-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathnested.html
1383 lines (1139 loc) · 98.5 KB
/
nested.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.nested — PyTorch 2.6 documentation</title>
<link rel="canonical" href="https://pytorch.org/docs/stable/_modules/torch/nested.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="stylesheet" href="../../_static/css/custom.css" type="text/css" />
<link rel="index" title="Index" href="../../genindex.html" />
<link rel="search" title="Search" href="../../search.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'>2.6 ▼</a>
</div>
<div id="searchBox">
<div class="searchbox" id="googleSearchBox">
<script async src="https://cse.google.com/cse.js?cx=e65585f8c3ea1440e"></script>
<div class="gcse-search"></div>
</div>
<div id="sphinxSearchBox" style="display: none;">
<div role="search">
<form id="rtd-search-form" class="wy-form" action="../../search.html" method="get">
<input type="text" name="q" placeholder="Search Docs" />
<input type="hidden" name="check_keywords" value="yes" />
<input type="hidden" name="area" value="default" />
</form>
</div>
</div>
</div>
<form id="searchForm">
<label style="margin-bottom: 1rem">
<input type="radio" name="searchType" value="google" checked>
Google Search
</label>
<label style="margin-bottom: 1rem">
<input type="radio" name="searchType" value="sphinx">
Classic Search
</label>
</form>
<script>
document.addEventListener('DOMContentLoaded', function() {
const searchForm = document.getElementById('searchForm');
const googleSearchBox = document.getElementById('googleSearchBox');
const sphinxSearchBox = document.getElementById('sphinxSearchBox');
// Function to toggle search box visibility
function toggleSearchBox(searchType) {
googleSearchBox.style.display = searchType === 'google' ? 'block' : 'none';
sphinxSearchBox.style.display = searchType === 'sphinx' ? 'block' : 'none';
}
// Determine the default search type
let defaultSearchType;
const currentUrl = window.location.href;
if (currentUrl.startsWith('https://pytorch.org/docs/stable')) {
// For the stable documentation, default to Google
defaultSearchType = localStorage.getItem('searchType') || 'google';
} else {
// For any other version, including docs-preview, default to Sphinx
defaultSearchType = 'sphinx';
}
// Set the default search type
document.querySelector(`input[name="searchType"][value="${defaultSearchType}"]`).checked = true;
toggleSearchBox(defaultSearchType);
// Event listener for changes in search type
searchForm.addEventListener('change', function(event) {
const selectedSearchType = event.target.value;
localStorage.setItem('searchType', selectedSearchType);
toggleSearchBox(selectedSearchType);
});
// Set placeholder text for Google search box
window.onload = function() {
var placeholderText = "Search Docs";
var googleSearchboxText = document.querySelector("#gsc-i-id1");
if (googleSearchboxText) {
googleSearchboxText.placeholder = placeholderText;
googleSearchboxText.style.fontFamily = 'FreightSans';
googleSearchboxText.style.fontSize = "1.2rem";
googleSearchboxText.style.color = '#262626';
}
};
});
</script>
<!--temporarily add a link to survey -->
<div class="survey-link">
<p><i class="fas fa-poll" aria-hidden="true">  </i><a href="https://forms.gle/tdrnwJhaQ9tUePxz9">Share Your Feedback</a> about our new search</p>
</div>
</div>
<p class="caption" role="heading"><span class="caption-text">Community</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../../community/build_ci_governance.html">PyTorch Governance | Build + CI</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../community/contribution_guide.html">PyTorch Contribution Guide</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../community/design.html">PyTorch Design Philosophy</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../community/governance.html">PyTorch Governance | Mechanics</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../community/persons_of_interest.html">PyTorch Governance | Maintainers</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">Developer Notes</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../../notes/amp_examples.html">Automatic Mixed Precision examples</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../notes/autograd.html">Autograd mechanics</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../notes/broadcasting.html">Broadcasting semantics</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../notes/cpu_threading_torchscript_inference.html">CPU threading and TorchScript inference</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../notes/cuda.html">CUDA semantics</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../notes/custom_operators.html">PyTorch Custom Operators Landing Page</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../notes/ddp.html">Distributed Data Parallel</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../notes/extending.html">Extending PyTorch</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../notes/extending.func.html">Extending torch.func with autograd.Function</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../notes/faq.html">Frequently Asked Questions</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../notes/fsdp.html">FSDP Notes</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../notes/get_start_xpu.html">Getting Started on Intel GPU</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../notes/gradcheck.html">Gradcheck mechanics</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../notes/hip.html">HIP (ROCm) semantics</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../notes/large_scale_deployments.html">Features for large-scale deployments</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../notes/modules.html">Modules</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../notes/mps.html">MPS backend</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../notes/multiprocessing.html">Multiprocessing best practices</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../notes/numerical_accuracy.html">Numerical accuracy</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../notes/randomness.html">Reproducibility</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../notes/serialization.html">Serialization semantics</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../notes/windows.html">Windows FAQ</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">Language Bindings</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../../cpp_index.html">C++</a></li>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/javadoc/">Javadoc</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../deploy.html">torch::deploy</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">Python API</span></p>
<ul>
<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="../../accelerator.html">torch.accelerator</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../cpu.html">torch.cpu</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../cuda.html">torch.cuda</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../torch_cuda_memory.html">Understanding CUDA Memory Usage</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../torch_cuda_memory.html#generating-a-snapshot">Generating a Snapshot</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../torch_cuda_memory.html#using-the-visualizer">Using the visualizer</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../torch_cuda_memory.html#snapshot-api-reference">Snapshot API Reference</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../mps.html">torch.mps</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../xpu.html">torch.xpu</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../mtia.html">torch.mtia</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../mtia.memory.html">torch.mtia.memory</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../meta.html">Meta device</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../backends.html">torch.backends</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../export.html">torch.export</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../distributed.html">torch.distributed</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../distributed.tensor.html">torch.distributed.tensor</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../distributed.algorithms.join.html">torch.distributed.algorithms.join</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../distributed.elastic.html">torch.distributed.elastic</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../fsdp.html">torch.distributed.fsdp</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../distributed.fsdp.fully_shard.html">torch.distributed.fsdp.fully_shard</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../distributed.tensor.parallel.html">torch.distributed.tensor.parallel</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../distributed.optim.html">torch.distributed.optim</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../distributed.pipelining.html">torch.distributed.pipelining</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../distributed.checkpoint.html">torch.distributed.checkpoint</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../distributions.html">torch.distributions</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../torch.compiler.html">torch.compiler</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../fft.html">torch.fft</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../func.html">torch.func</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../futures.html">torch.futures</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../fx.html">torch.fx</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../fx.experimental.html">torch.fx.experimental</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../hub.html">torch.hub</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../jit.html">torch.jit</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../linalg.html">torch.linalg</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../monitor.html">torch.monitor</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../signal.html">torch.signal</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../special.html">torch.special</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../torch.overrides.html">torch.overrides</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../package.html">torch.package</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../profiler.html">torch.profiler</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../nn.init.html">torch.nn.init</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../nn.attention.html">torch.nn.attention</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../onnx.html">torch.onnx</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../optim.html">torch.optim</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../complex_numbers.html">Complex Numbers</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../ddp_comm_hooks.html">DDP Communication Hooks</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../quantization.html">Quantization</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../rpc.html">Distributed RPC Framework</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../random.html">torch.random</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../masked.html">torch.masked</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../nested.html">torch.nested</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../size.html">torch.Size</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../sparse.html">torch.sparse</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../storage.html">torch.Storage</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../testing.html">torch.testing</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../utils.html">torch.utils</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../benchmark_utils.html">torch.utils.benchmark</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../bottleneck.html">torch.utils.bottleneck</a></li>
<li class="toctree-l1"><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="../../module_tracker.html">torch.utils.module_tracker</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../type_info.html">Type Info</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../named_tensor.html">Named Tensors</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../name_inference.html">Named Tensors operator coverage</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../config_mod.html">torch.__config__</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../future_mod.html">torch.__future__</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../logging.html">torch._logging</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../torch_environment_variables.html">Torch Environment Variables</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">Libraries</span></p>
<ul>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/audio/stable">torchaudio</a></li>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/data">TorchData</a></li>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/torchrec">TorchRec</a></li>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/serve">TorchServe</a></li>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/text/stable">torchtext</a></li>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/vision/stable">torchvision</a></li>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/xla/">PyTorch on XLA Devices</a></li>
</ul>
</div>
</div>
</nav>
<div class="pytorch-container">
<div class="pytorch-page-level-bar" id="pytorch-page-level-bar">
<div class="pytorch-breadcrumbs-wrapper">
<div role="navigation" aria-label="breadcrumbs navigation">
<ul class="pytorch-breadcrumbs">
<li>
<a href="../../index.html">
Docs
</a> >
</li>
<li><a href="../index.html">Module code</a> ></li>
<li><a href="../torch.html">torch</a> ></li>
<li>torch.nested</li>
<li class="pytorch-breadcrumbs-aside">
</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">
<h1>Source code for torch.nested</h1><div class="highlight"><pre>
<span></span><span class="c1"># mypy: allow-untyped-defs</span>
<span class="kn">from</span> <span class="nn">typing</span> <span class="kn">import</span> <span class="n">List</span><span class="p">,</span> <span class="n">Optional</span><span class="p">,</span> <span class="n">Tuple</span><span class="p">,</span> <span class="n">Union</span>
<span class="kn">import</span> <span class="nn">torch</span>
<span class="kn">import</span> <span class="nn">torch.nn.functional</span> <span class="k">as</span> <span class="nn">F</span>
<span class="kn">from</span> <span class="nn">torch</span> <span class="kn">import</span> <span class="n">SymInt</span><span class="p">,</span> <span class="n">Tensor</span>
<span class="kn">from</span> <span class="nn">torch._C</span> <span class="kn">import</span> <span class="n">_add_docstr</span><span class="p">,</span> <span class="n">_nested</span> <span class="c1"># type: ignore[attr-defined]</span>
<span class="kn">from</span> <span class="nn">torch.types</span> <span class="kn">import</span> <span class="n">_device</span> <span class="k">as</span> <span class="n">Device</span><span class="p">,</span> <span class="n">_dtype</span> <span class="k">as</span> <span class="n">DType</span>
<span class="n">__all__</span> <span class="o">=</span> <span class="p">[</span>
<span class="s2">"to_padded_tensor"</span><span class="p">,</span>
<span class="s2">"as_nested_tensor"</span><span class="p">,</span>
<span class="s2">"nested_tensor"</span><span class="p">,</span>
<span class="s2">"nested_tensor_from_jagged"</span><span class="p">,</span>
<span class="s2">"narrow"</span><span class="p">,</span>
<span class="s2">"masked_select"</span><span class="p">,</span>
<span class="p">]</span>
<span class="c1"># Allowlist these for weights_only load of NJT</span>
<span class="kn">from</span> <span class="nn">._internal.nested_tensor</span> <span class="kn">import</span> <span class="n">NestedTensor</span> <span class="k">as</span> <span class="n">_NestedTensor</span><span class="p">,</span> <span class="n">_rebuild_njt</span>
<span class="n">torch</span><span class="o">.</span><span class="n">serialization</span><span class="o">.</span><span class="n">add_safe_globals</span><span class="p">([</span><span class="n">_NestedTensor</span><span class="p">,</span> <span class="n">_rebuild_njt</span><span class="p">])</span>
<div class="viewcode-block" id="as_nested_tensor"><a class="viewcode-back" href="../../nested.html#torch.nested.as_nested_tensor">[docs]</a><span class="k">def</span> <span class="nf">as_nested_tensor</span><span class="p">(</span>
<span class="n">ts</span><span class="p">:</span> <span class="n">Union</span><span class="p">[</span><span class="n">Tensor</span><span class="p">,</span> <span class="n">List</span><span class="p">[</span><span class="n">Tensor</span><span class="p">],</span> <span class="n">Tuple</span><span class="p">[</span><span class="n">Tensor</span><span class="p">,</span> <span class="o">...</span><span class="p">]],</span>
<span class="n">dtype</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">DType</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
<span class="n">device</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">Device</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
<span class="n">layout</span><span class="o">=</span><span class="kc">None</span>
<span class="p">)</span> <span class="o">-></span> <span class="n">Tensor</span><span class="p">:</span>
<span class="w"> </span><span class="sa">r</span><span class="sd">"""</span>
<span class="sd"> Constructs a nested tensor preserving autograd history from a tensor or a list / tuple of</span>
<span class="sd"> tensors.</span>
<span class="sd"> If a nested tensor is passed, it will be returned directly unless the device / dtype / layout</span>
<span class="sd"> differ. Note that converting device / dtype will result in a copy, while converting layout</span>
<span class="sd"> is not currently supported by this function.</span>
<span class="sd"> If a non-nested tensor is passed, it is treated as a batch of constituents of consistent size.</span>
<span class="sd"> A copy will be incurred if the passed device / dtype differ from those of the input OR if</span>
<span class="sd"> the input is non-contiguous. Otherwise, the input's storage will be used directly.</span>
<span class="sd"> If a tensor list is provided, tensors in the list are always copied during construction of</span>
<span class="sd"> the nested tensor.</span>
<span class="sd"> Args:</span>
<span class="sd"> ts (Tensor or List[Tensor] or Tuple[Tensor]): a tensor to treat as a nested tensor OR a</span>
<span class="sd"> list / tuple of tensors with the same ndim</span>
<span class="sd"> Keyword arguments:</span>
<span class="sd"> dtype (:class:`torch.dtype`, optional): the desired type of returned nested tensor.</span>
<span class="sd"> Default: if None, same :class:`torch.dtype` as leftmost tensor in the list.</span>
<span class="sd"> device (:class:`torch.device`, optional): the desired device of returned nested tensor.</span>
<span class="sd"> Default: if None, same :class:`torch.device` as leftmost tensor in the list</span>
<span class="sd"> layout (:class:`torch.layout`, optional): the desired layout of returned nested tensor.</span>
<span class="sd"> Only strided and jagged layouts are supported. Default: if None, the strided layout.</span>
<span class="sd"> Example::</span>
<span class="sd"> >>> a = torch.arange(3, dtype=torch.float, requires_grad=True)</span>
<span class="sd"> >>> b = torch.arange(5, dtype=torch.float, requires_grad=True)</span>
<span class="sd"> >>> nt = torch.nested.as_nested_tensor([a, b])</span>
<span class="sd"> >>> nt.is_leaf</span>
<span class="sd"> False</span>
<span class="sd"> >>> fake_grad = torch.nested.nested_tensor([torch.ones_like(a), torch.zeros_like(b)])</span>
<span class="sd"> >>> nt.backward(fake_grad)</span>
<span class="sd"> >>> a.grad</span>
<span class="sd"> tensor([1., 1., 1.])</span>
<span class="sd"> >>> b.grad</span>
<span class="sd"> tensor([0., 0., 0., 0., 0.])</span>
<span class="sd"> >>> c = torch.randn(3, 5, requires_grad=True)</span>
<span class="sd"> >>> nt2 = torch.nested.as_nested_tensor(c)</span>
<span class="sd"> """</span>
<span class="n">is_tensor_list</span> <span class="o">=</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">ts</span><span class="p">,</span> <span class="p">(</span><span class="nb">list</span><span class="p">,</span> <span class="nb">tuple</span><span class="p">))</span> <span class="ow">and</span> <span class="nb">all</span><span class="p">(</span><span class="nb">isinstance</span><span class="p">(</span><span class="n">t</span><span class="p">,</span> <span class="n">Tensor</span><span class="p">)</span> <span class="k">for</span> <span class="n">t</span> <span class="ow">in</span> <span class="n">ts</span><span class="p">)</span>
<span class="k">if</span> <span class="ow">not</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">ts</span><span class="p">,</span> <span class="n">Tensor</span><span class="p">)</span> <span class="ow">and</span> <span class="ow">not</span> <span class="n">is_tensor_list</span><span class="p">:</span>
<span class="k">raise</span> <span class="ne">TypeError</span><span class="p">(</span>
<span class="s2">"as_nested_tensor(): Expected first argument to be a tensor or a list / tuple of tensors "</span>
<span class="p">)</span>
<span class="c1"># convert tuple -> list if needed</span>
<span class="k">if</span> <span class="n">is_tensor_list</span> <span class="ow">and</span> <span class="ow">not</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">ts</span><span class="p">,</span> <span class="nb">list</span><span class="p">):</span>
<span class="n">ts</span> <span class="o">=</span> <span class="nb">list</span><span class="p">(</span><span class="n">ts</span><span class="p">)</span>
<span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">ts</span><span class="p">,</span> <span class="n">Tensor</span><span class="p">)</span> <span class="ow">and</span> <span class="n">ts</span><span class="o">.</span><span class="n">dim</span><span class="p">()</span> <span class="o"><</span> <span class="mi">2</span><span class="p">:</span>
<span class="k">raise</span> <span class="ne">RuntimeError</span><span class="p">(</span><span class="s2">"as_nested_tensor(): Expected tensor argument to have dim() > 1"</span><span class="p">)</span>
<span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">ts</span><span class="p">,</span> <span class="n">Tensor</span><span class="p">)</span> <span class="ow">and</span> <span class="n">ts</span><span class="o">.</span><span class="n">is_nested</span><span class="p">:</span>
<span class="k">if</span> <span class="n">layout</span> <span class="o">==</span> <span class="n">ts</span><span class="o">.</span><span class="n">layout</span><span class="p">:</span>
<span class="c1"># return input directly or input copied to device / dtype</span>
<span class="k">return</span> <span class="n">ts</span><span class="o">.</span><span class="n">to</span><span class="p">(</span><span class="n">device</span><span class="o">=</span><span class="n">device</span><span class="p">,</span> <span class="n">dtype</span><span class="o">=</span><span class="n">dtype</span><span class="p">)</span>
<span class="k">else</span><span class="p">:</span>
<span class="c1"># TODO: Just use nt.to(layout=layout) when it exists.</span>
<span class="k">raise</span> <span class="ne">RuntimeError</span><span class="p">(</span>
<span class="s2">"as_nested_tensor(): Converting between nested tensor layouts is not supported"</span><span class="p">)</span>
<span class="k">if</span> <span class="n">layout</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
<span class="n">layout</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">strided</span>
<span class="k">if</span> <span class="n">layout</span> <span class="o">==</span> <span class="n">torch</span><span class="o">.</span><span class="n">strided</span><span class="p">:</span>
<span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">ts</span><span class="p">,</span> <span class="n">Tensor</span><span class="p">):</span>
<span class="c1"># contiguous() might be necessary to get flattened view.</span>
<span class="c1"># we could probably be more precise about when to do this as an optimization</span>
<span class="n">buffer</span> <span class="o">=</span> <span class="n">ts</span><span class="o">.</span><span class="n">contiguous</span><span class="p">()</span><span class="o">.</span><span class="n">view</span><span class="p">(</span><span class="o">-</span><span class="mi">1</span><span class="p">)</span><span class="o">.</span><span class="n">to</span><span class="p">(</span><span class="n">device</span><span class="o">=</span><span class="n">device</span><span class="p">,</span> <span class="n">dtype</span><span class="o">=</span><span class="n">dtype</span><span class="p">)</span>
<span class="n">nested_sizes</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">tensor</span><span class="p">([</span><span class="n">t</span><span class="o">.</span><span class="n">shape</span> <span class="k">for</span> <span class="n">t</span> <span class="ow">in</span> <span class="n">ts</span><span class="p">])</span>
<span class="k">return</span> <span class="n">torch</span><span class="o">.</span><span class="n">_nested_view_from_buffer</span><span class="p">(</span>
<span class="n">buffer</span><span class="p">,</span>
<span class="n">nested_sizes</span><span class="p">,</span>
<span class="o">*</span><span class="n">torch</span><span class="o">.</span><span class="n">_nested_compute_contiguous_strides_offsets</span><span class="p">(</span><span class="n">nested_sizes</span><span class="p">))</span>
<span class="k">else</span><span class="p">:</span>
<span class="k">assert</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">ts</span><span class="p">,</span> <span class="nb">list</span><span class="p">)</span>
<span class="k">return</span> <span class="n">torch</span><span class="o">.</span><span class="n">_nested_tensor_from_tensor_list</span><span class="p">(</span><span class="n">ts</span><span class="p">,</span> <span class="n">dtype</span><span class="p">,</span> <span class="kc">None</span><span class="p">,</span> <span class="n">device</span><span class="p">,</span> <span class="kc">None</span><span class="p">)</span>
<span class="k">elif</span> <span class="n">layout</span> <span class="o">==</span> <span class="n">torch</span><span class="o">.</span><span class="n">jagged</span><span class="p">:</span>
<span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">ts</span><span class="p">,</span> <span class="n">Tensor</span><span class="p">):</span>
<span class="k">if</span> <span class="n">device</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
<span class="n">device</span> <span class="o">=</span> <span class="n">ts</span><span class="o">.</span><span class="n">device</span>
<span class="c1"># contiguous() might be necessary to get flattened view.</span>
<span class="c1"># we could probably be more precise about when to do this as an optimization</span>
<span class="n">values</span> <span class="o">=</span> <span class="n">ts</span><span class="o">.</span><span class="n">contiguous</span><span class="p">()</span><span class="o">.</span><span class="n">flatten</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="o">.</span><span class="n">to</span><span class="p">(</span><span class="n">device</span><span class="o">=</span><span class="n">device</span><span class="p">,</span> <span class="n">dtype</span><span class="o">=</span><span class="n">dtype</span><span class="p">)</span>
<span class="n">batch_size</span> <span class="o">=</span> <span class="n">ts</span><span class="o">.</span><span class="n">shape</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span>
<span class="n">seq_len</span> <span class="o">=</span> <span class="n">ts</span><span class="o">.</span><span class="n">shape</span><span class="p">[</span><span class="mi">1</span><span class="p">]</span>
<span class="n">offsets</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">arange</span><span class="p">(</span><span class="mi">0</span><span class="p">,</span> <span class="n">batch_size</span> <span class="o">*</span> <span class="n">seq_len</span> <span class="o">+</span> <span class="mi">1</span><span class="p">,</span> <span class="n">seq_len</span><span class="p">,</span>
<span class="n">device</span><span class="o">=</span><span class="n">device</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">int64</span><span class="p">)</span>
<span class="kn">from</span> <span class="nn">torch.nested._internal.nested_tensor</span> <span class="kn">import</span> <span class="n">nested_view_from_values_offsets</span>
<span class="k">return</span> <span class="n">nested_view_from_values_offsets</span><span class="p">(</span>
<span class="n">values</span><span class="p">,</span> <span class="n">offsets</span><span class="p">,</span> <span class="n">min_seqlen</span><span class="o">=</span><span class="n">seq_len</span><span class="p">,</span> <span class="n">max_seqlen</span><span class="o">=</span><span class="n">seq_len</span>
<span class="p">)</span>
<span class="k">else</span><span class="p">:</span>
<span class="kn">from</span> <span class="nn">torch.nested._internal.nested_tensor</span> <span class="kn">import</span> <span class="n">jagged_from_list</span>
<span class="k">assert</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">ts</span><span class="p">,</span> <span class="nb">list</span><span class="p">)</span>
<span class="n">nt</span><span class="p">,</span> <span class="n">_</span> <span class="o">=</span> <span class="n">jagged_from_list</span><span class="p">(</span><span class="n">ts</span><span class="p">,</span> <span class="n">offsets</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span> <span class="n">device</span><span class="o">=</span><span class="n">device</span><span class="p">,</span> <span class="n">dtype</span><span class="o">=</span><span class="n">dtype</span><span class="p">)</span>
<span class="k">return</span> <span class="n">nt</span>
<span class="k">else</span><span class="p">:</span>
<span class="k">raise</span> <span class="ne">RuntimeError</span><span class="p">(</span><span class="sa">f</span><span class="s2">"Specified layout is unsupported for nested tensors: </span><span class="si">{</span><span class="n">layout</span><span class="si">}</span><span class="s2">"</span><span class="p">)</span></div>
<span class="c1"># Note: This not only adds doc strings for the nested ops, but</span>
<span class="c1"># also connects the torch.nested Python namespace to the torch._C._nested builtins.</span>
<span class="n">to_padded_tensor</span> <span class="o">=</span> <span class="n">_add_docstr</span><span class="p">(</span>
<span class="n">_nested</span><span class="o">.</span><span class="n">nested_to_padded_tensor</span><span class="p">,</span>
<span class="w"> </span><span class="sa">r</span><span class="sd">"""</span>
<span class="sd">to_padded_tensor(input, padding, output_size=None, out=None) -> Tensor</span>
<span class="sd">Returns a new (non-nested) Tensor by padding the :attr:`input` nested tensor.</span>
<span class="sd">The leading entries will be filled with the nested data,</span>
<span class="sd">while the trailing entries will be padded.</span>
<span class="sd">.. warning::</span>
<span class="sd"> :func:`to_padded_tensor` always copies the underlying data,</span>
<span class="sd"> since the nested and the non-nested tensors differ in memory layout.</span>
<span class="sd">Args:</span>
<span class="sd"> padding (float): The padding value for the trailing entries.</span>
<span class="sd">Keyword args:</span>
<span class="sd"> output_size (Tuple[int]): The size of the output tensor.</span>
<span class="sd"> If given, it must be large enough to contain all nested data;</span>
<span class="sd"> else, will infer by taking the max size of each nested sub-tensor along each dimension.</span>
<span class="sd"> out (Tensor, optional): the output tensor.</span>
<span class="sd">Example::</span>
<span class="sd"> >>> nt = torch.nested.nested_tensor([torch.randn((2, 5)), torch.randn((3, 4))])</span>
<span class="sd"> nested_tensor([</span>
<span class="sd"> tensor([[ 1.6862, -1.1282, 1.1031, 0.0464, -1.3276],</span>
<span class="sd"> [-1.9967, -1.0054, 1.8972, 0.9174, -1.4995]]),</span>
<span class="sd"> tensor([[-1.8546, -0.7194, -0.2918, -0.1846],</span>
<span class="sd"> [ 0.2773, 0.8793, -0.5183, -0.6447],</span>
<span class="sd"> [ 1.8009, 1.8468, -0.9832, -1.5272]])</span>
<span class="sd"> ])</span>
<span class="sd"> >>> pt_infer = torch.nested.to_padded_tensor(nt, 0.0)</span>
<span class="sd"> tensor([[[ 1.6862, -1.1282, 1.1031, 0.0464, -1.3276],</span>
<span class="sd"> [-1.9967, -1.0054, 1.8972, 0.9174, -1.4995],</span>
<span class="sd"> [ 0.0000, 0.0000, 0.0000, 0.0000, 0.0000]],</span>
<span class="sd"> [[-1.8546, -0.7194, -0.2918, -0.1846, 0.0000],</span>
<span class="sd"> [ 0.2773, 0.8793, -0.5183, -0.6447, 0.0000],</span>
<span class="sd"> [ 1.8009, 1.8468, -0.9832, -1.5272, 0.0000]]])</span>
<span class="sd"> >>> pt_large = torch.nested.to_padded_tensor(nt, 1.0, (2, 4, 6))</span>
<span class="sd"> tensor([[[ 1.6862, -1.1282, 1.1031, 0.0464, -1.3276, 1.0000],</span>
<span class="sd"> [-1.9967, -1.0054, 1.8972, 0.9174, -1.4995, 1.0000],</span>
<span class="sd"> [ 1.0000, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000],</span>
<span class="sd"> [ 1.0000, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000]],</span>
<span class="sd"> [[-1.8546, -0.7194, -0.2918, -0.1846, 1.0000, 1.0000],</span>
<span class="sd"> [ 0.2773, 0.8793, -0.5183, -0.6447, 1.0000, 1.0000],</span>
<span class="sd"> [ 1.8009, 1.8468, -0.9832, -1.5272, 1.0000, 1.0000],</span>
<span class="sd"> [ 1.0000, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000]]])</span>
<span class="sd"> >>> pt_small = torch.nested.to_padded_tensor(nt, 2.0, (2, 2, 2))</span>
<span class="sd"> RuntimeError: Value in output_size is less than NestedTensor padded size. Truncation is not supported.</span>
<span class="sd">"""</span><span class="p">,</span>
<span class="p">)</span>
<div class="viewcode-block" id="nested_tensor"><a class="viewcode-back" href="../../nested.html#torch.nested.nested_tensor">[docs]</a><span class="k">def</span> <span class="nf">nested_tensor</span><span class="p">(</span><span class="n">tensor_list</span><span class="p">,</span> <span class="o">*</span><span class="p">,</span> <span class="n">dtype</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span> <span class="n">layout</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span> <span class="n">device</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span> <span class="n">requires_grad</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span> <span class="n">pin_memory</span><span class="o">=</span><span class="kc">False</span><span class="p">)</span> <span class="o">-></span> <span class="n">Tensor</span><span class="p">:</span>
<span class="w"> </span><span class="sa">r</span><span class="sd">"""</span>
<span class="sd">Constructs a nested tensor with no autograd history (also known as a "leaf tensor", see</span>
<span class="sd">:ref:`Autograd mechanics <autograd-mechanics>`) from :attr:`tensor_list` a list of tensors.</span>
<span class="sd">Args:</span>
<span class="sd"> tensor_list (List[array_like]): a list of tensors, or anything that can be passed to torch.tensor,</span>
<span class="sd"> where each element of the list has the same dimensionality.</span>
<span class="sd">Keyword arguments:</span>
<span class="sd"> dtype (:class:`torch.dtype`, optional): the desired type of returned nested tensor.</span>
<span class="sd"> Default: if None, same :class:`torch.dtype` as leftmost tensor in the list.</span>
<span class="sd"> layout (:class:`torch.layout`, optional): the desired layout of returned nested tensor.</span>
<span class="sd"> Only strided and jagged layouts are supported. Default: if None, the strided layout.</span>
<span class="sd"> device (:class:`torch.device`, optional): the desired device of returned nested tensor.</span>
<span class="sd"> Default: if None, same :class:`torch.device` as leftmost tensor in the list</span>
<span class="sd"> requires_grad (bool, optional): If autograd should record operations on the</span>
<span class="sd"> returned nested tensor. Default: ``False``.</span>
<span class="sd"> pin_memory (bool, optional): If set, returned nested tensor would be allocated in</span>
<span class="sd"> the pinned memory. Works only for CPU tensors. Default: ``False``.</span>
<span class="sd">Example::</span>
<span class="sd"> >>> a = torch.arange(3, dtype=torch.float, requires_grad=True)</span>
<span class="sd"> >>> b = torch.arange(5, dtype=torch.float, requires_grad=True)</span>
<span class="sd"> >>> nt = torch.nested.nested_tensor([a, b], requires_grad=True)</span>
<span class="sd"> >>> nt.is_leaf</span>
<span class="sd"> True</span>
<span class="sd"> """</span>
<span class="k">if</span> <span class="n">layout</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
<span class="n">layout</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">strided</span>
<span class="k">if</span> <span class="n">layout</span> <span class="o">==</span> <span class="n">torch</span><span class="o">.</span><span class="n">strided</span><span class="p">:</span>
<span class="k">return</span> <span class="n">_nested</span><span class="o">.</span><span class="n">nested_tensor</span><span class="p">(</span>
<span class="n">tensor_list</span><span class="p">,</span>
<span class="n">dtype</span><span class="o">=</span><span class="n">dtype</span><span class="p">,</span>
<span class="n">device</span><span class="o">=</span><span class="n">device</span><span class="p">,</span>
<span class="n">requires_grad</span><span class="o">=</span><span class="n">requires_grad</span><span class="p">,</span>
<span class="n">pin_memory</span><span class="o">=</span><span class="n">pin_memory</span><span class="p">)</span>
<span class="k">elif</span> <span class="n">layout</span> <span class="o">==</span> <span class="n">torch</span><span class="o">.</span><span class="n">jagged</span><span class="p">:</span>
<span class="c1"># Need to wrap lists of scalars as tensors</span>
<span class="n">list_of_tensors</span> <span class="o">=</span> <span class="p">[</span><span class="n">t</span> <span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">t</span><span class="p">,</span> <span class="n">Tensor</span><span class="p">)</span> <span class="k">else</span> <span class="n">torch</span><span class="o">.</span><span class="n">as_tensor</span><span class="p">(</span><span class="n">t</span><span class="p">)</span> <span class="k">for</span> <span class="n">t</span> <span class="ow">in</span> <span class="n">tensor_list</span><span class="p">]</span>
<span class="kn">from</span> <span class="nn">torch.nested._internal.nested_tensor</span> <span class="kn">import</span> <span class="n">jagged_from_list</span>
<span class="k">with</span> <span class="n">torch</span><span class="o">.</span><span class="n">no_grad</span><span class="p">():</span>
<span class="n">nt</span><span class="p">,</span> <span class="n">_</span> <span class="o">=</span> <span class="n">jagged_from_list</span><span class="p">(</span><span class="n">list_of_tensors</span><span class="p">,</span> <span class="n">offsets</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span> <span class="n">device</span><span class="o">=</span><span class="n">device</span><span class="p">,</span> <span class="n">dtype</span><span class="o">=</span><span class="n">dtype</span><span class="p">)</span>
<span class="n">nt</span><span class="o">.</span><span class="n">requires_grad_</span><span class="p">(</span><span class="n">requires_grad</span><span class="p">)</span>
<span class="k">if</span> <span class="n">pin_memory</span><span class="p">:</span>
<span class="n">nt</span> <span class="o">=</span> <span class="n">nt</span><span class="o">.</span><span class="n">pin_memory</span><span class="p">()</span> <span class="c1"># type: ignore[assignment]</span>
<span class="k">return</span> <span class="n">nt</span>
<span class="k">else</span><span class="p">:</span>
<span class="k">raise</span> <span class="ne">RuntimeError</span><span class="p">(</span><span class="sa">f</span><span class="s2">"Specified layout is unsupported for nested tensors: </span><span class="si">{</span><span class="n">layout</span><span class="si">}</span><span class="s2">"</span><span class="p">)</span></div>
<span class="k">def</span> <span class="nf">narrow</span><span class="p">(</span><span class="n">tensor</span><span class="p">:</span> <span class="n">Tensor</span><span class="p">,</span> <span class="n">dim</span><span class="p">:</span> <span class="nb">int</span><span class="p">,</span> <span class="n">start</span><span class="p">:</span> <span class="n">Union</span><span class="p">[</span><span class="nb">int</span><span class="p">,</span> <span class="n">Tensor</span><span class="p">],</span> <span class="n">length</span><span class="p">:</span> <span class="n">Union</span><span class="p">[</span><span class="nb">int</span><span class="p">,</span> <span class="n">Tensor</span><span class="p">],</span> <span class="n">layout</span><span class="o">=</span><span class="n">torch</span><span class="o">.</span><span class="n">strided</span><span class="p">)</span> <span class="o">-></span> <span class="n">Tensor</span><span class="p">:</span>
<span class="w"> </span><span class="sa">r</span><span class="sd">"""</span>
<span class="sd">Constructs a nested tensor (which might be a view) from :attr:`tensor`, a strided tensor. This follows</span>
<span class="sd">similar semantics to torch.Tensor.narrow, where in the :attr:`dim`-th dimension the new nested tensor</span>
<span class="sd">shows only the elements in the interval `[start, start+length)`. As nested representations</span>
<span class="sd">allow for a different `start` and `length` at each 'row' of that dimension, :attr:`start` and :attr:`length`</span>
<span class="sd">can also be tensors of shape `tensor.shape[0]`.</span>
<span class="sd">There's some differences depending on the layout you use for the nested tensor. If using strided layout,</span>
<span class="sd">torch.narrow will do a copy of the narrowed data into a contiguous NT with strided layout, while</span>
<span class="sd">jagged layout narrow() will create a non-contiguous view of your original strided tensor. This particular</span>
<span class="sd">representation is really useful for representing kv-caches in Transformer models, as specialized</span>
<span class="sd">SDPA kernels can deal with format easily, resulting in performance improvements.</span>
<span class="sd">Args:</span>
<span class="sd"> tensor (:class:`torch.Tensor`): a strided tensor, which will be used as the underlying data</span>
<span class="sd"> for the nested tensor if using the jagged layout or will be copied for the strided layout.</span>
<span class="sd"> dim (int): the dimension where narrow will be applied. Only `dim=1` is supported for the</span>
<span class="sd"> jagged layout, while strided supports all dim</span>
<span class="sd"> start (Union[int, :class:`torch.Tensor`]): starting element for the narrow operation</span>
<span class="sd"> length (Union[int, :class:`torch.Tensor`]): number of elements taken during the narrow op</span>
<span class="sd">Keyword arguments:</span>
<span class="sd"> layout (:class:`torch.layout`, optional): the desired layout of returned nested tensor.</span>
<span class="sd"> Only strided and jagged layouts are supported. Default: if None, the strided layout.</span>
<span class="sd">Example::</span>
<span class="sd"> >>> starts = torch.tensor([0, 1, 2, 3, 4], dtype=torch.int64)</span>
<span class="sd"> >>> lengths = torch.tensor([3, 2, 2, 1, 5], dtype=torch.int64)</span>
<span class="sd"> >>> narrow_base = torch.randn(5, 10, 20)</span>
<span class="sd"> >>> nt_narrowed = torch.nested.narrow(narrow_base, 1, starts, lengths, layout=torch.jagged)</span>
<span class="sd"> >>> nt_narrowed.is_contiguous()</span>
<span class="sd"> False</span>
<span class="sd"> """</span>
<span class="k">if</span> <span class="ow">not</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">start</span><span class="p">,</span> <span class="p">(</span><span class="nb">int</span><span class="p">,</span> <span class="n">SymInt</span><span class="p">,</span> <span class="n">Tensor</span><span class="p">)):</span>
<span class="k">raise</span> <span class="ne">RuntimeError</span><span class="p">(</span><span class="s2">"start must be an integer or a tensor"</span><span class="p">)</span>
<span class="k">if</span> <span class="ow">not</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">length</span><span class="p">,</span> <span class="p">(</span><span class="nb">int</span><span class="p">,</span> <span class="n">SymInt</span><span class="p">,</span> <span class="n">Tensor</span><span class="p">)):</span>
<span class="k">raise</span> <span class="ne">RuntimeError</span><span class="p">(</span><span class="s2">"length must be an integer or a tensor"</span><span class="p">)</span>
<span class="k">if</span> <span class="n">layout</span> <span class="o">==</span> <span class="n">torch</span><span class="o">.</span><span class="n">strided</span><span class="p">:</span>
<span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">start</span><span class="p">,</span> <span class="n">Tensor</span><span class="p">)</span> <span class="ow">or</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">length</span><span class="p">,</span> <span class="n">Tensor</span><span class="p">):</span>
<span class="k">raise</span> <span class="ne">RuntimeError</span><span class="p">(</span><span class="s2">"start and length must be integers for the strided layout NT impl"</span><span class="p">)</span>
<span class="c1"># TODO: switch to as_nested_tensor(tensor) when it is available</span>
<span class="n">nt</span> <span class="o">=</span> <span class="n">as_nested_tensor</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">unbind</span><span class="p">(</span><span class="n">tensor</span><span class="p">),</span> <span class="n">layout</span><span class="o">=</span><span class="n">torch</span><span class="o">.</span><span class="n">strided</span><span class="p">)</span><span class="o">.</span><span class="n">narrow</span><span class="p">(</span><span class="n">dim</span><span class="p">,</span> <span class="n">start</span><span class="p">,</span> <span class="n">length</span><span class="p">)</span>
<span class="k">elif</span> <span class="n">layout</span> <span class="o">==</span> <span class="n">torch</span><span class="o">.</span><span class="n">jagged</span><span class="p">:</span>
<span class="k">if</span> <span class="n">dim</span> <span class="o">!=</span> <span class="mi">1</span><span class="p">:</span>
<span class="k">raise</span> <span class="ne">RuntimeError</span><span class="p">(</span><span class="s2">"jagged layout only supports dim=1"</span><span class="p">)</span>
<span class="kn">from</span> <span class="nn">torch.nested._internal.nested_tensor</span> <span class="kn">import</span> <span class="n">jagged_from_tensor_and_lengths</span>
<span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">start</span><span class="p">,</span> <span class="p">(</span><span class="nb">int</span><span class="p">,</span> <span class="n">SymInt</span><span class="p">)):</span>
<span class="n">start</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">tensor</span><span class="p">([</span><span class="n">start</span><span class="p">],</span> <span class="n">device</span><span class="o">=</span><span class="n">tensor</span><span class="o">.</span><span class="n">device</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">int64</span><span class="p">)</span>
<span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">length</span><span class="p">,</span> <span class="p">(</span><span class="nb">int</span><span class="p">,</span> <span class="n">SymInt</span><span class="p">)):</span>
<span class="n">length</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">tensor</span><span class="p">([</span><span class="n">length</span><span class="p">],</span> <span class="n">device</span><span class="o">=</span><span class="n">tensor</span><span class="o">.</span><span class="n">device</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">int64</span><span class="p">)</span>
<span class="n">nt</span><span class="p">,</span> <span class="n">_</span><span class="p">,</span> <span class="n">_</span> <span class="o">=</span> <span class="n">jagged_from_tensor_and_lengths</span><span class="p">(</span><span class="n">tensor</span><span class="p">,</span> <span class="n">start</span><span class="p">,</span> <span class="n">length</span><span class="p">)</span>
<span class="k">else</span><span class="p">:</span>
<span class="k">raise</span> <span class="ne">RuntimeError</span><span class="p">(</span><span class="sa">f</span><span class="s2">"Specified layout is unsupported for nested narrow: </span><span class="si">{</span><span class="n">layout</span><span class="si">}</span><span class="s2">"</span><span class="p">)</span>
<span class="k">return</span> <span class="n">nt</span>
<span class="k">def</span> <span class="nf">nested_tensor_from_jagged</span><span class="p">(</span>
<span class="n">values</span><span class="p">:</span> <span class="n">Tensor</span><span class="p">,</span>
<span class="n">offsets</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">Tensor</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
<span class="n">lengths</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">Tensor</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
<span class="n">jagged_dim</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">int</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
<span class="n">min_seqlen</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">int</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
<span class="n">max_seqlen</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">int</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
<span class="p">)</span> <span class="o">-></span> <span class="n">Tensor</span><span class="p">:</span>
<span class="w"> </span><span class="sa">r</span><span class="sd">"""</span>
<span class="sd">Constructs a jagged layout nested tensor from the given jagged components. The jagged layout</span>
<span class="sd">consists of a required values buffer with the jagged dimension packed into a single dimension.</span>
<span class="sd">The offsets / lengths metadata determines how this dimension is split into batch elements</span>
<span class="sd">and are expected to be allocated on the same device as the values buffer.</span>
<span class="sd">Expected metadata formats:</span>
<span class="sd"> * offsets: Indices within the packed dimension splitting it into heterogeneously-sized</span>
<span class="sd"> batch elements. Example: [0, 2, 3, 6] indicates that a packed jagged dim of size 6</span>
<span class="sd"> should be conceptually split into batch elements of length [2, 1, 3]. Note that both the</span>
<span class="sd"> beginning and ending offsets are required for kernel convenience (i.e. shape batch_size + 1).</span>
<span class="sd"> * lengths: Lengths of the individual batch elements; shape == batch_size. Example: [2, 1, 3]</span>
<span class="sd"> indicates that a packed jagged dim of size 6 should be conceptually split into batch</span>
<span class="sd"> elements of length [2, 1, 3].</span>
<span class="sd">Note that it can be useful to provide both offsets and lengths. This describes a nested tensor</span>
<span class="sd">with "holes", where the offsets indicate the start position of each batch item and the length</span>
<span class="sd">specifies the total number of elements (see example below).</span>
<span class="sd">The returned jagged layout nested tensor will be a view of the input values tensor.</span>
<span class="sd">Args:</span>
<span class="sd"> values (:class:`torch.Tensor`): The underlying buffer in the shape of</span>
<span class="sd"> (sum_B(*), D_1, ..., D_N). The jagged dimension is packed into a single dimension,</span>
<span class="sd"> with the offsets / lengths metadata used to distinguish batch elements.</span>
<span class="sd"> offsets (optional :class:`torch.Tensor`): Offsets into the jagged dimension of shape B + 1.</span>
<span class="sd"> lengths (optional :class:`torch.Tensor`): Lengths of the batch elements of shape B.</span>
<span class="sd"> jagged_dim (optional int): Indicates which dimension in values is the packed jagged</span>
<span class="sd"> dimension. If None, this is set to dim=1 (i.e. the dimension immediately following</span>
<span class="sd"> the batch dimension). Default: None</span>
<span class="sd"> min_seqlen (optional int): If set, uses the specified value as the cached minimum sequence</span>
<span class="sd"> length for the returned nested tensor. This can be a useful alternative to computing</span>
<span class="sd"> this value on-demand, possibly avoiding a GPU -> CPU sync. Default: None</span>
<span class="sd"> max_seqlen (optional int): If set, uses the specified value as the cached maximum sequence</span>
<span class="sd"> length for the returned nested tensor. This can be a useful alternative to computing</span>
<span class="sd"> this value on-demand, possibly avoiding a GPU -> CPU sync. Default: None</span>
<span class="sd">Example::</span>
<span class="sd"> >>> values = torch.randn(12, 5)</span>
<span class="sd"> >>> offsets = torch.tensor([0, 3, 5, 6, 10, 12])</span>
<span class="sd"> >>> nt = nested_tensor_from_jagged(values, offsets)</span>
<span class="sd"> >>> # 3D shape with the middle dimension jagged</span>
<span class="sd"> >>> nt.shape</span>
<span class="sd"> torch.Size([5, j2, 5])</span>
<span class="sd"> >>> # Length of each item in the batch:</span>
<span class="sd"> >>> offsets.diff()</span>
<span class="sd"> tensor([3, 2, 1, 4, 2])</span>
<span class="sd"> >>> values = torch.randn(6, 5)</span>
<span class="sd"> >>> offsets = torch.tensor([0, 2, 3, 6])</span>
<span class="sd"> >>> lengths = torch.tensor([1, 1, 2])</span>
<span class="sd"> >>> # NT with holes</span>
<span class="sd"> >>> nt = nested_tensor_from_jagged(values, offsets, lengths)</span>
<span class="sd"> >>> a, b, c = nt.unbind()</span>
<span class="sd"> >>> # Batch item 1 consists of indices [0, 1)</span>
<span class="sd"> >>> torch.equal(a, values[0:1, :])</span>
<span class="sd"> True</span>
<span class="sd"> >>> # Batch item 2 consists of indices [2, 3)</span>
<span class="sd"> >>> torch.equal(b, values[2:3, :])</span>
<span class="sd"> True</span>
<span class="sd"> >>> # Batch item 3 consists of indices [3, 5)</span>
<span class="sd"> >>> torch.equal(c, values[3:5, :])</span>
<span class="sd"> True</span>
<span class="sd"> """</span>
<span class="kn">from</span> <span class="nn">torch.fx._symbolic_trace</span> <span class="kn">import</span> <span class="n">is_fx_tracing</span>
<span class="k">if</span> <span class="n">is_fx_tracing</span><span class="p">():</span>
<span class="k">raise</span> <span class="ne">RuntimeError</span><span class="p">(</span>
<span class="s2">"torch.nested.nested_tensor_from_jagged does not support tracing with fx.symbolic_trace. "</span>
<span class="s2">"Use fx.wrap to wrap the function that calls nested_tensor_from_jagged."</span>
<span class="p">)</span>
<span class="k">if</span> <span class="n">offsets</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
<span class="k">if</span> <span class="n">lengths</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
<span class="k">raise</span> <span class="ne">RuntimeError</span><span class="p">(</span>
<span class="s2">"nested_tensor_from_jagged(): At least one of offsets or lengths is required."</span>
<span class="p">)</span>
<span class="k">else</span><span class="p">:</span>
<span class="c1"># TODO: Truly support offsets=None at some point?</span>
<span class="c1"># For now, just convert lengths -> offsets for kernel convenience</span>
<span class="n">offsets</span> <span class="o">=</span> <span class="n">F</span><span class="o">.</span><span class="n">pad</span><span class="p">(</span><span class="n">lengths</span><span class="o">.</span><span class="n">cumsum</span><span class="p">(</span><span class="mi">0</span><span class="p">),</span> <span class="p">(</span><span class="mi">1</span><span class="p">,</span> <span class="mi">0</span><span class="p">))</span>
<span class="n">lengths</span> <span class="o">=</span> <span class="kc">None</span>
<span class="k">if</span> <span class="n">jagged_dim</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
<span class="n">jagged_dim</span> <span class="o">=</span> <span class="mi">1</span>
<span class="kn">from</span> <span class="nn">torch.nested._internal.nested_tensor</span> <span class="kn">import</span> <span class="n">nested_view_from_values_offsets_lengths</span>
<span class="k">return</span> <span class="n">nested_view_from_values_offsets_lengths</span><span class="p">(</span>
<span class="n">values</span><span class="p">,</span> <span class="n">offsets</span><span class="p">,</span> <span class="n">lengths</span><span class="p">,</span> <span class="n">ragged_idx</span><span class="o">=</span><span class="n">jagged_dim</span><span class="p">,</span> <span class="n">min_seqlen</span><span class="o">=</span><span class="n">min_seqlen</span><span class="p">,</span> <span class="n">max_seqlen</span><span class="o">=</span><span class="n">max_seqlen</span><span class="p">)</span>
<span class="k">def</span> <span class="nf">masked_select</span><span class="p">(</span><span class="n">tensor</span><span class="p">:</span> <span class="n">Tensor</span><span class="p">,</span> <span class="n">mask</span><span class="p">:</span> <span class="n">Tensor</span><span class="p">)</span> <span class="o">-></span> <span class="n">Tensor</span><span class="p">:</span>
<span class="w"> </span><span class="sa">r</span><span class="sd">"""</span>
<span class="sd"> Constructs a nested tensor given a strided tensor input and a strided mask, the resulting jagged layout nested tensor</span>
<span class="sd"> will have values retain values where the mask is equal to True. The dimensionality of the mask is preserved and is</span>
<span class="sd"> represented with the offsets, this is unlike :func:`masked_select` where the output is collapsed to a 1D tensor.</span>
<span class="sd"> Args:</span>
<span class="sd"> tensor (:class:`torch.Tensor`): a strided tensor from which the jagged layout nested tensor is constructed from.</span>