forked from pytorch/pytorch.github.io
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathautograd.html
2212 lines (1981 loc) · 192 KB
/
autograd.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>Automatic differentiation package - torch.autograd — PyTorch 1.8.1 documentation</title>
<link rel="canonical" href="https://pytorch.org/docs/stable/autograd.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="https://cdn.jsdelivr.net/npm/[email protected]/dist/katex.min.css" type="text/css" />
<link rel="stylesheet" href="_static/css/jit.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="index" title="Index" href="genindex.html" />
<link rel="search" title="Search" href="search.html" />
<link rel="next" title="torch.cuda" href="cuda.html" />
<link rel="prev" title="Tensor Views" href="tensor_view.html" />
<script src="_static/js/modernizr.min.js"></script>
<!-- Preload the theme fonts -->
<link rel="preload" href="_static/fonts/FreightSans/freight-sans-book.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="_static/fonts/FreightSans/freight-sans-medium.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="_static/fonts/IBMPlexMono/IBMPlexMono-Medium.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="_static/fonts/FreightSans/freight-sans-bold.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="_static/fonts/FreightSans/freight-sans-medium-italic.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="_static/fonts/IBMPlexMono/IBMPlexMono-SemiBold.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<!-- Preload the katex fonts -->
<link rel="preload" href="https://cdn.jsdelivr.net/npm/[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>
<a href="https://pytorch.org/get-started">Get Started</a>
</li>
<li>
<a href="https://pytorch.org/ecosystem">Ecosystem</a>
</li>
<li>
<a href="https://pytorch.org/mobile">Mobile</a>
</li>
<li>
<a href="https://pytorch.org/blog/">Blog</a>
</li>
<li>
<a href="https://pytorch.org/tutorials">Tutorials</a>
</li>
<li class="active">
<div id="resourcesDropdownButton" data-toggle="resources-dropdown" class="resources-dropdown">
<a class="resource-option with-down-orange-arrow">
Docs
</a>
<div class="resources-dropdown-menu">
<a class="doc-dropdown-option nav-dropdown-item" href="https://pytorch.org/docs/stable/index.html">
<span class="dropdown-title">PyTorch</span>
<p></p>
</a>
<a class="doc-dropdown-option nav-dropdown-item" href="https://pytorch.org/audio/stable/index.html">
<span class="dropdown-title">torchaudio</span>
<p></p>
</a>
<a class="doc-dropdown-option nav-dropdown-item" href="https://pytorch.org/text/stable/index.html">
<span class="dropdown-title">torchtext</span>
<p></p>
</a>
<a class="doc-dropdown-option nav-dropdown-item" href="https://pytorch.org/vision/stable/index.html">
<span class="dropdown-title">torchvision</span>
<p></p>
</a>
<a class="doc-dropdown-option nav-dropdown-item" href="https://pytorch.org/elastic/">
<span class="dropdown-title">TorchElastic</span>
<p></p>
</a>
<a class="doc-dropdown-option nav-dropdown-item" href="https://pytorch.org/serve/">
<span class="dropdown-title">TorchServe</span>
<p></p>
</a>
<a class="doc-dropdown-option nav-dropdown-item" href="https://pytorch.org/xla">
<span class="dropdown-title">PyTorch on XLA Devices</span>
<p></p>
</a>
</div>
</li>
<li>
<div id="resourcesDropdownButton" data-toggle="resources-dropdown" class="resources-dropdown">
<a class="resource-option with-down-arrow">
Resources
</a>
<div class="resources-dropdown-menu">
<a class="nav-dropdown-item" href="https://pytorch.org/features">
<span class="dropdown-title">About</span>
<p>Learn about PyTorch’s features and capabilities</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://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://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/hub">
<span class="dropdown-title">Models (Beta)</span>
<p>Discover, publish, and reuse pre-trained models</p>
</a>
</div>
</div>
</li>
<li>
<a href="https://github.com/pytorch/pytorch">Github</a>
</li>
</ul>
</div>
<a class="main-menu-open-button" href="#" data-behavior="open-mobile-menu"></a>
</div>
</div>
</div>
<body class="pytorch-body">
<div class="table-of-contents-link-wrapper">
<span>Table of Contents</span>
<a href="#" class="toggle-table-of-contents" data-behavior="toggle-table-of-contents"></a>
</div>
<nav data-toggle="wy-nav-shift" class="pytorch-left-menu" id="pytorch-left-menu">
<div class="pytorch-side-scroll">
<div class="pytorch-menu pytorch-menu-vertical" data-spy="affix" role="navigation" aria-label="main navigation">
<div class="pytorch-left-menu-search">
<div class="version">
<a href='https://pytorch.org/docs/versions.html'>1.8.1 ▼</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"><span class="caption-text">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/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/faq.html">Frequently Asked Questions</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/multiprocessing.html">Multiprocessing best practices</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"><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>
</ul>
<p class="caption"><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 current"><a class="current reference internal" href="#">torch.autograd</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="amp.html">torch.cuda.amp</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="distributed.html">torch.distributed</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="fft.html">torch.fft</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="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="torch.overrides.html">torch.overrides</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="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="pipeline.html">Pipeline Parallelism</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="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="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"><a class="reference internal" href="checkpoint.html">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="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="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__.html">torch.__config__</a></li>
</ul>
<p class="caption"><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/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/elastic/">TorchElastic</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="http://pytorch.org/xla/">PyTorch on XLA Devices</a></li>
</ul>
<p class="caption"><span class="caption-text">Community</span></p>
<ul>
<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/governance.html">PyTorch Governance</a></li>
<li class="toctree-l1"><a class="reference internal" href="community/persons_of_interest.html">PyTorch Governance | Persons of Interest</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>Automatic differentiation package - torch.autograd</li>
<li class="pytorch-breadcrumbs-aside">
<a href="_sources/autograd.rst.txt" rel="nofollow"><img src="_static/images/view-page-source-icon.svg"></a>
</li>
</ul>
</div>
</div>
<div class="pytorch-shortcuts-wrapper" id="pytorch-shortcuts-wrapper">
Shortcuts
</div>
</div>
<section data-toggle="wy-nav-shift" id="pytorch-content-wrap" class="pytorch-content-wrap">
<div class="pytorch-content-left">
<div class="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="module-torch.autograd">
<span id="automatic-differentiation-package-torch-autograd"></span><h1>Automatic differentiation package - torch.autograd<a class="headerlink" href="#module-torch.autograd" title="Permalink to this headline">¶</a></h1>
<p><code class="docutils literal notranslate"><span class="pre">torch.autograd</span></code> provides classes and functions implementing automatic
differentiation of arbitrary scalar valued functions. It requires minimal
changes to the existing code - you only need to declare <code class="xref py py-class docutils literal notranslate"><span class="pre">Tensor</span></code> s
for which gradients should be computed with the <code class="docutils literal notranslate"><span class="pre">requires_grad=True</span></code> keyword.
As of now, we only support autograd for floating point <code class="xref py py-class docutils literal notranslate"><span class="pre">Tensor</span></code> types (
half, float, double and bfloat16) and complex <code class="xref py py-class docutils literal notranslate"><span class="pre">Tensor</span></code> types (cfloat, cdouble).</p>
<dl class="function">
<dt id="torch.autograd.backward">
<code class="sig-prename descclassname">torch.autograd.</code><code class="sig-name descname">backward</code><span class="sig-paren">(</span><em class="sig-param">tensors</em>, <em class="sig-param">grad_tensors=None</em>, <em class="sig-param">retain_graph=None</em>, <em class="sig-param">create_graph=False</em>, <em class="sig-param">grad_variables=None</em>, <em class="sig-param">inputs=None</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/autograd.html#backward"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.autograd.backward" title="Permalink to this definition">¶</a></dt>
<dd><p>Computes the sum of gradients of given tensors w.r.t. graph leaves.</p>
<p>The graph is differentiated using the chain rule. If any of <code class="docutils literal notranslate"><span class="pre">tensors</span></code>
are non-scalar (i.e. their data has more than one element) and require
gradient, then the Jacobian-vector product would be computed, in this
case the function additionally requires specifying <code class="docutils literal notranslate"><span class="pre">grad_tensors</span></code>.
It should be a sequence of matching length, that contains the “vector”
in the Jacobian-vector product, usually the gradient of the differentiated
function w.r.t. corresponding tensors (<code class="docutils literal notranslate"><span class="pre">None</span></code> is an acceptable value for
all tensors that don’t need gradient tensors).</p>
<p>This function accumulates gradients in the leaves - you might need to zero
<code class="docutils literal notranslate"><span class="pre">.grad</span></code> attributes or set them to <code class="docutils literal notranslate"><span class="pre">None</span></code> before calling it.
See <a class="reference internal" href="#default-grad-layouts"><span class="std std-ref">Default gradient layouts</span></a>
for details on the memory layout of accumulated gradients.</p>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>Using this method with <code class="docutils literal notranslate"><span class="pre">create_graph=True</span></code> will create a reference cycle
between the parameter and its gradient which can cause a memory leak.
We recommend using <code class="docutils literal notranslate"><span class="pre">autograd.grad</span></code> when creating the graph to avoid this.
If you have to use this function, make sure to reset the <code class="docutils literal notranslate"><span class="pre">.grad</span></code> fields of your
parameters to <code class="docutils literal notranslate"><span class="pre">None</span></code> after use to break the cycle and avoid the leak.</p>
</div>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>If you run any forward ops, create <code class="docutils literal notranslate"><span class="pre">grad_tensors</span></code>, and/or call <code class="docutils literal notranslate"><span class="pre">backward</span></code>
in a user-specified CUDA stream context, see
<a class="reference internal" href="notes/cuda.html#bwd-cuda-stream-semantics"><span class="std std-ref">Stream semantics of backward passes</span></a>.</p>
</div>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><ul class="simple">
<li><p><strong>tensors</strong> (<em>sequence of Tensor</em>) – Tensors of which the derivative will be
computed.</p></li>
<li><p><strong>grad_tensors</strong> (<em>sequence of</em><em> (</em><a class="reference internal" href="tensors.html#torch.Tensor" title="torch.Tensor"><em>Tensor</em></a><em> or </em><a class="reference external" href="https://docs.python.org/3/library/constants.html#None" title="(in Python v3.9)"><em>None</em></a><em>)</em>) – The “vector” in the Jacobian-vector
product, usually gradients w.r.t. each element of corresponding tensors.
None values can be specified for scalar Tensors or ones that don’t require
grad. If a None value would be acceptable for all grad_tensors, then this
argument is optional.</p></li>
<li><p><strong>retain_graph</strong> (<a class="reference external" href="https://docs.python.org/3/library/functions.html#bool" title="(in Python v3.9)"><em>bool</em></a><em>, </em><em>optional</em>) – If <code class="docutils literal notranslate"><span class="pre">False</span></code>, the graph used to compute the grad
will be freed. Note that in nearly all cases setting this option to <code class="docutils literal notranslate"><span class="pre">True</span></code>
is not needed and often can be worked around in a much more efficient
way. Defaults to the value of <code class="docutils literal notranslate"><span class="pre">create_graph</span></code>.</p></li>
<li><p><strong>create_graph</strong> (<a class="reference external" href="https://docs.python.org/3/library/functions.html#bool" title="(in Python v3.9)"><em>bool</em></a><em>, </em><em>optional</em>) – If <code class="docutils literal notranslate"><span class="pre">True</span></code>, graph of the derivative will
be constructed, allowing to compute higher order derivative products.
Defaults to <code class="docutils literal notranslate"><span class="pre">False</span></code>.</p></li>
<li><p><strong>inputs</strong> (<em>sequence of Tensor</em>) – Inputs w.r.t. which the gradient will be
accumulated into <code class="docutils literal notranslate"><span class="pre">.grad</span></code>. All other Tensors will be ignored. If not
provided, the gradient is accumulated into all the leaf Tensors that were
used to compute the attr::tensors. All the provided inputs must be leaf
Tensors.</p></li>
</ul>
</dd>
</dl>
</dd></dl>
<dl class="function">
<dt id="torch.autograd.grad">
<code class="sig-prename descclassname">torch.autograd.</code><code class="sig-name descname">grad</code><span class="sig-paren">(</span><em class="sig-param">outputs</em>, <em class="sig-param">inputs</em>, <em class="sig-param">grad_outputs=None</em>, <em class="sig-param">retain_graph=None</em>, <em class="sig-param">create_graph=False</em>, <em class="sig-param">only_inputs=True</em>, <em class="sig-param">allow_unused=False</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/autograd.html#grad"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.autograd.grad" title="Permalink to this definition">¶</a></dt>
<dd><p>Computes and returns the sum of gradients of outputs w.r.t. the inputs.</p>
<p><code class="docutils literal notranslate"><span class="pre">grad_outputs</span></code> should be a sequence of length matching <code class="docutils literal notranslate"><span class="pre">output</span></code>
containing the “vector” in Jacobian-vector product, usually the pre-computed
gradients w.r.t. each of the outputs. If an output doesn’t require_grad,
then the gradient can be <code class="docutils literal notranslate"><span class="pre">None</span></code>).</p>
<p>If <code class="docutils literal notranslate"><span class="pre">only_inputs</span></code> is <code class="docutils literal notranslate"><span class="pre">True</span></code>, the function will only return a list of gradients
w.r.t the specified inputs. If it’s <code class="docutils literal notranslate"><span class="pre">False</span></code>, then gradient w.r.t. all remaining
leaves will still be computed, and will be accumulated into their <code class="docutils literal notranslate"><span class="pre">.grad</span></code>
attribute.</p>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>If you run any forward ops, create <code class="docutils literal notranslate"><span class="pre">grad_outputs</span></code>, and/or call <code class="docutils literal notranslate"><span class="pre">grad</span></code>
in a user-specified CUDA stream context, see
<a class="reference internal" href="notes/cuda.html#bwd-cuda-stream-semantics"><span class="std std-ref">Stream semantics of backward passes</span></a>.</p>
</div>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><ul class="simple">
<li><p><strong>outputs</strong> (<em>sequence of Tensor</em>) – outputs of the differentiated function.</p></li>
<li><p><strong>inputs</strong> (<em>sequence of Tensor</em>) – Inputs w.r.t. which the gradient will be
returned (and not accumulated into <code class="docutils literal notranslate"><span class="pre">.grad</span></code>).</p></li>
<li><p><strong>grad_outputs</strong> (<em>sequence of Tensor</em>) – The “vector” in the Jacobian-vector product.
Usually gradients w.r.t. each output. None values can be specified for scalar
Tensors or ones that don’t require grad. If a None value would be acceptable
for all grad_tensors, then this argument is optional. Default: None.</p></li>
<li><p><strong>retain_graph</strong> (<a class="reference external" href="https://docs.python.org/3/library/functions.html#bool" title="(in Python v3.9)"><em>bool</em></a><em>, </em><em>optional</em>) – If <code class="docutils literal notranslate"><span class="pre">False</span></code>, the graph used to compute the grad
will be freed. Note that in nearly all cases setting this option to <code class="docutils literal notranslate"><span class="pre">True</span></code>
is not needed and often can be worked around in a much more efficient
way. Defaults to the value of <code class="docutils literal notranslate"><span class="pre">create_graph</span></code>.</p></li>
<li><p><strong>create_graph</strong> (<a class="reference external" href="https://docs.python.org/3/library/functions.html#bool" title="(in Python v3.9)"><em>bool</em></a><em>, </em><em>optional</em>) – If <code class="docutils literal notranslate"><span class="pre">True</span></code>, graph of the derivative will
be constructed, allowing to compute higher order derivative products.
Default: <code class="docutils literal notranslate"><span class="pre">False</span></code>.</p></li>
<li><p><strong>allow_unused</strong> (<a class="reference external" href="https://docs.python.org/3/library/functions.html#bool" title="(in Python v3.9)"><em>bool</em></a><em>, </em><em>optional</em>) – If <code class="docutils literal notranslate"><span class="pre">False</span></code>, specifying inputs that were not
used when computing outputs (and therefore their grad is always zero)
is an error. Defaults to <code class="docutils literal notranslate"><span class="pre">False</span></code>.</p></li>
</ul>
</dd>
</dl>
</dd></dl>
<div class="section" id="functional-higher-level-api">
<span id="functional-api"></span><h2>Functional higher level API<a class="headerlink" href="#functional-higher-level-api" title="Permalink to this headline">¶</a></h2>
<div class="admonition warning">
<p class="admonition-title">Warning</p>
<p>This API is in beta. Even though the function signatures are very unlikely to change, major
improvements to performances are planned before we consider this stable.</p>
</div>
<p>This section contains the higher level API for the autograd that builds on the basic API above
and allows you to compute jacobians, hessians, etc.</p>
<p>This API works with user-provided functions that take only Tensors as input and return
only Tensors.
If your function takes other arguments that are not Tensors or Tensors that don’t have requires_grad set,
you can use a lambda to capture them.
For example, for a function <code class="docutils literal notranslate"><span class="pre">f</span></code> that takes three inputs, a Tensor for which we want the jacobian, another
tensor that should be considered constant and a boolean flag as <code class="docutils literal notranslate"><span class="pre">f(input,</span> <span class="pre">constant,</span> <span class="pre">flag=flag)</span></code>
you can use it as <code class="docutils literal notranslate"><span class="pre">functional.jacobian(lambda</span> <span class="pre">x:</span> <span class="pre">f(x,</span> <span class="pre">constant,</span> <span class="pre">flag=flag),</span> <span class="pre">input)</span></code>.</p>
<dl class="function">
<dt id="torch.autograd.functional.jacobian">
<code class="sig-prename descclassname">torch.autograd.functional.</code><code class="sig-name descname">jacobian</code><span class="sig-paren">(</span><em class="sig-param">func</em>, <em class="sig-param">inputs</em>, <em class="sig-param">create_graph=False</em>, <em class="sig-param">strict=False</em>, <em class="sig-param">vectorize=False</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/autograd/functional.html#jacobian"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.autograd.functional.jacobian" title="Permalink to this definition">¶</a></dt>
<dd><p>Function that computes the Jacobian of a given function.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><ul class="simple">
<li><p><strong>func</strong> (<em>function</em>) – a Python function that takes Tensor inputs and returns
a tuple of Tensors or a Tensor.</p></li>
<li><p><strong>inputs</strong> (<em>tuple of Tensors</em><em> or </em><a class="reference internal" href="tensors.html#torch.Tensor" title="torch.Tensor"><em>Tensor</em></a>) – inputs to the function <code class="docutils literal notranslate"><span class="pre">func</span></code>.</p></li>
<li><p><strong>create_graph</strong> (<a class="reference external" href="https://docs.python.org/3/library/functions.html#bool" title="(in Python v3.9)"><em>bool</em></a><em>, </em><em>optional</em>) – If <code class="docutils literal notranslate"><span class="pre">True</span></code>, the Jacobian will be
computed in a differentiable manner. Note that when <code class="docutils literal notranslate"><span class="pre">strict</span></code> is
<code class="docutils literal notranslate"><span class="pre">False</span></code>, the result can not require gradients or be disconnected
from the inputs. Defaults to <code class="docutils literal notranslate"><span class="pre">False</span></code>.</p></li>
<li><p><strong>strict</strong> (<a class="reference external" href="https://docs.python.org/3/library/functions.html#bool" title="(in Python v3.9)"><em>bool</em></a><em>, </em><em>optional</em>) – If <code class="docutils literal notranslate"><span class="pre">True</span></code>, an error will be raised when we
detect that there exists an input such that all the outputs are
independent of it. If <code class="docutils literal notranslate"><span class="pre">False</span></code>, we return a Tensor of zeros as the
jacobian for said inputs, which is the expected mathematical value.
Defaults to <code class="docutils literal notranslate"><span class="pre">False</span></code>.</p></li>
<li><p><strong>vectorize</strong> (<a class="reference external" href="https://docs.python.org/3/library/functions.html#bool" title="(in Python v3.9)"><em>bool</em></a><em>, </em><em>optional</em>) – This feature is experimental, please use at
your own risk. When computing the jacobian, usually we invoke
<code class="docutils literal notranslate"><span class="pre">autograd.grad</span></code> once per row of the jacobian. If this flag is
<code class="docutils literal notranslate"><span class="pre">True</span></code>, we use the vmap prototype feature as the backend to
vectorize calls to <code class="docutils literal notranslate"><span class="pre">autograd.grad</span></code> so we only invoke it once
instead of once per row. This should lead to performance
improvements in many use cases, however, due to this feature
being incomplete, there may be performance cliffs. Please
use <cite>torch._C._debug_only_display_vmap_fallback_warnings(True)</cite>
to show any performance warnings and file us issues if
warnings exist for your use case. Defaults to <code class="docutils literal notranslate"><span class="pre">False</span></code>.</p></li>
</ul>
</dd>
<dt class="field-even">Returns</dt>
<dd class="field-even"><p>if there is a single
input and output, this will be a single Tensor containing the
Jacobian for the linearized inputs and output. If one of the two is
a tuple, then the Jacobian will be a tuple of Tensors. If both of
them are tuples, then the Jacobian will be a tuple of tuple of
Tensors where <code class="docutils literal notranslate"><span class="pre">Jacobian[i][j]</span></code> will contain the Jacobian of the
<code class="docutils literal notranslate"><span class="pre">i</span></code>th output and <code class="docutils literal notranslate"><span class="pre">j</span></code>th input and will have as size the
concatenation of the sizes of the corresponding output and the
corresponding input and will have same dtype and device as the
corresponding input.</p>
</dd>
<dt class="field-odd">Return type</dt>
<dd class="field-odd"><p>Jacobian (<a class="reference internal" href="tensors.html#torch.Tensor" title="torch.Tensor">Tensor</a> or nested tuple of Tensors)</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="k">def</span> <span class="nf">exp_reducer</span><span class="p">(</span><span class="n">x</span><span class="p">):</span>
<span class="gp">... </span> <span class="k">return</span> <span class="n">x</span><span class="o">.</span><span class="n">exp</span><span class="p">()</span><span class="o">.</span><span class="n">sum</span><span class="p">(</span><span class="n">dim</span><span class="o">=</span><span class="mi">1</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">inputs</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">2</span><span class="p">,</span> <span class="mi">2</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">jacobian</span><span class="p">(</span><span class="n">exp_reducer</span><span class="p">,</span> <span class="n">inputs</span><span class="p">)</span>
<span class="go">tensor([[[1.4917, 2.4352],</span>
<span class="go"> [0.0000, 0.0000]],</span>
<span class="go"> [[0.0000, 0.0000],</span>
<span class="go"> [2.4369, 2.3799]]])</span>
</pre></div>
</div>
<div class="doctest highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="n">jacobian</span><span class="p">(</span><span class="n">exp_reducer</span><span class="p">,</span> <span class="n">inputs</span><span class="p">,</span> <span class="n">create_graph</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span>
<span class="go">tensor([[[1.4917, 2.4352],</span>
<span class="go"> [0.0000, 0.0000]],</span>
<span class="go"> [[0.0000, 0.0000],</span>
<span class="go"> [2.4369, 2.3799]]], grad_fn=<ViewBackward>)</span>
</pre></div>
</div>
<div class="doctest highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="k">def</span> <span class="nf">exp_adder</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="mi">2</span> <span class="o">*</span> <span class="n">x</span><span class="o">.</span><span class="n">exp</span><span class="p">()</span> <span class="o">+</span> <span class="mi">3</span> <span class="o">*</span> <span class="n">y</span>
<span class="gp">>>> </span><span class="n">inputs</span> <span class="o">=</span> <span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">rand</span><span class="p">(</span><span class="mi">2</span><span class="p">),</span> <span class="n">torch</span><span class="o">.</span><span class="n">rand</span><span class="p">(</span><span class="mi">2</span><span class="p">))</span>
<span class="gp">>>> </span><span class="n">jacobian</span><span class="p">(</span><span class="n">exp_adder</span><span class="p">,</span> <span class="n">inputs</span><span class="p">)</span>
<span class="go">(tensor([[2.8052, 0.0000],</span>
<span class="go"> [0.0000, 3.3963]]),</span>
<span class="go"> tensor([[3., 0.],</span>
<span class="go"> [0., 3.]]))</span>
</pre></div>
</div>
</dd></dl>
<dl class="function">
<dt id="torch.autograd.functional.hessian">
<code class="sig-prename descclassname">torch.autograd.functional.</code><code class="sig-name descname">hessian</code><span class="sig-paren">(</span><em class="sig-param">func</em>, <em class="sig-param">inputs</em>, <em class="sig-param">create_graph=False</em>, <em class="sig-param">strict=False</em>, <em class="sig-param">vectorize=False</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/autograd/functional.html#hessian"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.autograd.functional.hessian" title="Permalink to this definition">¶</a></dt>
<dd><p>Function that computes the Hessian of a given scalar function.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><ul class="simple">
<li><p><strong>func</strong> (<em>function</em>) – a Python function that takes Tensor inputs and returns
a Tensor with a single element.</p></li>
<li><p><strong>inputs</strong> (<em>tuple of Tensors</em><em> or </em><a class="reference internal" href="tensors.html#torch.Tensor" title="torch.Tensor"><em>Tensor</em></a>) – inputs to the function <code class="docutils literal notranslate"><span class="pre">func</span></code>.</p></li>
<li><p><strong>create_graph</strong> (<a class="reference external" href="https://docs.python.org/3/library/functions.html#bool" title="(in Python v3.9)"><em>bool</em></a><em>, </em><em>optional</em>) – If <code class="docutils literal notranslate"><span class="pre">True</span></code>, the Hessian will be computed in
a differentiable manner. Note that when <code class="docutils literal notranslate"><span class="pre">strict</span></code> is <code class="docutils literal notranslate"><span class="pre">False</span></code>, the result can not
require gradients or be disconnected from the inputs.
Defaults to <code class="docutils literal notranslate"><span class="pre">False</span></code>.</p></li>
<li><p><strong>strict</strong> (<a class="reference external" href="https://docs.python.org/3/library/functions.html#bool" title="(in Python v3.9)"><em>bool</em></a><em>, </em><em>optional</em>) – If <code class="docutils literal notranslate"><span class="pre">True</span></code>, an error will be raised when we detect that there exists an input
such that all the outputs are independent of it. If <code class="docutils literal notranslate"><span class="pre">False</span></code>, we return a Tensor of zeros as the
hessian for said inputs, which is the expected mathematical value.
Defaults to <code class="docutils literal notranslate"><span class="pre">False</span></code>.</p></li>
<li><p><strong>vectorize</strong> (<a class="reference external" href="https://docs.python.org/3/library/functions.html#bool" title="(in Python v3.9)"><em>bool</em></a><em>, </em><em>optional</em>) – This feature is experimental, please use at
your own risk. When computing the hessian, usually we invoke
<code class="docutils literal notranslate"><span class="pre">autograd.grad</span></code> once per row of the hessian. If this flag is
<code class="docutils literal notranslate"><span class="pre">True</span></code>, we use the vmap prototype feature as the backend to
vectorize calls to <code class="docutils literal notranslate"><span class="pre">autograd.grad</span></code> so we only invoke it once
instead of once per row. This should lead to performance
improvements in many use cases, however, due to this feature
being incomplete, there may be performance cliffs. Please
use <cite>torch._C._debug_only_display_vmap_fallback_warnings(True)</cite>
to show any performance warnings and file us issues if
warnings exist for your use case. Defaults to <code class="docutils literal notranslate"><span class="pre">False</span></code>.</p></li>
</ul>
</dd>
<dt class="field-even">Returns</dt>
<dd class="field-even"><p>if there is a single input,
this will be a single Tensor containing the Hessian for the input.
If it is a tuple, then the Hessian will be a tuple of tuples where
<code class="docutils literal notranslate"><span class="pre">Hessian[i][j]</span></code> will contain the Hessian of the <code class="docutils literal notranslate"><span class="pre">i</span></code>th input
and <code class="docutils literal notranslate"><span class="pre">j</span></code>th input with size the sum of the size of the <code class="docutils literal notranslate"><span class="pre">i</span></code>th input plus
the size of the <code class="docutils literal notranslate"><span class="pre">j</span></code>th input. <code class="docutils literal notranslate"><span class="pre">Hessian[i][j]</span></code> will have the same
dtype and device as the corresponding <code class="docutils literal notranslate"><span class="pre">i</span></code>th input.</p>
</dd>
<dt class="field-odd">Return type</dt>
<dd class="field-odd"><p>Hessian (<a class="reference internal" href="tensors.html#torch.Tensor" title="torch.Tensor">Tensor</a> or a tuple of tuple of Tensors)</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="k">def</span> <span class="nf">pow_reducer</span><span class="p">(</span><span class="n">x</span><span class="p">):</span>
<span class="gp">... </span> <span class="k">return</span> <span class="n">x</span><span class="o">.</span><span class="n">pow</span><span class="p">(</span><span class="mi">3</span><span class="p">)</span><span class="o">.</span><span class="n">sum</span><span class="p">()</span>
<span class="gp">>>> </span><span class="n">inputs</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">2</span><span class="p">,</span> <span class="mi">2</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">hessian</span><span class="p">(</span><span class="n">pow_reducer</span><span class="p">,</span> <span class="n">inputs</span><span class="p">)</span>
<span class="go">tensor([[[[5.2265, 0.0000],</span>
<span class="go"> [0.0000, 0.0000]],</span>
<span class="go"> [[0.0000, 4.8221],</span>
<span class="go"> [0.0000, 0.0000]]],</span>
<span class="go"> [[[0.0000, 0.0000],</span>
<span class="go"> [1.9456, 0.0000]],</span>
<span class="go"> [[0.0000, 0.0000],</span>
<span class="go"> [0.0000, 3.2550]]]])</span>
</pre></div>
</div>
<div class="doctest highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="n">hessian</span><span class="p">(</span><span class="n">pow_reducer</span><span class="p">,</span> <span class="n">inputs</span><span class="p">,</span> <span class="n">create_graph</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span>
<span class="go">tensor([[[[5.2265, 0.0000],</span>
<span class="go"> [0.0000, 0.0000]],</span>
<span class="go"> [[0.0000, 4.8221],</span>
<span class="go"> [0.0000, 0.0000]]],</span>
<span class="go"> [[[0.0000, 0.0000],</span>
<span class="go"> [1.9456, 0.0000]],</span>
<span class="go"> [[0.0000, 0.0000],</span>
<span class="go"> [0.0000, 3.2550]]]], grad_fn=<ViewBackward>)</span>
</pre></div>
</div>
<div class="doctest highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="k">def</span> <span class="nf">pow_adder_reducer</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="p">(</span><span class="mi">2</span> <span class="o">*</span> <span class="n">x</span><span class="o">.</span><span class="n">pow</span><span class="p">(</span><span class="mi">2</span><span class="p">)</span> <span class="o">+</span> <span class="mi">3</span> <span class="o">*</span> <span class="n">y</span><span class="o">.</span><span class="n">pow</span><span class="p">(</span><span class="mi">2</span><span class="p">))</span><span class="o">.</span><span class="n">sum</span><span class="p">()</span>
<span class="gp">>>> </span><span class="n">inputs</span> <span class="o">=</span> <span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">rand</span><span class="p">(</span><span class="mi">2</span><span class="p">),</span> <span class="n">torch</span><span class="o">.</span><span class="n">rand</span><span class="p">(</span><span class="mi">2</span><span class="p">))</span>
<span class="gp">>>> </span><span class="n">hessian</span><span class="p">(</span><span class="n">pow_adder_reducer</span><span class="p">,</span> <span class="n">inputs</span><span class="p">)</span>
<span class="go">((tensor([[4., 0.],</span>
<span class="go"> [0., 4.]]),</span>
<span class="go"> tensor([[0., 0.],</span>
<span class="go"> [0., 0.]])),</span>
<span class="go"> (tensor([[0., 0.],</span>
<span class="go"> [0., 0.]]),</span>
<span class="go"> tensor([[6., 0.],</span>
<span class="go"> [0., 6.]])))</span>
</pre></div>
</div>
</dd></dl>
<dl class="function">
<dt id="torch.autograd.functional.vjp">
<code class="sig-prename descclassname">torch.autograd.functional.</code><code class="sig-name descname">vjp</code><span class="sig-paren">(</span><em class="sig-param">func</em>, <em class="sig-param">inputs</em>, <em class="sig-param">v=None</em>, <em class="sig-param">create_graph=False</em>, <em class="sig-param">strict=False</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/autograd/functional.html#vjp"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.autograd.functional.vjp" title="Permalink to this definition">¶</a></dt>
<dd><p>Function that computes the dot product between a vector <code class="docutils literal notranslate"><span class="pre">v</span></code> and the
Jacobian of the given function at the point given by the inputs.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><ul class="simple">
<li><p><strong>func</strong> (<em>function</em>) – a Python function that takes Tensor inputs and returns
a tuple of Tensors or a Tensor.</p></li>
<li><p><strong>inputs</strong> (<em>tuple of Tensors</em><em> or </em><a class="reference internal" href="tensors.html#torch.Tensor" title="torch.Tensor"><em>Tensor</em></a>) – inputs to the function <code class="docutils literal notranslate"><span class="pre">func</span></code>.</p></li>
<li><p><strong>v</strong> (<em>tuple of Tensors</em><em> or </em><a class="reference internal" href="tensors.html#torch.Tensor" title="torch.Tensor"><em>Tensor</em></a>) – The vector for which the vector
Jacobian product is computed. Must be the same size as the output
of <code class="docutils literal notranslate"><span class="pre">func</span></code>. This argument is optional when the output of <code class="docutils literal notranslate"><span class="pre">func</span></code>
contains a single element and (if it is not provided) will be set
as a Tensor containing a single <code class="docutils literal notranslate"><span class="pre">1</span></code>.</p></li>
<li><p><strong>create_graph</strong> (<a class="reference external" href="https://docs.python.org/3/library/functions.html#bool" title="(in Python v3.9)"><em>bool</em></a><em>, </em><em>optional</em>) – If <code class="docutils literal notranslate"><span class="pre">True</span></code>, both the output and result
will be computed in a differentiable way. Note that when <code class="docutils literal notranslate"><span class="pre">strict</span></code>
is <code class="docutils literal notranslate"><span class="pre">False</span></code>, the result can not require gradients or be
disconnected from the inputs. Defaults to <code class="docutils literal notranslate"><span class="pre">False</span></code>.</p></li>
<li><p><strong>strict</strong> (<a class="reference external" href="https://docs.python.org/3/library/functions.html#bool" title="(in Python v3.9)"><em>bool</em></a><em>, </em><em>optional</em>) – If <code class="docutils literal notranslate"><span class="pre">True</span></code>, an error will be raised when we
detect that there exists an input such that all the outputs are
independent of it. If <code class="docutils literal notranslate"><span class="pre">False</span></code>, we return a Tensor of zeros as the
vjp for said inputs, which is the expected mathematical value.
Defaults to <code class="docutils literal notranslate"><span class="pre">False</span></code>.</p></li>
</ul>
</dd>
<dt class="field-even">Returns</dt>
<dd class="field-even"><p><dl>
<dt>tuple with:</dt><dd><p>func_output (tuple of Tensors or Tensor): output of <code class="docutils literal notranslate"><span class="pre">func(inputs)</span></code></p>
<p>vjp (tuple of Tensors or Tensor): result of the dot product with
the same shape as the inputs.</p>
</dd>
</dl>
</p>
</dd>
<dt class="field-odd">Return type</dt>
<dd class="field-odd"><p>output (<a class="reference external" href="https://docs.python.org/3/library/stdtypes.html#tuple" title="(in Python v3.9)">tuple</a>)</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="k">def</span> <span class="nf">exp_reducer</span><span class="p">(</span><span class="n">x</span><span class="p">):</span>
<span class="gp">... </span> <span class="k">return</span> <span class="n">x</span><span class="o">.</span><span class="n">exp</span><span class="p">()</span><span class="o">.</span><span class="n">sum</span><span class="p">(</span><span class="n">dim</span><span class="o">=</span><span class="mi">1</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">inputs</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">4</span><span class="p">,</span> <span class="mi">4</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">v</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">ones</span><span class="p">(</span><span class="mi">4</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">vjp</span><span class="p">(</span><span class="n">exp_reducer</span><span class="p">,</span> <span class="n">inputs</span><span class="p">,</span> <span class="n">v</span><span class="p">)</span>
<span class="go">(tensor([5.7817, 7.2458, 5.7830, 6.7782]),</span>
<span class="go"> tensor([[1.4458, 1.3962, 1.3042, 1.6354],</span>
<span class="go"> [2.1288, 1.0652, 1.5483, 2.5035],</span>
<span class="go"> [2.2046, 1.1292, 1.1432, 1.3059],</span>
<span class="go"> [1.3225, 1.6652, 1.7753, 2.0152]]))</span>
</pre></div>
</div>
<div class="doctest highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="n">vjp</span><span class="p">(</span><span class="n">exp_reducer</span><span class="p">,</span> <span class="n">inputs</span><span class="p">,</span> <span class="n">v</span><span class="p">,</span> <span class="n">create_graph</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span>
<span class="go">(tensor([5.7817, 7.2458, 5.7830, 6.7782], grad_fn=<SumBackward1>),</span>
<span class="go"> tensor([[1.4458, 1.3962, 1.3042, 1.6354],</span>
<span class="go"> [2.1288, 1.0652, 1.5483, 2.5035],</span>
<span class="go"> [2.2046, 1.1292, 1.1432, 1.3059],</span>
<span class="go"> [1.3225, 1.6652, 1.7753, 2.0152]], grad_fn=<MulBackward0>))</span>
</pre></div>
</div>
<div class="doctest highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="k">def</span> <span class="nf">adder</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="mi">2</span> <span class="o">*</span> <span class="n">x</span> <span class="o">+</span> <span class="mi">3</span> <span class="o">*</span> <span class="n">y</span>
<span class="gp">>>> </span><span class="n">inputs</span> <span class="o">=</span> <span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">rand</span><span class="p">(</span><span class="mi">2</span><span class="p">),</span> <span class="n">torch</span><span class="o">.</span><span class="n">rand</span><span class="p">(</span><span class="mi">2</span><span class="p">))</span>
<span class="gp">>>> </span><span class="n">v</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">ones</span><span class="p">(</span><span class="mi">2</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">vjp</span><span class="p">(</span><span class="n">adder</span><span class="p">,</span> <span class="n">inputs</span><span class="p">,</span> <span class="n">v</span><span class="p">)</span>
<span class="go">(tensor([2.4225, 2.3340]),</span>
<span class="go"> (tensor([2., 2.]), tensor([3., 3.])))</span>
</pre></div>
</div>
</dd></dl>
<dl class="function">
<dt id="torch.autograd.functional.jvp">
<code class="sig-prename descclassname">torch.autograd.functional.</code><code class="sig-name descname">jvp</code><span class="sig-paren">(</span><em class="sig-param">func</em>, <em class="sig-param">inputs</em>, <em class="sig-param">v=None</em>, <em class="sig-param">create_graph=False</em>, <em class="sig-param">strict=False</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/autograd/functional.html#jvp"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.autograd.functional.jvp" title="Permalink to this definition">¶</a></dt>
<dd><p>Function that computes the dot product between the Jacobian of
the given function at the point given by the inputs and a vector <code class="docutils literal notranslate"><span class="pre">v</span></code>.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><ul class="simple">
<li><p><strong>func</strong> (<em>function</em>) – a Python function that takes Tensor inputs and returns
a tuple of Tensors or a Tensor.</p></li>
<li><p><strong>inputs</strong> (<em>tuple of Tensors</em><em> or </em><a class="reference internal" href="tensors.html#torch.Tensor" title="torch.Tensor"><em>Tensor</em></a>) – inputs to the function <code class="docutils literal notranslate"><span class="pre">func</span></code>.</p></li>
<li><p><strong>v</strong> (<em>tuple of Tensors</em><em> or </em><a class="reference internal" href="tensors.html#torch.Tensor" title="torch.Tensor"><em>Tensor</em></a>) – The vector for which the Jacobian
vector product is computed. Must be the same size as the input of
<code class="docutils literal notranslate"><span class="pre">func</span></code>. This argument is optional when the input to <code class="docutils literal notranslate"><span class="pre">func</span></code>
contains a single element and (if it is not provided) will be set
as a Tensor containing a single <code class="docutils literal notranslate"><span class="pre">1</span></code>.</p></li>
<li><p><strong>create_graph</strong> (<a class="reference external" href="https://docs.python.org/3/library/functions.html#bool" title="(in Python v3.9)"><em>bool</em></a><em>, </em><em>optional</em>) – If <code class="docutils literal notranslate"><span class="pre">True</span></code>, both the output and result
will be computed in a differentiable way. Note that when <code class="docutils literal notranslate"><span class="pre">strict</span></code>
is <code class="docutils literal notranslate"><span class="pre">False</span></code>, the result can not require gradients or be
disconnected from the inputs. Defaults to <code class="docutils literal notranslate"><span class="pre">False</span></code>.</p></li>
<li><p><strong>strict</strong> (<a class="reference external" href="https://docs.python.org/3/library/functions.html#bool" title="(in Python v3.9)"><em>bool</em></a><em>, </em><em>optional</em>) – If <code class="docutils literal notranslate"><span class="pre">True</span></code>, an error will be raised when we
detect that there exists an input such that all the outputs are
independent of it. If <code class="docutils literal notranslate"><span class="pre">False</span></code>, we return a Tensor of zeros as the
jvp for said inputs, which is the expected mathematical value.
Defaults to <code class="docutils literal notranslate"><span class="pre">False</span></code>.</p></li>
</ul>
</dd>
<dt class="field-even">Returns</dt>
<dd class="field-even"><p><dl>
<dt>tuple with:</dt><dd><p>func_output (tuple of Tensors or Tensor): output of <code class="docutils literal notranslate"><span class="pre">func(inputs)</span></code></p>
<p>jvp (tuple of Tensors or Tensor): result of the dot product with
the same shape as the output.</p>
</dd>
</dl>
</p>
</dd>
<dt class="field-odd">Return type</dt>
<dd class="field-odd"><p>output (<a class="reference external" href="https://docs.python.org/3/library/stdtypes.html#tuple" title="(in Python v3.9)">tuple</a>)</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="k">def</span> <span class="nf">exp_reducer</span><span class="p">(</span><span class="n">x</span><span class="p">):</span>
<span class="gp">... </span> <span class="k">return</span> <span class="n">x</span><span class="o">.</span><span class="n">exp</span><span class="p">()</span><span class="o">.</span><span class="n">sum</span><span class="p">(</span><span class="n">dim</span><span class="o">=</span><span class="mi">1</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">inputs</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">4</span><span class="p">,</span> <span class="mi">4</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">v</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">ones</span><span class="p">(</span><span class="mi">4</span><span class="p">,</span> <span class="mi">4</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">jvp</span><span class="p">(</span><span class="n">exp_reducer</span><span class="p">,</span> <span class="n">inputs</span><span class="p">,</span> <span class="n">v</span><span class="p">)</span>
<span class="go">(tensor([6.3090, 4.6742, 7.9114, 8.2106]),</span>
<span class="go"> tensor([6.3090, 4.6742, 7.9114, 8.2106]))</span>
</pre></div>
</div>
<div class="doctest highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="n">jvp</span><span class="p">(</span><span class="n">exp_reducer</span><span class="p">,</span> <span class="n">inputs</span><span class="p">,</span> <span class="n">v</span><span class="p">,</span> <span class="n">create_graph</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span>
<span class="go">(tensor([6.3090, 4.6742, 7.9114, 8.2106], grad_fn=<SumBackward1>),</span>
<span class="go"> tensor([6.3090, 4.6742, 7.9114, 8.2106], grad_fn=<SqueezeBackward1>))</span>
</pre></div>
</div>
<div class="doctest highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="k">def</span> <span class="nf">adder</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="mi">2</span> <span class="o">*</span> <span class="n">x</span> <span class="o">+</span> <span class="mi">3</span> <span class="o">*</span> <span class="n">y</span>
<span class="gp">>>> </span><span class="n">inputs</span> <span class="o">=</span> <span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">rand</span><span class="p">(</span><span class="mi">2</span><span class="p">),</span> <span class="n">torch</span><span class="o">.</span><span class="n">rand</span><span class="p">(</span><span class="mi">2</span><span class="p">))</span>
<span class="gp">>>> </span><span class="n">v</span> <span class="o">=</span> <span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">ones</span><span class="p">(</span><span class="mi">2</span><span class="p">),</span> <span class="n">torch</span><span class="o">.</span><span class="n">ones</span><span class="p">(</span><span class="mi">2</span><span class="p">))</span>
<span class="gp">>>> </span><span class="n">jvp</span><span class="p">(</span><span class="n">adder</span><span class="p">,</span> <span class="n">inputs</span><span class="p">,</span> <span class="n">v</span><span class="p">)</span>
<span class="go">(tensor([2.2399, 2.5005]),</span>
<span class="go"> tensor([5., 5.]))</span>
</pre></div>
</div>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>The jvp is currently computed by using the backward of the backward
(sometimes called the double backwards trick) as we don’t have support
for forward mode AD in PyTorch at the moment.</p>
</div>
</dd></dl>
<dl class="function">
<dt id="torch.autograd.functional.vhp">
<code class="sig-prename descclassname">torch.autograd.functional.</code><code class="sig-name descname">vhp</code><span class="sig-paren">(</span><em class="sig-param">func</em>, <em class="sig-param">inputs</em>, <em class="sig-param">v=None</em>, <em class="sig-param">create_graph=False</em>, <em class="sig-param">strict=False</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/autograd/functional.html#vhp"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.autograd.functional.vhp" title="Permalink to this definition">¶</a></dt>
<dd><p>Function that computes the dot product between a vector <code class="docutils literal notranslate"><span class="pre">v</span></code> and the
Hessian of a given scalar function at the point given by the inputs.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><ul class="simple">
<li><p><strong>func</strong> (<em>function</em>) – a Python function that takes Tensor inputs and returns
a Tensor with a single element.</p></li>
<li><p><strong>inputs</strong> (<em>tuple of Tensors</em><em> or </em><a class="reference internal" href="tensors.html#torch.Tensor" title="torch.Tensor"><em>Tensor</em></a>) – inputs to the function <code class="docutils literal notranslate"><span class="pre">func</span></code>.</p></li>
<li><p><strong>v</strong> (<em>tuple of Tensors</em><em> or </em><a class="reference internal" href="tensors.html#torch.Tensor" title="torch.Tensor"><em>Tensor</em></a>) – The vector for which the vector Hessian
product is computed. Must be the same size as the input of
<code class="docutils literal notranslate"><span class="pre">func</span></code>. This argument is optional when <code class="docutils literal notranslate"><span class="pre">func</span></code>’s input contains
a single element and (if it is not provided) will be set as a
Tensor containing a single <code class="docutils literal notranslate"><span class="pre">1</span></code>.</p></li>
<li><p><strong>create_graph</strong> (<a class="reference external" href="https://docs.python.org/3/library/functions.html#bool" title="(in Python v3.9)"><em>bool</em></a><em>, </em><em>optional</em>) – If <code class="docutils literal notranslate"><span class="pre">True</span></code>, both the output and result
will be computed in a differentiable way. Note that when <code class="docutils literal notranslate"><span class="pre">strict</span></code>
is <code class="docutils literal notranslate"><span class="pre">False</span></code>, the result can not require gradients or be
disconnected from the inputs.
Defaults to <code class="docutils literal notranslate"><span class="pre">False</span></code>.</p></li>
<li><p><strong>strict</strong> (<a class="reference external" href="https://docs.python.org/3/library/functions.html#bool" title="(in Python v3.9)"><em>bool</em></a><em>, </em><em>optional</em>) – If <code class="docutils literal notranslate"><span class="pre">True</span></code>, an error will be raised when we
detect that there exists an input such that all the outputs are
independent of it. If <code class="docutils literal notranslate"><span class="pre">False</span></code>, we return a Tensor of zeros as the
vhp for said inputs, which is the expected mathematical value.
Defaults to <code class="docutils literal notranslate"><span class="pre">False</span></code>.</p></li>
</ul>
</dd>
<dt class="field-even">Returns</dt>
<dd class="field-even"><p><dl>
<dt>tuple with:</dt><dd><p>func_output (tuple of Tensors or Tensor): output of <code class="docutils literal notranslate"><span class="pre">func(inputs)</span></code></p>
<p>vhp (tuple of Tensors or Tensor): result of the dot product with the
same shape as the inputs.</p>
</dd>
</dl>
</p>
</dd>
<dt class="field-odd">Return type</dt>
<dd class="field-odd"><p>output (<a class="reference external" href="https://docs.python.org/3/library/stdtypes.html#tuple" title="(in Python v3.9)">tuple</a>)</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="k">def</span> <span class="nf">pow_reducer</span><span class="p">(</span><span class="n">x</span><span class="p">):</span>
<span class="gp">... </span> <span class="k">return</span> <span class="n">x</span><span class="o">.</span><span class="n">pow</span><span class="p">(</span><span class="mi">3</span><span class="p">)</span><span class="o">.</span><span class="n">sum</span><span class="p">()</span>
<span class="gp">>>> </span><span class="n">inputs</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">2</span><span class="p">,</span> <span class="mi">2</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">v</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">ones</span><span class="p">(</span><span class="mi">2</span><span class="p">,</span> <span class="mi">2</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">vhp</span><span class="p">(</span><span class="n">pow_reducer</span><span class="p">,</span> <span class="n">inputs</span><span class="p">,</span> <span class="n">v</span><span class="p">)</span>
<span class="go">(tensor(0.5591),</span>
<span class="go"> tensor([[1.0689, 1.2431],</span>
<span class="go"> [3.0989, 4.4456]]))</span>
<span class="gp">>>> </span><span class="n">vhp</span><span class="p">(</span><span class="n">pow_reducer</span><span class="p">,</span> <span class="n">inputs</span><span class="p">,</span> <span class="n">v</span><span class="p">,</span> <span class="n">create_graph</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span>
<span class="go">(tensor(0.5591, grad_fn=<SumBackward0>),</span>
<span class="go"> tensor([[1.0689, 1.2431],</span>
<span class="go"> [3.0989, 4.4456]], grad_fn=<MulBackward0>))</span>
<span class="gp">>>> </span><span class="k">def</span> <span class="nf">pow_adder_reducer</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="p">(</span><span class="mi">2</span> <span class="o">*</span> <span class="n">x</span><span class="o">.</span><span class="n">pow</span><span class="p">(</span><span class="mi">2</span><span class="p">)</span> <span class="o">+</span> <span class="mi">3</span> <span class="o">*</span> <span class="n">y</span><span class="o">.</span><span class="n">pow</span><span class="p">(</span><span class="mi">2</span><span class="p">))</span><span class="o">.</span><span class="n">sum</span><span class="p">()</span>
<span class="gp">>>> </span><span class="n">inputs</span> <span class="o">=</span> <span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">rand</span><span class="p">(</span><span class="mi">2</span><span class="p">),</span> <span class="n">torch</span><span class="o">.</span><span class="n">rand</span><span class="p">(</span><span class="mi">2</span><span class="p">))</span>
<span class="gp">>>> </span><span class="n">v</span> <span class="o">=</span> <span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">zeros</span><span class="p">(</span><span class="mi">2</span><span class="p">),</span> <span class="n">torch</span><span class="o">.</span><span class="n">ones</span><span class="p">(</span><span class="mi">2</span><span class="p">))</span>
<span class="gp">>>> </span><span class="n">vhp</span><span class="p">(</span><span class="n">pow_adder_reducer</span><span class="p">,</span> <span class="n">inputs</span><span class="p">,</span> <span class="n">v</span><span class="p">)</span>
<span class="go">(tensor(4.8053),</span>
<span class="go"> (tensor([0., 0.]),</span>
<span class="go"> tensor([6., 6.])))</span>
</pre></div>
</div>
</dd></dl>
<dl class="function">
<dt id="torch.autograd.functional.hvp">
<code class="sig-prename descclassname">torch.autograd.functional.</code><code class="sig-name descname">hvp</code><span class="sig-paren">(</span><em class="sig-param">func</em>, <em class="sig-param">inputs</em>, <em class="sig-param">v=None</em>, <em class="sig-param">create_graph=False</em>, <em class="sig-param">strict=False</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/autograd/functional.html#hvp"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.autograd.functional.hvp" title="Permalink to this definition">¶</a></dt>
<dd><p>Function that computes the dot product between the Hessian of a given scalar
function and a vector <code class="docutils literal notranslate"><span class="pre">v</span></code> at the point given by the inputs.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><ul class="simple">
<li><p><strong>func</strong> (<em>function</em>) – a Python function that takes Tensor inputs and returns
a Tensor with a single element.</p></li>
<li><p><strong>inputs</strong> (<em>tuple of Tensors</em><em> or </em><a class="reference internal" href="tensors.html#torch.Tensor" title="torch.Tensor"><em>Tensor</em></a>) – inputs to the function <code class="docutils literal notranslate"><span class="pre">func</span></code>.</p></li>
<li><p><strong>v</strong> (<em>tuple of Tensors</em><em> or </em><a class="reference internal" href="tensors.html#torch.Tensor" title="torch.Tensor"><em>Tensor</em></a>) – The vector for which the Hessian vector
product is computed. Must be the same size as the input of
<code class="docutils literal notranslate"><span class="pre">func</span></code>. This argument is optional when <code class="docutils literal notranslate"><span class="pre">func</span></code>’s input contains
a single element and (if it is not provided) will be set as a
Tensor containing a single <code class="docutils literal notranslate"><span class="pre">1</span></code>.</p></li>
<li><p><strong>create_graph</strong> (<a class="reference external" href="https://docs.python.org/3/library/functions.html#bool" title="(in Python v3.9)"><em>bool</em></a><em>, </em><em>optional</em>) – If <code class="docutils literal notranslate"><span class="pre">True</span></code>, both the output and result will be
computed in a differentiable way. Note that when <code class="docutils literal notranslate"><span class="pre">strict</span></code> is
<code class="docutils literal notranslate"><span class="pre">False</span></code>, the result can not require gradients or be disconnected
from the inputs. Defaults to <code class="docutils literal notranslate"><span class="pre">False</span></code>.</p></li>
<li><p><strong>strict</strong> (<a class="reference external" href="https://docs.python.org/3/library/functions.html#bool" title="(in Python v3.9)"><em>bool</em></a><em>, </em><em>optional</em>) – If <code class="docutils literal notranslate"><span class="pre">True</span></code>, an error will be raised when we
detect that there exists an input such that all the outputs are
independent of it. If <code class="docutils literal notranslate"><span class="pre">False</span></code>, we return a Tensor of zeros as the
hvp for said inputs, which is the expected mathematical value.
Defaults to <code class="docutils literal notranslate"><span class="pre">False</span></code>.</p></li>
</ul>
</dd>
<dt class="field-even">Returns</dt>
<dd class="field-even"><p><dl>
<dt>tuple with:</dt><dd><p>func_output (tuple of Tensors or Tensor): output of <code class="docutils literal notranslate"><span class="pre">func(inputs)</span></code></p>
<p>hvp (tuple of Tensors or Tensor): result of the dot product with
the same shape as the inputs.</p>
</dd>
</dl>
</p>
</dd>
<dt class="field-odd">Return type</dt>
<dd class="field-odd"><p>output (<a class="reference external" href="https://docs.python.org/3/library/stdtypes.html#tuple" title="(in Python v3.9)">tuple</a>)</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="k">def</span> <span class="nf">pow_reducer</span><span class="p">(</span><span class="n">x</span><span class="p">):</span>
<span class="gp">... </span> <span class="k">return</span> <span class="n">x</span><span class="o">.</span><span class="n">pow</span><span class="p">(</span><span class="mi">3</span><span class="p">)</span><span class="o">.</span><span class="n">sum</span><span class="p">()</span>
<span class="gp">>>> </span><span class="n">inputs</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">2</span><span class="p">,</span> <span class="mi">2</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">v</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">ones</span><span class="p">(</span><span class="mi">2</span><span class="p">,</span> <span class="mi">2</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">hvp</span><span class="p">(</span><span class="n">pow_reducer</span><span class="p">,</span> <span class="n">inputs</span><span class="p">,</span> <span class="n">v</span><span class="p">)</span>
<span class="go">(tensor(0.1448),</span>
<span class="go"> tensor([[2.0239, 1.6456],</span>
<span class="go"> [2.4988, 1.4310]]))</span>
</pre></div>
</div>
<div class="doctest highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="n">hvp</span><span class="p">(</span><span class="n">pow_reducer</span><span class="p">,</span> <span class="n">inputs</span><span class="p">,</span> <span class="n">v</span><span class="p">,</span> <span class="n">create_graph</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span>
<span class="go">(tensor(0.1448, grad_fn=<SumBackward0>),</span>
<span class="go"> tensor([[2.0239, 1.6456],</span>
<span class="go"> [2.4988, 1.4310]], grad_fn=<MulBackward0>))</span>
</pre></div>
</div>
<div class="doctest highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="k">def</span> <span class="nf">pow_adder_reducer</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="p">(</span><span class="mi">2</span> <span class="o">*</span> <span class="n">x</span><span class="o">.</span><span class="n">pow</span><span class="p">(</span><span class="mi">2</span><span class="p">)</span> <span class="o">+</span> <span class="mi">3</span> <span class="o">*</span> <span class="n">y</span><span class="o">.</span><span class="n">pow</span><span class="p">(</span><span class="mi">2</span><span class="p">))</span><span class="o">.</span><span class="n">sum</span><span class="p">()</span>
<span class="gp">>>> </span><span class="n">inputs</span> <span class="o">=</span> <span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">rand</span><span class="p">(</span><span class="mi">2</span><span class="p">),</span> <span class="n">torch</span><span class="o">.</span><span class="n">rand</span><span class="p">(</span><span class="mi">2</span><span class="p">))</span>
<span class="gp">>>> </span><span class="n">v</span> <span class="o">=</span> <span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">zeros</span><span class="p">(</span><span class="mi">2</span><span class="p">),</span> <span class="n">torch</span><span class="o">.</span><span class="n">ones</span><span class="p">(</span><span class="mi">2</span><span class="p">))</span>
<span class="gp">>>> </span><span class="n">hvp</span><span class="p">(</span><span class="n">pow_adder_reducer</span><span class="p">,</span> <span class="n">inputs</span><span class="p">,</span> <span class="n">v</span><span class="p">)</span>
<span class="go">(tensor(2.3030),</span>
<span class="go"> (tensor([0., 0.]),</span>
<span class="go"> tensor([6., 6.])))</span>
</pre></div>
</div>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>This function is significantly slower than <cite>vhp</cite> due to backward mode AD constraints.
If your functions is twice continuously differentiable, then hvp = vhp.t(). So if you
know that your function satisfies this condition, you should use vhp instead that is
much faster with the current implementation.</p>
</div>
</dd></dl>
</div>
<div class="section" id="locally-disabling-gradient-computation">
<span id="locally-disable-grad"></span><h2>Locally disabling gradient computation<a class="headerlink" href="#locally-disabling-gradient-computation" title="Permalink to this headline">¶</a></h2>
<dl class="class">
<dt id="torch.autograd.no_grad">
<em class="property">class </em><code class="sig-prename descclassname">torch.autograd.</code><code class="sig-name descname">no_grad</code><a class="reference internal" href="_modules/torch/autograd/grad_mode.html#no_grad"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.autograd.no_grad" title="Permalink to this definition">¶</a></dt>
<dd><p>Context-manager that disabled gradient calculation.</p>
<p>Disabling gradient calculation is useful for inference, when you are sure
that you will not call <code class="xref py py-meth docutils literal notranslate"><span class="pre">Tensor.backward()</span></code>. It will reduce memory
consumption for computations that would otherwise have <cite>requires_grad=True</cite>.</p>
<p>In this mode, the result of every computation will have
<cite>requires_grad=False</cite>, even when the inputs have <cite>requires_grad=True</cite>.</p>
<p>This context manager is thread local; it will not affect computation
in other threads.</p>
<p>Also functions as a decorator. (Make sure to instantiate with parenthesis.)</p>
<p>Example:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></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">tensor</span><span class="p">([</span><span class="mi">1</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="k">with</span> <span class="n">torch</span><span class="o">.</span><span class="n">no_grad</span><span class="p">():</span>
<span class="gp">... </span> <span class="n">y</span> <span class="o">=</span> <span class="n">x</span> <span class="o">*</span> <span class="mi">2</span>
<span class="gp">>>> </span><span class="n">y</span><span class="o">.</span><span class="n">requires_grad</span>
<span class="go">False</span>
<span class="gp">>>> </span><span class="nd">@torch</span><span class="o">.</span><span class="n">no_grad</span><span class="p">()</span>
<span class="gp">... </span><span class="k">def</span> <span class="nf">doubler</span><span class="p">(</span><span class="n">x</span><span class="p">):</span>
<span class="gp">... </span> <span class="k">return</span> <span class="n">x</span> <span class="o">*</span> <span class="mi">2</span>
<span class="gp">>>> </span><span class="n">z</span> <span class="o">=</span> <span class="n">doubler</span><span class="p">(</span><span class="n">x</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">z</span><span class="o">.</span><span class="n">requires_grad</span>
<span class="go">False</span>
</pre></div>
</div>
</dd></dl>
<dl class="class">
<dt id="torch.autograd.enable_grad">
<em class="property">class </em><code class="sig-prename descclassname">torch.autograd.</code><code class="sig-name descname">enable_grad</code><a class="reference internal" href="_modules/torch/autograd/grad_mode.html#enable_grad"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.autograd.enable_grad" title="Permalink to this definition">¶</a></dt>
<dd><p>Context-manager that enables gradient calculation.</p>
<p>Enables gradient calculation, if it has been disabled via <a class="reference internal" href="#torch.autograd.no_grad" title="torch.autograd.no_grad"><code class="xref py py-class docutils literal notranslate"><span class="pre">no_grad</span></code></a>
or <a class="reference internal" href="#torch.autograd.set_grad_enabled" title="torch.autograd.set_grad_enabled"><code class="xref py py-class docutils literal notranslate"><span class="pre">set_grad_enabled</span></code></a>.</p>
<p>This context manager is thread local; it will not affect computation
in other threads.</p>
<p>Also functions as a decorator. (Make sure to instantiate with parenthesis.)</p>
<p>Example:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></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">tensor</span><span class="p">([</span><span class="mi">1</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="k">with</span> <span class="n">torch</span><span class="o">.</span><span class="n">no_grad</span><span class="p">():</span>
<span class="gp">... </span> <span class="k">with</span> <span class="n">torch</span><span class="o">.</span><span class="n">enable_grad</span><span class="p">():</span>
<span class="gp">... </span> <span class="n">y</span> <span class="o">=</span> <span class="n">x</span> <span class="o">*</span> <span class="mi">2</span>
<span class="gp">>>> </span><span class="n">y</span><span class="o">.</span><span class="n">requires_grad</span>
<span class="go">True</span>
<span class="gp">>>> </span><span class="n">y</span><span class="o">.</span><span class="n">backward</span><span class="p">()</span>
<span class="gp">>>> </span><span class="n">x</span><span class="o">.</span><span class="n">grad</span>
<span class="gp">>>> </span><span class="nd">@torch</span><span class="o">.</span><span class="n">enable_grad</span><span class="p">()</span>
<span class="gp">... </span><span class="k">def</span> <span class="nf">doubler</span><span class="p">(</span><span class="n">x</span><span class="p">):</span>
<span class="gp">... </span> <span class="k">return</span> <span class="n">x</span> <span class="o">*</span> <span class="mi">2</span>
<span class="gp">>>> </span><span class="k">with</span> <span class="n">torch</span><span class="o">.</span><span class="n">no_grad</span><span class="p">():</span>
<span class="gp">... </span> <span class="n">z</span> <span class="o">=</span> <span class="n">doubler</span><span class="p">(</span><span class="n">x</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">z</span><span class="o">.</span><span class="n">requires_grad</span>
<span class="go">True</span>