forked from pytorch/pytorch.github.io
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathautograd.html
797 lines (666 loc) · 60.7 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
<!DOCTYPE html>
<!--[if IE 8]><html class="no-js lt-ie9" lang="en" > <![endif]-->
<!--[if gt IE 8]><!--> <html class="no-js" lang="en" > <!--<![endif]-->
<head>
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>Autograd mechanics — PyTorch master documentation</title>
<link rel="stylesheet" href="../_static/css/theme.css" type="text/css" />
<link rel="stylesheet" href="../_static/pygments.css" type="text/css" />
<link rel="stylesheet" href="https://fonts.googleapis.com/css?family=Lato" type="text/css" />
<link rel="stylesheet" href="../_static/css/pytorch_theme.css" type="text/css" />
<link rel="index" title="Index" href="../genindex.html" />
<link rel="search" title="Search" href="../search.html" />
<link rel="next" title="Broadcasting semantics" href="broadcasting.html" />
<link rel="prev" title="PyTorch documentation" href="../index.html" />
<script src="../_static/js/modernizr.min.js"></script>
</head>
<body class="wy-body-for-nav">
<div class="wy-grid-for-nav">
<nav data-toggle="wy-nav-shift" class="wy-nav-side">
<div class="wy-side-scroll">
<div class="wy-side-nav-search">
<a href="../index.html">
<img src="../_static/pytorch-logo-dark.svg" class="logo" alt="Logo"/>
</a>
<div class="version">
0.3.1 <br/> <a href="https://pytorch.org/docs/versions.html"> version selector ▼</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>
<div class="wy-menu wy-menu-vertical" data-spy="affix" role="navigation" aria-label="main navigation">
<p class="caption"><span class="caption-text">Notes</span></p>
<ul class="current">
<li class="toctree-l1 current"><a class="current reference internal" href="#">Autograd mechanics</a><ul>
<li class="toctree-l2"><a class="reference internal" href="#excluding-subgraphs-from-backward">Excluding subgraphs from backward</a><ul>
<li class="toctree-l3"><a class="reference internal" href="#requires-grad"><code class="docutils literal"><span class="pre">requires_grad</span></code></a></li>
<li class="toctree-l3"><a class="reference internal" href="#volatile"><code class="docutils literal"><span class="pre">volatile</span></code></a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="#how-autograd-encodes-the-history">How autograd encodes the history</a></li>
<li class="toctree-l2"><a class="reference internal" href="#in-place-operations-on-variables">In-place operations on Variables</a></li>
<li class="toctree-l2"><a class="reference internal" href="#in-place-correctness-checks">In-place correctness checks</a></li>
</ul>
</li>
<li class="toctree-l1"><a class="reference internal" href="broadcasting.html">Broadcasting semantics</a><ul>
<li class="toctree-l2"><a class="reference internal" href="broadcasting.html#general-semantics">General semantics</a></li>
<li class="toctree-l2"><a class="reference internal" href="broadcasting.html#in-place-semantics">In-place semantics</a></li>
<li class="toctree-l2"><a class="reference internal" href="broadcasting.html#backwards-compatibility">Backwards compatibility</a></li>
</ul>
</li>
<li class="toctree-l1"><a class="reference internal" href="cuda.html">CUDA semantics</a><ul>
<li class="toctree-l2"><a class="reference internal" href="cuda.html#asynchronous-execution">Asynchronous execution</a><ul>
<li class="toctree-l3"><a class="reference internal" href="cuda.html#cuda-streams">CUDA streams</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="cuda.html#memory-management">Memory management</a></li>
<li class="toctree-l2"><a class="reference internal" href="cuda.html#best-practices">Best practices</a><ul>
<li class="toctree-l3"><a class="reference internal" href="cuda.html#device-agnostic-code">Device-agnostic code</a></li>
<li class="toctree-l3"><a class="reference internal" href="cuda.html#use-pinned-memory-buffers">Use pinned memory buffers</a></li>
<li class="toctree-l3"><a class="reference internal" href="cuda.html#use-nn-dataparallel-instead-of-multiprocessing">Use nn.DataParallel instead of multiprocessing</a></li>
</ul>
</li>
</ul>
</li>
<li class="toctree-l1"><a class="reference internal" href="extending.html">Extending PyTorch</a><ul>
<li class="toctree-l2"><a class="reference internal" href="extending.html#extending-torch-autograd">Extending <code class="docutils literal"><span class="pre">torch.autograd</span></code></a></li>
<li class="toctree-l2"><a class="reference internal" href="extending.html#extending-torch-nn">Extending <code class="docutils literal"><span class="pre">torch.nn</span></code></a><ul>
<li class="toctree-l3"><a class="reference internal" href="extending.html#adding-a-module">Adding a <code class="docutils literal"><span class="pre">Module</span></code></a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="extending.html#writing-custom-c-extensions">Writing custom C extensions</a></li>
</ul>
</li>
<li class="toctree-l1"><a class="reference internal" href="multiprocessing.html">Multiprocessing best practices</a><ul>
<li class="toctree-l2"><a class="reference internal" href="multiprocessing.html#sharing-cuda-tensors">Sharing CUDA tensors</a></li>
<li class="toctree-l2"><a class="reference internal" href="multiprocessing.html#best-practices-and-tips">Best practices and tips</a><ul>
<li class="toctree-l3"><a class="reference internal" href="multiprocessing.html#avoiding-and-fighting-deadlocks">Avoiding and fighting deadlocks</a></li>
<li class="toctree-l3"><a class="reference internal" href="multiprocessing.html#reuse-buffers-passed-through-a-queue">Reuse buffers passed through a Queue</a></li>
<li class="toctree-l3"><a class="reference internal" href="multiprocessing.html#asynchronous-multiprocess-training-e-g-hogwild">Asynchronous multiprocess training (e.g. Hogwild)</a><ul>
<li class="toctree-l4"><a class="reference internal" href="multiprocessing.html#hogwild">Hogwild</a></li>
</ul>
</li>
</ul>
</li>
</ul>
</li>
<li class="toctree-l1"><a class="reference internal" href="serialization.html">Serialization semantics</a><ul>
<li class="toctree-l2"><a class="reference internal" href="serialization.html#best-practices">Best practices</a><ul>
<li class="toctree-l3"><a class="reference internal" href="serialization.html#recommended-approach-for-saving-a-model">Recommended approach for saving a model</a></li>
</ul>
</li>
</ul>
</li>
</ul>
<p class="caption"><span class="caption-text">Package Reference</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../torch.html">torch</a><ul>
<li class="toctree-l2"><a class="reference internal" href="../torch.html#tensors">Tensors</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../torch.html#creation-ops">Creation Ops</a></li>
<li class="toctree-l3"><a class="reference internal" href="../torch.html#indexing-slicing-joining-mutating-ops">Indexing, Slicing, Joining, Mutating Ops</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../torch.html#random-sampling">Random sampling</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../torch.html#in-place-random-sampling">In-place random sampling</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../torch.html#serialization">Serialization</a></li>
<li class="toctree-l2"><a class="reference internal" href="../torch.html#parallelism">Parallelism</a></li>
<li class="toctree-l2"><a class="reference internal" href="../torch.html#math-operations">Math operations</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../torch.html#pointwise-ops">Pointwise Ops</a></li>
<li class="toctree-l3"><a class="reference internal" href="../torch.html#reduction-ops">Reduction Ops</a></li>
<li class="toctree-l3"><a class="reference internal" href="../torch.html#comparison-ops">Comparison Ops</a></li>
<li class="toctree-l3"><a class="reference internal" href="../torch.html#other-operations">Other Operations</a></li>
<li class="toctree-l3"><a class="reference internal" href="../torch.html#blas-and-lapack-operations">BLAS and LAPACK Operations</a></li>
</ul>
</li>
</ul>
</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="../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="../nn.html">torch.nn</a><ul>
<li class="toctree-l2"><a class="reference internal" href="../nn.html#parameters">Parameters</a></li>
<li class="toctree-l2"><a class="reference internal" href="../nn.html#containers">Containers</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#module"><span class="hidden-section">Module</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#sequential"><span class="hidden-section">Sequential</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#modulelist"><span class="hidden-section">ModuleList</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#parameterlist"><span class="hidden-section">ParameterList</span></a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../nn.html#convolution-layers">Convolution Layers</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#conv1d"><span class="hidden-section">Conv1d</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#conv2d"><span class="hidden-section">Conv2d</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#conv3d"><span class="hidden-section">Conv3d</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#convtranspose1d"><span class="hidden-section">ConvTranspose1d</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#convtranspose2d"><span class="hidden-section">ConvTranspose2d</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#convtranspose3d"><span class="hidden-section">ConvTranspose3d</span></a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../nn.html#pooling-layers">Pooling Layers</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#maxpool1d"><span class="hidden-section">MaxPool1d</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#maxpool2d"><span class="hidden-section">MaxPool2d</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#maxpool3d"><span class="hidden-section">MaxPool3d</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#maxunpool1d"><span class="hidden-section">MaxUnpool1d</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#maxunpool2d"><span class="hidden-section">MaxUnpool2d</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#maxunpool3d"><span class="hidden-section">MaxUnpool3d</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#avgpool1d"><span class="hidden-section">AvgPool1d</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#avgpool2d"><span class="hidden-section">AvgPool2d</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#avgpool3d"><span class="hidden-section">AvgPool3d</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#fractionalmaxpool2d"><span class="hidden-section">FractionalMaxPool2d</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#lppool2d"><span class="hidden-section">LPPool2d</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#adaptivemaxpool1d"><span class="hidden-section">AdaptiveMaxPool1d</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#adaptivemaxpool2d"><span class="hidden-section">AdaptiveMaxPool2d</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#adaptivemaxpool3d"><span class="hidden-section">AdaptiveMaxPool3d</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#adaptiveavgpool1d"><span class="hidden-section">AdaptiveAvgPool1d</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#adaptiveavgpool2d"><span class="hidden-section">AdaptiveAvgPool2d</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#adaptiveavgpool3d"><span class="hidden-section">AdaptiveAvgPool3d</span></a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../nn.html#padding-layers">Padding Layers</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#reflectionpad2d"><span class="hidden-section">ReflectionPad2d</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#replicationpad2d"><span class="hidden-section">ReplicationPad2d</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#replicationpad3d"><span class="hidden-section">ReplicationPad3d</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#zeropad2d"><span class="hidden-section">ZeroPad2d</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#constantpad2d"><span class="hidden-section">ConstantPad2d</span></a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../nn.html#non-linear-activations">Non-linear Activations</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#relu"><span class="hidden-section">ReLU</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#relu6"><span class="hidden-section">ReLU6</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#elu"><span class="hidden-section">ELU</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#selu"><span class="hidden-section">SELU</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#prelu"><span class="hidden-section">PReLU</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#leakyrelu"><span class="hidden-section">LeakyReLU</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#threshold"><span class="hidden-section">Threshold</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#hardtanh"><span class="hidden-section">Hardtanh</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#sigmoid"><span class="hidden-section">Sigmoid</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#tanh"><span class="hidden-section">Tanh</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#logsigmoid"><span class="hidden-section">LogSigmoid</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#softplus"><span class="hidden-section">Softplus</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#softshrink"><span class="hidden-section">Softshrink</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#softsign"><span class="hidden-section">Softsign</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#tanhshrink"><span class="hidden-section">Tanhshrink</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#softmin"><span class="hidden-section">Softmin</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#softmax"><span class="hidden-section">Softmax</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#softmax2d"><span class="hidden-section">Softmax2d</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#logsoftmax"><span class="hidden-section">LogSoftmax</span></a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../nn.html#normalization-layers">Normalization layers</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#batchnorm1d"><span class="hidden-section">BatchNorm1d</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#batchnorm2d"><span class="hidden-section">BatchNorm2d</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#batchnorm3d"><span class="hidden-section">BatchNorm3d</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#instancenorm1d"><span class="hidden-section">InstanceNorm1d</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#instancenorm2d"><span class="hidden-section">InstanceNorm2d</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#instancenorm3d"><span class="hidden-section">InstanceNorm3d</span></a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../nn.html#recurrent-layers">Recurrent layers</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#rnn"><span class="hidden-section">RNN</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#lstm"><span class="hidden-section">LSTM</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#gru"><span class="hidden-section">GRU</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#rnncell"><span class="hidden-section">RNNCell</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#lstmcell"><span class="hidden-section">LSTMCell</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#grucell"><span class="hidden-section">GRUCell</span></a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../nn.html#linear-layers">Linear layers</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#linear"><span class="hidden-section">Linear</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#bilinear"><span class="hidden-section">Bilinear</span></a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../nn.html#dropout-layers">Dropout layers</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#dropout"><span class="hidden-section">Dropout</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#dropout2d"><span class="hidden-section">Dropout2d</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#dropout3d"><span class="hidden-section">Dropout3d</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#alphadropout"><span class="hidden-section">AlphaDropout</span></a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../nn.html#sparse-layers">Sparse layers</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#embedding"><span class="hidden-section">Embedding</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#embeddingbag"><span class="hidden-section">EmbeddingBag</span></a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../nn.html#distance-functions">Distance functions</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#cosinesimilarity"><span class="hidden-section">CosineSimilarity</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#pairwisedistance"><span class="hidden-section">PairwiseDistance</span></a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../nn.html#loss-functions">Loss functions</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#l1loss"><span class="hidden-section">L1Loss</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#mseloss"><span class="hidden-section">MSELoss</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#crossentropyloss"><span class="hidden-section">CrossEntropyLoss</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#nllloss"><span class="hidden-section">NLLLoss</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#poissonnllloss"><span class="hidden-section">PoissonNLLLoss</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#nllloss2d"><span class="hidden-section">NLLLoss2d</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#kldivloss"><span class="hidden-section">KLDivLoss</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#bceloss"><span class="hidden-section">BCELoss</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#bcewithlogitsloss"><span class="hidden-section">BCEWithLogitsLoss</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#marginrankingloss"><span class="hidden-section">MarginRankingLoss</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#hingeembeddingloss"><span class="hidden-section">HingeEmbeddingLoss</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#multilabelmarginloss"><span class="hidden-section">MultiLabelMarginLoss</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#smoothl1loss"><span class="hidden-section">SmoothL1Loss</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#softmarginloss"><span class="hidden-section">SoftMarginLoss</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#multilabelsoftmarginloss"><span class="hidden-section">MultiLabelSoftMarginLoss</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#cosineembeddingloss"><span class="hidden-section">CosineEmbeddingLoss</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#multimarginloss"><span class="hidden-section">MultiMarginLoss</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#tripletmarginloss"><span class="hidden-section">TripletMarginLoss</span></a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../nn.html#vision-layers">Vision layers</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#pixelshuffle"><span class="hidden-section">PixelShuffle</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#upsample"><span class="hidden-section">Upsample</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#upsamplingnearest2d"><span class="hidden-section">UpsamplingNearest2d</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#upsamplingbilinear2d"><span class="hidden-section">UpsamplingBilinear2d</span></a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../nn.html#dataparallel-layers-multi-gpu-distributed">DataParallel layers (multi-GPU, distributed)</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#dataparallel"><span class="hidden-section">DataParallel</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#distributeddataparallel"><span class="hidden-section">DistributedDataParallel</span></a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../nn.html#utilities">Utilities</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#clip-grad-norm"><span class="hidden-section">clip_grad_norm</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#weight-norm"><span class="hidden-section">weight_norm</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#remove-weight-norm"><span class="hidden-section">remove_weight_norm</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#packedsequence"><span class="hidden-section">PackedSequence</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#pack-padded-sequence"><span class="hidden-section">pack_padded_sequence</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#pad-packed-sequence"><span class="hidden-section">pad_packed_sequence</span></a></li>
</ul>
</li>
</ul>
</li>
<li class="toctree-l1"><a class="reference internal" href="../nn.html#torch-nn-functional">torch.nn.functional</a><ul>
<li class="toctree-l2"><a class="reference internal" href="../nn.html#convolution-functions">Convolution functions</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#id14"><span class="hidden-section">conv1d</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#id15"><span class="hidden-section">conv2d</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#id16"><span class="hidden-section">conv3d</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#conv-transpose1d"><span class="hidden-section">conv_transpose1d</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#conv-transpose2d"><span class="hidden-section">conv_transpose2d</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#conv-transpose3d"><span class="hidden-section">conv_transpose3d</span></a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../nn.html#pooling-functions">Pooling functions</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#avg-pool1d"><span class="hidden-section">avg_pool1d</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#avg-pool2d"><span class="hidden-section">avg_pool2d</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#avg-pool3d"><span class="hidden-section">avg_pool3d</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#max-pool1d"><span class="hidden-section">max_pool1d</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#max-pool2d"><span class="hidden-section">max_pool2d</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#max-pool3d"><span class="hidden-section">max_pool3d</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#max-unpool1d"><span class="hidden-section">max_unpool1d</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#max-unpool2d"><span class="hidden-section">max_unpool2d</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#max-unpool3d"><span class="hidden-section">max_unpool3d</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#lp-pool2d"><span class="hidden-section">lp_pool2d</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#adaptive-max-pool1d"><span class="hidden-section">adaptive_max_pool1d</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#adaptive-max-pool2d"><span class="hidden-section">adaptive_max_pool2d</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#adaptive-max-pool3d"><span class="hidden-section">adaptive_max_pool3d</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#adaptive-avg-pool1d"><span class="hidden-section">adaptive_avg_pool1d</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#adaptive-avg-pool2d"><span class="hidden-section">adaptive_avg_pool2d</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#adaptive-avg-pool3d"><span class="hidden-section">adaptive_avg_pool3d</span></a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../nn.html#non-linear-activation-functions">Non-linear activation functions</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#id17"><span class="hidden-section">threshold</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#id18"><span class="hidden-section">relu</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#id19"><span class="hidden-section">hardtanh</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#id20"><span class="hidden-section">relu6</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#id21"><span class="hidden-section">elu</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#id22"><span class="hidden-section">selu</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#leaky-relu"><span class="hidden-section">leaky_relu</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#id23"><span class="hidden-section">prelu</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#rrelu"><span class="hidden-section">rrelu</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#glu"><span class="hidden-section">glu</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#id24"><span class="hidden-section">logsigmoid</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#hardshrink"><span class="hidden-section">hardshrink</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#id25"><span class="hidden-section">tanhshrink</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#id26"><span class="hidden-section">softsign</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#id27"><span class="hidden-section">softplus</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#id28"><span class="hidden-section">softmin</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#id29"><span class="hidden-section">softmax</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#id30"><span class="hidden-section">softshrink</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#log-softmax"><span class="hidden-section">log_softmax</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#id31"><span class="hidden-section">tanh</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#id32"><span class="hidden-section">sigmoid</span></a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../nn.html#normalization-functions">Normalization functions</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#batch-norm"><span class="hidden-section">batch_norm</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#normalize"><span class="hidden-section">normalize</span></a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../nn.html#linear-functions">Linear functions</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#id33"><span class="hidden-section">linear</span></a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../nn.html#dropout-functions">Dropout functions</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#id34"><span class="hidden-section">dropout</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#alpha-dropout"><span class="hidden-section">alpha_dropout</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#id35"><span class="hidden-section">dropout2d</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#id36"><span class="hidden-section">dropout3d</span></a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../nn.html#id37">Distance functions</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#pairwise-distance"><span class="hidden-section">pairwise_distance</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#cosine-similarity"><span class="hidden-section">cosine_similarity</span></a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../nn.html#id38">Loss functions</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#binary-cross-entropy"><span class="hidden-section">binary_cross_entropy</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#poisson-nll-loss"><span class="hidden-section">poisson_nll_loss</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#cosine-embedding-loss"><span class="hidden-section">cosine_embedding_loss</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#cross-entropy"><span class="hidden-section">cross_entropy</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#hinge-embedding-loss"><span class="hidden-section">hinge_embedding_loss</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#kl-div"><span class="hidden-section">kl_div</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#l1-loss"><span class="hidden-section">l1_loss</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#mse-loss"><span class="hidden-section">mse_loss</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#margin-ranking-loss"><span class="hidden-section">margin_ranking_loss</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#multilabel-margin-loss"><span class="hidden-section">multilabel_margin_loss</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#multilabel-soft-margin-loss"><span class="hidden-section">multilabel_soft_margin_loss</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#multi-margin-loss"><span class="hidden-section">multi_margin_loss</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#nll-loss"><span class="hidden-section">nll_loss</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#binary-cross-entropy-with-logits"><span class="hidden-section">binary_cross_entropy_with_logits</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#smooth-l1-loss"><span class="hidden-section">smooth_l1_loss</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#soft-margin-loss"><span class="hidden-section">soft_margin_loss</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#triplet-margin-loss"><span class="hidden-section">triplet_margin_loss</span></a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../nn.html#vision-functions">Vision functions</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#pixel-shuffle"><span class="hidden-section">pixel_shuffle</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#pad"><span class="hidden-section">pad</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#id40"><span class="hidden-section">upsample</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#upsample-nearest"><span class="hidden-section">upsample_nearest</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#upsample-bilinear"><span class="hidden-section">upsample_bilinear</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#grid-sample"><span class="hidden-section">grid_sample</span></a></li>
<li class="toctree-l3"><a class="reference internal" href="../nn.html#affine-grid"><span class="hidden-section">affine_grid</span></a></li>
</ul>
</li>
</ul>
</li>
<li class="toctree-l1"><a class="reference internal" href="../nn.html#torch-nn-init">torch.nn.init</a></li>
<li class="toctree-l1"><a class="reference internal" href="../optim.html">torch.optim</a><ul>
<li class="toctree-l2"><a class="reference internal" href="../optim.html#how-to-use-an-optimizer">How to use an optimizer</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../optim.html#constructing-it">Constructing it</a></li>
<li class="toctree-l3"><a class="reference internal" href="../optim.html#per-parameter-options">Per-parameter options</a></li>
<li class="toctree-l3"><a class="reference internal" href="../optim.html#taking-an-optimization-step">Taking an optimization step</a><ul>
<li class="toctree-l4"><a class="reference internal" href="../optim.html#optimizer-step"><code class="docutils literal"><span class="pre">optimizer.step()</span></code></a></li>
<li class="toctree-l4"><a class="reference internal" href="../optim.html#optimizer-step-closure"><code class="docutils literal"><span class="pre">optimizer.step(closure)</span></code></a></li>
</ul>
</li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../optim.html#algorithms">Algorithms</a></li>
<li class="toctree-l2"><a class="reference internal" href="../optim.html#how-to-adjust-learning-rate">How to adjust Learning Rate</a></li>
</ul>
</li>
<li class="toctree-l1"><a class="reference internal" href="../autograd.html">torch.autograd</a><ul>
<li class="toctree-l2"><a class="reference internal" href="../autograd.html#variable">Variable</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../autograd.html#api-compatibility">API compatibility</a></li>
<li class="toctree-l3"><a class="reference internal" href="../autograd.html#in-place-operations-on-variables">In-place operations on Variables</a></li>
<li class="toctree-l3"><a class="reference internal" href="../autograd.html#in-place-correctness-checks">In-place correctness checks</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../autograd.html#function"><span class="hidden-section">Function</span></a></li>
<li class="toctree-l2"><a class="reference internal" href="../autograd.html#profiler">Profiler</a></li>
</ul>
</li>
<li class="toctree-l1"><a class="reference internal" href="../distributions.html">torch.distributions</a><ul>
<li class="toctree-l2"><a class="reference internal" href="../distributions.html#distribution"><span class="hidden-section">Distribution</span></a></li>
<li class="toctree-l2"><a class="reference internal" href="../distributions.html#bernoulli"><span class="hidden-section">Bernoulli</span></a></li>
<li class="toctree-l2"><a class="reference internal" href="../distributions.html#categorical"><span class="hidden-section">Categorical</span></a></li>
<li class="toctree-l2"><a class="reference internal" href="../distributions.html#normal"><span class="hidden-section">Normal</span></a></li>
</ul>
</li>
<li class="toctree-l1"><a class="reference internal" href="../multiprocessing.html">torch.multiprocessing</a><ul>
<li class="toctree-l2"><a class="reference internal" href="../multiprocessing.html#strategy-management">Strategy management</a></li>
<li class="toctree-l2"><a class="reference internal" href="../multiprocessing.html#sharing-cuda-tensors">Sharing CUDA tensors</a></li>
<li class="toctree-l2"><a class="reference internal" href="../multiprocessing.html#sharing-strategies">Sharing strategies</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../multiprocessing.html#file-descriptor-file-descriptor">File descriptor - <code class="docutils literal"><span class="pre">file_descriptor</span></code></a></li>
<li class="toctree-l3"><a class="reference internal" href="../multiprocessing.html#file-system-file-system">File system - <code class="docutils literal"><span class="pre">file_system</span></code></a></li>
</ul>
</li>
</ul>
</li>
<li class="toctree-l1"><a class="reference internal" href="../distributed.html">torch.distributed</a><ul>
<li class="toctree-l2"><a class="reference internal" href="../distributed.html#basics">Basics</a></li>
<li class="toctree-l2"><a class="reference internal" href="../distributed.html#initialization">Initialization</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../distributed.html#tcp-initialization">TCP initialization</a></li>
<li class="toctree-l3"><a class="reference internal" href="../distributed.html#shared-file-system-initialization">Shared file-system initialization</a></li>
<li class="toctree-l3"><a class="reference internal" href="../distributed.html#environment-variable-initialization">Environment variable initialization</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../distributed.html#groups">Groups</a></li>
<li class="toctree-l2"><a class="reference internal" href="../distributed.html#point-to-point-communication">Point-to-point communication</a></li>
<li class="toctree-l2"><a class="reference internal" href="../distributed.html#collective-functions">Collective functions</a></li>
</ul>
</li>
<li class="toctree-l1"><a class="reference internal" href="../legacy.html">torch.legacy</a></li>
<li class="toctree-l1"><a class="reference internal" href="../cuda.html">torch.cuda</a><ul>
<li class="toctree-l2"><a class="reference internal" href="../cuda.html#random-number-generator">Random Number Generator</a></li>
<li class="toctree-l2"><a class="reference internal" href="../cuda.html#communication-collectives">Communication collectives</a></li>
<li class="toctree-l2"><a class="reference internal" href="../cuda.html#streams-and-events">Streams and events</a></li>
<li class="toctree-l2"><a class="reference internal" href="../cuda.html#memory-management">Memory management</a></li>
<li class="toctree-l2"><a class="reference internal" href="../cuda.html#nvidia-tools-extension-nvtx">NVIDIA Tools Extension (NVTX)</a></li>
</ul>
</li>
<li class="toctree-l1"><a class="reference internal" href="../ffi.html">torch.utils.ffi</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="../model_zoo.html">torch.utils.model_zoo</a></li>
<li class="toctree-l1"><a class="reference internal" href="../onnx.html">torch.onnx</a><ul>
<li class="toctree-l2"><a class="reference internal" href="../onnx.html#example-end-to-end-alexnet-from-pytorch-to-caffe2">Example: End-to-end AlexNet from PyTorch to Caffe2</a></li>
<li class="toctree-l2"><a class="reference internal" href="../onnx.html#limitations">Limitations</a></li>
<li class="toctree-l2"><a class="reference internal" href="../onnx.html#supported-operators">Supported operators</a></li>
<li class="toctree-l2"><a class="reference internal" href="../onnx.html#functions">Functions</a></li>
</ul>
</li>
</ul>
</div>
</div>
</nav>
<section data-toggle="wy-nav-shift" class="wy-nav-content-wrap">
<nav class="wy-nav-top" aria-label="top navigation">
<i data-toggle="wy-nav-top" class="fa fa-bars"></i>
<a href="../index.html">PyTorch</a>
</nav>
<div class="wy-nav-content">
<div class="rst-content">
<div role="navigation" aria-label="breadcrumbs navigation">
<ul class="wy-breadcrumbs">
<li><a href="../index.html">Docs</a> »</li>
<li>Autograd mechanics</li>
<li class="wy-breadcrumbs-aside">
<a href="../_sources/notes/autograd.rst.txt" rel="nofollow"> View page source</a>
</li>
</ul>
<hr/>
</div>
<div role="main" class="document" itemscope="itemscope" itemtype="http://schema.org/Article">
<div itemprop="articleBody">
<div class="section" id="autograd-mechanics">
<h1>Autograd mechanics<a class="headerlink" href="#autograd-mechanics" title="Permalink to this headline">¶</a></h1>
<p>This note will present an overview of how autograd works and records the
operations. It’s not strictly necessary to understand all this, but we recommend
getting familiar with it, as it will help you write more efficient, cleaner
programs, and can aid you in debugging.</p>
<div class="section" id="excluding-subgraphs-from-backward">
<span id="excluding-subgraphs"></span><h2>Excluding subgraphs from backward<a class="headerlink" href="#excluding-subgraphs-from-backward" title="Permalink to this headline">¶</a></h2>
<p>Every Variable has two flags: <code class="xref py py-attr docutils literal"><span class="pre">requires_grad</span></code> and <code class="xref py py-attr docutils literal"><span class="pre">volatile</span></code>.
They both allow for fine grained exclusion of subgraphs from gradient
computation and can increase efficiency.</p>
<div class="section" id="requires-grad">
<span id="excluding-requires-grad"></span><h3><code class="docutils literal"><span class="pre">requires_grad</span></code><a class="headerlink" href="#requires-grad" title="Permalink to this headline">¶</a></h3>
<p>If there’s a single input to an operation that requires gradient, its output
will also require gradient. Conversely, only if all inputs don’t require
gradient, the output also won’t require it. Backward computation is never
performed in the subgraphs, where all Variables didn’t require gradients.</p>
<div class="code highlight-default"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="n">x</span> <span class="o">=</span> <span class="n">Variable</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">randn</span><span class="p">(</span><span class="mi">5</span><span class="p">,</span> <span class="mi">5</span><span class="p">))</span>
<span class="gp">>>> </span><span class="n">y</span> <span class="o">=</span> <span class="n">Variable</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">randn</span><span class="p">(</span><span class="mi">5</span><span class="p">,</span> <span class="mi">5</span><span class="p">))</span>
<span class="gp">>>> </span><span class="n">z</span> <span class="o">=</span> <span class="n">Variable</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">randn</span><span class="p">(</span><span class="mi">5</span><span class="p">,</span> <span class="mi">5</span><span class="p">),</span> <span class="n">requires_grad</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">a</span> <span class="o">=</span> <span class="n">x</span> <span class="o">+</span> <span class="n">y</span>
<span class="gp">>>> </span><span class="n">a</span><span class="o">.</span><span class="n">requires_grad</span>
<span class="go">False</span>
<span class="gp">>>> </span><span class="n">b</span> <span class="o">=</span> <span class="n">a</span> <span class="o">+</span> <span class="n">z</span>
<span class="gp">>>> </span><span class="n">b</span><span class="o">.</span><span class="n">requires_grad</span>
<span class="go">True</span>
</pre></div>
</div>
<p>This is especially useful when you want to freeze part of your model, or you
know in advance that you’re not going to use gradients w.r.t. some parameters.
For example if you want to finetune a pretrained CNN, it’s enough to switch the
<code class="xref py py-attr docutils literal"><span class="pre">requires_grad</span></code> flags in the frozen base, and no intermediate buffers will
be saved, until the computation gets to the last layer, where the affine
transform will use weights that require gradient, and the output of the network
will also require them.</p>
<div class="code highlight-default"><div class="highlight"><pre><span></span><span class="n">model</span> <span class="o">=</span> <span class="n">torchvision</span><span class="o">.</span><span class="n">models</span><span class="o">.</span><span class="n">resnet18</span><span class="p">(</span><span class="n">pretrained</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span>
<span class="k">for</span> <span class="n">param</span> <span class="ow">in</span> <span class="n">model</span><span class="o">.</span><span class="n">parameters</span><span class="p">():</span>
<span class="n">param</span><span class="o">.</span><span class="n">requires_grad</span> <span class="o">=</span> <span class="kc">False</span>
<span class="c1"># Replace the last fully-connected layer</span>
<span class="c1"># Parameters of newly constructed modules have requires_grad=True by default</span>
<span class="n">model</span><span class="o">.</span><span class="n">fc</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">Linear</span><span class="p">(</span><span class="mi">512</span><span class="p">,</span> <span class="mi">100</span><span class="p">)</span>
<span class="c1"># Optimize only the classifier</span>
<span class="n">optimizer</span> <span class="o">=</span> <span class="n">optim</span><span class="o">.</span><span class="n">SGD</span><span class="p">(</span><span class="n">model</span><span class="o">.</span><span class="n">fc</span><span class="o">.</span><span class="n">parameters</span><span class="p">(),</span> <span class="n">lr</span><span class="o">=</span><span class="mf">1e-2</span><span class="p">,</span> <span class="n">momentum</span><span class="o">=</span><span class="mf">0.9</span><span class="p">)</span>
</pre></div>
</div>
</div>
<div class="section" id="volatile">
<h3><code class="docutils literal"><span class="pre">volatile</span></code><a class="headerlink" href="#volatile" title="Permalink to this headline">¶</a></h3>
<p>Volatile is recommended for purely inference mode, when you’re sure you won’t
be even calling <cite>.backward()</cite>. It’s more efficient than any other autograd
setting - it will use the absolute minimal amount of memory to evaluate the
model. <code class="docutils literal"><span class="pre">volatile</span></code> also determines that <code class="docutils literal"><span class="pre">requires_grad</span> <span class="pre">is</span> <span class="pre">False</span></code>.</p>
<p>Volatile differs from <a class="reference internal" href="#excluding-requires-grad"><span class="std std-ref">requires_grad</span></a> in how the flag propagates.
If there’s even a single volatile input to an operation, its output is also
going to be volatile. Volatility spreads across the graph much easier than
non-requiring gradient - you only need a <strong>single</strong> volatile leaf to have a
volatile output, while you need <strong>all</strong> leaves to not require gradient to
have an output that doesn’t require gradient. Using volatile flag you don’t
need to change any settings of your model parameters to use it for
inference. It’s enough to create a volatile input, and this will ensure that
no intermediate states are saved.</p>
<div class="code highlight-default"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="n">regular_input</span> <span class="o">=</span> <span class="n">Variable</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">randn</span><span class="p">(</span><span class="mi">1</span><span class="p">,</span> <span class="mi">3</span><span class="p">,</span> <span class="mi">227</span><span class="p">,</span> <span class="mi">227</span><span class="p">))</span>
<span class="gp">>>> </span><span class="n">volatile_input</span> <span class="o">=</span> <span class="n">Variable</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">randn</span><span class="p">(</span><span class="mi">1</span><span class="p">,</span> <span class="mi">3</span><span class="p">,</span> <span class="mi">227</span><span class="p">,</span> <span class="mi">227</span><span class="p">),</span> <span class="n">volatile</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">model</span> <span class="o">=</span> <span class="n">torchvision</span><span class="o">.</span><span class="n">models</span><span class="o">.</span><span class="n">resnet18</span><span class="p">(</span><span class="n">pretrained</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">model</span><span class="p">(</span><span class="n">regular_input</span><span class="p">)</span><span class="o">.</span><span class="n">requires_grad</span>
<span class="go">True</span>
<span class="gp">>>> </span><span class="n">model</span><span class="p">(</span><span class="n">volatile_input</span><span class="p">)</span><span class="o">.</span><span class="n">requires_grad</span>
<span class="go">False</span>
<span class="gp">>>> </span><span class="n">model</span><span class="p">(</span><span class="n">volatile_input</span><span class="p">)</span><span class="o">.</span><span class="n">volatile</span>
<span class="go">True</span>
<span class="gp">>>> </span><span class="n">model</span><span class="p">(</span><span class="n">volatile_input</span><span class="p">)</span><span class="o">.</span><span class="n">grad_fn</span> <span class="ow">is</span> <span class="kc">None</span>
<span class="go">True</span>
</pre></div>
</div>
</div>
</div>
<div class="section" id="how-autograd-encodes-the-history">
<h2>How autograd encodes the history<a class="headerlink" href="#how-autograd-encodes-the-history" title="Permalink to this headline">¶</a></h2>
<p>Autograd is reverse automatic differentiation system. Conceptually,
autograd records a graph recording all of the operations that created
the data as you execute operations, giving you a directed acyclic graph
whose leaves are the input variables and roots are the output variables.
By tracing this graph from roots to leaves, you can automatically
compute the gradients using the chain rule.</p>
<p>Internally, autograd represents this graph as a graph of
<code class="xref py py-class docutils literal"><span class="pre">Function</span></code> objects (really expressions), which can be
<code class="xref py py-meth docutils literal"><span class="pre">apply()</span></code> ed to compute the result of
evaluating the graph. When computing the forwards pass, autograd
simultaneously performs the requested computations and builds up a graph
representing the function that computes the gradient (the <code class="docutils literal"><span class="pre">.grad_fn</span></code>
attribute of each <code class="xref py py-class docutils literal"><span class="pre">Variable</span></code> is an entry point into this graph).
When the forwards pass is completed, we evaluate this graph in the
backwards pass to compute the gradients.</p>
<p>An important thing to note is that the graph is recreated from scratch at every
iteration, and this is exactly what allows for using arbitrary Python control
flow statements, that can change the overall shape and size of the graph at
every iteration. You don’t have to encode all possible paths before you
launch the training - what you run is what you differentiate.</p>
</div>
<div class="section" id="in-place-operations-on-variables">
<h2>In-place operations on Variables<a class="headerlink" href="#in-place-operations-on-variables" title="Permalink to this headline">¶</a></h2>
<p>Supporting in-place operations in autograd is a hard matter, and we discourage
their use in most cases. Autograd’s aggressive buffer freeing and reuse makes
it very efficient and there are very few occasions when in-place operations
actually lower memory usage by any significant amount. Unless you’re operating
under heavy memory pressure, you might never need to use them.</p>
<p>There are two main reasons that limit the applicability of in-place operations:</p>
<ol class="arabic simple">
<li>Overwriting values required to compute gradients. This is why variables don’t
support <code class="docutils literal"><span class="pre">log_</span></code>. Its gradient formula requires the original input, and while
it is possible to recreate it by computing the inverse operation, it is
numerically unstable, and requires additional work that often defeats the
purpose of using these functions.</li>
<li>Every in-place operation actually requires the implementation to rewrite the
computational graph. Out-of-place versions simply allocate new objects and
keep references to the old graph, while in-place operations, require
changing the creator of all inputs to the <code class="xref py py-class docutils literal"><span class="pre">Function</span></code> representing
this operation. This can be tricky, especially if there are many Variables
that reference the same storage (e.g. created by indexing or transposing),
and in-place functions will actually raise an error if the storage of
modified inputs is referenced by any other <code class="xref py py-class docutils literal"><span class="pre">Variable</span></code>.</li>
</ol>
</div>
<div class="section" id="in-place-correctness-checks">
<h2>In-place correctness checks<a class="headerlink" href="#in-place-correctness-checks" title="Permalink to this headline">¶</a></h2>
<p>Every variable keeps a version counter, that is incremented every time it’s
marked dirty in any operation. When a Function saves any tensors for backward,
a version counter of their containing Variable is saved as well. Once you access
<code class="docutils literal"><span class="pre">self.saved_tensors</span></code> it is checked, and if it’s greater than the saved value
an error is raised.</p>
</div>
</div>
</div>
</div>
<footer>
<div class="rst-footer-buttons" role="navigation" aria-label="footer navigation">
<a href="broadcasting.html" class="btn btn-neutral float-right" title="Broadcasting semantics" accesskey="n" rel="next">Next <span class="fa fa-arrow-circle-right"></span></a>
<a href="../index.html" class="btn btn-neutral" title="PyTorch documentation" accesskey="p" rel="prev"><span class="fa fa-arrow-circle-left"></span> Previous</a>
</div>
<hr/>
<div role="contentinfo">
<p>
© Copyright 2017, Torch Contributors.
</p>
</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>.
</footer>
</div>
</div>
</section>
</div>
<script type="text/javascript">
var DOCUMENTATION_OPTIONS = {
URL_ROOT:'../',
VERSION:'master',
LANGUAGE:'None',
COLLAPSE_INDEX:false,
FILE_SUFFIX:'.html',
HAS_SOURCE: true,
SOURCELINK_SUFFIX: '.txt'
};
</script>
<script type="text/javascript" src="../_static/jquery.js"></script>
<script type="text/javascript" src="../_static/underscore.js"></script>
<script type="text/javascript" src="../_static/doctools.js"></script>
<script type="text/javascript" src="https://cdn.mathjax.org/mathjax/latest/MathJax.js?config=TeX-AMS-MML_HTMLorMML"></script>
<script type="text/javascript" src="../_static/js/theme.js"></script>
<script type="text/javascript">
jQuery(function () {
SphinxRtdTheme.Navigation.enableSticky();
});
</script>
<script>
(function(i,s,o,g,r,a,m){i['GoogleAnalyticsObject']=r;i[r]=i[r]||function(){
(i[r].q=i[r].q||[]).push(arguments)},i[r].l=1*new Date();a=s.createElement(o),
m=s.getElementsByTagName(o)[0];a.async=1;a.src=g;m.parentNode.insertBefore(a,m)
})(window,document,'script','https://www.google-analytics.com/analytics.js','ga');
ga('create', 'UA-90545585-1', 'auto');
ga('send', 'pageview');
</script>
</body>
</html>