-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathnamed_tensor.html
1334 lines (1130 loc) · 105 KB
/
named_tensor.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>Named Tensors — PyTorch 2.3 documentation</title>
<link rel="canonical" href="https://pytorch.org/docs/stable/named_tensor.html"/>
<link rel="stylesheet" href="_static/css/theme.css" type="text/css" />
<!-- <link rel="stylesheet" href="_static/pygments.css" type="text/css" /> -->
<link rel="stylesheet" href="_static/pygments.css" type="text/css" />
<link rel="stylesheet" href="_static/css/theme.css" type="text/css" />
<link rel="stylesheet" href="_static/copybutton.css" type="text/css" />
<link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/katex.min.css" type="text/css" />
<link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/katex.min.css" type="text/css" />
<link rel="stylesheet" href="_static/katex-math.css" type="text/css" />
<link rel="stylesheet" href="_static/sphinx-dropdown.css" type="text/css" />
<link rel="stylesheet" href="_static/panels-bootstrap.min.css" type="text/css" />
<link rel="stylesheet" href="_static/css/jit.css" type="text/css" />
<link rel="index" title="Index" href="genindex.html" />
<link rel="search" title="Search" href="search.html" />
<link rel="next" title="Named Tensors operator coverage" href="name_inference.html" />
<link rel="prev" title="Type Info" href="type_info.html" />
<!-- Google Tag Manager -->
<script>(function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':
new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],
j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src=
'https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);
})(window,document,'script','dataLayer','GTM-T8XT4PS');</script>
<!-- End Google Tag Manager -->
<script src="_static/js/modernizr.min.js"></script>
<!-- Preload the theme fonts -->
<link rel="preload" href="_static/fonts/FreightSans/freight-sans-book.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="_static/fonts/FreightSans/freight-sans-medium.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="_static/fonts/IBMPlexMono/IBMPlexMono-Medium.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="_static/fonts/FreightSans/freight-sans-bold.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="_static/fonts/FreightSans/freight-sans-medium-italic.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="_static/fonts/IBMPlexMono/IBMPlexMono-SemiBold.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<!-- Preload the katex fonts -->
<link rel="preload" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/fonts/KaTeX_Math-Italic.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/fonts/KaTeX_Main-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/fonts/KaTeX_Main-Bold.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/fonts/KaTeX_Size1-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/fonts/KaTeX_Size4-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/fonts/KaTeX_Size2-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/fonts/KaTeX_Size3-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/fonts/KaTeX_Caligraphic-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="stylesheet" href="https://use.fontawesome.com/releases/v5.15.2/css/all.css" integrity="sha384-vSIIfh2YWi9wW0r9iZe7RJPrKwp6bG+s9QZMoITbCckVJqGCCRhc+ccxNcdpHuYu" crossorigin="anonymous">
</head>
<div class="container-fluid header-holder tutorials-header" id="header-holder">
<div class="container">
<div class="header-container">
<a class="header-logo" href="https://pytorch.org/" aria-label="PyTorch"></a>
<div class="main-menu">
<ul>
<li class="main-menu-item">
<div id="resourcesDropdownButton" data-toggle="resources-dropdown" class="resources-dropdown">
<a class="with-down-arrow">
Learn
</a>
<div class="resources-dropdown-menu">
<a class="nav-dropdown-item" href="https://pytorch.org/get-started">
<span class=dropdown-title>Get Started</span>
<p>Run PyTorch locally or get started quickly with one of the supported cloud platforms</p>
</a>
<a class="nav-dropdown-item" href="https://pytorch.org/tutorials">
<span class="dropdown-title">Tutorials</span>
<p>Whats new in PyTorch tutorials</p>
</a>
<a class="nav-dropdown-item" href="https://pytorch.org/tutorials/beginner/basics/intro.html">
<span class="dropdown-title">Learn the Basics</span>
<p>Familiarize yourself with PyTorch concepts and modules</p>
</a>
<a class="nav-dropdown-item" href="https://pytorch.org/tutorials/recipes/recipes_index.html">
<span class="dropdown-title">PyTorch Recipes</span>
<p>Bite-size, ready-to-deploy PyTorch code examples</p>
</a>
<a class="nav-dropdown-item" href="https://pytorch.org/tutorials/beginner/introyt.html">
<span class="dropdown-title">Intro to PyTorch - YouTube Series</span>
<p>Master PyTorch basics with our engaging YouTube tutorial series</p>
</a>
</div>
</div>
</li>
<li>
<div id="resourcesDropdownButton" data-toggle="resources-dropdown" class="resources-dropdown">
<a class="with-down-arrow">
Ecosystem
</a>
<div class="resources-dropdown-menu">
<a class="nav-dropdown-item" href="https://pytorch.org/ecosystem">
<span class="dropdown-title">Tools</span>
<p>Learn about the tools and frameworks in the PyTorch Ecosystem</p>
</a>
<a class="nav-dropdown-item" href="https://pytorch.org/#community-module">
<span class=dropdown-title>Community</span>
<p>Join the PyTorch developer community to contribute, learn, and get your questions answered</p>
</a>
<a class="nav-dropdown-item" href="https://discuss.pytorch.org/" target="_blank">
<span class=dropdown-title>Forums</span>
<p>A place to discuss PyTorch code, issues, install, research</p>
</a>
<a class="nav-dropdown-item" href="https://pytorch.org/resources">
<span class=dropdown-title>Developer Resources</span>
<p>Find resources and get questions answered</p>
</a>
<a class="nav-dropdown-item" href="https://pytorch.org/ecosystem/contributor-awards-2023">
<span class="dropdown-title">Contributor Awards - 2023</span>
<p>Award winners announced at this year's PyTorch Conference</p>
</a>
</div>
</div>
</li>
<li>
<div id="resourcesDropdownButton" data-toggle="resources-dropdown" class="resources-dropdown">
<a class="with-down-arrow">
Edge
</a>
<div class="resources-dropdown-menu">
<a class="nav-dropdown-item" href="https://pytorch.org/edge">
<span class="dropdown-title">About PyTorch Edge</span>
<p>Build innovative and privacy-aware AI experiences for edge devices</p>
</a>
<a class="nav-dropdown-item" href="https://pytorch.org/executorch-overview">
<span class="dropdown-title">ExecuTorch</span>
<p>End-to-end solution for enabling on-device inference capabilities across mobile and edge devices</p>
</a>
</div>
</div>
</li>
<li class="main-menu-item">
<div id="resourcesDropdownButton" data-toggle="resources-dropdown" class="resources-dropdown">
<a class="with-down-arrow">
Docs
</a>
<div class="resources-dropdown-menu">
<a class="nav-dropdown-item" href="https://pytorch.org/docs/stable/index.html">
<span class="dropdown-title">PyTorch</span>
<p>Explore the documentation for comprehensive guidance on how to use PyTorch</p>
</a>
<a class="nav-dropdown-item" href="https://pytorch.org/pytorch-domains">
<span class="dropdown-title">PyTorch Domains</span>
<p>Read the PyTorch Domains documentation to learn more about domain-specific libraries</p>
</a>
</div>
</div>
</li>
<li>
<div id="resourcesDropdownButton" data-toggle="resources-dropdown" class="resources-dropdown">
<a class="with-down-arrow">
Blogs & News
</a>
<div class="resources-dropdown-menu">
<a class="nav-dropdown-item" href="https://pytorch.org/blog/">
<span class="dropdown-title">PyTorch Blog</span>
<p>Catch up on the latest technical news and happenings</p>
</a>
<a class="nav-dropdown-item" href="https://pytorch.org/community-blog">
<span class="dropdown-title">Community Blog</span>
<p>Stories from the PyTorch ecosystem</p>
</a>
<a class="nav-dropdown-item" href="https://pytorch.org/videos">
<span class="dropdown-title">Videos</span>
<p>Learn about the latest PyTorch tutorials, new, and more </p>
<a class="nav-dropdown-item" href="https://pytorch.org/community-stories">
<span class="dropdown-title">Community Stories</span>
<p>Learn how our community solves real, everyday machine learning problems with PyTorch</p>
</a>
<a class="nav-dropdown-item" href="https://pytorch.org/events">
<span class="dropdown-title">Events</span>
<p>Find events, webinars, and podcasts</p>
</a>
</div>
</li>
<li>
<div id="resourcesDropdownButton" data-toggle="resources-dropdown" class="resources-dropdown">
<a class="with-down-arrow">
About
</a>
<div class="resources-dropdown-menu">
<a class="nav-dropdown-item" href="https://pytorch.org/foundation">
<span class="dropdown-title">PyTorch Foundation</span>
<p>Learn more about the PyTorch Foundation</p>
</a>
<a class="nav-dropdown-item" href="https://pytorch.org/governing-board">
<span class="dropdown-title">Governing Board</span>
<p></p>
</a>
</div>
</div>
</li>
<li class="main-menu-item">
<div class="no-dropdown">
<a href="https://pytorch.org/join" data-cta="join">
Become a Member
</a>
</div>
</li>
<li>
<div class="main-menu-item">
<a href="https://github.com/pytorch/pytorch" class="github-icon">
</a>
</div>
</li>
<!--- TODO: This block adds the search icon to the nav bar. We will enable it later.
<li>
<div class="main-menu-item">
<a href="https://github.com/pytorch/pytorch" class="search-icon">
</a>
</div>
</li>
--->
</ul>
</div>
<a class="main-menu-open-button" href="#" data-behavior="open-mobile-menu"></a>
</div>
</div>
</div>
<body class="pytorch-body">
<div class="table-of-contents-link-wrapper">
<span>Table of Contents</span>
<a href="#" class="toggle-table-of-contents" data-behavior="toggle-table-of-contents"></a>
</div>
<nav data-toggle="wy-nav-shift" class="pytorch-left-menu" id="pytorch-left-menu">
<div class="pytorch-side-scroll">
<div class="pytorch-menu pytorch-menu-vertical" data-spy="affix" role="navigation" aria-label="main navigation">
<div class="pytorch-left-menu-search">
<div class="version">
<a href='https://pytorch.org/docs/versions.html'>2.3 ▼</a>
</div>
<div role="search">
<form id="rtd-search-form" class="wy-form" action="search.html" method="get">
<input type="text" name="q" placeholder="Search Docs" />
<input type="hidden" name="check_keywords" value="yes" />
<input type="hidden" name="area" value="default" />
</form>
</div>
</div>
<p class="caption" role="heading"><span class="caption-text">Community</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="community/build_ci_governance.html">PyTorch Governance | Build + CI</a></li>
<li class="toctree-l1"><a class="reference internal" href="community/contribution_guide.html">PyTorch Contribution Guide</a></li>
<li class="toctree-l1"><a class="reference internal" href="community/design.html">PyTorch Design Philosophy</a></li>
<li class="toctree-l1"><a class="reference internal" href="community/governance.html">PyTorch Governance | Mechanics</a></li>
<li class="toctree-l1"><a class="reference internal" href="community/persons_of_interest.html">PyTorch Governance | Maintainers</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">Developer Notes</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="notes/amp_examples.html">CUDA Automatic Mixed Precision examples</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/autograd.html">Autograd mechanics</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/broadcasting.html">Broadcasting semantics</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/cpu_threading_torchscript_inference.html">CPU threading and TorchScript inference</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/cuda.html">CUDA semantics</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/ddp.html">Distributed Data Parallel</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/extending.html">Extending PyTorch</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/extending.func.html">Extending torch.func with autograd.Function</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/faq.html">Frequently Asked Questions</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/fsdp.html">FSDP Notes</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/gradcheck.html">Gradcheck mechanics</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/hip.html">HIP (ROCm) semantics</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/large_scale_deployments.html">Features for large-scale deployments</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/modules.html">Modules</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/mps.html">MPS backend</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/multiprocessing.html">Multiprocessing best practices</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/numerical_accuracy.html">Numerical accuracy</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/randomness.html">Reproducibility</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/serialization.html">Serialization semantics</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/windows.html">Windows FAQ</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">Language Bindings</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="cpp_index.html">C++</a></li>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/javadoc/">Javadoc</a></li>
<li class="toctree-l1"><a class="reference internal" href="deploy.html">torch::deploy</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">Python API</span></p>
<ul class="current">
<li class="toctree-l1"><a class="reference internal" href="torch.html">torch</a></li>
<li class="toctree-l1"><a class="reference internal" href="nn.html">torch.nn</a></li>
<li class="toctree-l1"><a class="reference internal" href="nn.functional.html">torch.nn.functional</a></li>
<li class="toctree-l1"><a class="reference internal" href="tensors.html">torch.Tensor</a></li>
<li class="toctree-l1"><a class="reference internal" href="tensor_attributes.html">Tensor Attributes</a></li>
<li class="toctree-l1"><a class="reference internal" href="tensor_view.html">Tensor Views</a></li>
<li class="toctree-l1"><a class="reference internal" href="amp.html">torch.amp</a></li>
<li class="toctree-l1"><a class="reference internal" href="autograd.html">torch.autograd</a></li>
<li class="toctree-l1"><a class="reference internal" href="library.html">torch.library</a></li>
<li class="toctree-l1"><a class="reference internal" href="cpu.html">torch.cpu</a></li>
<li class="toctree-l1"><a class="reference internal" href="cuda.html">torch.cuda</a></li>
<li class="toctree-l1"><a class="reference internal" href="torch_cuda_memory.html">Understanding CUDA Memory Usage</a></li>
<li class="toctree-l1"><a class="reference internal" href="torch_cuda_memory.html#generating-a-snapshot">Generating a Snapshot</a></li>
<li class="toctree-l1"><a class="reference internal" href="torch_cuda_memory.html#using-the-visualizer">Using the visualizer</a></li>
<li class="toctree-l1"><a class="reference internal" href="torch_cuda_memory.html#snapshot-api-reference">Snapshot API Reference</a></li>
<li class="toctree-l1"><a class="reference internal" href="mps.html">torch.mps</a></li>
<li class="toctree-l1"><a class="reference internal" href="xpu.html">torch.xpu</a></li>
<li class="toctree-l1"><a class="reference internal" href="meta.html">Meta device</a></li>
<li class="toctree-l1"><a class="reference internal" href="backends.html">torch.backends</a></li>
<li class="toctree-l1"><a class="reference internal" href="export.html">torch.export</a></li>
<li class="toctree-l1"><a class="reference internal" href="distributed.html">torch.distributed</a></li>
<li class="toctree-l1"><a class="reference internal" href="distributed.algorithms.join.html">torch.distributed.algorithms.join</a></li>
<li class="toctree-l1"><a class="reference internal" href="distributed.elastic.html">torch.distributed.elastic</a></li>
<li class="toctree-l1"><a class="reference internal" href="fsdp.html">torch.distributed.fsdp</a></li>
<li class="toctree-l1"><a class="reference internal" href="distributed.optim.html">torch.distributed.optim</a></li>
<li class="toctree-l1"><a class="reference internal" href="distributed.tensor.parallel.html">torch.distributed.tensor.parallel</a></li>
<li class="toctree-l1"><a class="reference internal" href="distributed.checkpoint.html">torch.distributed.checkpoint</a></li>
<li class="toctree-l1"><a class="reference internal" href="distributions.html">torch.distributions</a></li>
<li class="toctree-l1"><a class="reference internal" href="torch.compiler.html">torch.compiler</a></li>
<li class="toctree-l1"><a class="reference internal" href="fft.html">torch.fft</a></li>
<li class="toctree-l1"><a class="reference internal" href="func.html">torch.func</a></li>
<li class="toctree-l1"><a class="reference internal" href="futures.html">torch.futures</a></li>
<li class="toctree-l1"><a class="reference internal" href="fx.html">torch.fx</a></li>
<li class="toctree-l1"><a class="reference internal" href="fx.experimental.html">torch.fx.experimental</a></li>
<li class="toctree-l1"><a class="reference internal" href="hub.html">torch.hub</a></li>
<li class="toctree-l1"><a class="reference internal" href="jit.html">torch.jit</a></li>
<li class="toctree-l1"><a class="reference internal" href="linalg.html">torch.linalg</a></li>
<li class="toctree-l1"><a class="reference internal" href="monitor.html">torch.monitor</a></li>
<li class="toctree-l1"><a class="reference internal" href="signal.html">torch.signal</a></li>
<li class="toctree-l1"><a class="reference internal" href="special.html">torch.special</a></li>
<li class="toctree-l1"><a class="reference internal" href="torch.overrides.html">torch.overrides</a></li>
<li class="toctree-l1"><a class="reference internal" href="package.html">torch.package</a></li>
<li class="toctree-l1"><a class="reference internal" href="profiler.html">torch.profiler</a></li>
<li class="toctree-l1"><a class="reference internal" href="nn.init.html">torch.nn.init</a></li>
<li class="toctree-l1"><a class="reference internal" href="nn.attention.html">torch.nn.attention</a></li>
<li class="toctree-l1"><a class="reference internal" href="onnx.html">torch.onnx</a></li>
<li class="toctree-l1"><a class="reference internal" href="optim.html">torch.optim</a></li>
<li class="toctree-l1"><a class="reference internal" href="complex_numbers.html">Complex Numbers</a></li>
<li class="toctree-l1"><a class="reference internal" href="ddp_comm_hooks.html">DDP Communication Hooks</a></li>
<li class="toctree-l1"><a class="reference internal" href="pipeline.html">Pipeline Parallelism</a></li>
<li class="toctree-l1"><a class="reference internal" href="quantization.html">Quantization</a></li>
<li class="toctree-l1"><a class="reference internal" href="rpc.html">Distributed RPC Framework</a></li>
<li class="toctree-l1"><a class="reference internal" href="random.html">torch.random</a></li>
<li class="toctree-l1"><a class="reference internal" href="masked.html">torch.masked</a></li>
<li class="toctree-l1"><a class="reference internal" href="nested.html">torch.nested</a></li>
<li class="toctree-l1"><a class="reference internal" href="sparse.html">torch.sparse</a></li>
<li class="toctree-l1"><a class="reference internal" href="storage.html">torch.Storage</a></li>
<li class="toctree-l1"><a class="reference internal" href="testing.html">torch.testing</a></li>
<li class="toctree-l1"><a class="reference internal" href="utils.html">torch.utils</a></li>
<li class="toctree-l1"><a class="reference internal" href="benchmark_utils.html">torch.utils.benchmark</a></li>
<li class="toctree-l1"><a class="reference internal" href="bottleneck.html">torch.utils.bottleneck</a></li>
<li class="toctree-l1"><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="deterministic.html">torch.utils.deterministic</a></li>
<li class="toctree-l1"><a class="reference internal" href="jit_utils.html">torch.utils.jit</a></li>
<li class="toctree-l1"><a class="reference internal" href="dlpack.html">torch.utils.dlpack</a></li>
<li class="toctree-l1"><a class="reference internal" href="mobile_optimizer.html">torch.utils.mobile_optimizer</a></li>
<li class="toctree-l1"><a class="reference internal" href="model_zoo.html">torch.utils.model_zoo</a></li>
<li class="toctree-l1"><a class="reference internal" href="tensorboard.html">torch.utils.tensorboard</a></li>
<li class="toctree-l1"><a class="reference internal" href="type_info.html">Type Info</a></li>
<li class="toctree-l1 current"><a class="current reference internal" href="#">Named Tensors</a></li>
<li class="toctree-l1"><a class="reference internal" href="name_inference.html">Named Tensors operator coverage</a></li>
<li class="toctree-l1"><a class="reference internal" href="config_mod.html">torch.__config__</a></li>
<li class="toctree-l1"><a class="reference internal" href="future_mod.html">torch.__future__</a></li>
<li class="toctree-l1"><a class="reference internal" href="logging.html">torch._logging</a></li>
<li class="toctree-l1"><a class="reference internal" href="torch_environment_variables.html">Torch Environment Variables</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">Libraries</span></p>
<ul>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/audio/stable">torchaudio</a></li>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/data">TorchData</a></li>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/torchrec">TorchRec</a></li>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/serve">TorchServe</a></li>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/text/stable">torchtext</a></li>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/vision/stable">torchvision</a></li>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/xla/">PyTorch on XLA Devices</a></li>
</ul>
</div>
</div>
</nav>
<div class="pytorch-container">
<div class="pytorch-page-level-bar" id="pytorch-page-level-bar">
<div class="pytorch-breadcrumbs-wrapper">
<div role="navigation" aria-label="breadcrumbs navigation">
<ul class="pytorch-breadcrumbs">
<li>
<a href="index.html">
Docs
</a> >
</li>
<li>Named Tensors</li>
<li class="pytorch-breadcrumbs-aside">
<a href="_sources/named_tensor.rst.txt" rel="nofollow"><img src="_static/images/view-page-source-icon.svg"></a>
</li>
</ul>
</div>
</div>
<div class="pytorch-shortcuts-wrapper" id="pytorch-shortcuts-wrapper">
Shortcuts
</div>
</div>
<section data-toggle="wy-nav-shift" id="pytorch-content-wrap" class="pytorch-content-wrap">
<div class="pytorch-content-left">
<!-- Google Tag Manager (noscript) -->
<noscript><iframe src="https://www.googletagmanager.com/ns.html?id=GTM-T8XT4PS"
height="0" width="0" style="display:none;visibility:hidden"></iframe></noscript>
<!-- End Google Tag Manager (noscript) -->
<div class="rst-content">
<div role="main" class="main-content" itemscope="itemscope" itemtype="http://schema.org/Article">
<article itemprop="articleBody" id="pytorch-article" class="pytorch-article">
<div class="section" id="named-tensors">
<span id="named-tensors-doc"></span><h1>Named Tensors<a class="headerlink" href="#named-tensors" title="Permalink to this heading">¶</a></h1>
<p>Named Tensors allow users to give explicit names to tensor dimensions.
In most cases, operations that take dimension parameters will accept
dimension names, avoiding the need to track dimensions by position.
In addition, named tensors use names to automatically check that APIs
are being used correctly at runtime, providing extra safety. Names can
also be used to rearrange dimensions, for example, to support
“broadcasting by name” rather than “broadcasting by position”.</p>
<div class="admonition warning">
<p class="admonition-title">Warning</p>
<p>The named tensor API is a prototype feature and subject to change.</p>
</div>
<div class="section" id="creating-named-tensors">
<h2>Creating named tensors<a class="headerlink" href="#creating-named-tensors" title="Permalink to this heading">¶</a></h2>
<p>Factory functions now take a new <code class="xref py py-attr docutils literal notranslate"><span class="pre">names</span></code> argument that associates a name
with each dimension.</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="n">torch</span><span class="o">.</span><span class="n">zeros</span><span class="p">(</span><span class="mi">2</span><span class="p">,</span> <span class="mi">3</span><span class="p">,</span> <span class="n">names</span><span class="o">=</span><span class="p">(</span><span class="s1">'N'</span><span class="p">,</span> <span class="s1">'C'</span><span class="p">))</span>
<span class="go">tensor([[0., 0., 0.],</span>
<span class="go"> [0., 0., 0.]], names=('N', 'C'))</span>
</pre></div>
</div>
<p>Named dimensions, like regular Tensor dimensions, are ordered.
<code class="docutils literal notranslate"><span class="pre">tensor.names[i]</span></code> is the name of dimension <code class="docutils literal notranslate"><span class="pre">i</span></code> of <code class="docutils literal notranslate"><span class="pre">tensor</span></code>.</p>
<p>The following factory functions support named tensors:</p>
<ul class="simple">
<li><p><a class="reference internal" href="generated/torch.empty.html#torch.empty" title="torch.empty"><code class="xref py py-func docutils literal notranslate"><span class="pre">torch.empty()</span></code></a></p></li>
<li><p><a class="reference internal" href="generated/torch.rand.html#torch.rand" title="torch.rand"><code class="xref py py-func docutils literal notranslate"><span class="pre">torch.rand()</span></code></a></p></li>
<li><p><a class="reference internal" href="generated/torch.randn.html#torch.randn" title="torch.randn"><code class="xref py py-func docutils literal notranslate"><span class="pre">torch.randn()</span></code></a></p></li>
<li><p><a class="reference internal" href="generated/torch.ones.html#torch.ones" title="torch.ones"><code class="xref py py-func docutils literal notranslate"><span class="pre">torch.ones()</span></code></a></p></li>
<li><p><a class="reference internal" href="generated/torch.tensor.html#torch.tensor" title="torch.tensor"><code class="xref py py-func docutils literal notranslate"><span class="pre">torch.tensor()</span></code></a></p></li>
<li><p><a class="reference internal" href="generated/torch.zeros.html#torch.zeros" title="torch.zeros"><code class="xref py py-func docutils literal notranslate"><span class="pre">torch.zeros()</span></code></a></p></li>
</ul>
</div>
<div class="section" id="named-dimensions">
<h2>Named dimensions<a class="headerlink" href="#named-dimensions" title="Permalink to this heading">¶</a></h2>
<p>See <a class="reference internal" href="#torch.Tensor.names" title="torch.Tensor.names"><code class="xref py py-attr docutils literal notranslate"><span class="pre">names</span></code></a> for restrictions on tensor names.</p>
<p>Use <a class="reference internal" href="#torch.Tensor.names" title="torch.Tensor.names"><code class="xref py py-attr docutils literal notranslate"><span class="pre">names</span></code></a> to access the dimension names of a tensor and
<a class="reference internal" href="#torch.Tensor.rename" title="torch.Tensor.rename"><code class="xref py py-meth docutils literal notranslate"><span class="pre">rename()</span></code></a> to rename named dimensions.</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="n">imgs</span> <span class="o">=</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">2</span><span class="p">,</span> <span class="mi">2</span><span class="p">,</span> <span class="mi">3</span> <span class="p">,</span> <span class="n">names</span><span class="o">=</span><span class="p">(</span><span class="s1">'N'</span><span class="p">,</span> <span class="s1">'C'</span><span class="p">,</span> <span class="s1">'H'</span><span class="p">,</span> <span class="s1">'W'</span><span class="p">))</span>
<span class="gp">>>> </span><span class="n">imgs</span><span class="o">.</span><span class="n">names</span>
<span class="go">('N', 'C', 'H', 'W')</span>
<span class="gp">>>> </span><span class="n">renamed_imgs</span> <span class="o">=</span> <span class="n">imgs</span><span class="o">.</span><span class="n">rename</span><span class="p">(</span><span class="n">H</span><span class="o">=</span><span class="s1">'height'</span><span class="p">,</span> <span class="n">W</span><span class="o">=</span><span class="s1">'width'</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">renamed_imgs</span><span class="o">.</span><span class="n">names</span>
<span class="go">('N', 'C', 'height', 'width)</span>
</pre></div>
</div>
<p>Named tensors can coexist with unnamed tensors; named tensors are instances of
<a class="reference internal" href="tensors.html#torch.Tensor" title="torch.Tensor"><code class="xref py py-class docutils literal notranslate"><span class="pre">torch.Tensor</span></code></a>. Unnamed tensors have <code class="docutils literal notranslate"><span class="pre">None</span></code>-named dimensions. Named
tensors do not require all dimensions to be named.</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="n">imgs</span> <span class="o">=</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">2</span><span class="p">,</span> <span class="mi">2</span><span class="p">,</span> <span class="mi">3</span> <span class="p">,</span> <span class="n">names</span><span class="o">=</span><span class="p">(</span><span class="kc">None</span><span class="p">,</span> <span class="s1">'C'</span><span class="p">,</span> <span class="s1">'H'</span><span class="p">,</span> <span class="s1">'W'</span><span class="p">))</span>
<span class="gp">>>> </span><span class="n">imgs</span><span class="o">.</span><span class="n">names</span>
<span class="go">(None, 'C', 'H', 'W')</span>
</pre></div>
</div>
</div>
<div class="section" id="name-propagation-semantics">
<h2>Name propagation semantics<a class="headerlink" href="#name-propagation-semantics" title="Permalink to this heading">¶</a></h2>
<p>Named tensors use names to automatically check that APIs are being called
correctly at runtime. This occurs in a process called <em>name inference</em>.
More formally, name inference consists of the following two steps:</p>
<ul class="simple">
<li><p><strong>Check names</strong>: an operator may perform automatic checks at runtime that
check that certain dimension names must match.</p></li>
<li><p><strong>Propagate names</strong>: name inference propagates names to output tensors.</p></li>
</ul>
<p>All operations that support named tensors propagate names.</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="n">x</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">randn</span><span class="p">(</span><span class="mi">3</span><span class="p">,</span> <span class="mi">3</span><span class="p">,</span> <span class="n">names</span><span class="o">=</span><span class="p">(</span><span class="s1">'N'</span><span class="p">,</span> <span class="s1">'C'</span><span class="p">))</span>
<span class="gp">>>> </span><span class="n">x</span><span class="o">.</span><span class="n">abs</span><span class="p">()</span><span class="o">.</span><span class="n">names</span>
<span class="go">('N', 'C')</span>
</pre></div>
</div>
<div class="section" id="match-semantics">
<span id="match-semantics-doc"></span><h3>match semantics<a class="headerlink" href="#match-semantics" title="Permalink to this heading">¶</a></h3>
<p>Two names <em>match</em> if they are equal (string equality) or if at least one is <code class="docutils literal notranslate"><span class="pre">None</span></code>.
Nones are essentially a special “wildcard” name.</p>
<p><code class="docutils literal notranslate"><span class="pre">unify(A,</span> <span class="pre">B)</span></code> determines which of the names <code class="docutils literal notranslate"><span class="pre">A</span></code> and <code class="docutils literal notranslate"><span class="pre">B</span></code> to propagate to the outputs.
It returns the more <em>specific</em> of the two names, if they match. If the names do not match,
then it errors.</p>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>In practice, when working with named tensors, one should avoid having unnamed
dimensions because their handling can be complicated. It is recommended to lift
all unnamed dimensions to be named dimensions by using <a class="reference internal" href="#torch.Tensor.refine_names" title="torch.Tensor.refine_names"><code class="xref py py-meth docutils literal notranslate"><span class="pre">refine_names()</span></code></a>.</p>
</div>
</div>
<div class="section" id="basic-name-inference-rules">
<h3>Basic name inference rules<a class="headerlink" href="#basic-name-inference-rules" title="Permalink to this heading">¶</a></h3>
<p>Let’s see how <code class="docutils literal notranslate"><span class="pre">match</span></code> and <code class="docutils literal notranslate"><span class="pre">unify</span></code> are used in name inference in the case of
adding two one-dim tensors with no broadcasting.</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">x</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">randn</span><span class="p">(</span><span class="mi">3</span><span class="p">,</span> <span class="n">names</span><span class="o">=</span><span class="p">(</span><span class="s1">'X'</span><span class="p">,))</span>
<span class="n">y</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">randn</span><span class="p">(</span><span class="mi">3</span><span class="p">)</span>
<span class="n">z</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">randn</span><span class="p">(</span><span class="mi">3</span><span class="p">,</span> <span class="n">names</span><span class="o">=</span><span class="p">(</span><span class="s1">'Z'</span><span class="p">,))</span>
</pre></div>
</div>
<p><strong>Check names</strong>: check that the names of the two tensors <em>match</em>.</p>
<p>For the following examples:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="c1"># x + y # match('X', None) is True</span>
<span class="gp">>>> </span><span class="c1"># x + z # match('X', 'Z') is False</span>
<span class="gp">>>> </span><span class="c1"># x + x # match('X', 'X') is True</span>
<span class="gp">>>> </span><span class="n">x</span> <span class="o">+</span> <span class="n">z</span>
<span class="go">Error when attempting to broadcast dims ['X'] and dims ['Z']: dim 'X' and dim 'Z' are at the same position from the right but do not match.</span>
</pre></div>
</div>
<p><strong>Propagate names</strong>: <em>unify</em> the names to select which one to propagate.
In the case of <code class="docutils literal notranslate"><span class="pre">x</span> <span class="pre">+</span> <span class="pre">y</span></code>, <code class="docutils literal notranslate"><span class="pre">unify('X',</span> <span class="pre">None)</span> <span class="pre">=</span> <span class="pre">'X'</span></code> because <code class="docutils literal notranslate"><span class="pre">'X'</span></code> is more
specific than <code class="docutils literal notranslate"><span class="pre">None</span></code>.</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="p">(</span><span class="n">x</span> <span class="o">+</span> <span class="n">y</span><span class="p">)</span><span class="o">.</span><span class="n">names</span>
<span class="go">('X',)</span>
<span class="gp">>>> </span><span class="p">(</span><span class="n">x</span> <span class="o">+</span> <span class="n">x</span><span class="p">)</span><span class="o">.</span><span class="n">names</span>
<span class="go">('X',)</span>
</pre></div>
</div>
<p>For a comprehensive list of name inference rules, see <a class="reference internal" href="name_inference.html#name-inference-reference-doc"><span class="std std-ref">Named Tensors operator coverage</span></a>.
Here are two common operations that may be useful to go over:</p>
<ul class="simple">
<li><p>Binary arithmetic ops: <a class="reference internal" href="name_inference.html#unifies-names-from-inputs-doc"><span class="std std-ref">Unifies names from inputs</span></a></p></li>
<li><p>Matrix multiplication ops: <a class="reference internal" href="name_inference.html#contracts-away-dims-doc"><span class="std std-ref">Contracts away dims</span></a></p></li>
</ul>
</div>
</div>
<div class="section" id="explicit-alignment-by-names">
<h2>Explicit alignment by names<a class="headerlink" href="#explicit-alignment-by-names" title="Permalink to this heading">¶</a></h2>
<p>Use <a class="reference internal" href="#torch.Tensor.align_as" title="torch.Tensor.align_as"><code class="xref py py-meth docutils literal notranslate"><span class="pre">align_as()</span></code></a> or <a class="reference internal" href="#torch.Tensor.align_to" title="torch.Tensor.align_to"><code class="xref py py-meth docutils literal notranslate"><span class="pre">align_to()</span></code></a> to align tensor dimensions
by name to a specified ordering. This is useful for performing “broadcasting by names”.</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="c1"># This function is agnostic to the dimension ordering of `input`,</span>
<span class="c1"># as long as it has a `C` dimension somewhere.</span>
<span class="k">def</span> <span class="nf">scale_channels</span><span class="p">(</span><span class="nb">input</span><span class="p">,</span> <span class="n">scale</span><span class="p">):</span>
<span class="n">scale</span> <span class="o">=</span> <span class="n">scale</span><span class="o">.</span><span class="n">refine_names</span><span class="p">(</span><span class="s1">'C'</span><span class="p">)</span>
<span class="k">return</span> <span class="nb">input</span> <span class="o">*</span> <span class="n">scale</span><span class="o">.</span><span class="n">align_as</span><span class="p">(</span><span class="nb">input</span><span class="p">)</span>
<span class="o">>>></span> <span class="n">num_channels</span> <span class="o">=</span> <span class="mi">3</span>
<span class="o">>>></span> <span class="n">scale</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">randn</span><span class="p">(</span><span class="n">num_channels</span><span class="p">,</span> <span class="n">names</span><span class="o">=</span><span class="p">(</span><span class="s1">'C'</span><span class="p">,))</span>
<span class="o">>>></span> <span class="n">imgs</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">rand</span><span class="p">(</span><span class="mi">3</span><span class="p">,</span> <span class="mi">3</span><span class="p">,</span> <span class="mi">3</span><span class="p">,</span> <span class="n">num_channels</span><span class="p">,</span> <span class="n">names</span><span class="o">=</span><span class="p">(</span><span class="s1">'N'</span><span class="p">,</span> <span class="s1">'H'</span><span class="p">,</span> <span class="s1">'W'</span><span class="p">,</span> <span class="s1">'C'</span><span class="p">))</span>
<span class="o">>>></span> <span class="n">more_imgs</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">rand</span><span class="p">(</span><span class="mi">3</span><span class="p">,</span> <span class="n">num_channels</span><span class="p">,</span> <span class="mi">3</span><span class="p">,</span> <span class="mi">3</span><span class="p">,</span> <span class="n">names</span><span class="o">=</span><span class="p">(</span><span class="s1">'N'</span><span class="p">,</span> <span class="s1">'C'</span><span class="p">,</span> <span class="s1">'H'</span><span class="p">,</span> <span class="s1">'W'</span><span class="p">))</span>
<span class="o">>>></span> <span class="n">videos</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">randn</span><span class="p">(</span><span class="mi">3</span><span class="p">,</span> <span class="n">num_channels</span><span class="p">,</span> <span class="mi">3</span><span class="p">,</span> <span class="mi">3</span><span class="p">,</span> <span class="mi">3</span><span class="p">,</span> <span class="n">names</span><span class="o">=</span><span class="p">(</span><span class="s1">'N'</span><span class="p">,</span> <span class="s1">'C'</span><span class="p">,</span> <span class="s1">'H'</span><span class="p">,</span> <span class="s1">'W'</span><span class="p">,</span> <span class="s1">'D'</span><span class="p">)</span>
<span class="o">>>></span> <span class="n">scale_channels</span><span class="p">(</span><span class="n">imgs</span><span class="p">,</span> <span class="n">scale</span><span class="p">)</span>
<span class="o">>>></span> <span class="n">scale_channels</span><span class="p">(</span><span class="n">more_imgs</span><span class="p">,</span> <span class="n">scale</span><span class="p">)</span>
<span class="o">>>></span> <span class="n">scale_channels</span><span class="p">(</span><span class="n">videos</span><span class="p">,</span> <span class="n">scale</span><span class="p">)</span>
</pre></div>
</div>
</div>
<div class="section" id="manipulating-dimensions">
<h2>Manipulating dimensions<a class="headerlink" href="#manipulating-dimensions" title="Permalink to this heading">¶</a></h2>
<p>Use <a class="reference internal" href="#torch.Tensor.align_to" title="torch.Tensor.align_to"><code class="xref py py-meth docutils literal notranslate"><span class="pre">align_to()</span></code></a> to permute large amounts of dimensions without
mentioning all of them as in required by <a class="reference internal" href="generated/torch.Tensor.permute.html#torch.Tensor.permute" title="torch.Tensor.permute"><code class="xref py py-meth docutils literal notranslate"><span class="pre">permute()</span></code></a>.</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="n">tensor</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">randn</span><span class="p">(</span><span class="mi">2</span><span class="p">,</span> <span class="mi">2</span><span class="p">,</span> <span class="mi">2</span><span class="p">,</span> <span class="mi">2</span><span class="p">,</span> <span class="mi">2</span><span class="p">,</span> <span class="mi">2</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">named_tensor</span> <span class="o">=</span> <span class="n">tensor</span><span class="o">.</span><span class="n">refine_names</span><span class="p">(</span><span class="s1">'A'</span><span class="p">,</span> <span class="s1">'B'</span><span class="p">,</span> <span class="s1">'C'</span><span class="p">,</span> <span class="s1">'D'</span><span class="p">,</span> <span class="s1">'E'</span><span class="p">,</span> <span class="s1">'F'</span><span class="p">)</span>
<span class="go"># Move the F (dim 5) and E dimension (dim 4) to the front while keeping</span>
<span class="go"># the rest in the same order</span>
<span class="gp">>>> </span><span class="n">tensor</span><span class="o">.</span><span class="n">permute</span><span class="p">(</span><span class="mi">5</span><span class="p">,</span> <span class="mi">4</span><span class="p">,</span> <span class="mi">0</span><span class="p">,</span> <span class="mi">1</span><span class="p">,</span> <span class="mi">2</span><span class="p">,</span> <span class="mi">3</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">named_tensor</span><span class="o">.</span><span class="n">align_to</span><span class="p">(</span><span class="s1">'F'</span><span class="p">,</span> <span class="s1">'E'</span><span class="p">,</span> <span class="o">...</span><span class="p">)</span>
</pre></div>
</div>
<p>Use <a class="reference internal" href="generated/torch.Tensor.flatten.html#torch.Tensor.flatten" title="torch.Tensor.flatten"><code class="xref py py-meth docutils literal notranslate"><span class="pre">flatten()</span></code></a> and <a class="reference internal" href="generated/torch.Tensor.unflatten.html#torch.Tensor.unflatten" title="torch.Tensor.unflatten"><code class="xref py py-meth docutils literal notranslate"><span class="pre">unflatten()</span></code></a> to flatten and unflatten
dimensions, respectively. These methods are more verbose than <a class="reference internal" href="generated/torch.Tensor.view.html#torch.Tensor.view" title="torch.Tensor.view"><code class="xref py py-meth docutils literal notranslate"><span class="pre">view()</span></code></a>
and <a class="reference internal" href="generated/torch.Tensor.reshape.html#torch.Tensor.reshape" title="torch.Tensor.reshape"><code class="xref py py-meth docutils literal notranslate"><span class="pre">reshape()</span></code></a>, but have more semantic meaning to someone reading the code.</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="n">imgs</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">randn</span><span class="p">(</span><span class="mi">32</span><span class="p">,</span> <span class="mi">3</span><span class="p">,</span> <span class="mi">128</span><span class="p">,</span> <span class="mi">128</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">named_imgs</span> <span class="o">=</span> <span class="n">imgs</span><span class="o">.</span><span class="n">refine_names</span><span class="p">(</span><span class="s1">'N'</span><span class="p">,</span> <span class="s1">'C'</span><span class="p">,</span> <span class="s1">'H'</span><span class="p">,</span> <span class="s1">'W'</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">flat_imgs</span> <span class="o">=</span> <span class="n">imgs</span><span class="o">.</span><span class="n">view</span><span class="p">(</span><span class="mi">32</span><span class="p">,</span> <span class="o">-</span><span class="mi">1</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">named_flat_imgs</span> <span class="o">=</span> <span class="n">named_imgs</span><span class="o">.</span><span class="n">flatten</span><span class="p">([</span><span class="s1">'C'</span><span class="p">,</span> <span class="s1">'H'</span><span class="p">,</span> <span class="s1">'W'</span><span class="p">],</span> <span class="s1">'features'</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">named_flat_imgs</span><span class="o">.</span><span class="n">names</span>
<span class="go">('N', 'features')</span>
<span class="gp">>>> </span><span class="n">unflattened_named_imgs</span> <span class="o">=</span> <span class="n">named_flat_imgs</span><span class="o">.</span><span class="n">unflatten</span><span class="p">(</span><span class="s1">'features'</span><span class="p">,</span> <span class="p">[(</span><span class="s1">'C'</span><span class="p">,</span> <span class="mi">3</span><span class="p">),</span> <span class="p">(</span><span class="s1">'H'</span><span class="p">,</span> <span class="mi">128</span><span class="p">),</span> <span class="p">(</span><span class="s1">'W'</span><span class="p">,</span> <span class="mi">128</span><span class="p">)])</span>
<span class="gp">>>> </span><span class="n">unflattened_named_imgs</span><span class="o">.</span><span class="n">names</span>
<span class="go">('N', 'C', 'H', 'W')</span>
</pre></div>
</div>
</div>
<div class="section" id="autograd-support">
<span id="named-tensors-autograd-doc"></span><h2>Autograd support<a class="headerlink" href="#autograd-support" title="Permalink to this heading">¶</a></h2>
<p>Autograd currently supports named tensors in a limited manner: autograd ignores
names on all tensors. Gradient computation is still correct but we lose the
safety that names give us.</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="n">x</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">randn</span><span class="p">(</span><span class="mi">3</span><span class="p">,</span> <span class="n">names</span><span class="o">=</span><span class="p">(</span><span class="s1">'D'</span><span class="p">,))</span>
<span class="gp">>>> </span><span class="n">weight</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">randn</span><span class="p">(</span><span class="mi">3</span><span class="p">,</span> <span class="n">names</span><span class="o">=</span><span class="p">(</span><span class="s1">'D'</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">loss</span> <span class="o">=</span> <span class="p">(</span><span class="n">x</span> <span class="o">-</span> <span class="n">weight</span><span class="p">)</span><span class="o">.</span><span class="n">abs</span><span class="p">()</span>
<span class="gp">>>> </span><span class="n">grad_loss</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">randn</span><span class="p">(</span><span class="mi">3</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">loss</span><span class="o">.</span><span class="n">backward</span><span class="p">(</span><span class="n">grad_loss</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">weight</span><span class="o">.</span><span class="n">grad</span> <span class="c1"># Unnamed for now. Will be named in the future</span>
<span class="go">tensor([-1.8107, -0.6357, 0.0783])</span>
<span class="gp">>>> </span><span class="n">weight</span><span class="o">.</span><span class="n">grad</span><span class="o">.</span><span class="n">zero_</span><span class="p">()</span>
<span class="gp">>>> </span><span class="n">grad_loss</span> <span class="o">=</span> <span class="n">grad_loss</span><span class="o">.</span><span class="n">refine_names</span><span class="p">(</span><span class="s1">'C'</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">loss</span> <span class="o">=</span> <span class="p">(</span><span class="n">x</span> <span class="o">-</span> <span class="n">weight</span><span class="p">)</span><span class="o">.</span><span class="n">abs</span><span class="p">()</span>
<span class="go"># Ideally we'd check that the names of loss and grad_loss match but we don't yet.</span>
<span class="gp">>>> </span><span class="n">loss</span><span class="o">.</span><span class="n">backward</span><span class="p">(</span><span class="n">grad_loss</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">weight</span><span class="o">.</span><span class="n">grad</span>
<span class="go">tensor([-1.8107, -0.6357, 0.0783])</span>
</pre></div>
</div>
</div>
<div class="section" id="currently-supported-operations-and-subsystems">
<h2>Currently supported operations and subsystems<a class="headerlink" href="#currently-supported-operations-and-subsystems" title="Permalink to this heading">¶</a></h2>
<div class="section" id="operators">
<h3>Operators<a class="headerlink" href="#operators" title="Permalink to this heading">¶</a></h3>
<p>See <a class="reference internal" href="name_inference.html#name-inference-reference-doc"><span class="std std-ref">Named Tensors operator coverage</span></a> for a full list of the supported torch and
tensor operations. We do not yet support the following that is not covered by the link:</p>
<ul class="simple">
<li><p>indexing, advanced indexing.</p></li>
</ul>
<p>For <code class="docutils literal notranslate"><span class="pre">torch.nn.functional</span></code> operators, we support the following:</p>
<ul class="simple">
<li><p><a class="reference internal" href="generated/torch.nn.functional.relu.html#torch.nn.functional.relu" title="torch.nn.functional.relu"><code class="xref py py-func docutils literal notranslate"><span class="pre">torch.nn.functional.relu()</span></code></a></p></li>
<li><p><a class="reference internal" href="generated/torch.nn.functional.softmax.html#torch.nn.functional.softmax" title="torch.nn.functional.softmax"><code class="xref py py-func docutils literal notranslate"><span class="pre">torch.nn.functional.softmax()</span></code></a></p></li>
<li><p><a class="reference internal" href="generated/torch.nn.functional.log_softmax.html#torch.nn.functional.log_softmax" title="torch.nn.functional.log_softmax"><code class="xref py py-func docutils literal notranslate"><span class="pre">torch.nn.functional.log_softmax()</span></code></a></p></li>
<li><p><a class="reference internal" href="generated/torch.nn.functional.tanh.html#torch.nn.functional.tanh" title="torch.nn.functional.tanh"><code class="xref py py-func docutils literal notranslate"><span class="pre">torch.nn.functional.tanh()</span></code></a></p></li>
<li><p><a class="reference internal" href="generated/torch.nn.functional.sigmoid.html#torch.nn.functional.sigmoid" title="torch.nn.functional.sigmoid"><code class="xref py py-func docutils literal notranslate"><span class="pre">torch.nn.functional.sigmoid()</span></code></a></p></li>
<li><p><a class="reference internal" href="generated/torch.nn.functional.dropout.html#torch.nn.functional.dropout" title="torch.nn.functional.dropout"><code class="xref py py-func docutils literal notranslate"><span class="pre">torch.nn.functional.dropout()</span></code></a></p></li>
</ul>
</div>
<div class="section" id="subsystems">
<h3>Subsystems<a class="headerlink" href="#subsystems" title="Permalink to this heading">¶</a></h3>
<p>Autograd is supported, see <a class="reference internal" href="#named-tensors-autograd-doc"><span class="std std-ref">Autograd support</span></a>.
Because gradients are currently unnamed, optimizers may work but are untested.</p>
<p>NN modules are currently unsupported. This can lead to the following when calling
modules with named tensor inputs:</p>
<ul class="simple">
<li><p>NN module parameters are unnamed, so outputs may be partially named.</p></li>
<li><p>NN module forward passes have code that don’t support named tensors and will
error out appropriately.</p></li>
</ul>
<p>We also do not support the following subsystems, though some may work out
of the box:</p>
<ul class="simple">
<li><p>distributions</p></li>
<li><p>serialization (<a class="reference internal" href="generated/torch.load.html#torch.load" title="torch.load"><code class="xref py py-func docutils literal notranslate"><span class="pre">torch.load()</span></code></a>, <a class="reference internal" href="generated/torch.save.html#torch.save" title="torch.save"><code class="xref py py-func docutils literal notranslate"><span class="pre">torch.save()</span></code></a>)</p></li>
<li><p>multiprocessing</p></li>
<li><p>JIT</p></li>
<li><p>distributed</p></li>
<li><p>ONNX</p></li>
</ul>
<p>If any of these would help your use case, please
<a class="reference external" href="https://github.com/pytorch/pytorch/issues?q=is%3Aopen+is%3Aissue+label%3A%22module%3A+named+tensor%22">search if an issue has already been filed</a>
and if not, <a class="reference external" href="https://github.com/pytorch/pytorch/issues/new/choose">file one</a>.</p>
</div>
</div>
<div class="section" id="named-tensor-api-reference">
<h2>Named tensor API reference<a class="headerlink" href="#named-tensor-api-reference" title="Permalink to this heading">¶</a></h2>
<p>In this section please find the documentation for named tensor specific APIs.
For a comprehensive reference for how names are propagated through other PyTorch
operators, see <a class="reference internal" href="name_inference.html#name-inference-reference-doc"><span class="std std-ref">Named Tensors operator coverage</span></a>.</p>
<dl class="py class">
<dt class="sig sig-object py">
<em class="property"><span class="pre">class</span><span class="w"> </span></em><span class="sig-prename descclassname"><span class="pre">torch.</span></span><span class="sig-name descname"><span class="pre">Tensor</span></span></dt>
<dd><dl class="py attribute">
<dt class="sig sig-object py" id="torch.Tensor.names">
<span class="sig-name descname"><span class="pre">names</span></span><a class="headerlink" href="#torch.Tensor.names" title="Permalink to this definition">¶</a></dt>
<dd><p>Stores names for each of this tensor’s dimensions.</p>
<p><code class="docutils literal notranslate"><span class="pre">names[idx]</span></code> corresponds to the name of tensor dimension <code class="docutils literal notranslate"><span class="pre">idx</span></code>.
Names are either a string if the dimension is named or <code class="docutils literal notranslate"><span class="pre">None</span></code> if the
dimension is unnamed.</p>
<p>Dimension names may contain characters or underscore. Furthermore, a dimension
name must be a valid Python variable name (i.e., does not start with underscore).</p>
<p>Tensors may not have two named dimensions with the same name.</p>
<div class="admonition warning">
<p class="admonition-title">Warning</p>
<p>The named tensor API is experimental and subject to change.</p>
</div>
</dd></dl>
<dl class="py method">
<dt class="sig sig-object py" id="torch.Tensor.rename">
<span class="sig-name descname"><span class="pre">rename</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="o"><span class="pre">*</span></span><span class="n"><span class="pre">names</span></span></em>, <em class="sig-param"><span class="o"><span class="pre">**</span></span><span class="n"><span class="pre">rename_map</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/_tensor.html#Tensor.rename"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.Tensor.rename" title="Permalink to this definition">¶</a></dt>
<dd><p>Renames dimension names of <code class="xref py py-attr docutils literal notranslate"><span class="pre">self</span></code>.</p>
<p>There are two main usages:</p>
<p><code class="docutils literal notranslate"><span class="pre">self.rename(**rename_map)</span></code> returns a view on tensor that has dims
renamed as specified in the mapping <code class="xref py py-attr docutils literal notranslate"><span class="pre">rename_map</span></code>.</p>
<p><code class="docutils literal notranslate"><span class="pre">self.rename(*names)</span></code> returns a view on tensor, renaming all
dimensions positionally using <a class="reference internal" href="#torch.Tensor.names" title="torch.Tensor.names"><code class="xref py py-attr docutils literal notranslate"><span class="pre">names</span></code></a>.
Use <code class="docutils literal notranslate"><span class="pre">self.rename(None)</span></code> to drop names on a tensor.</p>
<p>One cannot specify both positional args <a class="reference internal" href="#torch.Tensor.names" title="torch.Tensor.names"><code class="xref py py-attr docutils literal notranslate"><span class="pre">names</span></code></a> and keyword args
<code class="xref py py-attr docutils literal notranslate"><span class="pre">rename_map</span></code>.</p>
<p>Examples:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="n">imgs</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">rand</span><span class="p">(</span><span class="mi">2</span><span class="p">,</span> <span class="mi">3</span><span class="p">,</span> <span class="mi">5</span><span class="p">,</span> <span class="mi">7</span><span class="p">,</span> <span class="n">names</span><span class="o">=</span><span class="p">(</span><span class="s1">'N'</span><span class="p">,</span> <span class="s1">'C'</span><span class="p">,</span> <span class="s1">'H'</span><span class="p">,</span> <span class="s1">'W'</span><span class="p">))</span>
<span class="gp">>>> </span><span class="n">renamed_imgs</span> <span class="o">=</span> <span class="n">imgs</span><span class="o">.</span><span class="n">rename</span><span class="p">(</span><span class="n">N</span><span class="o">=</span><span class="s1">'batch'</span><span class="p">,</span> <span class="n">C</span><span class="o">=</span><span class="s1">'channels'</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">renamed_imgs</span><span class="o">.</span><span class="n">names</span>
<span class="go">('batch', 'channels', 'H', 'W')</span>
<span class="gp">>>> </span><span class="n">renamed_imgs</span> <span class="o">=</span> <span class="n">imgs</span><span class="o">.</span><span class="n">rename</span><span class="p">(</span><span class="kc">None</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">renamed_imgs</span><span class="o">.</span><span class="n">names</span>
<span class="go">(None, None, None, None)</span>
<span class="gp">>>> </span><span class="n">renamed_imgs</span> <span class="o">=</span> <span class="n">imgs</span><span class="o">.</span><span class="n">rename</span><span class="p">(</span><span class="s1">'batch'</span><span class="p">,</span> <span class="s1">'channel'</span><span class="p">,</span> <span class="s1">'height'</span><span class="p">,</span> <span class="s1">'width'</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">renamed_imgs</span><span class="o">.</span><span class="n">names</span>
<span class="go">('batch', 'channel', 'height', 'width')</span>
</pre></div>
</div>
<div class="admonition warning">
<p class="admonition-title">Warning</p>
<p>The named tensor API is experimental and subject to change.</p>
</div>
</dd></dl>
<dl class="py method">
<dt class="sig sig-object py" id="torch.Tensor.rename_">
<span class="sig-name descname"><span class="pre">rename_</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="o"><span class="pre">*</span></span><span class="n"><span class="pre">names</span></span></em>, <em class="sig-param"><span class="o"><span class="pre">**</span></span><span class="n"><span class="pre">rename_map</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/_tensor.html#Tensor.rename_"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.Tensor.rename_" title="Permalink to this definition">¶</a></dt>
<dd><p>In-place version of <a class="reference internal" href="#torch.Tensor.rename" title="torch.Tensor.rename"><code class="xref py py-meth docutils literal notranslate"><span class="pre">rename()</span></code></a>.</p>
</dd></dl>
<dl class="py method">
<dt class="sig sig-object py" id="torch.Tensor.refine_names">
<span class="sig-name descname"><span class="pre">refine_names</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="o"><span class="pre">*</span></span><span class="n"><span class="pre">names</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/_tensor.html#Tensor.refine_names"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.Tensor.refine_names" title="Permalink to this definition">¶</a></dt>
<dd><p>Refines the dimension names of <code class="xref py py-attr docutils literal notranslate"><span class="pre">self</span></code> according to <a class="reference internal" href="#torch.Tensor.names" title="torch.Tensor.names"><code class="xref py py-attr docutils literal notranslate"><span class="pre">names</span></code></a>.</p>
<p>Refining is a special case of renaming that “lifts” unnamed dimensions.
A <code class="docutils literal notranslate"><span class="pre">None</span></code> dim can be refined to have any name; a named dim can only be
refined to have the same name.</p>
<p>Because named tensors can coexist with unnamed tensors, refining names
gives a nice way to write named-tensor-aware code that works with both
named and unnamed tensors.</p>
<p><a class="reference internal" href="#torch.Tensor.names" title="torch.Tensor.names"><code class="xref py py-attr docutils literal notranslate"><span class="pre">names</span></code></a> may contain up to one Ellipsis (<code class="docutils literal notranslate"><span class="pre">...</span></code>).
The Ellipsis is expanded greedily; it is expanded in-place to fill
<a class="reference internal" href="#torch.Tensor.names" title="torch.Tensor.names"><code class="xref py py-attr docutils literal notranslate"><span class="pre">names</span></code></a> to the same length as <code class="docutils literal notranslate"><span class="pre">self.dim()</span></code> using names from the
corresponding indices of <code class="docutils literal notranslate"><span class="pre">self.names</span></code>.</p>
<p>Python 2 does not support Ellipsis but one may use a string literal
instead (<code class="docutils literal notranslate"><span class="pre">'...'</span></code>).</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><p><strong>names</strong> (<em>iterable</em><em> of </em><a class="reference external" href="https://docs.python.org/3/library/stdtypes.html#str" title="(in Python v3.12)"><em>str</em></a>) – The desired names of the output tensor. May
contain up to one Ellipsis.</p>
</dd>
</dl>
<p>Examples:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="n">imgs</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">randn</span><span class="p">(</span><span class="mi">32</span><span class="p">,</span> <span class="mi">3</span><span class="p">,</span> <span class="mi">128</span><span class="p">,</span> <span class="mi">128</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">named_imgs</span> <span class="o">=</span> <span class="n">imgs</span><span class="o">.</span><span class="n">refine_names</span><span class="p">(</span><span class="s1">'N'</span><span class="p">,</span> <span class="s1">'C'</span><span class="p">,</span> <span class="s1">'H'</span><span class="p">,</span> <span class="s1">'W'</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">named_imgs</span><span class="o">.</span><span class="n">names</span>
<span class="go">('N', 'C', 'H', 'W')</span>
<span class="gp">>>> </span><span class="n">tensor</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">randn</span><span class="p">(</span><span class="mi">2</span><span class="p">,</span> <span class="mi">3</span><span class="p">,</span> <span class="mi">5</span><span class="p">,</span> <span class="mi">7</span><span class="p">,</span> <span class="mi">11</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">tensor</span> <span class="o">=</span> <span class="n">tensor</span><span class="o">.</span><span class="n">refine_names</span><span class="p">(</span><span class="s1">'A'</span><span class="p">,</span> <span class="o">...</span><span class="p">,</span> <span class="s1">'B'</span><span class="p">,</span> <span class="s1">'C'</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">tensor</span><span class="o">.</span><span class="n">names</span>
<span class="go">('A', None, None, 'B', 'C')</span>
</pre></div>
</div>
<div class="admonition warning">
<p class="admonition-title">Warning</p>
<p>The named tensor API is experimental and subject to change.</p>
</div>
</dd></dl>
<dl class="py method">
<dt class="sig sig-object py" id="torch.Tensor.align_as">
<span class="sig-name descname"><span class="pre">align_as</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">other</span></span></em><span class="sig-paren">)</span> <span class="sig-return"><span class="sig-return-icon">→</span> <span class="sig-return-typehint"><a class="reference internal" href="tensors.html#torch.Tensor" title="torch.Tensor"><span class="pre">Tensor</span></a></span></span><a class="headerlink" href="#torch.Tensor.align_as" title="Permalink to this definition">¶</a></dt>
<dd><p>Permutes the dimensions of the <code class="xref py py-attr docutils literal notranslate"><span class="pre">self</span></code> tensor to match the dimension order
in the <code class="xref py py-attr docutils literal notranslate"><span class="pre">other</span></code> tensor, adding size-one dims for any new names.</p>
<p>This operation is useful for explicit broadcasting by names (see examples).</p>
<p>All of the dims of <code class="xref py py-attr docutils literal notranslate"><span class="pre">self</span></code> must be named in order to use this method.
The resulting tensor is a view on the original tensor.</p>
<p>All dimension names of <code class="xref py py-attr docutils literal notranslate"><span class="pre">self</span></code> must be present in <code class="docutils literal notranslate"><span class="pre">other.names</span></code>.
<code class="xref py py-attr docutils literal notranslate"><span class="pre">other</span></code> may contain named dimensions that are not in <code class="docutils literal notranslate"><span class="pre">self.names</span></code>;
the output tensor has a size-one dimension for each of those new names.</p>
<p>To align a tensor to a specific order, use <a class="reference internal" href="#torch.Tensor.align_to" title="torch.Tensor.align_to"><code class="xref py py-meth docutils literal notranslate"><span class="pre">align_to()</span></code></a>.</p>
<p>Examples:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="c1"># Example 1: Applying a mask</span>
<span class="o">>>></span> <span class="n">mask</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">randint</span><span class="p">(</span><span class="mi">2</span><span class="p">,</span> <span class="p">[</span><span class="mi">127</span><span class="p">,</span> <span class="mi">128</span><span class="p">],</span> <span class="n">dtype</span><span class="o">=</span><span class="n">torch</span><span class="o">.</span><span class="n">bool</span><span class="p">)</span><span class="o">.</span><span class="n">refine_names</span><span class="p">(</span><span class="s1">'W'</span><span class="p">,</span> <span class="s1">'H'</span><span class="p">)</span>
<span class="o">>>></span> <span class="n">imgs</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">randn</span><span class="p">(</span><span class="mi">32</span><span class="p">,</span> <span class="mi">128</span><span class="p">,</span> <span class="mi">127</span><span class="p">,</span> <span class="mi">3</span><span class="p">,</span> <span class="n">names</span><span class="o">=</span><span class="p">(</span><span class="s1">'N'</span><span class="p">,</span> <span class="s1">'H'</span><span class="p">,</span> <span class="s1">'W'</span><span class="p">,</span> <span class="s1">'C'</span><span class="p">))</span>
<span class="o">>>></span> <span class="n">imgs</span><span class="o">.</span><span class="n">masked_fill_</span><span class="p">(</span><span class="n">mask</span><span class="o">.</span><span class="n">align_as</span><span class="p">(</span><span class="n">imgs</span><span class="p">),</span> <span class="mi">0</span><span class="p">)</span>
<span class="c1"># Example 2: Applying a per-channel-scale</span>
<span class="o">>>></span> <span class="k">def</span> <span class="nf">scale_channels</span><span class="p">(</span><span class="nb">input</span><span class="p">,</span> <span class="n">scale</span><span class="p">):</span>
<span class="o">>>></span> <span class="n">scale</span> <span class="o">=</span> <span class="n">scale</span><span class="o">.</span><span class="n">refine_names</span><span class="p">(</span><span class="s1">'C'</span><span class="p">)</span>
<span class="o">>>></span> <span class="k">return</span> <span class="nb">input</span> <span class="o">*</span> <span class="n">scale</span><span class="o">.</span><span class="n">align_as</span><span class="p">(</span><span class="nb">input</span><span class="p">)</span>
<span class="o">>>></span> <span class="n">num_channels</span> <span class="o">=</span> <span class="mi">3</span>
<span class="o">>>></span> <span class="n">scale</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">randn</span><span class="p">(</span><span class="n">num_channels</span><span class="p">,</span> <span class="n">names</span><span class="o">=</span><span class="p">(</span><span class="s1">'C'</span><span class="p">,))</span>
<span class="o">>>></span> <span class="n">imgs</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">rand</span><span class="p">(</span><span class="mi">32</span><span class="p">,</span> <span class="mi">128</span><span class="p">,</span> <span class="mi">128</span><span class="p">,</span> <span class="n">num_channels</span><span class="p">,</span> <span class="n">names</span><span class="o">=</span><span class="p">(</span><span class="s1">'N'</span><span class="p">,</span> <span class="s1">'H'</span><span class="p">,</span> <span class="s1">'W'</span><span class="p">,</span> <span class="s1">'C'</span><span class="p">))</span>
<span class="o">>>></span> <span class="n">more_imgs</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">rand</span><span class="p">(</span><span class="mi">32</span><span class="p">,</span> <span class="n">num_channels</span><span class="p">,</span> <span class="mi">128</span><span class="p">,</span> <span class="mi">128</span><span class="p">,</span> <span class="n">names</span><span class="o">=</span><span class="p">(</span><span class="s1">'N'</span><span class="p">,</span> <span class="s1">'C'</span><span class="p">,</span> <span class="s1">'H'</span><span class="p">,</span> <span class="s1">'W'</span><span class="p">))</span>
<span class="o">>>></span> <span class="n">videos</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">randn</span><span class="p">(</span><span class="mi">3</span><span class="p">,</span> <span class="n">num_channels</span><span class="p">,</span> <span class="mi">128</span><span class="p">,</span> <span class="mi">128</span><span class="p">,</span> <span class="mi">128</span><span class="p">,</span> <span class="n">names</span><span class="o">=</span><span class="p">(</span><span class="s1">'N'</span><span class="p">,</span> <span class="s1">'C'</span><span class="p">,</span> <span class="s1">'H'</span><span class="p">,</span> <span class="s1">'W'</span><span class="p">,</span> <span class="s1">'D'</span><span class="p">))</span>
<span class="c1"># scale_channels is agnostic to the dimension order of the input</span>
<span class="o">>>></span> <span class="n">scale_channels</span><span class="p">(</span><span class="n">imgs</span><span class="p">,</span> <span class="n">scale</span><span class="p">)</span>
<span class="o">>>></span> <span class="n">scale_channels</span><span class="p">(</span><span class="n">more_imgs</span><span class="p">,</span> <span class="n">scale</span><span class="p">)</span>
<span class="o">>>></span> <span class="n">scale_channels</span><span class="p">(</span><span class="n">videos</span><span class="p">,</span> <span class="n">scale</span><span class="p">)</span>
</pre></div>
</div>
<div class="admonition warning">
<p class="admonition-title">Warning</p>
<p>The named tensor API is experimental and subject to change.</p>
</div>
</dd></dl>
<dl class="py method">
<dt class="sig sig-object py" id="torch.Tensor.align_to">
<span class="sig-name descname"><span class="pre">align_to</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="o"><span class="pre">*</span></span><span class="n"><span class="pre">names</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/_tensor.html#Tensor.align_to"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.Tensor.align_to" title="Permalink to this definition">¶</a></dt>
<dd><p>Permutes the dimensions of the <code class="xref py py-attr docutils literal notranslate"><span class="pre">self</span></code> tensor to match the order
specified in <a class="reference internal" href="#torch.Tensor.names" title="torch.Tensor.names"><code class="xref py py-attr docutils literal notranslate"><span class="pre">names</span></code></a>, adding size-one dims for any new names.</p>
<p>All of the dims of <code class="xref py py-attr docutils literal notranslate"><span class="pre">self</span></code> must be named in order to use this method.
The resulting tensor is a view on the original tensor.</p>
<p>All dimension names of <code class="xref py py-attr docutils literal notranslate"><span class="pre">self</span></code> must be present in <a class="reference internal" href="#torch.Tensor.names" title="torch.Tensor.names"><code class="xref py py-attr docutils literal notranslate"><span class="pre">names</span></code></a>.
<a class="reference internal" href="#torch.Tensor.names" title="torch.Tensor.names"><code class="xref py py-attr docutils literal notranslate"><span class="pre">names</span></code></a> may contain additional names that are not in <code class="docutils literal notranslate"><span class="pre">self.names</span></code>;
the output tensor has a size-one dimension for each of those new names.</p>
<p><a class="reference internal" href="#torch.Tensor.names" title="torch.Tensor.names"><code class="xref py py-attr docutils literal notranslate"><span class="pre">names</span></code></a> may contain up to one Ellipsis (<code class="docutils literal notranslate"><span class="pre">...</span></code>).
The Ellipsis is expanded to be equal to all dimension names of <code class="xref py py-attr docutils literal notranslate"><span class="pre">self</span></code>
that are not mentioned in <a class="reference internal" href="#torch.Tensor.names" title="torch.Tensor.names"><code class="xref py py-attr docutils literal notranslate"><span class="pre">names</span></code></a>, in the order that they appear
in <code class="xref py py-attr docutils literal notranslate"><span class="pre">self</span></code>.</p>
<p>Python 2 does not support Ellipsis but one may use a string literal
instead (<code class="docutils literal notranslate"><span class="pre">'...'</span></code>).</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><p><strong>names</strong> (<em>iterable</em><em> of </em><a class="reference external" href="https://docs.python.org/3/library/stdtypes.html#str" title="(in Python v3.12)"><em>str</em></a>) – The desired dimension ordering of the
output tensor. May contain up to one Ellipsis that is expanded
to all unmentioned dim names of <code class="xref py py-attr docutils literal notranslate"><span class="pre">self</span></code>.</p>
</dd>
</dl>
<p>Examples:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="n">tensor</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">randn</span><span class="p">(</span><span class="mi">2</span><span class="p">,</span> <span class="mi">2</span><span class="p">,</span> <span class="mi">2</span><span class="p">,</span> <span class="mi">2</span><span class="p">,</span> <span class="mi">2</span><span class="p">,</span> <span class="mi">2</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">named_tensor</span> <span class="o">=</span> <span class="n">tensor</span><span class="o">.</span><span class="n">refine_names</span><span class="p">(</span><span class="s1">'A'</span><span class="p">,</span> <span class="s1">'B'</span><span class="p">,</span> <span class="s1">'C'</span><span class="p">,</span> <span class="s1">'D'</span><span class="p">,</span> <span class="s1">'E'</span><span class="p">,</span> <span class="s1">'F'</span><span class="p">)</span>
<span class="go"># Move the F and E dims to the front while keeping the rest in order</span>
<span class="gp">>>> </span><span class="n">named_tensor</span><span class="o">.</span><span class="n">align_to</span><span class="p">(</span><span class="s1">'F'</span><span class="p">,</span> <span class="s1">'E'</span><span class="p">,</span> <span class="o">...</span><span class="p">)</span>
</pre></div>
</div>
<div class="admonition warning">
<p class="admonition-title">Warning</p>
<p>The named tensor API is experimental and subject to change.</p>
</div>
</dd></dl>
<dl class="py method">
<dt class="sig sig-object py">
<span class="sig-name descname"><span class="pre">flatten</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">dims</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">out_dim</span></span></em><span class="sig-paren">)</span> <span class="sig-return"><span class="sig-return-icon">→</span> <span class="sig-return-typehint"><a class="reference internal" href="tensors.html#torch.Tensor" title="torch.Tensor"><span class="pre">Tensor</span></a></span></span></dt>
<dd><p>Flattens <code class="xref py py-attr docutils literal notranslate"><span class="pre">dims</span></code> into a single dimension with name <code class="xref py py-attr docutils literal notranslate"><span class="pre">out_dim</span></code>.</p>
<p>All of <cite>dims</cite> must be consecutive in order in the <code class="xref py py-attr docutils literal notranslate"><span class="pre">self</span></code> tensor,
but not necessary contiguous in memory.</p>
<p>Examples:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="n">imgs</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">randn</span><span class="p">(</span><span class="mi">32</span><span class="p">,</span> <span class="mi">3</span><span class="p">,</span> <span class="mi">128</span><span class="p">,</span> <span class="mi">128</span><span class="p">,</span> <span class="n">names</span><span class="o">=</span><span class="p">(</span><span class="s1">'N'</span><span class="p">,</span> <span class="s1">'C'</span><span class="p">,</span> <span class="s1">'H'</span><span class="p">,</span> <span class="s1">'W'</span><span class="p">))</span>
<span class="gp">>>> </span><span class="n">flat_imgs</span> <span class="o">=</span> <span class="n">imgs</span><span class="o">.</span><span class="n">flatten</span><span class="p">([</span><span class="s1">'C'</span><span class="p">,</span> <span class="s1">'H'</span><span class="p">,</span> <span class="s1">'W'</span><span class="p">],</span> <span class="s1">'features'</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">flat_imgs</span><span class="o">.</span><span class="n">names</span><span class="p">,</span> <span class="n">flat_imgs</span><span class="o">.</span><span class="n">shape</span>
<span class="go">(('N', 'features'), torch.Size([32, 49152]))</span>
</pre></div>
</div>
<div class="admonition warning">
<p class="admonition-title">Warning</p>
<p>The named tensor API is experimental and subject to change.</p>
</div>
</dd></dl>
</dd></dl>
</div>
</div>
</article>
</div>
<footer>
<div class="rst-footer-buttons" role="navigation" aria-label="footer navigation">
<a href="name_inference.html" class="btn btn-neutral float-right" title="Named Tensors operator coverage" accesskey="n" rel="next">Next <img src="_static/images/chevron-right-orange.svg" class="next-page"></a>
<a href="type_info.html" class="btn btn-neutral" title="Type Info" accesskey="p" rel="prev"><img src="_static/images/chevron-right-orange.svg" class="previous-page"> Previous</a>
</div>
<hr>
<div role="contentinfo">
<p>
© Copyright 2023, PyTorch Contributors.
</p>
</div>
<div>
Built with <a href="http://sphinx-doc.org/">Sphinx</a> using a <a href="https://github.com/rtfd/sphinx_rtd_theme">theme</a> provided by <a href="https://readthedocs.org">Read the Docs</a>.
</div>
</footer>
</div>
<script>
var match = window.location.href.match(/\/_[a-zA-Z0-9_]*.html|_dynamo/gi);
var url = window.location.href.lastIndexOf(match[match.length-1]);
if (url)