-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathlibrary.html
2158 lines (1961 loc) · 235 KB
/
library.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 charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>torch.library — PyTorch 2.7 documentation</title>
<link rel="canonical" href="https://pytorch.org/docs/stable/library.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" />
<link rel="next" title="torch.accelerator" href="accelerator.html" />
<link rel="prev" title="torch.autograd.graph.increment_version" href="generated/torch.autograd.graph.increment_version.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-2024">
<span class="dropdown-title">Contributor Awards - 2024</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>
<a class="nav-dropdown-item" href="https://pytorch.org/executorch/stable/index.html">
<span class="dropdown-title">ExecuTorch Docs</span>
</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>
<a class="nav-dropdown-item" href="https://pytorch.org/newsletter">
<span class="dropdown-title">Newsletter</span>
<p>Stay up-to-date with the latest updates</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>
</a>
<a class="nav-dropdown-item" href="https://pytorch.org/credits">
<span class="dropdown-title">Cloud Credit Program</span>
</a>
<a class="nav-dropdown-item" href="https://pytorch.org/tac">
<span class="dropdown-title">Technical Advisory Council</span>
</a>
<a class="nav-dropdown-item" href="https://pytorch.org/staff">
<span class="dropdown-title">Staff</span>
</a>
<a class="nav-dropdown-item" href="https://pytorch.org/contact-us">
<span class="dropdown-title">Contact Us</span>
</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'>2.7 ▼</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>
</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/libtorch_stable_abi.html">LibTorch Stable ABI</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 current"><a class="current reference internal" href="#">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>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/ao">torchao</a></li>
</ul>
</div>
</div>
</nav>
<div class="pytorch-container">
<div class="pytorch-page-level-bar" id="pytorch-page-level-bar">
<div class="pytorch-breadcrumbs-wrapper">
<div role="navigation" aria-label="breadcrumbs navigation">
<ul class="pytorch-breadcrumbs">
<li>
<a href="index.html">
Docs
</a> >
</li>
<li>torch.library</li>
<li class="pytorch-breadcrumbs-aside">
<a href="_sources/library.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="module-torch.library">
<span id="torch-library"></span><span id="torch-library-docs"></span><h1>torch.library<a class="headerlink" href="#module-torch.library" title="Permalink to this heading">¶</a></h1>
<p>torch.library is a collection of APIs for extending PyTorch’s core library
of operators. It contains utilities for testing custom operators, creating new
custom operators, and extending operators defined with PyTorch’s C++ operator
registration APIs (e.g. aten operators).</p>
<p>For a detailed guide on effectively using these APIs, please see
<a class="reference external" href="https://pytorch.org/tutorials/advanced/custom_ops_landing_page.html">PyTorch Custom Operators Landing Page</a>
for more details on how to effectively use these APIs.</p>
<div class="section" id="testing-custom-ops">
<h2>Testing custom ops<a class="headerlink" href="#testing-custom-ops" title="Permalink to this heading">¶</a></h2>
<p>Use <a class="reference internal" href="#torch.library.opcheck" title="torch.library.opcheck"><code class="xref py py-func docutils literal notranslate"><span class="pre">torch.library.opcheck()</span></code></a> to test custom ops for incorrect usage of the
Python torch.library and/or C++ TORCH_LIBRARY APIs. Also, if your operator supports
training, use <a class="reference internal" href="autograd.html#module-torch.autograd.gradcheck" title="torch.autograd.gradcheck"><code class="xref py py-func docutils literal notranslate"><span class="pre">torch.autograd.gradcheck()</span></code></a> to test that the gradients are
mathematically correct.</p>
<dl class="py function">
<dt class="sig sig-object py" id="torch.library.opcheck">
<span class="sig-prename descclassname"><span class="pre">torch.library.</span></span><span class="sig-name descname"><span class="pre">opcheck</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">op</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">args</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">kwargs</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em>, <em class="sig-param"><span class="o"><span class="pre">*</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">test_utils</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">('test_schema',</span> <span class="pre">'test_autograd_registration',</span> <span class="pre">'test_faketensor',</span> <span class="pre">'test_aot_dispatch_dynamic')</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">raise_exception</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">True</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">atol</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">rtol</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/library.html#opcheck"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="reference external" href="https://github.com/pytorch/pytorch/blob/v2.7.0/torch/library.py#L1445"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.library.opcheck" title="Permalink to this definition">¶</a></dt>
<dd><p>Given an operator and some sample arguments, tests if the operator is
registered correctly.</p>
<p>That is, when you use the torch.library/TORCH_LIBRARY APIs to create a
custom op, you specified metadata (e.g. mutability info) about the custom op
and these APIs require that the functions you pass them satisfy certain
properties (e.g. no data pointer access in the fake/meta/abstract kernel)
<code class="docutils literal notranslate"><span class="pre">opcheck</span></code> tests these metadata and properties.</p>
<p>Concretely, we test the following:</p>
<ul class="simple">
<li><p>test_schema: If the schema matches the implementation of
the operator. For example: if the schema specifies a Tensor is mutated,
then we check the implementation mutates the Tensor. If the schema
specifies that we return a new Tensor, then we check that the
implementation returns a new Tensor (instead of an existing one or
a view of an existing one).</p></li>
<li><p>test_autograd_registration: If the operator supports training
(autograd): we check that its autograd formula is registered via
torch.library.register_autograd or a manual registration to one
or more DispatchKey::Autograd keys. Any other DispatchKey-based
registrations may lead to undefined behavior.</p></li>
<li><p>test_faketensor: If the operator has a FakeTensor kernel
(and if it is correct). The FakeTensor kernel is necessary (
but not sufficient) for the operator to work with PyTorch compilation
APIs (torch.compile/export/FX). We check that a FakeTensor kernel
(also sometimes known as a meta kernel) was registered for the
operator and that it is correct. This test takes the result of
running the operator on real tensors and the result of running
the operator on FakeTensors and checks that they have the same
Tensor metadata (sizes/strides/dtype/device/etc).</p></li>
<li><p>test_aot_dispatch_dynamic: If the operator has correct behavior
with PyTorch compilation APIs (torch.compile/export/FX).
This checks that the outputs (and gradients, if applicable) are the
same under eager-mode PyTorch and torch.compile.
This test is a superset of <code class="docutils literal notranslate"><span class="pre">test_faketensor</span></code> and is an e2e test;
other things it tests are that the operator supports
functionalization and that the backward pass (if it exists) also
supports FakeTensor and functionalization.</p></li>
</ul>
<p>For best results, please call <code class="docutils literal notranslate"><span class="pre">opcheck</span></code> multiple times with a
representative set of inputs. If your operator supports
autograd, please use <code class="docutils literal notranslate"><span class="pre">opcheck</span></code> with inputs with <code class="docutils literal notranslate"><span class="pre">requires_grad</span> <span class="pre">=</span> <span class="pre">True</span></code>;
if your operator supports multiple devices (e.g. CPU and CUDA), please
use <code class="docutils literal notranslate"><span class="pre">opcheck</span></code> with inputs on all supported devices.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><ul class="simple">
<li><p><strong>op</strong> (<a class="reference external" href="https://docs.python.org/3/library/typing.html#typing.Union" title="(in Python v3.13)"><em>Union</em></a><em>[</em><em>OpOverload</em><em>, </em><em>OpOverloadPacket</em><em>, </em><a class="reference internal" href="#torch._library.custom_ops.CustomOpDef" title="torch._library.custom_ops.CustomOpDef"><em>CustomOpDef</em></a><em>]</em>) – The operator. Must either be a function decorated with
<a class="reference internal" href="#torch.library.custom_op" title="torch.library.custom_op"><code class="xref py py-func docutils literal notranslate"><span class="pre">torch.library.custom_op()</span></code></a> or an OpOverload/OpOverloadPacket
found in torch.ops.* (e.g. torch.ops.aten.sin, torch.ops.mylib.foo)</p></li>
<li><p><strong>args</strong> (<a class="reference external" href="https://docs.python.org/3/library/stdtypes.html#tuple" title="(in Python v3.13)"><em>tuple</em></a><em>[</em><a class="reference external" href="https://docs.python.org/3/library/typing.html#typing.Any" title="(in Python v3.13)"><em>Any</em></a><em>, </em><em>...</em><em>]</em>) – The args to the operator</p></li>
<li><p><strong>kwargs</strong> (<a class="reference external" href="https://docs.python.org/3/library/typing.html#typing.Optional" title="(in Python v3.13)"><em>Optional</em></a><em>[</em><a class="reference external" href="https://docs.python.org/3/library/stdtypes.html#dict" title="(in Python v3.13)"><em>dict</em></a><em>[</em><a class="reference external" href="https://docs.python.org/3/library/stdtypes.html#str" title="(in Python v3.13)"><em>str</em></a><em>, </em><a class="reference external" href="https://docs.python.org/3/library/typing.html#typing.Any" title="(in Python v3.13)"><em>Any</em></a><em>]</em><em>]</em>) – The kwargs to the operator</p></li>
<li><p><strong>test_utils</strong> (<a class="reference external" href="https://docs.python.org/3/library/typing.html#typing.Union" title="(in Python v3.13)"><em>Union</em></a><em>[</em><a class="reference external" href="https://docs.python.org/3/library/stdtypes.html#str" title="(in Python v3.13)"><em>str</em></a><em>, </em><a class="reference external" href="https://docs.python.org/3/library/collections.abc.html#collections.abc.Sequence" title="(in Python v3.13)"><em>Sequence</em></a><em>[</em><a class="reference external" href="https://docs.python.org/3/library/stdtypes.html#str" title="(in Python v3.13)"><em>str</em></a><em>]</em><em>]</em>) – Tests that we should run. Default: all of them.
Example: (“test_schema”, “test_faketensor”)</p></li>
<li><p><strong>raise_exception</strong> (<a class="reference external" href="https://docs.python.org/3/library/functions.html#bool" title="(in Python v3.13)"><em>bool</em></a>) – If we should raise an exception on the first
error. If False, we will return a dict with information
on if each test passed or not.</p></li>
<li><p><strong>rtol</strong> (<em>Optional</em><em>[</em><a class="reference external" href="https://docs.python.org/3/library/functions.html#float" title="(in Python v3.13)"><em>float</em></a><em>]</em>) – Relative tolerance for floating point comparisons.
If specified <code class="docutils literal notranslate"><span class="pre">atol</span></code> must also be specified.
If omitted, default values based on the <code class="docutils literal notranslate"><span class="pre">dtype</span></code> are selected
(see the table in <a class="reference internal" href="testing.html#torch.testing.assert_close" title="torch.testing.assert_close"><code class="xref py py-func docutils literal notranslate"><span class="pre">torch.testing.assert_close()</span></code></a>).</p></li>
<li><p><strong>atol</strong> (<em>Optional</em><em>[</em><a class="reference external" href="https://docs.python.org/3/library/functions.html#float" title="(in Python v3.13)"><em>float</em></a><em>]</em>) – Absolute tolerance for floating point comparisons.
If specified <code class="docutils literal notranslate"><span class="pre">rtol</span></code> must also be specified.
If omitted, default values based on the <code class="docutils literal notranslate"><span class="pre">dtype</span></code> are selected
(see the table in <a class="reference internal" href="testing.html#torch.testing.assert_close" title="torch.testing.assert_close"><code class="xref py py-func docutils literal notranslate"><span class="pre">torch.testing.assert_close()</span></code></a>).</p></li>
</ul>
</dd>
<dt class="field-even">Return type</dt>
<dd class="field-even"><p><a class="reference external" href="https://docs.python.org/3/library/stdtypes.html#dict" title="(in Python v3.13)">dict</a>[<a class="reference external" href="https://docs.python.org/3/library/stdtypes.html#str" title="(in Python v3.13)">str</a>, <a class="reference external" href="https://docs.python.org/3/library/stdtypes.html#str" title="(in Python v3.13)">str</a>]</p>
</dd>
</dl>
<div class="admonition warning">
<p class="admonition-title">Warning</p>
<p>opcheck and <a class="reference internal" href="autograd.html#module-torch.autograd.gradcheck" title="torch.autograd.gradcheck"><code class="xref py py-func docutils literal notranslate"><span class="pre">torch.autograd.gradcheck()</span></code></a> test different things;
opcheck tests if your usage of torch.library APIs is correct while
<a class="reference internal" href="autograd.html#module-torch.autograd.gradcheck" title="torch.autograd.gradcheck"><code class="xref py py-func docutils literal notranslate"><span class="pre">torch.autograd.gradcheck()</span></code></a> tests if your autograd formula is
mathematically correct. Use both to test custom ops that support
gradient computation.</p>
</div>
<p class="rubric">Example</p>
<div class="doctest highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="nd">@torch</span><span class="o">.</span><span class="n">library</span><span class="o">.</span><span class="n">custom_op</span><span class="p">(</span><span class="s2">"mylib::numpy_mul"</span><span class="p">,</span> <span class="n">mutates_args</span><span class="o">=</span><span class="p">())</span>
<span class="gp">>>> </span><span class="k">def</span> <span class="nf">numpy_mul</span><span class="p">(</span><span class="n">x</span><span class="p">:</span> <span class="n">Tensor</span><span class="p">,</span> <span class="n">y</span><span class="p">:</span> <span class="nb">float</span><span class="p">)</span> <span class="o">-></span> <span class="n">Tensor</span><span class="p">:</span>
<span class="gp">>>> </span> <span class="n">x_np</span> <span class="o">=</span> <span class="n">x</span><span class="o">.</span><span class="n">numpy</span><span class="p">(</span><span class="n">force</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span>
<span class="gp">>>> </span> <span class="n">z_np</span> <span class="o">=</span> <span class="n">x_np</span> <span class="o">*</span> <span class="n">y</span>
<span class="gp">>>> </span> <span class="k">return</span> <span class="n">torch</span><span class="o">.</span><span class="n">from_numpy</span><span class="p">(</span><span class="n">z_np</span><span class="p">)</span><span class="o">.</span><span class="n">to</span><span class="p">(</span><span class="n">x</span><span class="o">.</span><span class="n">device</span><span class="p">)</span>
<span class="go">>>></span>
<span class="gp">>>> </span><span class="nd">@numpy_mul</span><span class="o">.</span><span class="n">register_fake</span>
<span class="gp">>>> </span><span class="k">def</span> <span class="nf">_</span><span class="p">(</span><span class="n">x</span><span class="p">,</span> <span class="n">y</span><span class="p">):</span>
<span class="gp">>>> </span> <span class="k">return</span> <span class="n">torch</span><span class="o">.</span><span class="n">empty_like</span><span class="p">(</span><span class="n">x</span><span class="p">)</span>
<span class="go">>>></span>
<span class="gp">>>> </span><span class="k">def</span> <span class="nf">setup_context</span><span class="p">(</span><span class="n">ctx</span><span class="p">,</span> <span class="n">inputs</span><span class="p">,</span> <span class="n">output</span><span class="p">):</span>
<span class="gp">>>> </span> <span class="n">y</span><span class="p">,</span> <span class="o">=</span> <span class="n">inputs</span>
<span class="gp">>>> </span> <span class="n">ctx</span><span class="o">.</span><span class="n">y</span> <span class="o">=</span> <span class="n">y</span>
<span class="go">>>></span>
<span class="gp">>>> </span><span class="k">def</span> <span class="nf">backward</span><span class="p">(</span><span class="n">ctx</span><span class="p">,</span> <span class="n">grad</span><span class="p">):</span>
<span class="gp">>>> </span> <span class="k">return</span> <span class="n">grad</span> <span class="o">*</span> <span class="n">ctx</span><span class="o">.</span><span class="n">y</span><span class="p">,</span> <span class="kc">None</span>
<span class="go">>>></span>
<span class="gp">>>> </span><span class="n">numpy_mul</span><span class="o">.</span><span class="n">register_autograd</span><span class="p">(</span><span class="n">backward</span><span class="p">,</span> <span class="n">setup_context</span><span class="o">=</span><span class="n">setup_context</span><span class="p">)</span>
<span class="go">>>></span>
<span class="gp">>>> </span><span class="n">sample_inputs</span> <span class="o">=</span> <span class="p">[</span>
<span class="gp">>>> </span> <span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">randn</span><span class="p">(</span><span class="mi">3</span><span class="p">),</span> <span class="mf">3.14</span><span class="p">),</span>
<span class="gp">>>> </span> <span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">randn</span><span class="p">(</span><span class="mi">2</span><span class="p">,</span> <span class="mi">3</span><span class="p">,</span> <span class="n">device</span><span class="o">=</span><span class="s1">'cuda'</span><span class="p">),</span> <span class="mf">2.718</span><span class="p">),</span>
<span class="gp">>>> </span> <span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">randn</span><span class="p">(</span><span class="mi">1</span><span class="p">,</span> <span class="mi">10</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="mf">1.234</span><span class="p">),</span>
<span class="gp">>>> </span> <span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">randn</span><span class="p">(</span><span class="mi">64</span><span class="p">,</span> <span class="mi">64</span><span class="p">,</span> <span class="n">device</span><span class="o">=</span><span class="s1">'cuda'</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="mf">90.18</span><span class="p">),</span>
<span class="gp">>>> </span><span class="p">]</span>
<span class="go">>>></span>
<span class="gp">>>> </span><span class="k">for</span> <span class="n">args</span> <span class="ow">in</span> <span class="n">sample_inputs</span><span class="p">:</span>
<span class="gp">>>> </span> <span class="n">torch</span><span class="o">.</span><span class="n">library</span><span class="o">.</span><span class="n">opcheck</span><span class="p">(</span><span class="n">numpy_mul</span><span class="p">,</span> <span class="n">args</span><span class="p">)</span>
</pre></div>
</div>
</dd></dl>
</div>
<div class="section" id="creating-new-custom-ops-in-python">
<h2>Creating new custom ops in Python<a class="headerlink" href="#creating-new-custom-ops-in-python" title="Permalink to this heading">¶</a></h2>
<p>Use <a class="reference internal" href="#torch.library.custom_op" title="torch.library.custom_op"><code class="xref py py-func docutils literal notranslate"><span class="pre">torch.library.custom_op()</span></code></a> to create new custom ops.</p>
<dl class="py function">
<dt class="sig sig-object py" id="torch.library.custom_op">
<span class="sig-prename descclassname"><span class="pre">torch.library.</span></span><span class="sig-name descname"><span class="pre">custom_op</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">name</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">fn</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em>, <em class="sig-param"><span class="o"><span class="pre">/</span></span></em>, <em class="sig-param"><span class="o"><span class="pre">*</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">mutates_args</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">device_types</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">schema</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em><span class="sig-paren">)</span><a class="reference external" href="https://github.com/pytorch/pytorch/blob/v2.7.0/torch/_library/custom_ops.py#L48"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.library.custom_op" title="Permalink to this definition">¶</a></dt>
<dd><p>Wraps a function into custom operator.</p>
<p>Reasons why you may want to create a custom op include:
- Wrapping a third-party library or custom kernel to work with PyTorch
subsystems like Autograd.
- Preventing torch.compile/export/FX tracing from peeking inside your function.</p>
<p>This API is used as a decorator around a function (please see examples).
The provided function must have type hints; these are needed to interface
with PyTorch’s various subsystems.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><ul class="simple">
<li><p><strong>name</strong> (<a class="reference external" href="https://docs.python.org/3/library/stdtypes.html#str" title="(in Python v3.13)"><em>str</em></a>) – A name for the custom op that looks like “{namespace}::{name}”,
e.g. “mylib::my_linear”. The name is used as the op’s stable identifier
in PyTorch subsystems (e.g. torch.export, FX graphs).
To avoid name collisions, please use your project name as the namespace;
e.g. all custom ops in pytorch/fbgemm use “fbgemm” as the namespace.</p></li>
<li><p><strong>mutates_args</strong> (<em>Iterable</em><em>[</em><a class="reference external" href="https://docs.python.org/3/library/stdtypes.html#str" title="(in Python v3.13)"><em>str</em></a><em>] or </em><em>"unknown"</em>) – The names of args that the function mutates.
This MUST be accurate, otherwise, the behavior is undefined. If “unknown”,
it pessimistically assumes that all inputs to the operator are being mutated.</p></li>
<li><p><strong>device_types</strong> (<em>None</em><em> | </em><a class="reference external" href="https://docs.python.org/3/library/stdtypes.html#str" title="(in Python v3.13)"><em>str</em></a><em> | </em><em>Sequence</em><em>[</em><a class="reference external" href="https://docs.python.org/3/library/stdtypes.html#str" title="(in Python v3.13)"><em>str</em></a><em>]</em>) – The device type(s) the function
is valid for. If no device type is provided, then the function
is used as the default implementation for all device types.
Examples: “cpu”, “cuda”.
When registering a device-specific implementation for an operator that accepts no Tensors,
we require the operator to have a “device: torch.device argument”.</p></li>
<li><p><strong>schema</strong> (<em>None</em><em> | </em><a class="reference external" href="https://docs.python.org/3/library/stdtypes.html#str" title="(in Python v3.13)"><em>str</em></a>) – A schema string for the operator. If None
(recommended) we’ll infer a schema for the operator from its type
annotations. We recommend letting us infer a schema unless you
have a specific reason not to.
Example: “(Tensor x, int y) -> (Tensor, Tensor)”.</p></li>
</ul>
</dd>
<dt class="field-even">Return type</dt>
<dd class="field-even"><p><a class="reference external" href="https://docs.python.org/3/library/typing.html#typing.Union" title="(in Python v3.13)"><em>Union</em></a>[<a class="reference external" href="https://docs.python.org/3/library/typing.html#typing.Callable" title="(in Python v3.13)"><em>Callable</em></a>[[<a class="reference external" href="https://docs.python.org/3/library/typing.html#typing.Callable" title="(in Python v3.13)"><em>Callable</em></a>[[…], <a class="reference external" href="https://docs.python.org/3/library/functions.html#object" title="(in Python v3.13)">object</a>]], <a class="reference internal" href="#torch._library.custom_ops.CustomOpDef" title="torch._library.custom_ops.CustomOpDef"><em>CustomOpDef</em></a>], <a class="reference internal" href="#torch._library.custom_ops.CustomOpDef" title="torch._library.custom_ops.CustomOpDef"><em>CustomOpDef</em></a>]</p>
</dd>
</dl>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>We recommend not passing in a <code class="docutils literal notranslate"><span class="pre">schema</span></code> arg and instead letting us infer
it from the type annotations. It is error-prone to write your own schema.
You may wish to provide your own schema if our interpretation of
the type annotation is not what you want.
For more info on how to write a schema string, see
<a class="reference external" href="https://github.com/pytorch/pytorch/blob/main/aten/src/ATen/native/README.md#func">here</a></p>
</div>
<dl>
<dt>Examples::</dt><dd><div class="doctest highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="kn">import</span> <span class="nn">torch</span>
<span class="gp">>>> </span><span class="kn">from</span> <span class="nn">torch</span> <span class="kn">import</span> <span class="n">Tensor</span>
<span class="gp">>>> </span><span class="kn">from</span> <span class="nn">torch.library</span> <span class="kn">import</span> <span class="n">custom_op</span>
<span class="gp">>>> </span><span class="kn">import</span> <span class="nn">numpy</span> <span class="k">as</span> <span class="nn">np</span>
<span class="go">>>></span>
<span class="gp">>>> </span><span class="nd">@custom_op</span><span class="p">(</span><span class="s2">"mylib::numpy_sin"</span><span class="p">,</span> <span class="n">mutates_args</span><span class="o">=</span><span class="p">())</span>
<span class="gp">>>> </span><span class="k">def</span> <span class="nf">numpy_sin</span><span class="p">(</span><span class="n">x</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="gp">>>> </span> <span class="n">x_np</span> <span class="o">=</span> <span class="n">x</span><span class="o">.</span><span class="n">cpu</span><span class="p">()</span><span class="o">.</span><span class="n">numpy</span><span class="p">()</span>
<span class="gp">>>> </span> <span class="n">y_np</span> <span class="o">=</span> <span class="n">np</span><span class="o">.</span><span class="n">sin</span><span class="p">(</span><span class="n">x_np</span><span class="p">)</span>
<span class="gp">>>> </span> <span class="k">return</span> <span class="n">torch</span><span class="o">.</span><span class="n">from_numpy</span><span class="p">(</span><span class="n">y_np</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">x</span><span class="o">.</span><span class="n">device</span><span class="p">)</span>
<span class="go">>>></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="gp">>>> </span><span class="n">y</span> <span class="o">=</span> <span class="n">numpy_sin</span><span class="p">(</span><span class="n">x</span><span class="p">)</span>
<span class="gp">>>> </span><span class="k">assert</span> <span class="n">torch</span><span class="o">.</span><span class="n">allclose</span><span class="p">(</span><span class="n">y</span><span class="p">,</span> <span class="n">x</span><span class="o">.</span><span class="n">sin</span><span class="p">())</span>
<span class="go">>>></span>
<span class="gp">>>> </span><span class="c1"># Example of a custom op that only works for one device type.</span>
<span class="gp">>>> </span><span class="nd">@custom_op</span><span class="p">(</span><span class="s2">"mylib::numpy_sin_cpu"</span><span class="p">,</span> <span class="n">mutates_args</span><span class="o">=</span><span class="p">(),</span> <span class="n">device_types</span><span class="o">=</span><span class="s2">"cpu"</span><span class="p">)</span>
<span class="gp">>>> </span><span class="k">def</span> <span class="nf">numpy_sin_cpu</span><span class="p">(</span><span class="n">x</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="gp">>>> </span> <span class="n">x_np</span> <span class="o">=</span> <span class="n">x</span><span class="o">.</span><span class="n">numpy</span><span class="p">()</span>
<span class="gp">>>> </span> <span class="n">y_np</span> <span class="o">=</span> <span class="n">np</span><span class="o">.</span><span class="n">sin</span><span class="p">(</span><span class="n">x_np</span><span class="p">)</span>
<span class="gp">>>> </span> <span class="k">return</span> <span class="n">torch</span><span class="o">.</span><span class="n">from_numpy</span><span class="p">(</span><span class="n">y_np</span><span class="p">)</span>
<span class="go">>>></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="gp">>>> </span><span class="n">y</span> <span class="o">=</span> <span class="n">numpy_sin_cpu</span><span class="p">(</span><span class="n">x</span><span class="p">)</span>
<span class="gp">>>> </span><span class="k">assert</span> <span class="n">torch</span><span class="o">.</span><span class="n">allclose</span><span class="p">(</span><span class="n">y</span><span class="p">,</span> <span class="n">x</span><span class="o">.</span><span class="n">sin</span><span class="p">())</span>
<span class="go">>>></span>
<span class="gp">>>> </span><span class="c1"># Example of a custom op that mutates an input</span>
<span class="gp">>>> </span><span class="nd">@custom_op</span><span class="p">(</span><span class="s2">"mylib::numpy_sin_inplace"</span><span class="p">,</span> <span class="n">mutates_args</span><span class="o">=</span><span class="p">{</span><span class="s2">"x"</span><span class="p">},</span> <span class="n">device_types</span><span class="o">=</span><span class="s2">"cpu"</span><span class="p">)</span>
<span class="gp">>>> </span><span class="k">def</span> <span class="nf">numpy_sin_inplace</span><span class="p">(</span><span class="n">x</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="gp">>>> </span> <span class="n">x_np</span> <span class="o">=</span> <span class="n">x</span><span class="o">.</span><span class="n">numpy</span><span class="p">()</span>
<span class="gp">>>> </span> <span class="n">np</span><span class="o">.</span><span class="n">sin</span><span class="p">(</span><span class="n">x_np</span><span class="p">,</span> <span class="n">out</span><span class="o">=</span><span class="n">x_np</span><span class="p">)</span>
<span class="go">>>></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="gp">>>> </span><span class="n">expected</span> <span class="o">=</span> <span class="n">x</span><span class="o">.</span><span class="n">sin</span><span class="p">()</span>
<span class="gp">>>> </span><span class="n">numpy_sin_inplace</span><span class="p">(</span><span class="n">x</span><span class="p">)</span>
<span class="gp">>>> </span><span class="k">assert</span> <span class="n">torch</span><span class="o">.</span><span class="n">allclose</span><span class="p">(</span><span class="n">x</span><span class="p">,</span> <span class="n">expected</span><span class="p">)</span>
<span class="go">>>></span>
<span class="gp">>>> </span><span class="c1"># Example of a factory function</span>
<span class="gp">>>> </span><span class="nd">@torch</span><span class="o">.</span><span class="n">library</span><span class="o">.</span><span class="n">custom_op</span><span class="p">(</span><span class="s2">"mylib::bar"</span><span class="p">,</span> <span class="n">mutates_args</span><span class="o">=</span><span class="p">{},</span> <span class="n">device_types</span><span class="o">=</span><span class="s2">"cpu"</span><span class="p">)</span>
<span class="gp">>>> </span><span class="k">def</span> <span class="nf">bar</span><span class="p">(</span><span class="n">device</span><span class="p">:</span> <span class="n">torch</span><span class="o">.</span><span class="n">device</span><span class="p">)</span> <span class="o">-></span> <span class="n">Tensor</span><span class="p">:</span>
<span class="gp">>>> </span> <span class="k">return</span> <span class="n">torch</span><span class="o">.</span><span class="n">ones</span><span class="p">(</span><span class="mi">3</span><span class="p">)</span>
<span class="go">>>></span>
<span class="gp">>>> </span><span class="n">bar</span><span class="p">(</span><span class="s2">"cpu"</span><span class="p">)</span>
</pre></div>
</div>
</dd>
</dl>
</dd></dl>
<dl class="py function">
<dt class="sig sig-object py" id="torch.library.triton_op">
<span class="sig-prename descclassname"><span class="pre">torch.library.</span></span><span class="sig-name descname"><span class="pre">triton_op</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">name</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">fn</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em>, <em class="sig-param"><span class="o"><span class="pre">/</span></span></em>, <em class="sig-param"><span class="o"><span class="pre">*</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">mutates_args</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">schema</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em><span class="sig-paren">)</span><a class="reference external" href="https://github.com/pytorch/pytorch/blob/v2.7.0/torch/_library/triton.py#L12"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.library.triton_op" title="Permalink to this definition">¶</a></dt>
<dd><p>Create a custom operator whose implementation is backed by 1+ triton kernels.</p>
<p>This is a more structured way of using triton kernels with PyTorch.
Prefer using triton kernels with no <code class="docutils literal notranslate"><span class="pre">torch.library</span></code> custom operator wrappers
(like <a class="reference internal" href="#torch.library.custom_op" title="torch.library.custom_op"><code class="xref py py-func docutils literal notranslate"><span class="pre">torch.library.custom_op()</span></code></a>, <a class="reference internal" href="#torch.library.triton_op" title="torch.library.triton_op"><code class="xref py py-func docutils literal notranslate"><span class="pre">torch.library.triton_op()</span></code></a>) because
that is simpler;
only use <a class="reference internal" href="#torch.library.custom_op" title="torch.library.custom_op"><code class="xref py py-func docutils literal notranslate"><span class="pre">torch.library.custom_op()</span></code></a>/<a class="reference internal" href="#torch.library.triton_op" title="torch.library.triton_op"><code class="xref py py-func docutils literal notranslate"><span class="pre">torch.library.triton_op()</span></code></a> if you
want to create an operator that behaves like PyTorch built-in operators.
For example, you may use a <code class="docutils literal notranslate"><span class="pre">torch.library</span></code> wrapper API to define the
behavior of the triton kernel when passed a tensor subclass or under
a TorchDispatchMode.</p>
<p>Use <a class="reference internal" href="#torch.library.triton_op" title="torch.library.triton_op"><code class="xref py py-func docutils literal notranslate"><span class="pre">torch.library.triton_op()</span></code></a> instead of <a class="reference internal" href="#torch.library.custom_op" title="torch.library.custom_op"><code class="xref py py-func docutils literal notranslate"><span class="pre">torch.library.custom_op()</span></code></a>
when the implementation
consists of 1+ triton kernels. <a class="reference internal" href="#torch.library.custom_op" title="torch.library.custom_op"><code class="xref py py-func docutils literal notranslate"><span class="pre">torch.library.custom_op()</span></code></a> treats
custom operators as opaque (<a class="reference internal" href="generated/torch.compile.html#torch.compile" title="torch.compile"><code class="xref py py-func docutils literal notranslate"><span class="pre">torch.compile()</span></code></a> and
<a class="reference internal" href="export.html#torch.export.export" title="torch.export.export"><code class="xref py py-func docutils literal notranslate"><span class="pre">torch.export.export()</span></code></a> will never trace into them), but <code class="docutils literal notranslate"><span class="pre">triton_op</span></code>
makes the implementation visible to these subsystems, allowing them
to optimize the triton kernel(s).</p>
<p>Note that <code class="docutils literal notranslate"><span class="pre">fn</span></code> must only consist of calls to PyTorch-understood
operators and triton kernels. Any triton kernels called inside <code class="docutils literal notranslate"><span class="pre">fn</span></code>
must be wrapped in a call to <a class="reference internal" href="#torch.library.wrap_triton" title="torch.library.wrap_triton"><code class="xref py py-func docutils literal notranslate"><span class="pre">torch.library.wrap_triton()</span></code></a>.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><ul class="simple">
<li><p><strong>name</strong> (<a class="reference external" href="https://docs.python.org/3/library/stdtypes.html#str" title="(in Python v3.13)"><em>str</em></a>) – A name for the custom op that looks like “{namespace}::{name}”,
e.g. “mylib::my_linear”. The name is used as the op’s stable identifier
in PyTorch subsystems (e.g. torch.export, FX graphs).
To avoid name collisions, please use your project name as the namespace;
e.g. all custom ops in pytorch/fbgemm use “fbgemm” as the namespace.</p></li>
<li><p><strong>mutates_args</strong> (<em>Iterable</em><em>[</em><a class="reference external" href="https://docs.python.org/3/library/stdtypes.html#str" title="(in Python v3.13)"><em>str</em></a><em>] or </em><em>"unknown"</em>) – The names of args that the function mutates.
This MUST be accurate, otherwise, the behavior is undefined. If “unknown”,
it pessimistically assumes that all inputs to the operator are being mutated.</p></li>
<li><p><strong>schema</strong> (<em>None</em><em> | </em><a class="reference external" href="https://docs.python.org/3/library/stdtypes.html#str" title="(in Python v3.13)"><em>str</em></a>) – A schema string for the operator. If None
(recommended) we’ll infer a schema for the operator from its type
annotations. We recommend letting us infer a schema unless you
have a specific reason not to.
Example: “(Tensor x, int y) -> (Tensor, Tensor)”.</p></li>
</ul>
</dd>
<dt class="field-even">Return type</dt>
<dd class="field-even"><p><a class="reference external" href="https://docs.python.org/3/library/typing.html#typing.Callable" title="(in Python v3.13)"><em>Callable</em></a></p>
</dd>
</dl>
<p>Example:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="kn">import</span> <span class="nn">torch</span>
<span class="gp">>>> </span><span class="kn">from</span> <span class="nn">torch.library</span> <span class="kn">import</span> <span class="n">triton_op</span><span class="p">,</span> <span class="n">wrap_triton</span>
<span class="go">>>></span>
<span class="gp">>>> </span><span class="kn">import</span> <span class="nn">triton</span>
<span class="gp">>>> </span><span class="kn">from</span> <span class="nn">triton</span> <span class="kn">import</span> <span class="n">language</span> <span class="k">as</span> <span class="n">tl</span>
<span class="go">>>></span>
<span class="gp">>>> </span><span class="nd">@triton</span><span class="o">.</span><span class="n">jit</span>
<span class="gp">>>> </span><span class="k">def</span> <span class="nf">add_kernel</span><span class="p">(</span>
<span class="gp">>>> </span> <span class="n">in_ptr0</span><span class="p">,</span>
<span class="gp">>>> </span> <span class="n">in_ptr1</span><span class="p">,</span>
<span class="gp">>>> </span> <span class="n">out_ptr</span><span class="p">,</span>
<span class="gp">>>> </span> <span class="n">n_elements</span><span class="p">,</span>
<span class="gp">>>> </span> <span class="n">BLOCK_SIZE</span><span class="p">:</span> <span class="s2">"tl.constexpr"</span><span class="p">,</span>
<span class="gp">>>> </span><span class="p">):</span>
<span class="gp">>>> </span> <span class="n">pid</span> <span class="o">=</span> <span class="n">tl</span><span class="o">.</span><span class="n">program_id</span><span class="p">(</span><span class="n">axis</span><span class="o">=</span><span class="mi">0</span><span class="p">)</span>
<span class="gp">>>> </span> <span class="n">block_start</span> <span class="o">=</span> <span class="n">pid</span> <span class="o">*</span> <span class="n">BLOCK_SIZE</span>
<span class="gp">>>> </span> <span class="n">offsets</span> <span class="o">=</span> <span class="n">block_start</span> <span class="o">+</span> <span class="n">tl</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">BLOCK_SIZE</span><span class="p">)</span>
<span class="gp">>>> </span> <span class="n">mask</span> <span class="o">=</span> <span class="n">offsets</span> <span class="o"><</span> <span class="n">n_elements</span>
<span class="gp">>>> </span> <span class="n">x</span> <span class="o">=</span> <span class="n">tl</span><span class="o">.</span><span class="n">load</span><span class="p">(</span><span class="n">in_ptr0</span> <span class="o">+</span> <span class="n">offsets</span><span class="p">,</span> <span class="n">mask</span><span class="o">=</span><span class="n">mask</span><span class="p">)</span>
<span class="gp">>>> </span> <span class="n">y</span> <span class="o">=</span> <span class="n">tl</span><span class="o">.</span><span class="n">load</span><span class="p">(</span><span class="n">in_ptr1</span> <span class="o">+</span> <span class="n">offsets</span><span class="p">,</span> <span class="n">mask</span><span class="o">=</span><span class="n">mask</span><span class="p">)</span>
<span class="gp">>>> </span> <span class="n">output</span> <span class="o">=</span> <span class="n">x</span> <span class="o">+</span> <span class="n">y</span>
<span class="gp">>>> </span> <span class="n">tl</span><span class="o">.</span><span class="n">store</span><span class="p">(</span><span class="n">out_ptr</span> <span class="o">+</span> <span class="n">offsets</span><span class="p">,</span> <span class="n">output</span><span class="p">,</span> <span class="n">mask</span><span class="o">=</span><span class="n">mask</span><span class="p">)</span>
<span class="go">>>></span>
<span class="gp">>>> </span><span class="nd">@triton_op</span><span class="p">(</span><span class="s2">"mylib::add"</span><span class="p">,</span> <span class="n">mutates_args</span><span class="o">=</span><span class="p">{})</span>
<span class="gp">>>> </span><span class="k">def</span> <span class="nf">add</span><span class="p">(</span><span class="n">x</span><span class="p">:</span> <span class="n">torch</span><span class="o">.</span><span class="n">Tensor</span><span class="p">,</span> <span class="n">y</span><span class="p">:</span> <span class="n">torch</span><span class="o">.</span><span class="n">Tensor</span><span class="p">)</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="gp">>>> </span> <span class="n">output</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">empty_like</span><span class="p">(</span><span class="n">x</span><span class="p">)</span>
<span class="gp">>>> </span> <span class="n">n_elements</span> <span class="o">=</span> <span class="n">output</span><span class="o">.</span><span class="n">numel</span><span class="p">()</span>
<span class="go">>>></span>
<span class="gp">>>> </span> <span class="k">def</span> <span class="nf">grid</span><span class="p">(</span><span class="n">meta</span><span class="p">):</span>
<span class="gp">>>> </span> <span class="k">return</span> <span class="p">(</span><span class="n">triton</span><span class="o">.</span><span class="n">cdiv</span><span class="p">(</span><span class="n">n_elements</span><span class="p">,</span> <span class="n">meta</span><span class="p">[</span><span class="s2">"BLOCK_SIZE"</span><span class="p">]),)</span>
<span class="go">>>></span>
<span class="gp">>>> </span> <span class="c1"># NB: we need to wrap the triton kernel in a call to wrap_triton</span>
<span class="gp">>>> </span> <span class="n">wrap_triton</span><span class="p">(</span><span class="n">add_kernel</span><span class="p">)[</span><span class="n">grid</span><span class="p">](</span><span class="n">x</span><span class="p">,</span> <span class="n">y</span><span class="p">,</span> <span class="n">output</span><span class="p">,</span> <span class="n">n_elements</span><span class="p">,</span> <span class="mi">16</span><span class="p">)</span>
<span class="gp">>>> </span> <span class="k">return</span> <span class="n">output</span>
<span class="go">>>></span>
<span class="gp">>>> </span><span class="nd">@torch</span><span class="o">.</span><span class="n">compile</span>
<span class="gp">>>> </span><span class="k">def</span> <span class="nf">f</span><span class="p">(</span><span class="n">x</span><span class="p">,</span> <span class="n">y</span><span class="p">):</span>
<span class="gp">>>> </span> <span class="k">return</span> <span class="n">add</span><span class="p">(</span><span class="n">x</span><span class="p">,</span> <span class="n">y</span><span class="p">)</span>
<span class="go">>>></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">device</span><span class="o">=</span><span class="s2">"cuda"</span><span class="p">)</span>
<span class="gp">>>> </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">device</span><span class="o">=</span><span class="s2">"cuda"</span><span class="p">)</span>
<span class="go">>>></span>
<span class="gp">>>> </span><span class="n">z</span> <span class="o">=</span> <span class="n">f</span><span class="p">(</span><span class="n">x</span><span class="p">,</span> <span class="n">y</span><span class="p">)</span>
<span class="gp">>>> </span><span class="k">assert</span> <span class="n">torch</span><span class="o">.</span><span class="n">allclose</span><span class="p">(</span><span class="n">z</span><span class="p">,</span> <span class="n">x</span> <span class="o">+</span> <span class="n">y</span><span class="p">)</span>
</pre></div>
</div>
</dd></dl>
<dl class="py function">
<dt class="sig sig-object py" id="torch.library.wrap_triton">
<span class="sig-prename descclassname"><span class="pre">torch.library.</span></span><span class="sig-name descname"><span class="pre">wrap_triton</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">triton_kernel</span></span></em>, <em class="sig-param"><span class="o"><span class="pre">/</span></span></em><span class="sig-paren">)</span><a class="reference external" href="https://github.com/pytorch/pytorch/blob/v2.7.0/torch/_library/triton.py#L200"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.library.wrap_triton" title="Permalink to this definition">¶</a></dt>
<dd><p>Allows capture of a triton kernel into a graph via make_fx or
non-strict <code class="docutils literal notranslate"><span class="pre">torch.export</span></code>.</p>
<p>These technologies perform Dispatcher-based tracing (via
<code class="docutils literal notranslate"><span class="pre">__torch_dispatch__</span></code>) and cannot see calls to raw triton kernels.
The <code class="docutils literal notranslate"><span class="pre">wrap_triton</span></code> API wraps a triton kernel into a callable that
can actually be traced into a graph.</p>
<p>Please use this API together with <a class="reference internal" href="#torch.library.triton_op" title="torch.library.triton_op"><code class="xref py py-func docutils literal notranslate"><span class="pre">torch.library.triton_op()</span></code></a>.</p>
<p class="rubric">Examples</p>
<div class="doctest highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="kn">import</span> <span class="nn">torch</span>
<span class="gp">>>> </span><span class="kn">import</span> <span class="nn">triton</span>
<span class="gp">>>> </span><span class="kn">from</span> <span class="nn">triton</span> <span class="kn">import</span> <span class="n">language</span> <span class="k">as</span> <span class="n">tl</span>
<span class="gp">>>> </span><span class="kn">from</span> <span class="nn">torch.fx.experimental.proxy_tensor</span> <span class="kn">import</span> <span class="n">make_fx</span>
<span class="gp">>>> </span><span class="kn">from</span> <span class="nn">torch.library</span> <span class="kn">import</span> <span class="n">wrap_triton</span>
<span class="go">>>></span>
<span class="gp">>>> </span><span class="nd">@triton</span><span class="o">.</span><span class="n">jit</span>
<span class="gp">>>> </span><span class="k">def</span> <span class="nf">add_kernel</span><span class="p">(</span>
<span class="gp">>>> </span> <span class="n">in_ptr0</span><span class="p">,</span>
<span class="gp">>>> </span> <span class="n">in_ptr1</span><span class="p">,</span>
<span class="gp">>>> </span> <span class="n">out_ptr</span><span class="p">,</span>
<span class="gp">>>> </span> <span class="n">n_elements</span><span class="p">,</span>
<span class="gp">>>> </span> <span class="n">BLOCK_SIZE</span><span class="p">:</span> <span class="s2">"tl.constexpr"</span><span class="p">,</span>
<span class="gp">>>> </span><span class="p">):</span>
<span class="gp">>>> </span> <span class="n">pid</span> <span class="o">=</span> <span class="n">tl</span><span class="o">.</span><span class="n">program_id</span><span class="p">(</span><span class="n">axis</span><span class="o">=</span><span class="mi">0</span><span class="p">)</span>
<span class="gp">>>> </span> <span class="n">block_start</span> <span class="o">=</span> <span class="n">pid</span> <span class="o">*</span> <span class="n">BLOCK_SIZE</span>
<span class="gp">>>> </span> <span class="n">offsets</span> <span class="o">=</span> <span class="n">block_start</span> <span class="o">+</span> <span class="n">tl</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">BLOCK_SIZE</span><span class="p">)</span>
<span class="gp">>>> </span> <span class="n">mask</span> <span class="o">=</span> <span class="n">offsets</span> <span class="o"><</span> <span class="n">n_elements</span>
<span class="gp">>>> </span> <span class="n">x</span> <span class="o">=</span> <span class="n">tl</span><span class="o">.</span><span class="n">load</span><span class="p">(</span><span class="n">in_ptr0</span> <span class="o">+</span> <span class="n">offsets</span><span class="p">,</span> <span class="n">mask</span><span class="o">=</span><span class="n">mask</span><span class="p">)</span>
<span class="gp">>>> </span> <span class="n">y</span> <span class="o">=</span> <span class="n">tl</span><span class="o">.</span><span class="n">load</span><span class="p">(</span><span class="n">in_ptr1</span> <span class="o">+</span> <span class="n">offsets</span><span class="p">,</span> <span class="n">mask</span><span class="o">=</span><span class="n">mask</span><span class="p">)</span>
<span class="gp">>>> </span> <span class="n">output</span> <span class="o">=</span> <span class="n">x</span> <span class="o">+</span> <span class="n">y</span>
<span class="gp">>>> </span> <span class="n">tl</span><span class="o">.</span><span class="n">store</span><span class="p">(</span><span class="n">out_ptr</span> <span class="o">+</span> <span class="n">offsets</span><span class="p">,</span> <span class="n">output</span><span class="p">,</span> <span class="n">mask</span><span class="o">=</span><span class="n">mask</span><span class="p">)</span>
<span class="go">>>></span>
<span class="gp">>>> </span><span class="k">def</span> <span class="nf">add</span><span class="p">(</span><span class="n">x</span><span class="p">,</span> <span class="n">y</span><span class="p">):</span>
<span class="gp">>>> </span> <span class="n">output</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">empty_like</span><span class="p">(</span><span class="n">x</span><span class="p">)</span>
<span class="gp">>>> </span> <span class="n">n_elements</span> <span class="o">=</span> <span class="n">output</span><span class="o">.</span><span class="n">numel</span><span class="p">()</span>
<span class="go">>>></span>
<span class="gp">>>> </span> <span class="k">def</span> <span class="nf">grid_fn</span><span class="p">(</span><span class="n">meta</span><span class="p">):</span>
<span class="gp">>>> </span> <span class="k">return</span> <span class="p">(</span><span class="n">triton</span><span class="o">.</span><span class="n">cdiv</span><span class="p">(</span><span class="n">n_elements</span><span class="p">,</span> <span class="n">meta</span><span class="p">[</span><span class="s2">"BLOCK_SIZE"</span><span class="p">]),)</span>
<span class="go">>>></span>
<span class="gp">>>> </span> <span class="n">wrap_triton</span><span class="p">(</span><span class="n">add_kernel</span><span class="p">)[</span><span class="n">grid_fn</span><span class="p">](</span><span class="n">x</span><span class="p">,</span> <span class="n">y</span><span class="p">,</span> <span class="n">output</span><span class="p">,</span> <span class="n">n_elements</span><span class="p">,</span> <span class="mi">16</span><span class="p">)</span>
<span class="gp">>>> </span> <span class="k">return</span> <span class="n">output</span>
<span class="go">>>></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">device</span><span class="o">=</span><span class="s2">"cuda"</span><span class="p">)</span>
<span class="gp">>>> </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">device</span><span class="o">=</span><span class="s2">"cuda"</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">gm</span> <span class="o">=</span> <span class="n">make_fx</span><span class="p">(</span><span class="n">add</span><span class="p">)(</span><span class="n">x</span><span class="p">,</span> <span class="n">y</span><span class="p">)</span>
<span class="gp">>>> </span><span class="nb">print</span><span class="p">(</span><span class="n">gm</span><span class="o">.</span><span class="n">code</span><span class="p">)</span>
<span class="gp">>>> </span><span class="c1"># def forward(self, x_1, y_1):</span>
<span class="gp">>>> </span><span class="c1"># empty_like = torch.ops.aten.empty_like.default(x_1, pin_memory = False)</span>
<span class="gp">>>> </span><span class="c1"># triton_kernel_wrapper_mutation_proxy = triton_kernel_wrapper_mutation(</span>
<span class="gp">>>> </span><span class="c1"># kernel_idx = 0, constant_args_idx = 0,</span>
<span class="gp">>>> </span><span class="c1"># grid = [(1, 1, 1)], kwargs = {</span>
<span class="gp">>>> </span><span class="c1"># 'in_ptr0': x_1, 'in_ptr1': y_1, 'out_ptr': empty_like,</span>
<span class="gp">>>> </span><span class="c1"># 'n_elements': 3, 'BLOCK_SIZE': 16</span>
<span class="gp">>>> </span><span class="c1"># })</span>
<span class="gp">>>> </span><span class="c1"># return empty_like</span>
</pre></div>
</div>
<dl class="field-list simple">
<dt class="field-odd">Return type</dt>
<dd class="field-odd"><p><a class="reference external" href="https://docs.python.org/3/library/typing.html#typing.Any" title="(in Python v3.13)"><em>Any</em></a></p>
</dd>
</dl>
</dd></dl>
</div>
<div class="section" id="extending-custom-ops-created-from-python-or-c">
<h2>Extending custom ops (created from Python or C++)<a class="headerlink" href="#extending-custom-ops-created-from-python-or-c" title="Permalink to this heading">¶</a></h2>
<p>Use the register.* methods, such as <a class="reference internal" href="#torch.library.register_kernel" title="torch.library.register_kernel"><code class="xref py py-func docutils literal notranslate"><span class="pre">torch.library.register_kernel()</span></code></a> and
<a class="reference internal" href="#torch.library.register_fake" title="torch.library.register_fake"><code class="xref py py-func docutils literal notranslate"><span class="pre">torch.library.register_fake()</span></code></a>, to add implementations
for any operators (they may have been created using <a class="reference internal" href="#torch.library.custom_op" title="torch.library.custom_op"><code class="xref py py-func docutils literal notranslate"><span class="pre">torch.library.custom_op()</span></code></a> or
via PyTorch’s C++ operator registration APIs).</p>
<dl class="py function">
<dt class="sig sig-object py" id="torch.library.register_kernel">
<span class="sig-prename descclassname"><span class="pre">torch.library.</span></span><span class="sig-name descname"><span class="pre">register_kernel</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">op</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">device_types</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">func</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em>, <em class="sig-param"><span class="o"><span class="pre">/</span></span></em>, <em class="sig-param"><span class="o"><span class="pre">*</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">lib</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/library.html#register_kernel"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="reference external" href="https://github.com/pytorch/pytorch/blob/v2.7.0/torch/library.py#L763"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.library.register_kernel" title="Permalink to this definition">¶</a></dt>
<dd><p>Register an implementation for a device type for this operator.</p>
<p>Some valid device_types are: “cpu”, “cuda”, “xla”, “mps”, “ipu”, “xpu”.
This API may be used as a decorator.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><ul class="simple">
<li><p><strong>op</strong> (<a class="reference external" href="https://docs.python.org/3/library/stdtypes.html#str" title="(in Python v3.13)"><em>str</em></a><em> | </em><em>OpOverload</em>) – The operator to register an impl to.</p></li>