-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathcuda.html
1544 lines (1297 loc) · 99 KB
/
cuda.html
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
<!DOCTYPE html>
<!--[if IE 8]><html class="no-js lt-ie9" lang="en" > <![endif]-->
<!--[if gt IE 8]><!--> <html class="no-js" lang="en" > <!--<![endif]-->
<head>
<meta name="robots" content="noindex">
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>torch.cuda — PyTorch master documentation</title>
<link rel="canonical" href="https://pytorch.org/docs/stable/cuda.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="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="Distributed communication package - torch.distributed" href="distributed.html" />
<link rel="prev" title="Automatic differentiation package - torch.autograd" href="autograd.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">
</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>
<div class="ecosystem-dropdown">
<a id="dropdownMenuButton" data-toggle="ecosystem-dropdown">
Ecosystem
</a>
<div class="ecosystem-dropdown-menu">
<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>
<a class="nav-dropdown-item" href="https://pytorch.org/ecosystem">
<span class=dropdown-title>Tools & Libraries</span>
<p>Explore the ecosystem of tools and libraries</p>
</a>
</div>
</div>
</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">
<a href="https://pytorch.org/docs/stable/index.html">Docs</a>
</li>
<li>
<div class="resources-dropdown">
<a id="resourcesDropdownButton" data-toggle="resources-dropdown">
Resources
</a>
<div class="resources-dropdown-menu">
<a class="nav-dropdown-item" href="https://pytorch.org/resources"">
<span class=dropdown-title>Developer Resources</span>
<p>Find resources and get questions answered</p>
</a>
<a class="nav-dropdown-item" href="https://pytorch.org/features">
<span class=dropdown-title>About</span>
<p>Learn about PyTorch’s features and capabilities</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='http://pytorch.org/docs/versions.html'>1.4.0 ▼</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/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/distributed_autograd.html">Distributed Autograd Design</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/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/rref.html">Remote Reference Protocol</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>
<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">Language Bindings</span></p>
<ul>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/cppdocs/">C++ API</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="autograd.html">torch.autograd</a></li>
<li class="toctree-l1 current"><a class="current reference internal" href="#">torch.cuda</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="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="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="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="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="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">torchvision Reference</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="torchvision/index.html">torchvision</a></li>
</ul>
<p class="caption"><span class="caption-text">torchaudio Reference</span></p>
<ul>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/audio">torchaudio</a></li>
</ul>
<p class="caption"><span class="caption-text">torchtext Reference</span></p>
<ul>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/text">torchtext</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>torch.cuda</li>
<li class="pytorch-breadcrumbs-aside">
<a href="_sources/cuda.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.cuda">
<span id="torch-cuda"></span><h1>torch.cuda<a class="headerlink" href="#module-torch.cuda" title="Permalink to this headline">¶</a></h1>
<p>This package adds support for CUDA tensor types, that implement the same
function as CPU tensors, but they utilize GPUs for computation.</p>
<p>It is lazily initialized, so you can always import it, and use
<a class="reference internal" href="#torch.cuda.is_available" title="torch.cuda.is_available"><code class="xref py py-func docutils literal notranslate"><span class="pre">is_available()</span></code></a> to determine if your system supports CUDA.</p>
<p><a class="reference internal" href="notes/cuda.html#cuda-semantics"><span class="std std-ref">CUDA semantics</span></a> has more details about working with CUDA.</p>
<dl class="function">
<dt id="torch.cuda.current_blas_handle">
<code class="sig-prename descclassname">torch.cuda.</code><code class="sig-name descname">current_blas_handle</code><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/cuda.html#current_blas_handle"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.cuda.current_blas_handle" title="Permalink to this definition">¶</a></dt>
<dd><p>Returns cublasHandle_t pointer to current cuBLAS handle</p>
</dd></dl>
<dl class="function">
<dt id="torch.cuda.current_device">
<code class="sig-prename descclassname">torch.cuda.</code><code class="sig-name descname">current_device</code><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/cuda.html#current_device"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.cuda.current_device" title="Permalink to this definition">¶</a></dt>
<dd><p>Returns the index of a currently selected device.</p>
</dd></dl>
<dl class="function">
<dt id="torch.cuda.current_stream">
<code class="sig-prename descclassname">torch.cuda.</code><code class="sig-name descname">current_stream</code><span class="sig-paren">(</span><em class="sig-param">device=None</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/cuda.html#current_stream"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.cuda.current_stream" title="Permalink to this definition">¶</a></dt>
<dd><p>Returns the currently selected <a class="reference internal" href="#torch.cuda.Stream" title="torch.cuda.Stream"><code class="xref py py-class docutils literal notranslate"><span class="pre">Stream</span></code></a> for a given device.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><p><strong>device</strong> (<a class="reference internal" href="tensor_attributes.html#torch.torch.device" title="torch.torch.device"><em>torch.device</em></a><em> or </em><em>python:int</em><em>, </em><em>optional</em>) – selected device. Returns
the currently selected <a class="reference internal" href="#torch.cuda.Stream" title="torch.cuda.Stream"><code class="xref py py-class docutils literal notranslate"><span class="pre">Stream</span></code></a> for the current device, given
by <a class="reference internal" href="#torch.cuda.current_device" title="torch.cuda.current_device"><code class="xref py py-func docutils literal notranslate"><span class="pre">current_device()</span></code></a>, if <a class="reference internal" href="#torch.cuda.device" title="torch.cuda.device"><code class="xref py py-attr docutils literal notranslate"><span class="pre">device</span></code></a> is <code class="docutils literal notranslate"><span class="pre">None</span></code>
(default).</p>
</dd>
</dl>
</dd></dl>
<dl class="function">
<dt id="torch.cuda.default_stream">
<code class="sig-prename descclassname">torch.cuda.</code><code class="sig-name descname">default_stream</code><span class="sig-paren">(</span><em class="sig-param">device=None</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/cuda.html#default_stream"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.cuda.default_stream" title="Permalink to this definition">¶</a></dt>
<dd><p>Returns the default <a class="reference internal" href="#torch.cuda.Stream" title="torch.cuda.Stream"><code class="xref py py-class docutils literal notranslate"><span class="pre">Stream</span></code></a> for a given device.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><p><strong>device</strong> (<a class="reference internal" href="tensor_attributes.html#torch.torch.device" title="torch.torch.device"><em>torch.device</em></a><em> or </em><em>python:int</em><em>, </em><em>optional</em>) – selected device. Returns
the default <a class="reference internal" href="#torch.cuda.Stream" title="torch.cuda.Stream"><code class="xref py py-class docutils literal notranslate"><span class="pre">Stream</span></code></a> for the current device, given by
<a class="reference internal" href="#torch.cuda.current_device" title="torch.cuda.current_device"><code class="xref py py-func docutils literal notranslate"><span class="pre">current_device()</span></code></a>, if <a class="reference internal" href="#torch.cuda.device" title="torch.cuda.device"><code class="xref py py-attr docutils literal notranslate"><span class="pre">device</span></code></a> is <code class="docutils literal notranslate"><span class="pre">None</span></code>
(default).</p>
</dd>
</dl>
</dd></dl>
<dl class="class">
<dt id="torch.cuda.device">
<em class="property">class </em><code class="sig-prename descclassname">torch.cuda.</code><code class="sig-name descname">device</code><span class="sig-paren">(</span><em class="sig-param">device</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/cuda.html#device"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.cuda.device" title="Permalink to this definition">¶</a></dt>
<dd><p>Context-manager that changes the selected device.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><p><strong>device</strong> (<a class="reference internal" href="tensor_attributes.html#torch.torch.device" title="torch.torch.device"><em>torch.device</em></a><em> or </em><em>python:int</em>) – device index to select. It’s a no-op if
this argument is a negative integer or <code class="docutils literal notranslate"><span class="pre">None</span></code>.</p>
</dd>
</dl>
</dd></dl>
<dl class="function">
<dt id="torch.cuda.device_count">
<code class="sig-prename descclassname">torch.cuda.</code><code class="sig-name descname">device_count</code><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/cuda.html#device_count"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.cuda.device_count" title="Permalink to this definition">¶</a></dt>
<dd><p>Returns the number of GPUs available.</p>
</dd></dl>
<dl class="class">
<dt id="torch.cuda.device_of">
<em class="property">class </em><code class="sig-prename descclassname">torch.cuda.</code><code class="sig-name descname">device_of</code><span class="sig-paren">(</span><em class="sig-param">obj</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/cuda.html#device_of"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.cuda.device_of" title="Permalink to this definition">¶</a></dt>
<dd><p>Context-manager that changes the current device to that of given object.</p>
<p>You can use both tensors and storages as arguments. If a given object is
not allocated on a GPU, this is a no-op.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><p><strong>obj</strong> (<a class="reference internal" href="tensors.html#torch.Tensor" title="torch.Tensor"><em>Tensor</em></a><em> or </em><em>Storage</em>) – object allocated on the selected device.</p>
</dd>
</dl>
</dd></dl>
<dl class="function">
<dt id="torch.cuda.get_device_capability">
<code class="sig-prename descclassname">torch.cuda.</code><code class="sig-name descname">get_device_capability</code><span class="sig-paren">(</span><em class="sig-param">device=None</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/cuda.html#get_device_capability"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.cuda.get_device_capability" title="Permalink to this definition">¶</a></dt>
<dd><p>Gets the cuda capability of a device.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><p><strong>device</strong> (<a class="reference internal" href="tensor_attributes.html#torch.torch.device" title="torch.torch.device"><em>torch.device</em></a><em> or </em><em>python:int</em><em>, </em><em>optional</em>) – device for which to return the
device capability. This function is a no-op if this argument is
a negative integer. It uses the current device, given by
<a class="reference internal" href="#torch.cuda.current_device" title="torch.cuda.current_device"><code class="xref py py-func docutils literal notranslate"><span class="pre">current_device()</span></code></a>, if <a class="reference internal" href="#torch.cuda.device" title="torch.cuda.device"><code class="xref py py-attr docutils literal notranslate"><span class="pre">device</span></code></a> is <code class="docutils literal notranslate"><span class="pre">None</span></code>
(default).</p>
</dd>
<dt class="field-even">Returns</dt>
<dd class="field-even"><p>the major and minor cuda capability of the device</p>
</dd>
<dt class="field-odd">Return type</dt>
<dd class="field-odd"><p>tuple(int, int)</p>
</dd>
</dl>
</dd></dl>
<dl class="function">
<dt id="torch.cuda.get_device_name">
<code class="sig-prename descclassname">torch.cuda.</code><code class="sig-name descname">get_device_name</code><span class="sig-paren">(</span><em class="sig-param">device=None</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/cuda.html#get_device_name"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.cuda.get_device_name" title="Permalink to this definition">¶</a></dt>
<dd><p>Gets the name of a device.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><p><strong>device</strong> (<a class="reference internal" href="tensor_attributes.html#torch.torch.device" title="torch.torch.device"><em>torch.device</em></a><em> or </em><em>python:int</em><em>, </em><em>optional</em>) – device for which to return the
name. This function is a no-op if this argument is a negative
integer. It uses the current device, given by <a class="reference internal" href="#torch.cuda.current_device" title="torch.cuda.current_device"><code class="xref py py-func docutils literal notranslate"><span class="pre">current_device()</span></code></a>,
if <a class="reference internal" href="#torch.cuda.device" title="torch.cuda.device"><code class="xref py py-attr docutils literal notranslate"><span class="pre">device</span></code></a> is <code class="docutils literal notranslate"><span class="pre">None</span></code> (default).</p>
</dd>
</dl>
</dd></dl>
<dl class="function">
<dt id="torch.cuda.init">
<code class="sig-prename descclassname">torch.cuda.</code><code class="sig-name descname">init</code><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/cuda.html#init"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.cuda.init" title="Permalink to this definition">¶</a></dt>
<dd><p>Initialize PyTorch’s CUDA state. You may need to call
this explicitly if you are interacting with PyTorch via
its C API, as Python bindings for CUDA functionality will not
be until this initialization takes place. Ordinary users
should not need this, as all of PyTorch’s CUDA methods
automatically initialize CUDA state on-demand.</p>
<p>Does nothing if the CUDA state is already initialized.</p>
</dd></dl>
<dl class="function">
<dt id="torch.cuda.ipc_collect">
<code class="sig-prename descclassname">torch.cuda.</code><code class="sig-name descname">ipc_collect</code><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/cuda.html#ipc_collect"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.cuda.ipc_collect" title="Permalink to this definition">¶</a></dt>
<dd><p>Force collects GPU memory after it has been released by CUDA IPC.</p>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>Checks if any sent CUDA tensors could be cleaned from the memory. Force
closes shared memory file used for reference counting if there is no
active counters. Useful when the producer process stopped actively sending
tensors and want to release unused memory.</p>
</div>
</dd></dl>
<dl class="function">
<dt id="torch.cuda.is_available">
<code class="sig-prename descclassname">torch.cuda.</code><code class="sig-name descname">is_available</code><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/cuda.html#is_available"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.cuda.is_available" title="Permalink to this definition">¶</a></dt>
<dd><p>Returns a bool indicating if CUDA is currently available.</p>
</dd></dl>
<dl class="function">
<dt id="torch.cuda.is_initialized">
<code class="sig-prename descclassname">torch.cuda.</code><code class="sig-name descname">is_initialized</code><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/cuda.html#is_initialized"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.cuda.is_initialized" title="Permalink to this definition">¶</a></dt>
<dd><p>Returns whether PyTorch’s CUDA state has been initialized.</p>
</dd></dl>
<dl class="function">
<dt id="torch.cuda.set_device">
<code class="sig-prename descclassname">torch.cuda.</code><code class="sig-name descname">set_device</code><span class="sig-paren">(</span><em class="sig-param">device</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/cuda.html#set_device"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.cuda.set_device" title="Permalink to this definition">¶</a></dt>
<dd><p>Sets the current device.</p>
<p>Usage of this function is discouraged in favor of <a class="reference internal" href="#torch.cuda.device" title="torch.cuda.device"><code class="xref any py py-class docutils literal notranslate"><span class="pre">device</span></code></a>. In most
cases it’s better to use <code class="docutils literal notranslate"><span class="pre">CUDA_VISIBLE_DEVICES</span></code> environmental variable.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><p><strong>device</strong> (<a class="reference internal" href="tensor_attributes.html#torch.torch.device" title="torch.torch.device"><em>torch.device</em></a><em> or </em><em>python:int</em>) – selected device. This function is a no-op
if this argument is negative.</p>
</dd>
</dl>
</dd></dl>
<dl class="function">
<dt id="torch.cuda.stream">
<code class="sig-prename descclassname">torch.cuda.</code><code class="sig-name descname">stream</code><span class="sig-paren">(</span><em class="sig-param">stream</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/cuda.html#stream"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.cuda.stream" title="Permalink to this definition">¶</a></dt>
<dd><p>Context-manager that selects a given stream.</p>
<p>All CUDA kernels queued within its context will be enqueued on a selected
stream.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><p><strong>stream</strong> (<a class="reference internal" href="#torch.cuda.Stream" title="torch.cuda.Stream"><em>Stream</em></a>) – selected stream. This manager is a no-op if it’s
<code class="docutils literal notranslate"><span class="pre">None</span></code>.</p>
</dd>
</dl>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>Streams are per-device. If the selected stream is not on the
current device, this function will also change the current device to
match the stream.</p>
</div>
</dd></dl>
<dl class="function">
<dt id="torch.cuda.synchronize">
<code class="sig-prename descclassname">torch.cuda.</code><code class="sig-name descname">synchronize</code><span class="sig-paren">(</span><em class="sig-param">device=None</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/cuda.html#synchronize"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.cuda.synchronize" title="Permalink to this definition">¶</a></dt>
<dd><p>Waits for all kernels in all streams on a CUDA device to complete.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><p><strong>device</strong> (<a class="reference internal" href="tensor_attributes.html#torch.torch.device" title="torch.torch.device"><em>torch.device</em></a><em> or </em><em>python:int</em><em>, </em><em>optional</em>) – device for which to synchronize.
It uses the current device, given by <a class="reference internal" href="#torch.cuda.current_device" title="torch.cuda.current_device"><code class="xref py py-func docutils literal notranslate"><span class="pre">current_device()</span></code></a>,
if <a class="reference internal" href="#torch.cuda.device" title="torch.cuda.device"><code class="xref py py-attr docutils literal notranslate"><span class="pre">device</span></code></a> is <code class="docutils literal notranslate"><span class="pre">None</span></code> (default).</p>
</dd>
</dl>
</dd></dl>
<div class="section" id="random-number-generator">
<h2>Random Number Generator<a class="headerlink" href="#random-number-generator" title="Permalink to this headline">¶</a></h2>
<dl class="function">
<dt id="torch.cuda.get_rng_state">
<code class="sig-prename descclassname">torch.cuda.</code><code class="sig-name descname">get_rng_state</code><span class="sig-paren">(</span><em class="sig-param">device='cuda'</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/cuda/random.html#get_rng_state"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.cuda.get_rng_state" title="Permalink to this definition">¶</a></dt>
<dd><p>Returns the random number generator state of the specified GPU as a ByteTensor.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><p><strong>device</strong> (<a class="reference internal" href="tensor_attributes.html#torch.torch.device" title="torch.torch.device"><em>torch.device</em></a><em> or </em><em>python:int</em><em>, </em><em>optional</em>) – The device to return the RNG state of.
Default: <code class="docutils literal notranslate"><span class="pre">'cuda'</span></code> (i.e., <code class="docutils literal notranslate"><span class="pre">torch.device('cuda')</span></code>, the current CUDA device).</p>
</dd>
</dl>
<div class="admonition warning">
<p class="admonition-title">Warning</p>
<p>This function eagerly initializes CUDA.</p>
</div>
</dd></dl>
<dl class="function">
<dt id="torch.cuda.get_rng_state_all">
<code class="sig-prename descclassname">torch.cuda.</code><code class="sig-name descname">get_rng_state_all</code><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/cuda/random.html#get_rng_state_all"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.cuda.get_rng_state_all" title="Permalink to this definition">¶</a></dt>
<dd><p>Returns a tuple of ByteTensor representing the random number states of all devices.</p>
</dd></dl>
<dl class="function">
<dt id="torch.cuda.set_rng_state">
<code class="sig-prename descclassname">torch.cuda.</code><code class="sig-name descname">set_rng_state</code><span class="sig-paren">(</span><em class="sig-param">new_state</em>, <em class="sig-param">device='cuda'</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/cuda/random.html#set_rng_state"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.cuda.set_rng_state" title="Permalink to this definition">¶</a></dt>
<dd><p>Sets the random number generator state of the specified GPU.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><ul class="simple">
<li><p><strong>new_state</strong> (<em>torch.ByteTensor</em>) – The desired state</p></li>
<li><p><strong>device</strong> (<a class="reference internal" href="tensor_attributes.html#torch.torch.device" title="torch.torch.device"><em>torch.device</em></a><em> or </em><em>python:int</em><em>, </em><em>optional</em>) – The device to set the RNG state.
Default: <code class="docutils literal notranslate"><span class="pre">'cuda'</span></code> (i.e., <code class="docutils literal notranslate"><span class="pre">torch.device('cuda')</span></code>, the current CUDA device).</p></li>
</ul>
</dd>
</dl>
</dd></dl>
<dl class="function">
<dt id="torch.cuda.set_rng_state_all">
<code class="sig-prename descclassname">torch.cuda.</code><code class="sig-name descname">set_rng_state_all</code><span class="sig-paren">(</span><em class="sig-param">new_states</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/cuda/random.html#set_rng_state_all"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.cuda.set_rng_state_all" title="Permalink to this definition">¶</a></dt>
<dd><p>Sets the random number generator state of all devices.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><p><strong>new_state</strong> (<em>tuple of torch.ByteTensor</em>) – The desired state for each device</p>
</dd>
</dl>
</dd></dl>
<dl class="function">
<dt id="torch.cuda.manual_seed">
<code class="sig-prename descclassname">torch.cuda.</code><code class="sig-name descname">manual_seed</code><span class="sig-paren">(</span><em class="sig-param">seed</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/cuda/random.html#manual_seed"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.cuda.manual_seed" title="Permalink to this definition">¶</a></dt>
<dd><p>Sets the seed for generating random numbers for the current GPU.
It’s safe to call this function if CUDA is not available; in that
case, it is silently ignored.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><p><strong>seed</strong> (<em>python:int</em>) – The desired seed.</p>
</dd>
</dl>
<div class="admonition warning">
<p class="admonition-title">Warning</p>
<p>If you are working with a multi-GPU model, this function is insufficient
to get determinism. To seed all GPUs, use <a class="reference internal" href="#torch.cuda.manual_seed_all" title="torch.cuda.manual_seed_all"><code class="xref py py-func docutils literal notranslate"><span class="pre">manual_seed_all()</span></code></a>.</p>
</div>
</dd></dl>
<dl class="function">
<dt id="torch.cuda.manual_seed_all">
<code class="sig-prename descclassname">torch.cuda.</code><code class="sig-name descname">manual_seed_all</code><span class="sig-paren">(</span><em class="sig-param">seed</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/cuda/random.html#manual_seed_all"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.cuda.manual_seed_all" title="Permalink to this definition">¶</a></dt>
<dd><p>Sets the seed for generating random numbers on all GPUs.
It’s safe to call this function if CUDA is not available; in that
case, it is silently ignored.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><p><strong>seed</strong> (<em>python:int</em>) – The desired seed.</p>
</dd>
</dl>
</dd></dl>
<dl class="function">
<dt id="torch.cuda.seed">
<code class="sig-prename descclassname">torch.cuda.</code><code class="sig-name descname">seed</code><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/cuda/random.html#seed"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.cuda.seed" title="Permalink to this definition">¶</a></dt>
<dd><p>Sets the seed for generating random numbers to a random number for the current GPU.
It’s safe to call this function if CUDA is not available; in that
case, it is silently ignored.</p>
<div class="admonition warning">
<p class="admonition-title">Warning</p>
<p>If you are working with a multi-GPU model, this function will only initialize
the seed on one GPU. To initialize all GPUs, use <a class="reference internal" href="#torch.cuda.seed_all" title="torch.cuda.seed_all"><code class="xref py py-func docutils literal notranslate"><span class="pre">seed_all()</span></code></a>.</p>
</div>
</dd></dl>
<dl class="function">
<dt id="torch.cuda.seed_all">
<code class="sig-prename descclassname">torch.cuda.</code><code class="sig-name descname">seed_all</code><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/cuda/random.html#seed_all"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.cuda.seed_all" title="Permalink to this definition">¶</a></dt>
<dd><p>Sets the seed for generating random numbers to a random number on all GPUs.
It’s safe to call this function if CUDA is not available; in that
case, it is silently ignored.</p>
</dd></dl>
<dl class="function">
<dt id="torch.cuda.initial_seed">
<code class="sig-prename descclassname">torch.cuda.</code><code class="sig-name descname">initial_seed</code><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/cuda/random.html#initial_seed"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.cuda.initial_seed" title="Permalink to this definition">¶</a></dt>
<dd><p>Returns the current random seed of the current GPU.</p>
<div class="admonition warning">
<p class="admonition-title">Warning</p>
<p>This function eagerly initializes CUDA.</p>
</div>
</dd></dl>
</div>
<div class="section" id="communication-collectives">
<h2>Communication collectives<a class="headerlink" href="#communication-collectives" title="Permalink to this headline">¶</a></h2>
<dl class="function">
<dt id="torch.cuda.comm.broadcast">
<code class="sig-prename descclassname">torch.cuda.comm.</code><code class="sig-name descname">broadcast</code><span class="sig-paren">(</span><em class="sig-param">tensor</em>, <em class="sig-param">devices</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/cuda/comm.html#broadcast"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.cuda.comm.broadcast" title="Permalink to this definition">¶</a></dt>
<dd><p>Broadcasts a tensor to a number of GPUs.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><ul class="simple">
<li><p><strong>tensor</strong> (<a class="reference internal" href="tensors.html#torch.Tensor" title="torch.Tensor"><em>Tensor</em></a>) – tensor to broadcast.</p></li>
<li><p><strong>devices</strong> (<em>Iterable</em>) – an iterable of devices among which to broadcast.
Note that it should be like (src, dst1, dst2, …), the first element
of which is the source device to broadcast from.</p></li>
</ul>
</dd>
<dt class="field-even">Returns</dt>
<dd class="field-even"><p>A tuple containing copies of the <code class="docutils literal notranslate"><span class="pre">tensor</span></code>, placed on devices
corresponding to indices from <code class="docutils literal notranslate"><span class="pre">devices</span></code>.</p>
</dd>
</dl>
</dd></dl>
<dl class="function">
<dt id="torch.cuda.comm.broadcast_coalesced">
<code class="sig-prename descclassname">torch.cuda.comm.</code><code class="sig-name descname">broadcast_coalesced</code><span class="sig-paren">(</span><em class="sig-param">tensors</em>, <em class="sig-param">devices</em>, <em class="sig-param">buffer_size=10485760</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/cuda/comm.html#broadcast_coalesced"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.cuda.comm.broadcast_coalesced" title="Permalink to this definition">¶</a></dt>
<dd><p>Broadcasts a sequence tensors to the specified GPUs.
Small tensors are first coalesced into a buffer to reduce the number
of synchronizations.</p>
<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</em>) – tensors to broadcast.</p></li>
<li><p><strong>devices</strong> (<em>Iterable</em>) – an iterable of devices among which to broadcast.
Note that it should be like (src, dst1, dst2, …), the first element
of which is the source device to broadcast from.</p></li>
<li><p><strong>buffer_size</strong> (<em>python:int</em>) – maximum size of the buffer used for coalescing</p></li>
</ul>
</dd>
<dt class="field-even">Returns</dt>
<dd class="field-even"><p>A tuple containing copies of the <code class="docutils literal notranslate"><span class="pre">tensor</span></code>, placed on devices
corresponding to indices from <code class="docutils literal notranslate"><span class="pre">devices</span></code>.</p>
</dd>
</dl>
</dd></dl>
<dl class="function">
<dt id="torch.cuda.comm.reduce_add">
<code class="sig-prename descclassname">torch.cuda.comm.</code><code class="sig-name descname">reduce_add</code><span class="sig-paren">(</span><em class="sig-param">inputs</em>, <em class="sig-param">destination=None</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/cuda/comm.html#reduce_add"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.cuda.comm.reduce_add" title="Permalink to this definition">¶</a></dt>
<dd><p>Sums tensors from multiple GPUs.</p>
<p>All inputs should have matching shapes.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><ul class="simple">
<li><p><strong>inputs</strong> (<em>Iterable</em><em>[</em><a class="reference internal" href="tensors.html#torch.Tensor" title="torch.Tensor"><em>Tensor</em></a><em>]</em>) – an iterable of tensors to add.</p></li>
<li><p><strong>destination</strong> (<em>python:int</em><em>, </em><em>optional</em>) – a device on which the output will be
placed (default: current device).</p></li>
</ul>
</dd>
<dt class="field-even">Returns</dt>
<dd class="field-even"><p>A tensor containing an elementwise sum of all inputs, placed on the
<code class="docutils literal notranslate"><span class="pre">destination</span></code> device.</p>
</dd>
</dl>
</dd></dl>
<dl class="function">
<dt id="torch.cuda.comm.scatter">
<code class="sig-prename descclassname">torch.cuda.comm.</code><code class="sig-name descname">scatter</code><span class="sig-paren">(</span><em class="sig-param">tensor</em>, <em class="sig-param">devices</em>, <em class="sig-param">chunk_sizes=None</em>, <em class="sig-param">dim=0</em>, <em class="sig-param">streams=None</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/cuda/comm.html#scatter"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.cuda.comm.scatter" title="Permalink to this definition">¶</a></dt>
<dd><p>Scatters tensor across multiple GPUs.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><ul class="simple">
<li><p><strong>tensor</strong> (<a class="reference internal" href="tensors.html#torch.Tensor" title="torch.Tensor"><em>Tensor</em></a>) – tensor to scatter.</p></li>
<li><p><strong>devices</strong> (<em>Iterable</em><em>[</em><em>python:int</em><em>]</em>) – iterable of ints, specifying among which
devices the tensor should be scattered.</p></li>
<li><p><strong>chunk_sizes</strong> (<em>Iterable</em><em>[</em><em>python:int</em><em>]</em><em>, </em><em>optional</em>) – sizes of chunks to be placed on
each device. It should match <code class="docutils literal notranslate"><span class="pre">devices</span></code> in length and sum to
<code class="docutils literal notranslate"><span class="pre">tensor.size(dim)</span></code>. If not specified, the tensor will be divided
into equal chunks.</p></li>
<li><p><strong>dim</strong> (<em>python:int</em><em>, </em><em>optional</em>) – A dimension along which to chunk the tensor.</p></li>
</ul>
</dd>
<dt class="field-even">Returns</dt>
<dd class="field-even"><p>A tuple containing chunks of the <code class="docutils literal notranslate"><span class="pre">tensor</span></code>, spread across given
<code class="docutils literal notranslate"><span class="pre">devices</span></code>.</p>
</dd>
</dl>
</dd></dl>
<dl class="function">
<dt id="torch.cuda.comm.gather">
<code class="sig-prename descclassname">torch.cuda.comm.</code><code class="sig-name descname">gather</code><span class="sig-paren">(</span><em class="sig-param">tensors</em>, <em class="sig-param">dim=0</em>, <em class="sig-param">destination=None</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/cuda/comm.html#gather"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.cuda.comm.gather" title="Permalink to this definition">¶</a></dt>
<dd><p>Gathers tensors from multiple GPUs.</p>
<p>Tensor sizes in all dimension different than <code class="docutils literal notranslate"><span class="pre">dim</span></code> have to match.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><ul class="simple">
<li><p><strong>tensors</strong> (<em>Iterable</em><em>[</em><a class="reference internal" href="tensors.html#torch.Tensor" title="torch.Tensor"><em>Tensor</em></a><em>]</em>) – iterable of tensors to gather.</p></li>
<li><p><strong>dim</strong> (<em>python:int</em>) – a dimension along which the tensors will be concatenated.</p></li>
<li><p><strong>destination</strong> (<em>python:int</em><em>, </em><em>optional</em>) – output device (-1 means CPU, default:
current device)</p></li>
</ul>
</dd>
<dt class="field-even">Returns</dt>
<dd class="field-even"><p>A tensor located on <code class="docutils literal notranslate"><span class="pre">destination</span></code> device, that is a result of
concatenating <code class="docutils literal notranslate"><span class="pre">tensors</span></code> along <code class="docutils literal notranslate"><span class="pre">dim</span></code>.</p>
</dd>
</dl>
</dd></dl>
</div>
<div class="section" id="streams-and-events">
<h2>Streams and events<a class="headerlink" href="#streams-and-events" title="Permalink to this headline">¶</a></h2>
<dl class="class">
<dt id="torch.cuda.Stream">
<em class="property">class </em><code class="sig-prename descclassname">torch.cuda.</code><code class="sig-name descname">Stream</code><a class="reference internal" href="_modules/torch/cuda/streams.html#Stream"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.cuda.Stream" title="Permalink to this definition">¶</a></dt>
<dd><p>Wrapper around a CUDA stream.</p>
<p>A CUDA stream is a linear sequence of execution that belongs to a specific
device, independent from other streams. See <a class="reference internal" href="notes/cuda.html#cuda-semantics"><span class="std std-ref">CUDA semantics</span></a> for
details.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><ul class="simple">
<li><p><strong>device</strong> (<a class="reference internal" href="tensor_attributes.html#torch.torch.device" title="torch.torch.device"><em>torch.device</em></a><em> or </em><em>python:int</em><em>, </em><em>optional</em>) – a device on which to allocate
the stream. If <a class="reference internal" href="#torch.cuda.device" title="torch.cuda.device"><code class="xref py py-attr docutils literal notranslate"><span class="pre">device</span></code></a> is <code class="docutils literal notranslate"><span class="pre">None</span></code> (default) or a negative
integer, this will use the current device.</p></li>
<li><p><strong>priority</strong> (<em>python:int</em><em>, </em><em>optional</em>) – priority of the stream. Lower numbers
represent higher priorities.</p></li>
</ul>
</dd>
</dl>
<dl class="method">
<dt id="torch.cuda.Stream.query">
<code class="sig-name descname">query</code><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/cuda/streams.html#Stream.query"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.cuda.Stream.query" title="Permalink to this definition">¶</a></dt>
<dd><p>Checks if all the work submitted has been completed.</p>
<dl class="field-list simple">
<dt class="field-odd">Returns</dt>
<dd class="field-odd"><p>A boolean indicating if all kernels in this stream are completed.</p>
</dd>
</dl>
</dd></dl>
<dl class="method">
<dt id="torch.cuda.Stream.record_event">
<code class="sig-name descname">record_event</code><span class="sig-paren">(</span><em class="sig-param">event=None</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/cuda/streams.html#Stream.record_event"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.cuda.Stream.record_event" title="Permalink to this definition">¶</a></dt>
<dd><p>Records an event.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><p><strong>event</strong> (<a class="reference internal" href="#torch.cuda.Event" title="torch.cuda.Event"><em>Event</em></a><em>, </em><em>optional</em>) – event to record. If not given, a new one
will be allocated.</p>
</dd>
<dt class="field-even">Returns</dt>
<dd class="field-even"><p>Recorded event.</p>
</dd>
</dl>
</dd></dl>
<dl class="method">
<dt id="torch.cuda.Stream.synchronize">
<code class="sig-name descname">synchronize</code><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/cuda/streams.html#Stream.synchronize"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.cuda.Stream.synchronize" title="Permalink to this definition">¶</a></dt>
<dd><p>Wait for all the kernels in this stream to complete.</p>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>This is a wrapper around <code class="docutils literal notranslate"><span class="pre">cudaStreamSynchronize()</span></code>: see
<a class="reference external" href="http://docs.nvidia.com/cuda/cuda-runtime-api/group__CUDART__STREAM.html">CUDA Stream documentation</a> for more info.</p>
</div>
</dd></dl>
<dl class="method">
<dt id="torch.cuda.Stream.wait_event">
<code class="sig-name descname">wait_event</code><span class="sig-paren">(</span><em class="sig-param">event</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/cuda/streams.html#Stream.wait_event"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.cuda.Stream.wait_event" title="Permalink to this definition">¶</a></dt>
<dd><p>Makes all future work submitted to the stream wait for an event.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><p><strong>event</strong> (<a class="reference internal" href="#torch.cuda.Event" title="torch.cuda.Event"><em>Event</em></a>) – an event to wait for.</p>
</dd>
</dl>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>This is a wrapper around <code class="docutils literal notranslate"><span class="pre">cudaStreamWaitEvent()</span></code>: see
<a class="reference external" href="http://docs.nvidia.com/cuda/cuda-runtime-api/group__CUDART__STREAM.html">CUDA Stream documentation</a> for more info.</p>
<p>This function returns without waiting for <code class="xref py py-attr docutils literal notranslate"><span class="pre">event</span></code>: only future
operations are affected.</p>
</div>
</dd></dl>
<dl class="method">
<dt id="torch.cuda.Stream.wait_stream">
<code class="sig-name descname">wait_stream</code><span class="sig-paren">(</span><em class="sig-param">stream</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/cuda/streams.html#Stream.wait_stream"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.cuda.Stream.wait_stream" title="Permalink to this definition">¶</a></dt>
<dd><p>Synchronizes with another stream.</p>
<p>All future work submitted to this stream will wait until all kernels
submitted to a given stream at the time of call complete.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><p><strong>stream</strong> (<a class="reference internal" href="#torch.cuda.Stream" title="torch.cuda.Stream"><em>Stream</em></a>) – a stream to synchronize.</p>
</dd>
</dl>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>This function returns without waiting for currently enqueued
kernels in <a class="reference internal" href="#torch.cuda.stream" title="torch.cuda.stream"><code class="xref py py-attr docutils literal notranslate"><span class="pre">stream</span></code></a>: only future operations are affected.</p>
</div>
</dd></dl>
</dd></dl>
<dl class="class">
<dt id="torch.cuda.Event">
<em class="property">class </em><code class="sig-prename descclassname">torch.cuda.</code><code class="sig-name descname">Event</code><a class="reference internal" href="_modules/torch/cuda/streams.html#Event"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.cuda.Event" title="Permalink to this definition">¶</a></dt>
<dd><p>Wrapper around a CUDA event.</p>
<p>CUDA events are synchronization markers that can be used to monitor the
device’s progress, to accurately measure timing, and to synchronize CUDA
streams.</p>
<p>The underlying CUDA events are lazily initialized when the event is first
recorded or exported to another process. After creation, only streams on the
same device may record the event. However, streams on any device can wait on
the event.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><ul class="simple">
<li><p><strong>enable_timing</strong> (<em>bool</em><em>, </em><em>optional</em>) – indicates if the event should measure time
(default: <code class="docutils literal notranslate"><span class="pre">False</span></code>)</p></li>
<li><p><strong>blocking</strong> (<em>bool</em><em>, </em><em>optional</em>) – if <code class="docutils literal notranslate"><span class="pre">True</span></code>, <a class="reference internal" href="#torch.cuda.Event.wait" title="torch.cuda.Event.wait"><code class="xref py py-meth docutils literal notranslate"><span class="pre">wait()</span></code></a> will be blocking (default: <code class="docutils literal notranslate"><span class="pre">False</span></code>)</p></li>
<li><p><strong>interprocess</strong> (<em>bool</em>) – if <code class="docutils literal notranslate"><span class="pre">True</span></code>, the event can be shared between processes
(default: <code class="docutils literal notranslate"><span class="pre">False</span></code>)</p></li>
</ul>
</dd>
</dl>
<dl class="method">
<dt id="torch.cuda.Event.elapsed_time">
<code class="sig-name descname">elapsed_time</code><span class="sig-paren">(</span><em class="sig-param">end_event</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/cuda/streams.html#Event.elapsed_time"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.cuda.Event.elapsed_time" title="Permalink to this definition">¶</a></dt>
<dd><p>Returns the time elapsed in milliseconds after the event was
recorded and before the end_event was recorded.</p>
</dd></dl>
<dl class="method">
<dt id="torch.cuda.Event.from_ipc_handle">
<em class="property">classmethod </em><code class="sig-name descname">from_ipc_handle</code><span class="sig-paren">(</span><em class="sig-param">device</em>, <em class="sig-param">handle</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/cuda/streams.html#Event.from_ipc_handle"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.cuda.Event.from_ipc_handle" title="Permalink to this definition">¶</a></dt>
<dd><p>Reconstruct an event from an IPC handle on the given device.</p>
</dd></dl>
<dl class="method">
<dt id="torch.cuda.Event.ipc_handle">
<code class="sig-name descname">ipc_handle</code><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/cuda/streams.html#Event.ipc_handle"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.cuda.Event.ipc_handle" title="Permalink to this definition">¶</a></dt>
<dd><p>Returns an IPC handle of this event. If not recorded yet, the event
will use the current device.</p>
</dd></dl>
<dl class="method">
<dt id="torch.cuda.Event.query">
<code class="sig-name descname">query</code><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/cuda/streams.html#Event.query"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.cuda.Event.query" title="Permalink to this definition">¶</a></dt>
<dd><p>Checks if all work currently captured by event has completed.</p>
<dl class="field-list simple">
<dt class="field-odd">Returns</dt>
<dd class="field-odd"><p>A boolean indicating if all work currently captured by event has
completed.</p>
</dd>
</dl>
</dd></dl>
<dl class="method">
<dt id="torch.cuda.Event.record">
<code class="sig-name descname">record</code><span class="sig-paren">(</span><em class="sig-param">stream=None</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/cuda/streams.html#Event.record"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.cuda.Event.record" title="Permalink to this definition">¶</a></dt>
<dd><p>Records the event in a given stream.</p>
<p>Uses <code class="docutils literal notranslate"><span class="pre">torch.cuda.current_stream()</span></code> if no stream is specified. The
stream’s device must match the event’s device.</p>
</dd></dl>
<dl class="method">
<dt id="torch.cuda.Event.synchronize">
<code class="sig-name descname">synchronize</code><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/cuda/streams.html#Event.synchronize"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.cuda.Event.synchronize" title="Permalink to this definition">¶</a></dt>
<dd><p>Waits for the event to complete.</p>
<p>Waits until the completion of all work currently captured in this event.
This prevents the CPU thread from proceeding until the event completes.</p>
<blockquote>
<div><div class="admonition note">
<p class="admonition-title">Note</p>
<p>This is a wrapper around <code class="docutils literal notranslate"><span class="pre">cudaEventSynchronize()</span></code>: see
<a class="reference external" href="https://docs.nvidia.com/cuda/cuda-runtime-api/group__CUDART__EVENT.html">CUDA Event documentation</a> for more info.</p>
</div>
</div></blockquote>
</dd></dl>
<dl class="method">
<dt id="torch.cuda.Event.wait">
<code class="sig-name descname">wait</code><span class="sig-paren">(</span><em class="sig-param">stream=None</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/cuda/streams.html#Event.wait"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.cuda.Event.wait" title="Permalink to this definition">¶</a></dt>
<dd><p>Makes all future work submitted to the given stream wait for this
event.</p>
<p>Use <code class="docutils literal notranslate"><span class="pre">torch.cuda.current_stream()</span></code> if no stream is specified.</p>
</dd></dl>
</dd></dl>
</div>
<div class="section" id="memory-management">
<h2>Memory management<a class="headerlink" href="#memory-management" title="Permalink to this headline">¶</a></h2>
<dl class="function">
<dt id="torch.cuda.empty_cache">
<code class="sig-prename descclassname">torch.cuda.</code><code class="sig-name descname">empty_cache</code><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/cuda/memory.html#empty_cache"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.cuda.empty_cache" title="Permalink to this definition">¶</a></dt>
<dd><p>Releases all unoccupied cached memory currently held by the caching
allocator so that those can be used in other GPU application and visible in
<cite>nvidia-smi</cite>.</p>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p><a class="reference internal" href="#torch.cuda.empty_cache" title="torch.cuda.empty_cache"><code class="xref py py-func docutils literal notranslate"><span class="pre">empty_cache()</span></code></a> doesn’t increase the amount of GPU
memory available for PyTorch. However, it may help reduce fragmentation
of GPU memory in certain cases. See <a class="reference internal" href="notes/cuda.html#cuda-memory-management"><span class="std std-ref">Memory management</span></a> for
more details about GPU memory management.</p>
</div>
</dd></dl>
<dl class="function">
<dt id="torch.cuda.memory_stats">
<code class="sig-prename descclassname">torch.cuda.</code><code class="sig-name descname">memory_stats</code><span class="sig-paren">(</span><em class="sig-param">device=None</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/cuda/memory.html#memory_stats"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.cuda.memory_stats" title="Permalink to this definition">¶</a></dt>
<dd><p>Returns a dictionary of CUDA memory allocator statistics for a
given device.</p>
<p>The return value of this function is a dictionary of statistics, each of
which is a non-negative integer.</p>
<p>Core statistics:</p>
<ul class="simple">
<li><p><code class="docutils literal notranslate"><span class="pre">"allocated.{all,large_pool,small_pool}.{current,peak,allocated,freed}"</span></code>:
number of allocation requests received by the memory allocator.</p></li>
<li><p><code class="docutils literal notranslate"><span class="pre">"allocated_bytes.{all,large_pool,small_pool}.{current,peak,allocated,freed}"</span></code>:
amount of allocated memory.</p></li>
<li><p><code class="docutils literal notranslate"><span class="pre">"segment.{all,large_pool,small_pool}.{current,peak,allocated,freed}"</span></code>:
number of reserved segments from <code class="docutils literal notranslate"><span class="pre">cudaMalloc()</span></code>.</p></li>
<li><p><code class="docutils literal notranslate"><span class="pre">"reserved_bytes.{all,large_pool,small_pool}.{current,peak,allocated,freed}"</span></code>:
amount of reserved memory.</p></li>
<li><p><code class="docutils literal notranslate"><span class="pre">"active.{all,large_pool,small_pool}.{current,peak,allocated,freed}"</span></code>:
number of active memory blocks.</p></li>
<li><p><code class="docutils literal notranslate"><span class="pre">"active_bytes.{all,large_pool,small_pool}.{current,peak,allocated,freed}"</span></code>:
amount of active memory.</p></li>
<li><p><code class="docutils literal notranslate"><span class="pre">"inactive_split.{all,large_pool,small_pool}.{current,peak,allocated,freed}"</span></code>:
number of inactive, non-releasable memory blocks.</p></li>
<li><p><code class="docutils literal notranslate"><span class="pre">"inactive_split_bytes.{all,large_pool,small_pool}.{current,peak,allocated,freed}"</span></code>:
amount of inactive, non-releasable memory.</p></li>
</ul>
<p>For these core statistics, values are broken down as follows.</p>
<p>Pool type:</p>
<ul class="simple">
<li><p><code class="docutils literal notranslate"><span class="pre">all</span></code>: combined statistics across all memory pools.</p></li>
<li><p><code class="docutils literal notranslate"><span class="pre">large_pool</span></code>: statistics for the large allocation pool
(as of October 2019, for size >= 1MB allocations).</p></li>
<li><p><code class="docutils literal notranslate"><span class="pre">small_pool</span></code>: statistics for the small allocation pool
(as of October 2019, for size < 1MB allocations).</p></li>
</ul>
<p>Metric type:</p>
<ul class="simple">
<li><p><code class="docutils literal notranslate"><span class="pre">current</span></code>: current value of this metric.</p></li>
<li><p><code class="docutils literal notranslate"><span class="pre">peak</span></code>: maximum value of this metric.</p></li>
<li><p><code class="docutils literal notranslate"><span class="pre">allocated</span></code>: historical total increase in this metric.</p></li>
<li><p><code class="docutils literal notranslate"><span class="pre">freed</span></code>: historical total decrease in this metric.</p></li>
</ul>
<p>In addition to the core statistics, we also provide some simple event