forked from pytorch/pytorch.github.io
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathquantization-support.html
1068 lines (876 loc) · 77.5 KB
/
quantization-support.html
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
<!DOCTYPE html>
<!--[if IE 8]><html class="no-js lt-ie9" lang="en" > <![endif]-->
<!--[if gt IE 8]><!--> <html class="no-js" lang="en" > <!--<![endif]-->
<head>
<meta name="robots" content="noindex">
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>Quantization Operation coverage — PyTorch 1.7.1 documentation</title>
<link rel="canonical" href="https://pytorch.org/docs/stable/quantization-support.html"/>
<link rel="stylesheet" href="_static/css/theme.css" type="text/css" />
<!-- <link rel="stylesheet" href="_static/pygments.css" type="text/css" /> -->
<link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/katex.min.css" type="text/css" />
<link rel="stylesheet" href="_static/css/jit.css" type="text/css" />
<link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/katex.min.css" type="text/css" />
<link rel="stylesheet" href="_static/katex-math.css" type="text/css" />
<link rel="index" title="Index" href="genindex.html" />
<link rel="search" title="Search" href="search.html" />
<script src="_static/js/modernizr.min.js"></script>
<!-- Preload the theme fonts -->
<link rel="preload" href="_static/fonts/FreightSans/freight-sans-book.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="_static/fonts/FreightSans/freight-sans-medium.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="_static/fonts/IBMPlexMono/IBMPlexMono-Medium.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="_static/fonts/FreightSans/freight-sans-bold.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="_static/fonts/FreightSans/freight-sans-medium-italic.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="_static/fonts/IBMPlexMono/IBMPlexMono-SemiBold.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<!-- Preload the katex fonts -->
<link rel="preload" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/fonts/KaTeX_Math-Italic.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/fonts/KaTeX_Main-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/fonts/KaTeX_Main-Bold.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/fonts/KaTeX_Size1-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/fonts/KaTeX_Size4-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/fonts/KaTeX_Size2-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/fonts/KaTeX_Size3-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/fonts/KaTeX_Caligraphic-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
</head>
<div class="container-fluid header-holder tutorials-header" id="header-holder">
<div class="container">
<div class="header-container">
<a class="header-logo" href="https://pytorch.org/" aria-label="PyTorch"></a>
<div class="main-menu">
<ul>
<li>
<a href="https://pytorch.org/get-started">Get Started</a>
</li>
<li>
<a href="https://pytorch.org/ecosystem">Ecosystem</a>
</li>
<li>
<a href="https://pytorch.org/mobile">Mobile</a>
</li>
<li>
<a href="https://pytorch.org/blog/">Blog</a>
</li>
<li>
<a href="https://pytorch.org/tutorials">Tutorials</a>
</li>
<li class="active">
<div id="resourcesDropdownButton" data-toggle="resources-dropdown" class="resources-dropdown">
<a class="resource-option with-down-orange-arrow">
Docs
</a>
<div class="resources-dropdown-menu">
<a class="doc-dropdown-option nav-dropdown-item" href="https://pytorch.org/docs/1.7.1/index.html">
<span class="dropdown-title">PyTorch</span>
<p></p>
</a>
<a class="doc-dropdown-option nav-dropdown-item" href="https://pytorch.org/audio/0.7.0/index.html">
<span class="dropdown-title">torchaudio</span>
<p></p>
</a>
<a class="doc-dropdown-option nav-dropdown-item" href="https://pytorch.org/text/0.8.1/index.html">
<span class="dropdown-title">torchtext</span>
<p></p>
</a>
<a class="doc-dropdown-option nav-dropdown-item" href="https://pytorch.org/vision/0.8/">
<span class="dropdown-title">torchvision</span>
<p></p>
</a>
<a class="doc-dropdown-option nav-dropdown-item" href="https://pytorch.org/elastic/">
<span class="dropdown-title">TorchElastic</span>
<p></p>
</a>
<a class="doc-dropdown-option nav-dropdown-item" href="https://pytorch.org/serve/">
<span class="dropdown-title">TorchServe</span>
<p></p>
</a>
<a class="doc-dropdown-option nav-dropdown-item" href="https://pytorch.org/xla">
<span class="dropdown-title">PyTorch on XLA Devices</span>
<p></p>
</a>
</div>
</li>
<li>
<div id="resourcesDropdownButton" data-toggle="resources-dropdown" class="resources-dropdown">
<a class="resource-option with-down-arrow">
Resources
</a>
<div class="resources-dropdown-menu">
<a class="nav-dropdown-item" href="https://pytorch.org/features">
<span class="dropdown-title">About</span>
<p>Learn about PyTorch’s features and capabilities</p>
</a>
<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://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://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/hub">
<span class="dropdown-title">Models (Beta)</span>
<p>Discover, publish, and reuse pre-trained models</p>
</a>
</div>
</div>
</li>
<li>
<a href="https://github.com/pytorch/pytorch">Github</a>
</li>
</ul>
</div>
<a class="main-menu-open-button" href="#" data-behavior="open-mobile-menu"></a>
</div>
</div>
</div>
<body class="pytorch-body">
<div class="table-of-contents-link-wrapper">
<span>Table of Contents</span>
<a href="#" class="toggle-table-of-contents" data-behavior="toggle-table-of-contents"></a>
</div>
<nav data-toggle="wy-nav-shift" class="pytorch-left-menu" id="pytorch-left-menu">
<div class="pytorch-side-scroll">
<div class="pytorch-menu pytorch-menu-vertical" data-spy="affix" role="navigation" aria-label="main navigation">
<div class="pytorch-left-menu-search">
<div class="version">
<a href='http://pytorch.org/docs/versions.html'>1.7.1 ▼</a>
</div>
<div role="search">
<form id="rtd-search-form" class="wy-form" action="search.html" method="get">
<input type="text" name="q" placeholder="Search Docs" />
<input type="hidden" name="check_keywords" value="yes" />
<input type="hidden" name="area" value="default" />
</form>
</div>
</div>
<p class="caption"><span class="caption-text">Notes</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="notes/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/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/faq.html">Frequently Asked Questions</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/large_scale_deployments.html">Features for large-scale deployments</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/multiprocessing.html">Multiprocessing best practices</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/randomness.html">Reproducibility</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/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"><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>
</ul>
<p class="caption"><span class="caption-text">Python API</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="torch.html">torch</a></li>
<li class="toctree-l1"><a class="reference internal" href="nn.html">torch.nn</a></li>
<li class="toctree-l1"><a class="reference internal" href="nn.functional.html">torch.nn.functional</a></li>
<li class="toctree-l1"><a class="reference internal" href="tensors.html">torch.Tensor</a></li>
<li class="toctree-l1"><a class="reference internal" href="tensor_attributes.html">Tensor Attributes</a></li>
<li class="toctree-l1"><a class="reference internal" href="tensor_view.html">Tensor Views</a></li>
<li class="toctree-l1"><a class="reference internal" href="autograd.html">torch.autograd</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="amp.html">torch.cuda.amp</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="distributed.html">torch.distributed</a></li>
<li class="toctree-l1"><a class="reference internal" href="distributions.html">torch.distributions</a></li>
<li class="toctree-l1"><a class="reference internal" href="fft.html">torch.fft</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="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="nn.init.html">torch.nn.init</a></li>
<li class="toctree-l1"><a class="reference internal" href="onnx.html">torch.onnx</a></li>
<li class="toctree-l1"><a class="reference internal" href="optim.html">torch.optim</a></li>
<li class="toctree-l1"><a class="reference internal" href="complex_numbers.html">Complex Numbers</a></li>
<li class="toctree-l1"><a class="reference internal" href="quantization.html">Quantization</a></li>
<li class="toctree-l1"><a class="reference internal" href="rpc.html">Distributed RPC Framework</a></li>
<li class="toctree-l1"><a class="reference internal" href="random.html">torch.random</a></li>
<li class="toctree-l1"><a class="reference internal" href="sparse.html">torch.sparse</a></li>
<li class="toctree-l1"><a class="reference internal" href="storage.html">torch.Storage</a></li>
<li class="toctree-l1"><a class="reference internal" href="bottleneck.html">torch.utils.bottleneck</a></li>
<li class="toctree-l1"><a class="reference internal" href="checkpoint.html">torch.utils.checkpoint</a></li>
<li class="toctree-l1"><a class="reference internal" href="cpp_extension.html">torch.utils.cpp_extension</a></li>
<li class="toctree-l1"><a class="reference internal" href="data.html">torch.utils.data</a></li>
<li class="toctree-l1"><a class="reference internal" href="dlpack.html">torch.utils.dlpack</a></li>
<li class="toctree-l1"><a class="reference internal" href="mobile_optimizer.html">torch.utils.mobile_optimizer</a></li>
<li class="toctree-l1"><a class="reference internal" href="model_zoo.html">torch.utils.model_zoo</a></li>
<li class="toctree-l1"><a class="reference internal" href="tensorboard.html">torch.utils.tensorboard</a></li>
<li class="toctree-l1"><a class="reference internal" href="type_info.html">Type Info</a></li>
<li class="toctree-l1"><a class="reference internal" href="named_tensor.html">Named Tensors</a></li>
<li class="toctree-l1"><a class="reference internal" href="name_inference.html">Named Tensors operator coverage</a></li>
<li class="toctree-l1"><a class="reference internal" href="__config__.html">torch.__config__</a></li>
</ul>
<p class="caption"><span class="caption-text">Libraries</span></p>
<ul>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/audio">torchaudio</a></li>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/text">torchtext</a></li>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/vision">torchvision</a></li>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/elastic/">TorchElastic</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="http://pytorch.org/xla/">PyTorch on XLA Devices</a></li>
</ul>
<p class="caption"><span class="caption-text">Community</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="community/contribution_guide.html">PyTorch Contribution Guide</a></li>
<li class="toctree-l1"><a class="reference internal" href="community/governance.html">PyTorch Governance</a></li>
<li class="toctree-l1"><a class="reference internal" href="community/persons_of_interest.html">PyTorch Governance | Persons of Interest</a></li>
</ul>
</div>
</div>
</nav>
<div class="pytorch-container">
<div class="pytorch-page-level-bar" id="pytorch-page-level-bar">
<div class="pytorch-breadcrumbs-wrapper">
<div role="navigation" aria-label="breadcrumbs navigation">
<ul class="pytorch-breadcrumbs">
<li>
<a href="index.html">
Docs
</a> >
</li>
<li>Quantization Operation coverage</li>
<li class="pytorch-breadcrumbs-aside">
<a href="_sources/quantization-support.rst.txt" rel="nofollow"><img src="_static/images/view-page-source-icon.svg"></a>
</li>
</ul>
</div>
</div>
<div class="pytorch-shortcuts-wrapper" id="pytorch-shortcuts-wrapper">
Shortcuts
</div>
</div>
<section data-toggle="wy-nav-shift" id="pytorch-content-wrap" class="pytorch-content-wrap">
<div class="pytorch-content-left">
<div class="rst-content">
<div role="main" class="main-content" itemscope="itemscope" itemtype="http://schema.org/Article">
<article itemprop="articleBody" id="pytorch-article" class="pytorch-article">
<div class="section" id="quantization-operation-coverage">
<h1>Quantization Operation coverage<a class="headerlink" href="#quantization-operation-coverage" title="Permalink to this headline">¶</a></h1>
<p>Quantized Tensors support a limited subset of data manipulation methods of the
regular full-precision tensor. For NN operators included in PyTorch, we
restrict support to:</p>
<blockquote>
<div><ol class="arabic simple">
<li><p>8 bit weights (data_type = qint8)</p></li>
<li><p>8 bit activations (data_type = quint8)</p></li>
</ol>
</div></blockquote>
<p>Note that operator implementations currently only
support per channel quantization for weights of the <strong>conv</strong> and <strong>linear</strong>
operators. Furthermore the minimum and the maximum of the input data is
mapped linearly to the minimum and the maximum of the quantized data
type such that zero is represented with no quantization error.</p>
<p>Additional data types and quantization schemes can be implemented through
the <a class="reference external" href="https://pytorch.org/tutorials/advanced/torch_script_custom_ops.html">custom operator mechanism</a>.</p>
<p>Many operations for quantized tensors are available under the same API as full
float version in <code class="docutils literal notranslate"><span class="pre">torch</span></code> or <code class="docutils literal notranslate"><span class="pre">torch.nn</span></code>. Quantized version of NN modules that
perform re-quantization are available in <code class="docutils literal notranslate"><span class="pre">torch.nn.quantized</span></code>. Those
operations explicitly take output quantization parameters (scale and zero_point) in
the operation signature.</p>
<p>In addition, we also support fused versions corresponding to common fusion
patterns that impact quantization at: <cite>torch.nn.intrinsic.quantized</cite>.</p>
<p>For quantization aware training, we support modules prepared for quantization
aware training at <cite>torch.nn.qat</cite> and <cite>torch.nn.intrinsic.qat</cite></p>
<p>The following operation list is sufficient to cover typical CNN and RNN models</p>
<div class="section" id="quantized-torch-tensor-operations">
<h2>Quantized <code class="docutils literal notranslate"><span class="pre">torch.Tensor</span></code> operations<a class="headerlink" href="#quantized-torch-tensor-operations" title="Permalink to this headline">¶</a></h2>
<p>Operations that are available from the <code class="docutils literal notranslate"><span class="pre">torch</span></code> namespace or as methods on
Tensor for quantized tensors:</p>
<ul class="simple">
<li><p><a class="reference internal" href="generated/torch.quantize_per_tensor.html#torch.quantize_per_tensor" title="torch.quantize_per_tensor"><code class="xref py py-func docutils literal notranslate"><span class="pre">quantize_per_tensor()</span></code></a> - Convert float tensor to quantized tensor
with per-tensor scale and zero point</p></li>
<li><p><a class="reference internal" href="generated/torch.quantize_per_channel.html#torch.quantize_per_channel" title="torch.quantize_per_channel"><code class="xref py py-func docutils literal notranslate"><span class="pre">quantize_per_channel()</span></code></a> - Convert float tensor to quantized
tensor with per-channel scale and zero point</p></li>
<li><p>View-based operations like <a class="reference internal" href="tensors.html#torch.Tensor.view" title="torch.Tensor.view"><code class="xref py py-meth docutils literal notranslate"><span class="pre">view()</span></code></a>,
<a class="reference internal" href="tensors.html#torch.Tensor.as_strided" title="torch.Tensor.as_strided"><code class="xref py py-meth docutils literal notranslate"><span class="pre">as_strided()</span></code></a>, <a class="reference internal" href="tensors.html#torch.Tensor.expand" title="torch.Tensor.expand"><code class="xref py py-meth docutils literal notranslate"><span class="pre">expand()</span></code></a>,
<a class="reference internal" href="tensors.html#torch.Tensor.flatten" title="torch.Tensor.flatten"><code class="xref py py-meth docutils literal notranslate"><span class="pre">flatten()</span></code></a>, <a class="reference internal" href="tensors.html#torch.Tensor.select" title="torch.Tensor.select"><code class="xref py py-meth docutils literal notranslate"><span class="pre">select()</span></code></a>, python-style
indexing, etc - work as on regular tensor (if quantization is not
per-channel)</p></li>
<li><dl class="simple">
<dt>Comparators</dt><dd><ul>
<li><p><a class="reference internal" href="tensors.html#torch.Tensor.ne" title="torch.Tensor.ne"><code class="xref py py-meth docutils literal notranslate"><span class="pre">ne()</span></code></a> — Not equal</p></li>
<li><p><a class="reference internal" href="tensors.html#torch.Tensor.eq" title="torch.Tensor.eq"><code class="xref py py-meth docutils literal notranslate"><span class="pre">eq()</span></code></a> — Equal</p></li>
<li><p><a class="reference internal" href="tensors.html#torch.Tensor.ge" title="torch.Tensor.ge"><code class="xref py py-meth docutils literal notranslate"><span class="pre">ge()</span></code></a> — Greater or equal</p></li>
<li><p><a class="reference internal" href="tensors.html#torch.Tensor.le" title="torch.Tensor.le"><code class="xref py py-meth docutils literal notranslate"><span class="pre">le()</span></code></a> — Less or equal</p></li>
<li><p><a class="reference internal" href="tensors.html#torch.Tensor.gt" title="torch.Tensor.gt"><code class="xref py py-meth docutils literal notranslate"><span class="pre">gt()</span></code></a> — Greater</p></li>
<li><p><a class="reference internal" href="tensors.html#torch.Tensor.lt" title="torch.Tensor.lt"><code class="xref py py-meth docutils literal notranslate"><span class="pre">lt()</span></code></a> — Less</p></li>
</ul>
</dd>
</dl>
</li>
<li><p><a class="reference internal" href="tensors.html#torch.Tensor.copy_" title="torch.Tensor.copy_"><code class="xref py py-meth docutils literal notranslate"><span class="pre">copy_()</span></code></a> — Copies src to self in-place</p></li>
<li><p><a class="reference internal" href="tensors.html#torch.Tensor.clone" title="torch.Tensor.clone"><code class="xref py py-meth docutils literal notranslate"><span class="pre">clone()</span></code></a> — Returns a deep copy of the passed-in tensor</p></li>
<li><p><a class="reference internal" href="tensors.html#torch.Tensor.dequantize" title="torch.Tensor.dequantize"><code class="xref py py-meth docutils literal notranslate"><span class="pre">dequantize()</span></code></a> — Convert quantized tensor to float tensor</p></li>
<li><p><a class="reference internal" href="tensors.html#torch.Tensor.equal" title="torch.Tensor.equal"><code class="xref py py-meth docutils literal notranslate"><span class="pre">equal()</span></code></a> — Compares two tensors, returns true if
quantization parameters and all integer elements are the same</p></li>
<li><p><a class="reference internal" href="tensors.html#torch.Tensor.int_repr" title="torch.Tensor.int_repr"><code class="xref py py-meth docutils literal notranslate"><span class="pre">int_repr()</span></code></a> — Prints the underlying integer representation
of the quantized tensor</p></li>
<li><p><a class="reference internal" href="tensors.html#torch.Tensor.max" title="torch.Tensor.max"><code class="xref py py-meth docutils literal notranslate"><span class="pre">max()</span></code></a> — Returns the maximum value of the tensor (reduction only)</p></li>
<li><p><a class="reference internal" href="tensors.html#torch.Tensor.mean" title="torch.Tensor.mean"><code class="xref py py-meth docutils literal notranslate"><span class="pre">mean()</span></code></a> — Mean function. Supported variants: reduction, dim, out</p></li>
<li><p><a class="reference internal" href="tensors.html#torch.Tensor.min" title="torch.Tensor.min"><code class="xref py py-meth docutils literal notranslate"><span class="pre">min()</span></code></a> — Returns the minimum value of the tensor (reduction only)</p></li>
<li><p><a class="reference internal" href="tensors.html#torch.Tensor.q_scale" title="torch.Tensor.q_scale"><code class="xref py py-meth docutils literal notranslate"><span class="pre">q_scale()</span></code></a> — Returns the scale of the per-tensor quantized tensor</p></li>
<li><p><a class="reference internal" href="tensors.html#torch.Tensor.q_zero_point" title="torch.Tensor.q_zero_point"><code class="xref py py-meth docutils literal notranslate"><span class="pre">q_zero_point()</span></code></a> — Returns the zero_point of the per-tensor
quantized zero point</p></li>
<li><p><a class="reference internal" href="tensors.html#torch.Tensor.q_per_channel_scales" title="torch.Tensor.q_per_channel_scales"><code class="xref py py-meth docutils literal notranslate"><span class="pre">q_per_channel_scales()</span></code></a> — Returns the scales of the
per-channel quantized tensor</p></li>
<li><p><a class="reference internal" href="tensors.html#torch.Tensor.q_per_channel_zero_points" title="torch.Tensor.q_per_channel_zero_points"><code class="xref py py-meth docutils literal notranslate"><span class="pre">q_per_channel_zero_points()</span></code></a> — Returns the zero points of
the per-channel quantized tensor</p></li>
<li><p><a class="reference internal" href="tensors.html#torch.Tensor.q_per_channel_axis" title="torch.Tensor.q_per_channel_axis"><code class="xref py py-meth docutils literal notranslate"><span class="pre">q_per_channel_axis()</span></code></a> — Returns the channel axis of the
per-channel quantized tensor</p></li>
<li><p><a class="reference internal" href="tensors.html#torch.Tensor.resize_" title="torch.Tensor.resize_"><code class="xref py py-meth docutils literal notranslate"><span class="pre">resize_()</span></code></a> — In-place resize</p></li>
<li><p><a class="reference internal" href="tensors.html#torch.Tensor.sort" title="torch.Tensor.sort"><code class="xref py py-meth docutils literal notranslate"><span class="pre">sort()</span></code></a> — Sorts the tensor</p></li>
<li><p><a class="reference internal" href="tensors.html#torch.Tensor.topk" title="torch.Tensor.topk"><code class="xref py py-meth docutils literal notranslate"><span class="pre">topk()</span></code></a> — Returns k largest values of a tensor</p></li>
</ul>
</div>
<div class="section" id="torch-nn-functional">
<h2><code class="docutils literal notranslate"><span class="pre">torch.nn.functional</span></code><a class="headerlink" href="#torch-nn-functional" title="Permalink to this headline">¶</a></h2>
<p>Basic activations are supported.</p>
<ul class="simple">
<li><p><a class="reference internal" href="nn.functional.html#torch.nn.functional.relu" title="torch.nn.functional.relu"><code class="xref py py-meth docutils literal notranslate"><span class="pre">relu()</span></code></a> — Rectified linear unit (copy)</p></li>
<li><p><a class="reference internal" href="nn.functional.html#torch.nn.functional.relu_" title="torch.nn.functional.relu_"><code class="xref py py-meth docutils literal notranslate"><span class="pre">relu_()</span></code></a> — Rectified linear unit (inplace)</p></li>
<li><p><a class="reference internal" href="nn.functional.html#torch.nn.functional.elu" title="torch.nn.functional.elu"><code class="xref py py-meth docutils literal notranslate"><span class="pre">elu()</span></code></a> - ELU</p></li>
<li><p><a class="reference internal" href="nn.functional.html#torch.nn.functional.max_pool2d" title="torch.nn.functional.max_pool2d"><code class="xref py py-meth docutils literal notranslate"><span class="pre">max_pool2d()</span></code></a> - Maximum pooling</p></li>
<li><p><a class="reference internal" href="nn.functional.html#torch.nn.functional.adaptive_avg_pool2d" title="torch.nn.functional.adaptive_avg_pool2d"><code class="xref py py-meth docutils literal notranslate"><span class="pre">adaptive_avg_pool2d()</span></code></a> - Adaptive average pooling</p></li>
<li><p><a class="reference internal" href="nn.functional.html#torch.nn.functional.avg_pool2d" title="torch.nn.functional.avg_pool2d"><code class="xref py py-meth docutils literal notranslate"><span class="pre">avg_pool2d()</span></code></a> - Average pooling</p></li>
<li><p><a class="reference internal" href="nn.functional.html#torch.nn.functional.interpolate" title="torch.nn.functional.interpolate"><code class="xref py py-meth docutils literal notranslate"><span class="pre">interpolate()</span></code></a> - Interpolation</p></li>
<li><p><a class="reference internal" href="nn.functional.html#torch.nn.functional.hardsigmoid" title="torch.nn.functional.hardsigmoid"><code class="xref py py-meth docutils literal notranslate"><span class="pre">hardsigmoid()</span></code></a> - Hardsigmoid</p></li>
<li><p><a class="reference internal" href="nn.functional.html#torch.nn.functional.hardswish" title="torch.nn.functional.hardswish"><code class="xref py py-meth docutils literal notranslate"><span class="pre">hardswish()</span></code></a> - Hardswish</p></li>
<li><p><a class="reference internal" href="nn.functional.html#torch.nn.functional.hardtanh" title="torch.nn.functional.hardtanh"><code class="xref py py-meth docutils literal notranslate"><span class="pre">hardtanh()</span></code></a> - Hardtanh</p></li>
<li><p><a class="reference internal" href="nn.functional.html#torch.nn.functional.upsample" title="torch.nn.functional.upsample"><code class="xref py py-meth docutils literal notranslate"><span class="pre">upsample()</span></code></a> - Upsampling</p></li>
<li><p><a class="reference internal" href="nn.functional.html#torch.nn.functional.upsample_bilinear" title="torch.nn.functional.upsample_bilinear"><code class="xref py py-meth docutils literal notranslate"><span class="pre">upsample_bilinear()</span></code></a> - Bilinear Upsampling</p></li>
<li><p><a class="reference internal" href="nn.functional.html#torch.nn.functional.upsample_nearest" title="torch.nn.functional.upsample_nearest"><code class="xref py py-meth docutils literal notranslate"><span class="pre">upsample_nearest()</span></code></a> - Upsampling Nearest</p></li>
</ul>
</div>
<div class="section" id="torch-nn-intrinsic">
<h2><code class="docutils literal notranslate"><span class="pre">torch.nn.intrinsic</span></code><a class="headerlink" href="#torch-nn-intrinsic" title="Permalink to this headline">¶</a></h2>
<p>Fused modules are provided for common patterns in CNNs. Combining several
operations together (like convolution and relu) allows for better quantization
accuracy</p>
<ul class="simple">
<li><p><cite>torch.nn.intrinsic</cite> — float versions of the modules, can be swapped with
quantized version 1 to 1:</p>
<ul>
<li><p><a class="reference internal" href="torch.nn.intrinsic.html#torch.nn.intrinsic.ConvBn1d" title="torch.nn.intrinsic.ConvBn1d"><code class="xref py py-class docutils literal notranslate"><span class="pre">ConvBn1d</span></code></a> — Conv1d + BatchNorm1d</p></li>
<li><p><a class="reference internal" href="torch.nn.intrinsic.html#torch.nn.intrinsic.ConvBn2d" title="torch.nn.intrinsic.ConvBn2d"><code class="xref py py-class docutils literal notranslate"><span class="pre">ConvBn2d</span></code></a> — Conv2d + BatchNorm</p></li>
<li><p><a class="reference internal" href="torch.nn.intrinsic.html#torch.nn.intrinsic.ConvBnReLU1d" title="torch.nn.intrinsic.ConvBnReLU1d"><code class="xref py py-class docutils literal notranslate"><span class="pre">ConvBnReLU1d</span></code></a> — Conv1d + BatchNorm1d + ReLU</p></li>
<li><p><a class="reference internal" href="torch.nn.intrinsic.html#torch.nn.intrinsic.ConvBnReLU2d" title="torch.nn.intrinsic.ConvBnReLU2d"><code class="xref py py-class docutils literal notranslate"><span class="pre">ConvBnReLU2d</span></code></a> — Conv2d + BatchNorm + ReLU</p></li>
<li><p><a class="reference internal" href="torch.nn.intrinsic.html#torch.nn.intrinsic.ConvReLU1d" title="torch.nn.intrinsic.ConvReLU1d"><code class="xref py py-class docutils literal notranslate"><span class="pre">ConvReLU1d</span></code></a> — Conv1d + ReLU</p></li>
<li><p><a class="reference internal" href="torch.nn.intrinsic.html#torch.nn.intrinsic.ConvReLU2d" title="torch.nn.intrinsic.ConvReLU2d"><code class="xref py py-class docutils literal notranslate"><span class="pre">ConvReLU2d</span></code></a> — Conv2d + ReLU</p></li>
<li><p><code class="xref py py-class docutils literal notranslate"><span class="pre">ConvReLU3d</span></code> — Conv3d + ReLU</p></li>
<li><p><code class="xref py py-class docutils literal notranslate"><span class="pre">LinearReLU</span></code> — Linear + ReLU</p></li>
</ul>
</li>
<li><p><cite>torch.nn.intrinsic.qat</cite> — versions of layers for quantization-aware training:</p>
<ul>
<li><p><a class="reference internal" href="torch.nn.intrinsic.qat.html#torch.nn.intrinsic.qat.ConvBn2d" title="torch.nn.intrinsic.qat.ConvBn2d"><code class="xref py py-class docutils literal notranslate"><span class="pre">ConvBn2d</span></code></a> — Conv2d + BatchNorm</p></li>
<li><p><a class="reference internal" href="torch.nn.intrinsic.qat.html#torch.nn.intrinsic.qat.ConvBnReLU2d" title="torch.nn.intrinsic.qat.ConvBnReLU2d"><code class="xref py py-class docutils literal notranslate"><span class="pre">ConvBnReLU2d</span></code></a> — Conv2d + BatchNorm + ReLU</p></li>
<li><p><a class="reference internal" href="torch.nn.intrinsic.qat.html#torch.nn.intrinsic.qat.ConvReLU2d" title="torch.nn.intrinsic.qat.ConvReLU2d"><code class="xref py py-class docutils literal notranslate"><span class="pre">ConvReLU2d</span></code></a> — Conv2d + ReLU</p></li>
<li><p><a class="reference internal" href="torch.nn.intrinsic.qat.html#torch.nn.intrinsic.qat.LinearReLU" title="torch.nn.intrinsic.qat.LinearReLU"><code class="xref py py-class docutils literal notranslate"><span class="pre">LinearReLU</span></code></a> — Linear + ReLU</p></li>
</ul>
</li>
<li><p><cite>torch.nn.intrinsic.quantized</cite> — quantized version of fused layers for
inference (no BatchNorm variants as it’s usually folded into convolution for
inference):</p>
<ul>
<li><p><a class="reference internal" href="torch.nn.intrinsic.quantized.html#torch.nn.intrinsic.quantized.LinearReLU" title="torch.nn.intrinsic.quantized.LinearReLU"><code class="xref py py-class docutils literal notranslate"><span class="pre">LinearReLU</span></code></a> — Linear + ReLU</p></li>
<li><p><code class="xref py py-class docutils literal notranslate"><span class="pre">ConvReLU1d</span></code> — 1D Convolution + ReLU</p></li>
<li><p><a class="reference internal" href="torch.nn.intrinsic.quantized.html#torch.nn.intrinsic.quantized.ConvReLU2d" title="torch.nn.intrinsic.quantized.ConvReLU2d"><code class="xref py py-class docutils literal notranslate"><span class="pre">ConvReLU2d</span></code></a> — 2D Convolution + ReLU</p></li>
<li><p><a class="reference internal" href="torch.nn.intrinsic.quantized.html#torch.nn.intrinsic.quantized.ConvReLU3d" title="torch.nn.intrinsic.quantized.ConvReLU3d"><code class="xref py py-class docutils literal notranslate"><span class="pre">ConvReLU3d</span></code></a> — 3D Convolution + ReLU</p></li>
</ul>
</li>
</ul>
</div>
<div class="section" id="torch-nn-qat">
<h2><cite>torch.nn.qat</cite><a class="headerlink" href="#torch-nn-qat" title="Permalink to this headline">¶</a></h2>
<p>Layers for the quantization-aware training</p>
<ul class="simple">
<li><p><a class="reference internal" href="torch.nn.qat.html#torch.nn.qat.Linear" title="torch.nn.qat.Linear"><code class="xref py py-class docutils literal notranslate"><span class="pre">Linear</span></code></a> — Linear (fully-connected) layer</p></li>
<li><p><a class="reference internal" href="torch.nn.qat.html#torch.nn.qat.Conv2d" title="torch.nn.qat.Conv2d"><code class="xref py py-class docutils literal notranslate"><span class="pre">Conv2d</span></code></a> — 2D convolution</p></li>
</ul>
</div>
<div class="section" id="torch-quantization">
<h2><cite>torch.quantization</cite><a class="headerlink" href="#torch-quantization" title="Permalink to this headline">¶</a></h2>
<ul class="simple">
<li><p>Functions for eager mode quantization:</p>
<ul>
<li><p><a class="reference internal" href="torch.quantization.html#torch.quantization.add_observer_" title="torch.quantization.add_observer_"><code class="xref py py-func docutils literal notranslate"><span class="pre">add_observer_()</span></code></a> — Adds observer for the leaf
modules (if quantization configuration is provided)</p></li>
<li><p><a class="reference internal" href="torch.quantization.html#torch.quantization.add_quant_dequant" title="torch.quantization.add_quant_dequant"><code class="xref py py-func docutils literal notranslate"><span class="pre">add_quant_dequant()</span></code></a>— Wraps the leaf child module using <a class="reference internal" href="torch.quantization.html#torch.quantization.QuantWrapper" title="torch.quantization.QuantWrapper"><code class="xref py py-class docutils literal notranslate"><span class="pre">QuantWrapper</span></code></a></p></li>
<li><p><a class="reference internal" href="torch.quantization.html#torch.quantization.convert" title="torch.quantization.convert"><code class="xref py py-func docutils literal notranslate"><span class="pre">convert()</span></code></a> — Converts float module with
observers into its quantized counterpart. Must have quantization
configuration</p></li>
<li><p><a class="reference internal" href="torch.quantization.html#torch.quantization.get_observer_dict" title="torch.quantization.get_observer_dict"><code class="xref py py-func docutils literal notranslate"><span class="pre">get_observer_dict()</span></code></a> — Traverses the module
children and collects all observers into a <code class="docutils literal notranslate"><span class="pre">dict</span></code></p></li>
<li><p><a class="reference internal" href="torch.quantization.html#torch.quantization.prepare" title="torch.quantization.prepare"><code class="xref py py-func docutils literal notranslate"><span class="pre">prepare()</span></code></a> — Prepares a copy of a model for
quantization</p></li>
<li><p><a class="reference internal" href="torch.quantization.html#torch.quantization.prepare_qat" title="torch.quantization.prepare_qat"><code class="xref py py-func docutils literal notranslate"><span class="pre">prepare_qat()</span></code></a> — Prepares a copy of a model for
quantization aware training</p></li>
<li><p><a class="reference internal" href="torch.quantization.html#torch.quantization.propagate_qconfig_" title="torch.quantization.propagate_qconfig_"><code class="xref py py-func docutils literal notranslate"><span class="pre">propagate_qconfig_()</span></code></a> — Propagates quantization
configurations through the module hierarchy and assign them to each leaf
module</p></li>
<li><p><a class="reference internal" href="torch.quantization.html#torch.quantization.quantize" title="torch.quantization.quantize"><code class="xref py py-func docutils literal notranslate"><span class="pre">quantize()</span></code></a> — Function for eager mode post training static quantization</p></li>
<li><p><a class="reference internal" href="torch.quantization.html#torch.quantization.quantize_dynamic" title="torch.quantization.quantize_dynamic"><code class="xref py py-func docutils literal notranslate"><span class="pre">quantize_dynamic()</span></code></a> — Function for eager mode post training dynamic quantization</p></li>
<li><p><a class="reference internal" href="torch.quantization.html#torch.quantization.quantize_qat" title="torch.quantization.quantize_qat"><code class="xref py py-func docutils literal notranslate"><span class="pre">quantize_qat()</span></code></a> — Function for eager mode quantization aware training function</p></li>
<li><p><a class="reference internal" href="torch.quantization.html#torch.quantization.swap_module" title="torch.quantization.swap_module"><code class="xref py py-func docutils literal notranslate"><span class="pre">swap_module()</span></code></a> — Swaps the module with its
quantized counterpart (if quantizable and if it has an observer)</p></li>
<li><p><a class="reference internal" href="torch.quantization.html#torch.quantization.default_eval_fn" title="torch.quantization.default_eval_fn"><code class="xref py py-func docutils literal notranslate"><span class="pre">default_eval_fn()</span></code></a> — Default evaluation function
used by the <a class="reference internal" href="torch.quantization.html#torch.quantization.quantize" title="torch.quantization.quantize"><code class="xref py py-func docutils literal notranslate"><span class="pre">torch.quantization.quantize()</span></code></a></p></li>
<li><p><a class="reference internal" href="torch.quantization.html#torch.quantization.fuse_modules" title="torch.quantization.fuse_modules"><code class="xref py py-func docutils literal notranslate"><span class="pre">fuse_modules()</span></code></a></p></li>
</ul>
</li>
<li><p>Functions for graph mode quantization:</p>
<ul>
<li><p><code class="xref py py-func docutils literal notranslate"><span class="pre">quantize_jit()</span></code> - Function for graph mode post training static quantization</p></li>
<li><p><code class="xref py py-func docutils literal notranslate"><span class="pre">quantize_dynamic_jit()</span></code> - Function for graph mode post training dynamic quantization</p></li>
</ul>
</li>
<li><dl class="simple">
<dt>Quantization configurations</dt><dd><ul>
<li><p><a class="reference internal" href="torch.quantization.html#torch.quantization.QConfig" title="torch.quantization.QConfig"><code class="xref py py-class docutils literal notranslate"><span class="pre">QConfig</span></code></a> — Quantization configuration class</p></li>
<li><p><code class="xref py py-attr docutils literal notranslate"><span class="pre">default_qconfig</span></code> — Same as
<code class="docutils literal notranslate"><span class="pre">QConfig(activation=default_observer,</span> <span class="pre">weight=default_weight_observer)</span></code>
(See <code class="xref py py-class docutils literal notranslate"><span class="pre">QConfig</span></code>)</p></li>
<li><p><code class="xref py py-attr docutils literal notranslate"><span class="pre">default_qat_qconfig</span></code> — Same as
<code class="docutils literal notranslate"><span class="pre">QConfig(activation=default_fake_quant,</span>
<span class="pre">weight=default_weight_fake_quant)</span></code> (See
<code class="xref py py-class docutils literal notranslate"><span class="pre">QConfig</span></code>)</p></li>
<li><p><code class="xref py py-attr docutils literal notranslate"><span class="pre">default_dynamic_qconfig</span></code> — Same as
<code class="docutils literal notranslate"><span class="pre">QConfigDynamic(weight=default_weight_observer)</span></code> (See
<code class="xref py py-class docutils literal notranslate"><span class="pre">QConfigDynamic</span></code>)</p></li>
<li><p><code class="xref py py-attr docutils literal notranslate"><span class="pre">float16_dynamic_qconfig</span></code> — Same as
<code class="docutils literal notranslate"><span class="pre">QConfigDynamic(weight=NoopObserver.with_args(dtype=torch.float16))</span></code>
(See <code class="xref py py-class docutils literal notranslate"><span class="pre">QConfigDynamic</span></code>)</p></li>
</ul>
</dd>
</dl>
</li>
<li><dl class="simple">
<dt>Stubs</dt><dd><ul>
<li><p><a class="reference internal" href="torch.quantization.html#torch.quantization.DeQuantStub" title="torch.quantization.DeQuantStub"><code class="xref py py-class docutils literal notranslate"><span class="pre">DeQuantStub</span></code></a> - placeholder module for
dequantize() operation in float-valued models</p></li>
<li><p><a class="reference internal" href="torch.quantization.html#torch.quantization.QuantStub" title="torch.quantization.QuantStub"><code class="xref py py-class docutils literal notranslate"><span class="pre">QuantStub</span></code></a> - placeholder module for
quantize() operation in float-valued models</p></li>
<li><p><a class="reference internal" href="torch.quantization.html#torch.quantization.QuantWrapper" title="torch.quantization.QuantWrapper"><code class="xref py py-class docutils literal notranslate"><span class="pre">QuantWrapper</span></code></a> — wraps the module to be
quantized. Inserts the <a class="reference internal" href="torch.quantization.html#torch.quantization.QuantStub" title="torch.quantization.QuantStub"><code class="xref py py-class docutils literal notranslate"><span class="pre">QuantStub</span></code></a> and</p></li>
<li><p><a class="reference internal" href="torch.quantization.html#torch.quantization.DeQuantStub" title="torch.quantization.DeQuantStub"><code class="xref py py-class docutils literal notranslate"><span class="pre">DeQuantStub</span></code></a></p></li>
</ul>
</dd>
</dl>
</li>
<li><p>Observers for computing the quantization parameters</p>
<ul>
<li><p>Default Observers. The rest of observers are available from
<code class="docutils literal notranslate"><span class="pre">torch.quantization.observer</span></code>:</p>
<ul>
<li><p><code class="xref py py-attr docutils literal notranslate"><span class="pre">default_observer</span></code> — Same as <code class="docutils literal notranslate"><span class="pre">MinMaxObserver.with_args(reduce_range=True)</span></code></p></li>
<li><p><code class="xref py py-attr docutils literal notranslate"><span class="pre">default_weight_observer</span></code> — Same as <code class="docutils literal notranslate"><span class="pre">MinMaxObserver.with_args(dtype=torch.qint8,</span> <span class="pre">qscheme=torch.per_tensor_symmetric)</span></code></p></li>
</ul>
</li>
<li><p><code class="xref py py-class docutils literal notranslate"><span class="pre">Observer</span></code> — Abstract base class for observers</p></li>
<li><p><a class="reference internal" href="torch.quantization.html#torch.quantization.MinMaxObserver" title="torch.quantization.MinMaxObserver"><code class="xref py py-class docutils literal notranslate"><span class="pre">MinMaxObserver</span></code></a> — Derives the quantization
parameters from the running minimum and maximum of the observed tensor inputs
(per tensor variant)</p></li>
<li><p><a class="reference internal" href="torch.quantization.html#torch.quantization.MovingAverageMinMaxObserver" title="torch.quantization.MovingAverageMinMaxObserver"><code class="xref py py-class docutils literal notranslate"><span class="pre">MovingAverageMinMaxObserver</span></code></a> — Derives the
quantization parameters from the running averages of the minimums and
maximums of the observed tensor inputs (per tensor variant)</p></li>
<li><p><a class="reference internal" href="torch.quantization.html#torch.quantization.PerChannelMinMaxObserver" title="torch.quantization.PerChannelMinMaxObserver"><code class="xref py py-class docutils literal notranslate"><span class="pre">PerChannelMinMaxObserver</span></code></a> — Derives the
quantization parameters from the running minimum and maximum of the observed
tensor inputs (per channel variant)</p></li>
<li><p><a class="reference internal" href="torch.quantization.html#torch.quantization.MovingAveragePerChannelMinMaxObserver" title="torch.quantization.MovingAveragePerChannelMinMaxObserver"><code class="xref py py-class docutils literal notranslate"><span class="pre">MovingAveragePerChannelMinMaxObserver</span></code></a> — Derives
the quantization parameters from the running averages of the minimums and
maximums of the observed tensor inputs (per channel variant)</p></li>
<li><p><a class="reference internal" href="torch.quantization.html#torch.quantization.HistogramObserver" title="torch.quantization.HistogramObserver"><code class="xref py py-class docutils literal notranslate"><span class="pre">HistogramObserver</span></code></a> — Derives the quantization
parameters by creating a histogram of running minimums and maximums.</p></li>
</ul>
</li>
<li><dl class="simple">
<dt>Observers that do not compute the quantization parameters:</dt><dd><ul>
<li><p><a class="reference internal" href="torch.quantization.html#torch.quantization.RecordingObserver" title="torch.quantization.RecordingObserver"><code class="xref py py-class docutils literal notranslate"><span class="pre">RecordingObserver</span></code></a> — Records all incoming
tensors. Used for debugging only.</p></li>
<li><p><a class="reference internal" href="torch.quantization.html#torch.quantization.NoopObserver" title="torch.quantization.NoopObserver"><code class="xref py py-class docutils literal notranslate"><span class="pre">NoopObserver</span></code></a> — Pass-through observer. Used
for situation when there are no quantization parameters (i.e.
quantization to <code class="docutils literal notranslate"><span class="pre">float16</span></code>)</p></li>
</ul>
</dd>
</dl>
</li>
<li><dl class="simple">
<dt>FakeQuantize module</dt><dd><ul>
<li><p><a class="reference internal" href="torch.quantization.html#torch.quantization.FakeQuantize" title="torch.quantization.FakeQuantize"><code class="xref py py-class docutils literal notranslate"><span class="pre">FakeQuantize</span></code></a> — Module for simulating the
quantization/dequantization at training time</p></li>
</ul>
</dd>
</dl>
</li>
</ul>
</div>
<div class="section" id="torch-nn-quantized">
<h2><cite>torch.nn.quantized</cite><a class="headerlink" href="#torch-nn-quantized" title="Permalink to this headline">¶</a></h2>
<p>Quantized version of standard NN layers.</p>
<ul class="simple">
<li><p><a class="reference internal" href="torch.nn.quantized.html#torch.nn.quantized.Quantize" title="torch.nn.quantized.Quantize"><code class="xref py py-class docutils literal notranslate"><span class="pre">Quantize</span></code></a> — Quantization layer, used to
automatically replace <a class="reference internal" href="torch.quantization.html#torch.quantization.QuantStub" title="torch.quantization.QuantStub"><code class="xref py py-class docutils literal notranslate"><span class="pre">QuantStub</span></code></a></p></li>
<li><p><a class="reference internal" href="torch.nn.quantized.html#torch.nn.quantized.DeQuantize" title="torch.nn.quantized.DeQuantize"><code class="xref py py-class docutils literal notranslate"><span class="pre">DeQuantize</span></code></a> — Dequantization layer, used to
replace <a class="reference internal" href="torch.quantization.html#torch.quantization.DeQuantStub" title="torch.quantization.DeQuantStub"><code class="xref py py-class docutils literal notranslate"><span class="pre">DeQuantStub</span></code></a></p></li>
<li><p><a class="reference internal" href="torch.nn.quantized.html#torch.nn.quantized.FloatFunctional" title="torch.nn.quantized.FloatFunctional"><code class="xref py py-class docutils literal notranslate"><span class="pre">FloatFunctional</span></code></a> — Wrapper class to make
stateless float operations stateful so that they can be replaced with
quantized versions</p></li>
<li><p><a class="reference internal" href="torch.nn.quantized.html#torch.nn.quantized.QFunctional" title="torch.nn.quantized.QFunctional"><code class="xref py py-class docutils literal notranslate"><span class="pre">QFunctional</span></code></a> — Wrapper class for quantized
versions of stateless operations like <code class="docutils literal notranslate"><span class="pre">torch.add</span></code></p></li>
<li><p><a class="reference internal" href="torch.nn.quantized.html#torch.nn.quantized.Conv1d" title="torch.nn.quantized.Conv1d"><code class="xref py py-class docutils literal notranslate"><span class="pre">Conv1d</span></code></a> — 1D convolution</p></li>
<li><p><a class="reference internal" href="torch.nn.quantized.html#torch.nn.quantized.Conv2d" title="torch.nn.quantized.Conv2d"><code class="xref py py-class docutils literal notranslate"><span class="pre">Conv2d</span></code></a> — 2D convolution</p></li>
<li><p><a class="reference internal" href="torch.nn.quantized.html#torch.nn.quantized.Conv3d" title="torch.nn.quantized.Conv3d"><code class="xref py py-class docutils literal notranslate"><span class="pre">Conv3d</span></code></a> — 3D convolution</p></li>
<li><p><a class="reference internal" href="torch.nn.quantized.html#torch.nn.quantized.Linear" title="torch.nn.quantized.Linear"><code class="xref py py-class docutils literal notranslate"><span class="pre">Linear</span></code></a> — Linear (fully-connected) layer</p></li>
<li><p><a class="reference internal" href="generated/torch.nn.MaxPool2d.html#torch.nn.MaxPool2d" title="torch.nn.MaxPool2d"><code class="xref py py-class docutils literal notranslate"><span class="pre">MaxPool2d</span></code></a> — 2D max pooling</p></li>
<li><p><a class="reference internal" href="torch.nn.quantized.html#torch.nn.quantized.ReLU" title="torch.nn.quantized.ReLU"><code class="xref py py-class docutils literal notranslate"><span class="pre">ReLU</span></code></a> — Rectified linear unit</p></li>
<li><p><a class="reference internal" href="torch.nn.quantized.html#torch.nn.quantized.ReLU6" title="torch.nn.quantized.ReLU6"><code class="xref py py-class docutils literal notranslate"><span class="pre">ReLU6</span></code></a> — Rectified linear unit with cut-off at
quantized representation of 6</p></li>
<li><p><a class="reference internal" href="torch.nn.quantized.html#torch.nn.quantized.ELU" title="torch.nn.quantized.ELU"><code class="xref py py-class docutils literal notranslate"><span class="pre">ELU</span></code></a> — ELU</p></li>
<li><p><a class="reference internal" href="torch.nn.quantized.html#torch.nn.quantized.Hardswish" title="torch.nn.quantized.Hardswish"><code class="xref py py-class docutils literal notranslate"><span class="pre">Hardswish</span></code></a> — Hardswish</p></li>
<li><p><a class="reference internal" href="torch.nn.quantized.html#torch.nn.quantized.BatchNorm2d" title="torch.nn.quantized.BatchNorm2d"><code class="xref py py-class docutils literal notranslate"><span class="pre">BatchNorm2d</span></code></a> — BatchNorm2d. <em>Note: this module is usually fused with Conv or Linear. Performance on ARM is not optimized</em>.</p></li>
<li><p><a class="reference internal" href="torch.nn.quantized.html#torch.nn.quantized.BatchNorm3d" title="torch.nn.quantized.BatchNorm3d"><code class="xref py py-class docutils literal notranslate"><span class="pre">BatchNorm3d</span></code></a> — BatchNorm3d. <em>Note: this module is usually fused with Conv or Linear. Performance on ARM is not optimized</em>.</p></li>
<li><p><a class="reference internal" href="torch.nn.quantized.html#torch.nn.quantized.LayerNorm" title="torch.nn.quantized.LayerNorm"><code class="xref py py-class docutils literal notranslate"><span class="pre">LayerNorm</span></code></a> — LayerNorm. <em>Note: performance on ARM is not optimized</em>.</p></li>
<li><p><a class="reference internal" href="torch.nn.quantized.html#torch.nn.quantized.GroupNorm" title="torch.nn.quantized.GroupNorm"><code class="xref py py-class docutils literal notranslate"><span class="pre">GroupNorm</span></code></a> — GroupNorm. <em>Note: performance on ARM is not optimized</em>.</p></li>
<li><p><a class="reference internal" href="torch.nn.quantized.html#torch.nn.quantized.InstanceNorm1d" title="torch.nn.quantized.InstanceNorm1d"><code class="xref py py-class docutils literal notranslate"><span class="pre">InstanceNorm1d</span></code></a> — InstanceNorm1d. <em>Note: performance on ARM is not optimized</em>.</p></li>
<li><p><a class="reference internal" href="torch.nn.quantized.html#torch.nn.quantized.InstanceNorm2d" title="torch.nn.quantized.InstanceNorm2d"><code class="xref py py-class docutils literal notranslate"><span class="pre">InstanceNorm2d</span></code></a> — InstanceNorm2d. <em>Note: performance on ARM is not optimized</em>.</p></li>
<li><p><a class="reference internal" href="torch.nn.quantized.html#torch.nn.quantized.InstanceNorm3d" title="torch.nn.quantized.InstanceNorm3d"><code class="xref py py-class docutils literal notranslate"><span class="pre">InstanceNorm3d</span></code></a> — InstanceNorm3d. <em>Note: performance on ARM is not optimized</em>.</p></li>
</ul>
</div>
<div class="section" id="torch-nn-quantized-dynamic">
<h2><cite>torch.nn.quantized.dynamic</cite><a class="headerlink" href="#torch-nn-quantized-dynamic" title="Permalink to this headline">¶</a></h2>
<p>Layers used in dynamically quantized models (i.e. quantized only on weights)</p>
<ul class="simple">
<li><p><a class="reference internal" href="torch.nn.quantized.dynamic.html#torch.nn.quantized.dynamic.Linear" title="torch.nn.quantized.dynamic.Linear"><code class="xref py py-class docutils literal notranslate"><span class="pre">Linear</span></code></a> — Linear (fully-connected) layer</p></li>
<li><p><a class="reference internal" href="torch.nn.quantized.dynamic.html#torch.nn.quantized.dynamic.LSTM" title="torch.nn.quantized.dynamic.LSTM"><code class="xref py py-class docutils literal notranslate"><span class="pre">LSTM</span></code></a> — Long-Short Term Memory RNN module</p></li>
<li><p><a class="reference internal" href="torch.nn.quantized.dynamic.html#torch.nn.quantized.dynamic.LSTMCell" title="torch.nn.quantized.dynamic.LSTMCell"><code class="xref py py-class docutils literal notranslate"><span class="pre">LSTMCell</span></code></a> — LSTM Cell</p></li>
<li><p><a class="reference internal" href="torch.nn.quantized.dynamic.html#torch.nn.quantized.dynamic.GRUCell" title="torch.nn.quantized.dynamic.GRUCell"><code class="xref py py-class docutils literal notranslate"><span class="pre">GRUCell</span></code></a> — GRU Cell</p></li>
<li><p><a class="reference internal" href="torch.nn.quantized.dynamic.html#torch.nn.quantized.dynamic.RNNCell" title="torch.nn.quantized.dynamic.RNNCell"><code class="xref py py-class docutils literal notranslate"><span class="pre">RNNCell</span></code></a> — RNN Cell</p></li>
</ul>
</div>
<div class="section" id="torch-nn-quantized-functional">
<h2><cite>torch.nn.quantized.functional</cite><a class="headerlink" href="#torch-nn-quantized-functional" title="Permalink to this headline">¶</a></h2>
<p>Functional versions of quantized NN layers (many of them accept explicit
quantization output parameters)</p>
<ul class="simple">
<li><p><a class="reference internal" href="torch.nn.quantized.html#torch.nn.quantized.functional.adaptive_avg_pool2d" title="torch.nn.quantized.functional.adaptive_avg_pool2d"><code class="xref py py-func docutils literal notranslate"><span class="pre">adaptive_avg_pool2d()</span></code></a> — 2D adaptive average pooling</p></li>
<li><p><a class="reference internal" href="torch.nn.quantized.html#torch.nn.quantized.functional.avg_pool2d" title="torch.nn.quantized.functional.avg_pool2d"><code class="xref py py-func docutils literal notranslate"><span class="pre">avg_pool2d()</span></code></a> — 2D average pooling</p></li>
<li><p><code class="xref py py-func docutils literal notranslate"><span class="pre">avg_pool3d()</span></code> — 3D average pooling</p></li>
<li><p><a class="reference internal" href="torch.nn.quantized.html#torch.nn.quantized.functional.conv1d" title="torch.nn.quantized.functional.conv1d"><code class="xref py py-func docutils literal notranslate"><span class="pre">conv1d()</span></code></a> — 1D convolution</p></li>
<li><p><a class="reference internal" href="torch.nn.quantized.html#torch.nn.quantized.functional.conv2d" title="torch.nn.quantized.functional.conv2d"><code class="xref py py-func docutils literal notranslate"><span class="pre">conv2d()</span></code></a> — 2D convolution</p></li>
<li><p><a class="reference internal" href="torch.nn.quantized.html#torch.nn.quantized.functional.conv3d" title="torch.nn.quantized.functional.conv3d"><code class="xref py py-func docutils literal notranslate"><span class="pre">conv3d()</span></code></a> — 3D convolution</p></li>
<li><p><a class="reference internal" href="torch.nn.quantized.html#torch.nn.quantized.functional.interpolate" title="torch.nn.quantized.functional.interpolate"><code class="xref py py-func docutils literal notranslate"><span class="pre">interpolate()</span></code></a> — Down-/up- sampler</p></li>
<li><p><a class="reference internal" href="torch.nn.quantized.html#torch.nn.quantized.functional.linear" title="torch.nn.quantized.functional.linear"><code class="xref py py-func docutils literal notranslate"><span class="pre">linear()</span></code></a> — Linear (fully-connected) op</p></li>
<li><p><a class="reference internal" href="torch.nn.quantized.html#torch.nn.quantized.functional.max_pool2d" title="torch.nn.quantized.functional.max_pool2d"><code class="xref py py-func docutils literal notranslate"><span class="pre">max_pool2d()</span></code></a> — 2D max pooling</p></li>
<li><p><a class="reference internal" href="torch.nn.quantized.html#torch.nn.quantized.functional.relu" title="torch.nn.quantized.functional.relu"><code class="xref py py-func docutils literal notranslate"><span class="pre">relu()</span></code></a> — Rectified linear unit</p></li>
<li><p><code class="xref py py-func docutils literal notranslate"><span class="pre">elu()</span></code> — ELU</p></li>
<li><p><code class="xref py py-func docutils literal notranslate"><span class="pre">hardsigmoid()</span></code> — Hardsigmoid</p></li>
<li><p><a class="reference internal" href="torch.nn.quantized.html#torch.nn.quantized.functional.hardswish" title="torch.nn.quantized.functional.hardswish"><code class="xref py py-func docutils literal notranslate"><span class="pre">hardswish()</span></code></a> — Hardswish</p></li>
<li><p><code class="xref py py-func docutils literal notranslate"><span class="pre">hardtanh()</span></code> — Hardtanh</p></li>
<li><p><a class="reference internal" href="torch.nn.quantized.html#torch.nn.quantized.functional.upsample" title="torch.nn.quantized.functional.upsample"><code class="xref py py-func docutils literal notranslate"><span class="pre">upsample()</span></code></a> — Upsampler. Will be
deprecated in favor of <a class="reference internal" href="torch.nn.quantized.html#torch.nn.quantized.functional.interpolate" title="torch.nn.quantized.functional.interpolate"><code class="xref py py-func docutils literal notranslate"><span class="pre">interpolate()</span></code></a></p></li>
<li><p><a class="reference internal" href="torch.nn.quantized.html#torch.nn.quantized.functional.upsample_bilinear" title="torch.nn.quantized.functional.upsample_bilinear"><code class="xref py py-func docutils literal notranslate"><span class="pre">upsample_bilinear()</span></code></a> — Bilinear
upsampler. Will be deprecated in favor of</p></li>
<li><p><a class="reference internal" href="torch.nn.quantized.html#torch.nn.quantized.functional.interpolate" title="torch.nn.quantized.functional.interpolate"><code class="xref py py-func docutils literal notranslate"><span class="pre">interpolate()</span></code></a></p></li>
<li><p><a class="reference internal" href="torch.nn.quantized.html#torch.nn.quantized.functional.upsample_nearest" title="torch.nn.quantized.functional.upsample_nearest"><code class="xref py py-func docutils literal notranslate"><span class="pre">upsample_nearest()</span></code></a> — Nearest neighbor
upsampler. Will be deprecated in favor of</p></li>
<li><p><a class="reference internal" href="torch.nn.quantized.html#torch.nn.quantized.functional.interpolate" title="torch.nn.quantized.functional.interpolate"><code class="xref py py-func docutils literal notranslate"><span class="pre">interpolate()</span></code></a></p></li>
</ul>
</div>
<div class="section" id="quantized-dtypes-and-quantization-schemes">
<h2>Quantized dtypes and quantization schemes<a class="headerlink" href="#quantized-dtypes-and-quantization-schemes" title="Permalink to this headline">¶</a></h2>
<ul class="simple">
<li><p><code class="xref py py-attr docutils literal notranslate"><span class="pre">torch.qscheme</span></code> — Type to describe the quantization scheme of a tensor.
Supported types:</p>
<ul>
<li><p><code class="xref py py-attr docutils literal notranslate"><span class="pre">torch.per_tensor_affine</span></code> — per tensor, asymmetric</p></li>
<li><p><code class="xref py py-attr docutils literal notranslate"><span class="pre">torch.per_channel_affine</span></code> — per channel, asymmetric</p></li>
<li><p><code class="xref py py-attr docutils literal notranslate"><span class="pre">torch.per_tensor_symmetric</span></code> — per tensor, symmetric</p></li>
<li><p><code class="xref py py-attr docutils literal notranslate"><span class="pre">torch.per_channel_symmetric</span></code> — per tensor, symmetric</p></li>
</ul>
</li>
<li><p><code class="docutils literal notranslate"><span class="pre">torch.dtype</span></code> — Type to describe the data. Supported types:</p>
<ul>
<li><p><code class="xref py py-attr docutils literal notranslate"><span class="pre">torch.quint8</span></code> — 8-bit unsigned integer</p></li>
<li><p><code class="xref py py-attr docutils literal notranslate"><span class="pre">torch.qint8</span></code> — 8-bit signed integer</p></li>
<li><p><code class="xref py py-attr docutils literal notranslate"><span class="pre">torch.qint32</span></code> — 32-bit signed integer</p></li>
</ul>
</li>
</ul>
</div>
</div>
</article>
</div>
<footer>
<hr>
<div role="contentinfo">
<p>
© Copyright 2019, Torch Contributors.
</p>
</div>
<div>
Built with <a href="http://sphinx-doc.org/">Sphinx</a> using a <a href="https://github.com/rtfd/sphinx_rtd_theme">theme</a> provided by <a href="https://readthedocs.org">Read the Docs</a>.
</div>
</footer>
</div>
</div>
<div class="pytorch-content-right" id="pytorch-content-right">
<div class="pytorch-right-menu" id="pytorch-right-menu">
<div class="pytorch-side-scroll" id="pytorch-side-scroll-right">
<ul>
<li><a class="reference internal" href="#">Quantization Operation coverage</a><ul>
<li><a class="reference internal" href="#quantized-torch-tensor-operations">Quantized <code class="docutils literal notranslate"><span class="pre">torch.Tensor</span></code> operations</a></li>
<li><a class="reference internal" href="#torch-nn-functional"><code class="docutils literal notranslate"><span class="pre">torch.nn.functional</span></code></a></li>
<li><a class="reference internal" href="#torch-nn-intrinsic"><code class="docutils literal notranslate"><span class="pre">torch.nn.intrinsic</span></code></a></li>
<li><a class="reference internal" href="#torch-nn-qat"><cite>torch.nn.qat</cite></a></li>
<li><a class="reference internal" href="#torch-quantization"><cite>torch.quantization</cite></a></li>
<li><a class="reference internal" href="#torch-nn-quantized"><cite>torch.nn.quantized</cite></a></li>
<li><a class="reference internal" href="#torch-nn-quantized-dynamic"><cite>torch.nn.quantized.dynamic</cite></a></li>
<li><a class="reference internal" href="#torch-nn-quantized-functional"><cite>torch.nn.quantized.functional</cite></a></li>
<li><a class="reference internal" href="#quantized-dtypes-and-quantization-schemes">Quantized dtypes and quantization schemes</a></li>
</ul>
</li>
</ul>
</div>
</div>
</div>
</section>
</div>
<script type="text/javascript" id="documentation_options" data-url_root="./" src="_static/documentation_options.js"></script>
<script src="_static/jquery.js"></script>
<script src="_static/underscore.js"></script>
<script src="_static/doctools.js"></script>
<script src="_static/language_data.js"></script>
<script type="text/javascript" src="_static/js/vendor/popper.min.js"></script>
<script type="text/javascript" src="_static/js/vendor/bootstrap.min.js"></script>
<script src="https://cdnjs.cloudflare.com/ajax/libs/list.js/1.5.0/list.min.js"></script>
<script type="text/javascript" src="_static/js/theme.js"></script>
<script type="text/javascript">
jQuery(function () {
SphinxRtdTheme.Navigation.enable(true);
});
</script>
<script>
(function(i,s,o,g,r,a,m){i['GoogleAnalyticsObject']=r;i[r]=i[r]||function(){
(i[r].q=i[r].q||[]).push(arguments)},i[r].l=1*new Date();a=s.createElement(o),
m=s.getElementsByTagName(o)[0];a.async=1;a.src=g;m.parentNode.insertBefore(a,m)
})(window,document,'script','https://www.google-analytics.com/analytics.js','ga');
ga('create', 'UA-90545585-1', 'auto');
ga('send', 'pageview');
</script>
<script async src="https://www.googletagmanager.com/gtag/js?id=UA-117752657-2"></script>
<script>
window.dataLayer = window.dataLayer || [];
function gtag(){dataLayer.push(arguments);}
gtag('js', new Date());
gtag('config', 'UA-117752657-2');
</script>
<img height="1" width="1" style="border-style:none;" alt="" src="https://www.googleadservices.com/pagead/conversion/795629140/?label=txkmCPmdtosBENSssfsC&guid=ON&script=0"/>
<!-- Begin Footer -->
<div class="container-fluid docs-tutorials-resources" id="docs-tutorials-resources">
<div class="container">
<div class="row">
<div class="col-md-4 text-center">
<h2>Docs</h2>
<p>Access comprehensive developer documentation for PyTorch</p>
<a class="with-right-arrow" href="https://pytorch.org/docs/1.7.1/index.html">View Docs</a>
</div>
<div class="col-md-4 text-center">
<h2>Tutorials</h2>
<p>Get in-depth tutorials for beginners and advanced developers</p>
<a class="with-right-arrow" href="https://pytorch.org/tutorials">View Tutorials</a>
</div>
<div class="col-md-4 text-center">
<h2>Resources</h2>
<p>Find development resources and get your questions answered</p>
<a class="with-right-arrow" href="https://pytorch.org/resources">View Resources</a>
</div>
</div>
</div>
</div>
<footer class="site-footer">
<div class="container footer-container">
<div class="footer-logo-wrapper">
<a href="https://pytorch.org/" class="footer-logo"></a>
</div>
<div class="footer-links-wrapper">
<div class="footer-links-col">
<ul>
<li class="list-title"><a href="https://pytorch.org/">PyTorch</a></li>
<li><a href="https://pytorch.org/get-started">Get Started</a></li>
<li><a href="https://pytorch.org/features">Features</a></li>
<li><a href="https://pytorch.org/ecosystem">Ecosystem</a></li>
<li><a href="https://pytorch.org/blog/">Blog</a></li>
<li><a href="https://github.com/pytorch/pytorch/blob/master/CONTRIBUTING.md">Contributing</a></li>
</ul>
</div>
<div class="footer-links-col">
<ul>
<li class="list-title"><a href="https://pytorch.org/resources">Resources</a></li>
<li><a href="https://pytorch.org/tutorials">Tutorials</a></li>
<li><a href="https://pytorch.org/docs/1.7.1/index.html">Docs</a></li>
<li><a href="https://discuss.pytorch.org" target="_blank">Discuss</a></li>
<li><a href="https://github.com/pytorch/pytorch/issues" target="_blank">Github Issues</a></li>
<li><a href="https://pytorch.org/assets/brand-guidelines/PyTorch-Brand-Guidelines.pdf" target="_blank">Brand Guidelines</a></li>
</ul>
</div>
<div class="footer-links-col follow-us-col">
<ul>
<li class="list-title">Stay Connected</li>
<li>
<div id="mc_embed_signup">
<form
action="https://twitter.us14.list-manage.com/subscribe/post?u=75419c71fe0a935e53dfa4a3f&id=91d0dccd39"
method="post"
id="mc-embedded-subscribe-form"
name="mc-embedded-subscribe-form"
class="email-subscribe-form validate"
target="_blank"
novalidate>
<div id="mc_embed_signup_scroll" class="email-subscribe-form-fields-wrapper">
<div class="mc-field-group">
<label for="mce-EMAIL" style="display:none;">Email Address</label>
<input type="email" value="" name="EMAIL" class="required email" id="mce-EMAIL" placeholder="Email Address">
</div>
<div id="mce-responses" class="clear">
<div class="response" id="mce-error-response" style="display:none"></div>
<div class="response" id="mce-success-response" style="display:none"></div>
</div> <!-- real people should not fill this in and expect good things - do not remove this or risk form bot signups-->
<div style="position: absolute; left: -5000px;" aria-hidden="true"><input type="text" name="b_75419c71fe0a935e53dfa4a3f_91d0dccd39" tabindex="-1" value=""></div>
<div class="clear">
<input type="submit" value="" name="subscribe" id="mc-embedded-subscribe" class="button email-subscribe-button">
</div>
</div>
</form>
</div>
</li>
</ul>
<div class="footer-social-icons">
<a href="https://www.facebook.com/pytorch" target="_blank" class="facebook"></a>
<a href="https://twitter.com/pytorch" target="_blank" class="twitter"></a>
<a href="https://www.youtube.com/pytorch" target="_blank" class="youtube"></a>
</div>
</div>
</div>
</div>
</footer>
<div class="cookie-banner-wrapper">
<div class="container">
<p class="gdpr-notice">To analyze traffic and optimize your experience, we serve cookies on this site. By clicking or navigating, you agree to allow our usage of cookies. As the current maintainers of this site, Facebook’s Cookies Policy applies. Learn more, including about available controls: <a href="https://www.facebook.com/policies/cookies/">Cookies Policy</a>.</p>
<img class="close-button" src="_static/images/pytorch-x.svg">
</div>
</div>
<!-- End Footer -->
<!-- Begin Mobile Menu -->
<div class="mobile-main-menu">
<div class="container-fluid">
<div class="container">
<div class="mobile-main-menu-header-container">
<a class="header-logo" href="https://pytorch.org/" aria-label="PyTorch"></a>
<a class="main-menu-close-button" href="#" data-behavior="close-mobile-menu"></a>
</div>
</div>
</div>
<div class="mobile-main-menu-links-container">
<div class="main-menu">
<ul>
<li>
<a href="https://pytorch.org/get-started">Get Started</a>
</li>
<li>
<a href="https://pytorch.org/ecosystem">Ecosystem</a>
</li>
<li>
<a href="https://pytorch.org/mobile">Mobile</a>
</li>
<li>
<a href="https://pytorch.org/hub">PyTorch Hub</a>
</li>
<li>
<a href="https://pytorch.org/blog/">Blog</a>
</li>
<li>
<a href="https://pytorch.org/tutorials">Tutorials</a>
</li>
<li class="resources-mobile-menu-title" class="active">
Docs
</li>
<ul class="resources-mobile-menu-items">
<li>
<a href="https://pytorch.org/docs/1.7.1/index.html">PyTorch</a>
</li>
<li>
<a href="https://pytorch.org/audio/0.7.0/index.html">torchaudio</a>
</li>
<li>
<a href="https://pytorch.org/text/0.8.1/index.html">torchtext</a>
</li>
<li>
<a href="https://pytorch.org/vision/0.8/">torchvision</a>
</li>
<li>
<a href="https://pytorch.org/elastic/">TorchElastic</a>