a class="reference internal" href="random.html">torch.random</a></li>
<li class="toctree-l1"><a class="reference internal" href="masked.html">torch.masked</a></li>
<li class="toctree-l1"><a class="reference internal" href="nested.html">torch.nested</a></li>
<li class="toctree-l1"><a class="reference internal" href="size.html">torch.Size</a></li>
<li class="toctree-l1"><a class="reference internal" href="sparse.html">torch.sparse</a></li>
<li class="toctree-l1"><a class="reference internal" href="storage.html">torch.Storage</a></li>
<li class="toctree-l1"><a class="reference internal" href="testing.html">torch.testing</a></li>
<li class="toctree-l1"><a class="reference internal" href="utils.html">torch.utils</a></li>
<li class="toctree-l1"><a class="reference internal" href="benchmark_utils.html">torch.utils.benchmark</a></li>
<li class="toctree-l1"><a class="reference internal" href="bottleneck.html">torch.utils.bottleneck</a></li>
<li class="toctree-l1 current"><a class="current reference internal" href="#">torch.utils.checkpoint</a></li>
<li class="toctree-l1"><a class="reference internal" href="cpp_extension.html">torch.utils.cpp_extension</a></li>
<li class="toctree-l1"><a class="reference internal" href="data.html">torch.utils.data</a></li>
<li class="toctree-l1"><a class="reference internal" href="deterministic.html">torch.utils.deterministic</a></li>
<li class="toctree-l1"><a class="reference internal" href="jit_utils.html">torch.utils.jit</a></li>
<li class="toctree-l1"><a class="reference internal" href="dlpack.html">torch.utils.dlpack</a></li>
<li class="toctree-l1"><a class="reference internal" href="mobile_optimizer.html">torch.utils.mobile_optimizer</a></li>
<li class="toctree-l1"><a class="reference internal" href="model_zoo.html">torch.utils.model_zoo</a></li>
<li class="toctree-l1"><a class="reference internal" href="tensorboard.html">torch.utils.tensorboard</a></li>
<li class="toctree-l1"><a class="reference internal" href="module_tracker.html">torch.utils.module_tracker</a></li>
<li class="toctree-l1"><a class="reference internal" href="type_info.html">Type Info</a></li>
<li class="toctree-l1"><a class="reference internal" href="named_tensor.html">Named Tensors</a></li>
<li class="toctree-l1"><a class="reference internal" href="name_inference.html">Named Tensors operator coverage</a></li>
<li class="toctree-l1"><a class="reference internal" href="config_mod.html">torch.__config__</a></li>
<li class="toctree-l1"><a class="reference internal" href="future_mod.html">torch.__future__</a></li>
<li class="toctree-l1"><a class="reference internal" href="logging.html">torch._logging</a></li>
<li class="toctree-l1"><a class="reference internal" href="torch_environment_variables.html">Torch Environment Variables</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">Libraries</span></p>
<ul>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/audio/stable">torchaudio</a></li>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/data">TorchData</a></li>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/torchrec">TorchRec</a></li>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/serve">TorchServe</a></li>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/text/stable">torchtext</a></li>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/vision/stable">torchvision</a></li>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/xla/">PyTorch on XLA Devices</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">
Docs
</a> >
</li>
<li>torch.utils.checkpoint</li>
<li class="pytorch-breadcrumbs-aside">
<a href="_sources/checkpoint.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">
<!-- Google Tag Manager (noscript) -->
<noscript><iframe src="https://www.googletagmanager.com/ns.html?id=GTM-T8XT4PS"
height="0" width="0" style="display:none;visibility:hidden"></iframe></noscript>
<!-- End Google Tag Manager (noscript) -->
<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="torch-utils-checkpoint">
<h1>torch.utils.checkpoint<a class="headerlink" href="#torch-utils-checkpoint" title="Permalink to this heading">¶</a></h1>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>Checkpointing is implemented by rerunning a forward-pass segment for
each checkpointed segment during backward propagation. This can cause persistent
states like the RNG state to be more advanced than they would without
checkpointing. By default, checkpointing includes logic to juggle
the RNG state such that checkpointed passes making use of RNG
(through dropout for example) have deterministic output as
compared to non-checkpointed passes. The logic to stash and restore
RNG states can incur a moderate performance hit depending on the runtime
of checkpointed operations. If deterministic output compared to
non-checkpointed passes is not required, supply <code class="docutils literal notranslate"><span class="pre">preserve_rng_state=False</span></code>
to <code class="docutils literal notranslate"><span class="pre">checkpoint</span></code> or <code class="docutils literal notranslate"><span class="pre">checkpoint_sequential</span></code> to omit stashing and
restoring the RNG state during each checkpoint.</p>
<p>The stashing logic saves and restores the RNG state for CPU and another
device type (infer the device type from Tensor arguments excluding CPU
tensors by <code class="docutils literal notranslate"><span class="pre">_infer_device_type</span></code>) to the <code class="docutils literal notranslate"><span class="pre">run_fn</span></code>. If there are multiple
device, device state will only be saved for devices of a single device type,
and the remaining devices will be ignored. Consequently, if any checkpointed
functions involve randomness, this may result in incorrect gradients. (Note
that if CUDA devices are among the devices detected, it will be prioritized;
otherwise, the first device encountered will be selected.) If there are no
CPU-tensors, the default device type state (default value is <cite>cuda</cite>, and it
could be set to other device by <code class="docutils literal notranslate"><span class="pre">DefaultDeviceType</span></code>) will be saved and restored.
However, the logic has no way to anticipate if the user will move
Tensors to a new device within the <code class="docutils literal notranslate"><span class="pre">run_fn</span></code> itself. Therefore, if you move
Tensors to a new device (“new” meaning not belonging to the set of
[current device + devices of Tensor arguments]) within <code class="docutils literal notranslate"><span class="pre">run_fn</span></code>, deterministic
output compared to non-checkpointed passes is never guaranteed.</p>
</div>
<dl class="py function">
<dt class="sig sig-object py" id="torch.utils.checkpoint.checkpoint">
<span class="sig-prename descclassname"><span class="pre">torch.utils.checkpoint.</span></span><span class="sig-name descname"><span class="pre">checkpoint</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">function</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">*args</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">use_reentrant=None</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">context_fn=<function</span> <span class="pre">noop_context_fn></span></span></em>, <em class="sig-param"><span class="n"><span class="pre">determinism_check='default'</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">debug=False</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">**kwargs</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/utils/checkpoint.html#checkpoint"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.utils.checkpoint.checkpoint" title="Permalink to this definition">¶</a></dt>
<dd><p>Checkpoint a model or part of the model.</p>
<p>Activation checkpointing is a technique that trades compute for memory.
Instead of keeping tensors needed for backward alive until they are used in
gradient computation during backward, forward computation in checkpointed
regions omits saving tensors for backward and recomputes them during the
backward pass. Activation checkpointing can be applied to any part of a
model.</p>
<p>There are currently two checkpointing implementations available, determined
by the <code class="xref py py-attr docutils literal notranslate"><span class="pre">use_reentrant</span></code> parameter. It is recommended that you use
<code class="docutils literal notranslate"><span class="pre">use_reentrant=False</span></code>. Please refer the note below for a discussion of
their differences.</p>
<div class="admonition warning">
<p class="admonition-title">Warning</p>
<p>If the <code class="xref py py-attr docutils literal notranslate"><span class="pre">function</span></code> invocation during the backward pass differs
from the forward pass, e.g., due to a global variable, the checkpointed
version may not be equivalent, potentially causing an
error being raised or leading to silently incorrect gradients.</p>
</div>
<div class="admonition warning">
<p class="admonition-title">Warning</p>
<p>The <code class="docutils literal notranslate"><span class="pre">use_reentrant</span></code> parameter should be passed explicitly. In version
2.4 we will raise an exception if <code class="docutils literal notranslate"><span class="pre">use_reentrant</span></code> is not passed.
If you are using the <code class="docutils literal notranslate"><span class="pre">use_reentrant=True</span></code> variant, please refer to the
note below for important considerations and potential limitations.</p>
</div>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>The reentrant variant of checkpoint (<code class="docutils literal notranslate"><span class="pre">use_reentrant=True</span></code>) and
the non-reentrant variant of checkpoint (<code class="docutils literal notranslate"><span class="pre">use_reentrant=False</span></code>)
differ in the following ways:</p>
<ul class="simple">
<li><p>Non-reentrant checkpoint stops recomputation as soon as all needed
intermediate activations have been recomputed. This feature is enabled
by default, but can be disabled with <code class="xref py py-func docutils literal notranslate"><span class="pre">set_checkpoint_early_stop()</span></code>.
Reentrant checkpoint always recomputes <code class="xref py py-attr docutils literal notranslate"><span class="pre">function</span></code> in its
entirety during the backward pass.</p></li>
<li><p>The reentrant variant does not record the autograd graph during the
forward pass, as it runs with the forward pass under
<a class="reference internal" href="generated/torch.no_grad.html#torch.no_grad" title="torch.no_grad"><code class="xref py py-func docutils literal notranslate"><span class="pre">torch.no_grad()</span></code></a>. The non-reentrant version does record the
autograd graph, allowing one to perform backward on the graph within
checkpointed regions.</p></li>
<li><p>The reentrant checkpoint only supports the
<a class="reference internal" href="generated/torch.autograd.backward.html#torch.autograd.backward" title="torch.autograd.backward"><code class="xref py py-func docutils literal notranslate"><span class="pre">torch.autograd.backward()</span></code></a> API for the backward pass without its
<cite>inputs</cite> argument, while the non-reentrant version supports all ways
of performing the backward pass.</p></li>
<li><p>At least one input and output must have <code class="docutils literal notranslate"><span class="pre">requires_grad=True</span></code> for the
reentrant variant. If this condition is unmet, the checkpointed part
of the model will not have gradients. The non-reentrant version does
not have this requirement.</p></li>
<li><p>The reentrant version does not consider tensors in nested structures
(e.g., custom objects, lists, dicts, etc) as participating in
autograd, while the non-reentrant version does.</p></li>
<li><p>The reentrant checkpoint does not support checkpointed regions with
detached tensors from the computational graph, whereas the
non-reentrant version does. For the reentrant variant, if the
checkpointed segment contains tensors detached using <code class="docutils literal notranslate"><span class="pre">detach()</span></code> or
with <a class="reference internal" href="generated/torch.no_grad.html#torch.no_grad" title="torch.no_grad"><code class="xref py py-func docutils literal notranslate"><span class="pre">torch.no_grad()</span></code></a>, the backward pass will raise an error.
This is because <code class="docutils literal notranslate"><span class="pre">checkpoint</span></code> makes all the outputs require gradients
and this causes issues when a tensor is defined to have no gradient in
the model. To avoid this, detach the tensors outside of the
<code class="docutils literal notranslate"><span class="pre">checkpoint</span></code> function.</p></li>
</ul>
</div>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><ul class="simple">
<li><p><strong>function</strong> – describes what to run in the forward pass of the model or
part of the model. It should also know how to handle the inputs
passed as the tuple. For example, in LSTM, if user passes
<code class="docutils literal notranslate"><span class="pre">(activation,</span> <span class="pre">hidden)</span></code>, <code class="xref py py-attr docutils literal notranslate"><span class="pre">function</span></code> should correctly use the
first input as <code class="docutils literal notranslate"><span class="pre">activation</span></code> and the second input as <code class="docutils literal notranslate"><span class="pre">hidden</span></code></p></li>
<li><p><strong>preserve_rng_state</strong> (<a class="reference external" href="https://docs.python.org/3/library/functions.html#bool" title="(in Python v3.13)"><em>bool</em></a><em>, </em><em>optional</em>) – Omit stashing and restoring
the RNG state during each checkpoint. Note that under torch.compile,
this flag doesn’t take effect and we always preserve RNG state.
Default: <code class="docutils literal notranslate"><span class="pre">True</span></code></p></li>
<li><p><strong>use_reentrant</strong> (<a class="reference external" href="https://docs.python.org/3/library/functions.html#bool" title="(in Python v3.13)"><em>bool</em></a>) – specify whether to use the activation checkpoint variant that
requires reentrant autograd. This parameter should be passed
explicitly. In version 2.5 we will raise an exception if
<code class="docutils literal notranslate"><span class="pre">use_reentrant</span></code> is not passed. If <code class="docutils literal notranslate"><span class="pre">use_reentrant=False</span></code>,
<code class="docutils literal notranslate"><span class="pre">checkpoint</span></code> will use an implementation that does not require
reentrant autograd. This allows <code class="docutils literal notranslate"><span class="pre">checkpoint</span></code> to support additional
functionality, such as working as expected with
<code class="docutils literal notranslate"><span class="pre">torch.autograd.grad</span></code> and support for keyword arguments input into
the checkpointed function.</p></li>
<li><p><strong>context_fn</strong> (<em>Callable</em><em>, </em><em>optional</em>) – A callable returning a tuple of two
context managers. The function and its recomputation will be run
under the first and second context managers respectively.
This argument is only supported if <code class="docutils literal notranslate"><span class="pre">use_reentrant=False</span></code>.</p></li>
<li><p><strong>determinism_check</strong> (<a class="reference external" href="https://docs.python.org/3/library/stdtypes.html#str" title="(in Python v3.13)"><em>str</em></a><em>, </em><em>optional</em>) – A string specifying the determinism
check to perform. By default it is set to <code class="docutils literal notranslate"><span class="pre">"default"</span></code> which
compares the shapes, dtypes, and devices of the recomputed tensors
against those the saved tensors. To turn off this check, specify
<code class="docutils literal notranslate"><span class="pre">"none"</span></code>. Currently these are the only two supported values.
Please open an issue if you would like to see more determinism
checks. This argument is only supported if <code class="docutils literal notranslate"><span class="pre">use_reentrant=False</span></code>,
if <code class="docutils literal notranslate"><span class="pre">use_reentrant=True</span></code>, the determinism check is always disabled.</p></li>
<li><p><strong>debug</strong> (<a class="reference external" href="https://docs.python.org/3/library/functions.html#bool" title="(in Python v3.13)"><em>bool</em></a><em>, </em><em>optional</em>) – If <code class="docutils literal notranslate"><span class="pre">True</span></code>, error messages will also include
a trace of the operators ran during the original forward computation
as well as the recomputation. This argument is only supported if
<code class="docutils literal notranslate"><span class="pre">use_reentrant=False</span></code>.</p></li>
<li><p><strong>args</strong> – tuple containing inputs to the <code class="xref py py-attr docutils literal notranslate"><span class="pre">function</span></code></p></li>
</ul>
</dd>
<dt class="field-even">Returns</dt>
<dd class="field-even"><p>Output of running <code class="xref py py-attr docutils literal notranslate"><span class="pre">function</span></code> on <code class="xref py py-attr docutils literal notranslate"><span class="pre">*args</span></code></p>
</dd>
</dl>
</dd></dl>
<dl class="py function">
<dt class="sig sig-object py" id="torch.utils.checkpoint.checkpoint_sequential">
<span class="sig-prename descclassname"><span class="pre">torch.utils.checkpoint.</span></span><span class="sig-name descname"><span class="pre">checkpoint_sequential</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">functions</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">segments</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">input</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">use_reentrant</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em>, <em class="sig-param"><span class="o"><span class="pre">**</span></span><span class="n"><span class="pre">kwargs</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/utils/checkpoint.html#checkpoint_sequential"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.utils.checkpoint.checkpoint_sequential" title="Permalink to this definition">¶</a></dt>
<dd><p>Checkpoint a sequential model to save memory.</p>
<p>Sequential models execute a list of modules/functions in order
(sequentially). Therefore, we can divide such a model in various segments
and checkpoint each segment. All segments except the last will not store
the intermediate activations. The inputs of each checkpointed segment will
be saved for re-running the segment in the backward pass.</p>
<div class="admonition warning">
<p class="admonition-title">Warning</p>
<p>The <code class="docutils literal notranslate"><span class="pre">use_reentrant</span></code> parameter should be passed explicitly. In version
2.4 we will raise an exception if <code class="docutils literal notranslate"><span class="pre">use_reentrant</span></code> is not passed.
If you are using the <code class="docutils literal notranslate"><span class="pre">use_reentrant=True`</span> <span class="pre">variant,</span> <span class="pre">please</span> <span class="pre">see</span>
<span class="pre">:func:`~torch.utils.checkpoint.checkpoint`</span> <span class="pre">for</span>
<span class="pre">the</span> <span class="pre">important</span> <span class="pre">considerations</span> <span class="pre">and</span> <span class="pre">limitations</span> <span class="pre">of</span> <span class="pre">this</span> <span class="pre">variant.</span> <span class="pre">It</span> <span class="pre">is</span>
<span class="pre">recommended</span> <span class="pre">that</span> <span class="pre">you</span> <span class="pre">use</span> <span class="pre">``use_reentrant=False</span></code>.</p>
</div>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><ul class="simple">
<li><p><strong>functions</strong> – A <a class="reference internal" href="generated/torch.nn.Sequential.html#torch.nn.Sequential" title="torch.nn.Sequential"><code class="xref py py-class docutils literal notranslate"><span class="pre">torch.nn.Sequential</span></code></a> or the list of modules or
functions (comprising the model) to run sequentially.</p></li>
<li><p><strong>segments</strong> – Number of chunks to create in the model</p></li>
<li><p><strong>input</strong> – A Tensor that is input to <code class="xref py py-attr docutils literal notranslate"><span class="pre">functions</span></code></p></li>
<li><p><strong>preserve_rng_state</strong> (<a class="reference external" href="https://docs.python.org/3/library/functions.html#bool" title="(in Python v3.13)"><em>bool</em></a><em>, </em><em>optional</em>) – Omit stashing and restoring
the RNG state during each checkpoint.
Default: <code class="docutils literal notranslate"><span class="pre">True</span></code></p></li>
<li><p><strong>use_reentrant</strong> (<a class="reference external" href="https://docs.python.org/3/library/functions.html#bool" title="(in Python v3.13)"><em>bool</em></a>) – specify whether to use the activation checkpoint variant that
requires reentrant autograd. This parameter should be passed
explicitly. In version 2.5 we will raise an exception if
<code class="docutils literal notranslate"><span class="pre">use_reentrant</span></code> is not passed. If <code class="docutils literal notranslate"><span class="pre">use_reentrant=False</span></code>,
<code class="docutils literal notranslate"><span class="pre">checkpoint</span></code> will use an implementation that does not require
reentrant autograd. This allows <code class="docutils literal notranslate"><span class="pre">checkpoint</span></code> to support additional
functionality, such as working as expected with
<code class="docutils literal notranslate"><span class="pre">torch.autograd.grad</span></code> and support for keyword arguments input into
the checkpointed function.</p></li>
</ul>
</dd>
<dt class="field-even">Returns</dt>
<dd class="field-even"><p>Output of running <code class="xref py py-attr docutils literal notranslate"><span class="pre">functions</span></code> sequentially on <code class="xref py py-attr docutils literal notranslate"><span class="pre">*inputs</span></code></p>
</dd>
</dl>
<p class="rubric">Example</p>
<div class="doctest highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="n">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="o">...</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">input_var</span> <span class="o">=</span> <span class="n">checkpoint_sequential</span><span class="p">(</span><span class="n">model</span><span class="p">,</span> <span class="n">chunks</span><span class="p">,</span> <span class="n">input_var</span><span class="p">)</span>
</pre></div>
</div>
</dd></dl>
<dl class="py function">
<dt class="sig sig-object py" id="torch.utils.checkpoint.set_checkpoint_debug_enabled">
<span class="sig-prename descclassname"><span class="pre">torch.utils.checkpoint.</span></span><span class="sig-name descname"><span class="pre">set_checkpoint_debug_enabled</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">enabled</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/utils/checkpoint.html#set_checkpoint_debug_enabled"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.utils.checkpoint.set_checkpoint_debug_enabled" title="Permalink to this definition">¶</a></dt>
<dd><p>Context manager that sets whether checkpoint should print additional debug
information when running. See the <code class="docutils literal notranslate"><span class="pre">debug</span></code> flag for
<a class="reference internal" href="#torch.utils.checkpoint.checkpoint" title="torch.utils.checkpoint.checkpoint"><code class="xref py py-func docutils literal notranslate"><span class="pre">checkpoint()</span></code></a> for more information. Note that
when set, this context manager overrides the value of <code class="docutils literal notranslate"><span class="pre">debug</span></code> passed to
checkpoint. To defer to the local setting, pass <code class="docutils literal notranslate"><span class="pre">None</span></code> to this context.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><p><strong>enabled</strong> (<a class="reference external" href="https://docs.python.org/3/library/functions.html#bool" title="(in Python v3.13)"><em>bool</em></a>) – Whether checkpoint should print debug information.
Default is ‘None’.</p>
</dd>
</dl>
</dd></dl>
<dl class="py class">
<dt class="sig sig-object py" id="torch.utils.checkpoint.CheckpointPolicy">
<em class="property"><span class="pre">class</span><span class="w"> </span></em><span class="sig-prename descclassname"><span class="pre">torch.utils.checkpoint.</span></span><span class="sig-name descname"><span class="pre">CheckpointPolicy</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">value</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/utils/checkpoint.html#CheckpointPolicy"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.utils.checkpoint.CheckpointPolicy" title="Permalink to this definition">¶</a></dt>
<dd><p>Enum for specifying the policy for checkpointing during backpropagation.</p>
<p>The following policies are supported:</p>
<ul class="simple">
<li><p><code class="docutils literal notranslate"><span class="pre">{MUST,PREFER}_SAVE</span></code>: The operation’s output will be saved during the forward
pass and will not be recomputed during the backward pass</p></li>
<li><p><code class="docutils literal notranslate"><span class="pre">{MUST,PREFER}_RECOMPUTE</span></code>: The operation’s output will not be saved during the
forward pass and will be recomputed during the backward pass</p></li>
</ul>
<p>Use <code class="docutils literal notranslate"><span class="pre">MUST_*</span></code> over <code class="docutils literal notranslate"><span class="pre">PREFER_*</span></code> to indicate that the policy should not be overridden
by other subsystems like <cite>torch.compile</cite>.</p>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>A policy function that always returns <code class="docutils literal notranslate"><span class="pre">PREFER_RECOMPUTE</span></code> is
equivalent to vanilla checkpointing.</p>
<p>A policy function that returns <code class="docutils literal notranslate"><span class="pre">PREFER_SAVE</span></code> every op is
NOT equivalent to not using checkpointing. Using such a policy would
save additional tensors not limited to ones that are actually needed for
gradient computation.</p>
</div>
</dd></dl>
<dl class="py class">
<dt class="sig sig-object py" id="torch.utils.checkpoint.SelectiveCheckpointContext">
<em class="property"><span class="pre">class</span><span class="w"> </span></em><span class="sig-prename descclassname"><span class="pre">torch.utils.checkpoint.</span></span><span class="sig-name descname"><span class="pre">SelectiveCheckpointContext</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="o"><span class="pre">*</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">is_recompute</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/utils/checkpoint.html#SelectiveCheckpointContext"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.utils.checkpoint.SelectiveCheckpointContext" title="Permalink to this definition">¶</a></dt>
<dd><p>Context passed to policy function during selective checkpointing.</p>
<p>This class is used to pass relevant metadata to the policy function during
selective checkpointing. The metadata includes whether the current invocation
of the policy function is during recomputation or not.</p>
<p class="rubric">Example</p>
<div class="doctest highlight-default notranslate"><div class="highlight"><pre><span></span><span class="go">>>></span>
<span class="gp">>>> </span><span class="k">def</span> <span class="nf">policy_fn</span><span class="p">(</span><span class="n">ctx</span><span class="p">,</span> <span class="n">op</span><span class="p">,</span> <span class="o">*</span><span class="n">args</span><span class="p">,</span> <span class="o">**</span><span class="n">kwargs</span><span class="p">):</span>
<span class="gp">>>> </span> <span class="nb">print</span><span class="p">(</span><span class="n">ctx</span><span class="o">.</span><span class="n">is_recompute</span><span class="p">)</span>
<span class="go">>>></span>
<span class="gp">>>> </span><span class="n">context_fn</span> <span class="o">=</span> <span class="n">functools</span><span class="o">.</span><span class="n">partial</span><span class="p">(</span><span class="n">create_selective_checkpoint_contexts</span><span class="p">,</span> <span class="n">policy_fn</span><span class="p">)</span>
<span class="go">>>></span>
<span class="gp">>>> </span><span class="n">out</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">utils</span><span class="o">.</span><span class="n">checkpoint</span><span class="o">.</span><span class="n">checkpoint</span><span class="p">(</span>
<span class="gp">>>> </span> <span class="n">fn</span><span class="p">,</span> <span class="n">x</span><span class="p">,</span> <span class="n">y</span><span class="p">,</span>
<span class="gp">>>> </span> <span class="n">use_reentrant</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="gp">>>> </span> <span class="n">context_fn</span><span class="o">=</span><span class="n">context_fn</span><span class="p">,</span>
<span class="gp">>>> </span><span class="p">)</span>
</pre></div>
</div>
</dd></dl>
<dl class="py function">
<dt class="sig sig-object py" id="torch.utils.checkpoint.create_selective_checkpoint_contexts">
<span class="sig-prename descclassname"><span class="pre">torch.utils.checkpoint.</span></span><span class="sig-name descname"><span class="pre">create_selective_checkpoint_contexts</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">policy_fn_or_list</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">allow_cache_entry_mutation</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">False</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/utils/checkpoint.html#create_selective_checkpoint_contexts"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.utils.checkpoint.create_selective_checkpoint_contexts" title="Permalink to this definition">¶</a></dt>
<dd><p>Helper to avoid recomputing certain ops during activation checkpointing.</p>
<p>Use this with <cite>torch.utils.checkpoint.checkpoint</cite> to control which
operations are recomputed during the backward pass.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><ul class="simple">
<li><p><strong>policy_fn_or_list</strong> (<em>Callable</em><em> or </em><em>List</em>) – <ul>
<li><p>If a policy function is provided, it should accept a
<a class="reference internal" href="#torch.utils.checkpoint.SelectiveCheckpointContext" title="torch.utils.checkpoint.SelectiveCheckpointContext"><code class="xref py py-class docutils literal notranslate"><span class="pre">SelectiveCheckpointContext</span></code></a>, the <code class="xref py py-class docutils literal notranslate"><span class="pre">OpOverload</span></code>, args and
kwargs to the op, and return a <a class="reference internal" href="#torch.utils.checkpoint.CheckpointPolicy" title="torch.utils.checkpoint.CheckpointPolicy"><code class="xref py py-class docutils literal notranslate"><span class="pre">CheckpointPolicy</span></code></a> enum value
indicating whether the execution of the op should be recomputed or not.</p></li>
<li><p>If a list of operations is provided, it is equivalent to a policy
returning <cite>CheckpointPolicy.MUST_SAVE</cite> for the specified
operations and <cite>CheckpointPolicy.PREFER_RECOMPUTE</cite> for all other
operations.</p></li>
</ul>
</p></li>
<li><p><strong>allow_cache_entry_mutation</strong> (<a class="reference external" href="https://docs.python.org/3/library/functions.html#bool" title="(in Python v3.13)"><em>bool</em></a><em>, </em><em>optional</em>) – By default, an error is
raised if any tensors cached by selective activation checkpoint are
mutated in order to ensure correctness. If set to <cite>True</cite>, this check
is disabled.</p></li>
</ul>
</dd>
<dt class="field-even">Returns</dt>
<dd class="field-even"><p>A tuple of two context managers.</p>
</dd>
</dl>
<p class="rubric">Example</p>
<div class="doctest highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="kn">import</span> <span class="nn">functools</span>
<span class="go">>>></span>
<span class="gp">>>> </span><span class="n">x</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">rand</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">requires_grad</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">y</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">rand</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">requires_grad</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span>
<span class="go">>>></span>
<span class="gp">>>> </span><span class="n">ops_to_save</span> <span class="o">=</span> <span class="p">[</span>
<span class="gp">>>> </span> <span class="n">torch</span><span class="o">.</span><span class="n">ops</span><span class="o">.</span><span class="n">aten</span><span class="o">.</span><span class="n">mm</span><span class="o">.</span><span class="n">default</span><span class="p">,</span>
<span class="gp">>>> </span><span class="p">]</span>
<span class="go">>>></span>
<span class="gp">>>> </span><span class="k">def</span> <span class="nf">policy_fn</span><span class="p">(</span><span class="n">ctx</span><span class="p">,</span> <span class="n">op</span><span class="p">,</span> <span class="o">*</span><span class="n">args</span><span class="p">,</span> <span class="o">**</span><span class="n">kwargs</span><span class="p">):</span>
<span class="gp">>>> </span> <span class="k">if</span> <span class="n">op</span> <span class="ow">in</span> <span class="n">ops_to_save</span><span class="p">:</span>
<span class="gp">>>> </span> <span class="k">return</span> <span class="n">CheckpointPolicy</span><span class="o">.</span><span class="n">MUST_SAVE</span>
<span class="gp">>>> </span> <span class="k">else</span><span class="p">:</span>
<span class="gp">>>> </span> <span class="k">return</span> <span class="n">CheckpointPolicy</span><span class="o">.</span><span class="n">PREFER_RECOMPUTE</span>
<span class="go">>>></span>
<span class="gp">>>> </span><span class="n">context_fn</span> <span class="o">=</span> <span class="n">functools</span><span class="o">.</span><span class="n">partial</span><span class="p">(</span><span class="n">create_selective_checkpoint_contexts</span><span class="p">,</span> <span class="n">policy_fn</span><span class="p">)</span>
<span class="go">>>></span>
<span class="gp">>>> </span><span class="c1"># or equivalently</span>
<span class="gp">>>> </span><span class="n">context_fn</span> <span class="o">=</span> <span class="n">functools</span><span class="o">.</span><span class="n">partial</span><span class="p">(</span><span class="n">create_selective_checkpoint_contexts</span><span class="p">,</span> <span class="n">ops_to_save</span><span class="p">)</span>
<span class="go">>>></span>
<span class="gp">>>> </span><span class="k">def</span> <span class="nf">fn</span><span class="p">(</span><span class="n">x</span><span class="p">,</span> <span class="n">y</span><span class="p">):</span>
<span class="gp">>>> </span> <span class="k">return</span> <span class="n">torch</span><span class="o">.</span><span class="n">sigmoid</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">matmul</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">matmul</span><span class="p">(</span><span class="n">x</span><span class="p">,</span> <span class="n">y</span><span class="p">),</span> <span class="n">y</span><span class="p">))</span> <span class="o">*</span> <span class="n">y</span>
<span class="go">>>></span>
<span class="gp">>>> </span><span class="n">out</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">utils</span><span class="o">.</span><span class="n">checkpoint</span><span class="o">.</span><span class="n">checkpoint</span><span class="p">(</span>
<span class="gp">>>> </span> <span class="n">fn</span><span class="p">,</span> <span class="n">x</span><span class="p">,</span> <span class="n">y</span><span class="p">,</span>
<span class="gp">>>> </span> <span class="n">use_reentrant</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="gp">>>> </span> <span class="n">context_fn</span><span class="o">=</span><span class="n">context_fn</span><span class="p">,</span>
<span class="gp">>>> </span><span class="p">)</span>
</pre></div>
</div>
</dd></dl>
</div>
</article>
</div>
<footer>
<div class="rst-footer-buttons" role="navigation" aria-label="footer navigation">
<a href="cpp_extension.html" class="btn btn-neutral float-right" title="torch.utils.cpp_extension" accesskey="n" rel="next">Next <img src="_static/images/chevron-right-orange.svg" class="next-page"></a>
<a href="bottleneck.html" class="btn btn-neutral" title="torch.utils.bottleneck" accesskey="p" rel="prev"><img src="_static/images/chevron-right-orange.svg" class="previous-page"> Previous</a>
</div>
<hr>
<div role="contentinfo">
<p>
© Copyright 2023, PyTorch Contributors.
</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>
var match = window.location.href.match(/\/_[a-zA-Z0-9_]*.html|_dynamo/gi);
var url = window.location.href.lastIndexOf(match[match.length-1]);
if (url)
{
var div = '<div class="admonition note"><p class="admonition-title">Note</p><p><i class="fa fa-exclamation-circle" aria-hidden="true"> </i> This page describes an internal API which is not intended to be used outside of the PyTorch codebase and can be modified or removed without notice.</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="#">torch.utils.checkpoint</a><ul>
<li><a class="reference internal" href="#torch.utils.checkpoint.checkpoint"><code class="docutils literal notranslate"><span class="pre">checkpoint()</span></code></a></li>
<li><a class="reference internal" href="#torch.utils.checkpoint.checkpoint_sequential"><code class="docutils literal notranslate"><span class="pre">checkpoint_sequential()</span></code></a></li>
<li><a class="reference internal" href="#torch.utils.checkpoint.set_checkpoint_debug_enabled"><code class="docutils literal notranslate"><span class="pre">set_checkpoint_debug_enabled()</span></code></a></li>
<li><a class="reference internal" href="#torch.utils.checkpoint.CheckpointPolicy"><code class="docutils literal notranslate"><span class="pre">CheckpointPolicy</span></code></a></li>
<li><a class="reference internal" href="#torch.utils.checkpoint.SelectiveCheckpointContext"><code class="docutils literal notranslate"><span class="pre">SelectiveCheckpointContext</span></code></a></li>
<li><a class="reference internal" href="#torch.utils.checkpoint.create_selective_checkpoint_contexts"><code class="docutils literal notranslate"><span class="pre">create_selective_checkpoint_contexts()</span></code></a></li>
</ul>
</li>
</ul>
</div>
</div>
</div>
</section>
</div>
<script type="text/javascript" id="documentation_options" data-url_root="./" src="_static/documentation_options.js"></script>
<script 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/sphinx_highlight.js"></script>
<script src="_static/clipboard.min.js"></script>
<script src="_static/copybutton.js"></script>
<script type="text/javascript" src="_static/js/vendor/popper.min.js"></script>
<script type="text/javascript" src="_static/js/vendor/bootstrap.min.js"></script>
<script src="https://cdnjs.cloudflare.com/ajax/libs/list.js/1.5.0/list.min.js"></script>
<script type="text/javascript" src="_static/js/theme.js"></script>
<script type="text/javascript">
jQuery(function () {
SphinxRtdTheme.Navigation.enable(true);
});
</script>
<script script type="text/javascript">
var collapsedSections = ['Developer Notes', 'Language Bindings', 'Libraries', 'Community'];
</script>
<img height="1" width="1" style="border-style:none;" alt="" src="https://www.googleadservices.com/pagead/conversion/795629140/?label=txkmCPmdtosBENSssfsC&guid=ON&script=0"/>
<!-- Begin Footer -->
<div class="container-fluid docs-tutorials-resources" id="docs-tutorials-resources">
<div class="container">
<div class="row">
<div class="col-md-4 text-center">
<h2>Docs</h2>
<p>Access comprehensive developer documentation for PyTorch</p>
<a class="with-right-arrow" href="https://pytorch.org/docs/stable/index.html">View Docs</a>
</div>
<div class="col-md-4 text-center">
<h2>Tutorials</h2>
<p>Get in-depth tutorials for beginners and advanced developers</p>
<a class="with-right-arrow" href="https://pytorch.org/tutorials">View Tutorials</a>
</div>
<div class="col-md-4 text-center">
<h2>Resources</h2>
<p>Find development resources and get your questions answered</p>
<a class="with-right-arrow" href="https://pytorch.org/resources">View Resources</a>
</div>
</div>
</div>
</div>
<footer class="site-footer">
<div class="container footer-container">
<div class="footer-logo-wrapper">
<a href="https://pytorch.org/" class="footer-logo"></a>
</div>
<div class="footer-links-wrapper">
<div class="footer-links-col">
<ul>
<li class="list-title"><a href="https://pytorch.org/">PyTorch</a></li>
<li><a href="https://pytorch.org/get-started">Get Started</a></li>
<li><a href="https://pytorch.org/features">Features</a></li>
<li><a href="https://pytorch.org/ecosystem">Ecosystem</a></li>
<li><a href="https://pytorch.org/blog/">Blog</a></li>
<li><a href="https://github.com/pytorch/pytorch/blob/master/CONTRIBUTING.md">Contributing</a></li>
</ul>
</div>
<div class="footer-links-col">
<ul>
<li class="list-title"><a href="https://pytorch.org/resources">Resources</a></li>
<li><a href="https://pytorch.org/tutorials">Tutorials</a></li>
<li><a href="https://pytorch.org/docs/stable/index.html">Docs</a></li>
<li><a href="https://discuss.pytorch.org" target="_blank">Discuss</a></li>
<li><a href="https://github.com/pytorch/pytorch/issues" target="_blank">Github Issues</a></li>
<li><a href="https://pytorch.org/assets/brand-guidelines/PyTorch-Brand-Guidelines.pdf" target="_blank">Brand Guidelines</a></li>
</ul>
</div>
<div class="footer-links-col">