-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathfaq.html
803 lines (609 loc) · 44.9 KB
/
faq.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
<!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>Frequently Asked Questions — PyTorch 1.7.0 documentation</title>
<link rel="canonical" href="https://pytorch.org/docs/stable/notes/faq.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" />
<link rel="next" title="Features for large-scale deployments" href="large_scale_deployments.html" />
<link rel="prev" title="Extending PyTorch" href="extending.html" />
<script src="../_static/js/modernizr.min.js"></script>
<!-- Preload the theme fonts -->
<link rel="preload" href="../_static/fonts/FreightSans/freight-sans-book.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="../_static/fonts/FreightSans/freight-sans-medium.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="../_static/fonts/IBMPlexMono/IBMPlexMono-Medium.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="../_static/fonts/FreightSans/freight-sans-bold.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="../_static/fonts/FreightSans/freight-sans-medium-italic.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="../_static/fonts/IBMPlexMono/IBMPlexMono-SemiBold.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<!-- Preload the katex fonts -->
<link rel="preload" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/fonts/KaTeX_Math-Italic.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/fonts/KaTeX_Main-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/fonts/KaTeX_Main-Bold.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/fonts/KaTeX_Size1-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/fonts/KaTeX_Size4-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/fonts/KaTeX_Size2-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/fonts/KaTeX_Size3-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/fonts/KaTeX_Caligraphic-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
</head>
<div class="container-fluid header-holder tutorials-header" id="header-holder">
<div class="container">
<div class="header-container">
<a class="header-logo" href="https://pytorch.org/" aria-label="PyTorch"></a>
<div class="main-menu">
<ul>
<li>
<a href="https://pytorch.org/get-started">Get Started</a>
</li>
<li>
<div class="ecosystem-dropdown">
<a id="dropdownMenuButton" data-toggle="ecosystem-dropdown">
Ecosystem
</a>
<div class="ecosystem-dropdown-menu">
<a class="nav-dropdown-item" href="https://pytorch.org/hub"">
<span class=dropdown-title>Models (Beta)</span>
<p>Discover, publish, and reuse pre-trained models</p>
</a>
<a class="nav-dropdown-item" href="https://pytorch.org/ecosystem">
<span class=dropdown-title>Tools & Libraries</span>
<p>Explore the ecosystem of tools and libraries</p>
</a>
</div>
</div>
</li>
<li>
<a href="https://pytorch.org/mobile">Mobile</a>
</li>
<li>
<a href="https://pytorch.org/blog/">Blog</a>
</li>
<li>
<a href="https://pytorch.org/tutorials">Tutorials</a>
</li>
<li class="active">
<a href="https://pytorch.org/docs/stable/index.html">Docs</a>
</li>
<li>
<div class="resources-dropdown">
<a id="resourcesDropdownButton" data-toggle="resources-dropdown">
Resources
</a>
<div class="resources-dropdown-menu">
<a class="nav-dropdown-item" href="https://pytorch.org/resources"">
<span class=dropdown-title>Developer Resources</span>
<p>Find resources and get questions answered</p>
</a>
<a class="nav-dropdown-item" href="https://pytorch.org/features">
<span class=dropdown-title>About</span>
<p>Learn about PyTorch’s features and capabilities</p>
</a>
</div>
</div>
</li>
<li>
<a href="https://github.com/pytorch/pytorch">Github</a>
</li>
</ul>
</div>
<a class="main-menu-open-button" href="#" data-behavior="open-mobile-menu"></a>
</div>
</div>
</div>
<body class="pytorch-body">
<div class="table-of-contents-link-wrapper">
<span>Table of Contents</span>
<a href="#" class="toggle-table-of-contents" data-behavior="toggle-table-of-contents"></a>
</div>
<nav data-toggle="wy-nav-shift" class="pytorch-left-menu" id="pytorch-left-menu">
<div class="pytorch-side-scroll">
<div class="pytorch-menu pytorch-menu-vertical" data-spy="affix" role="navigation" aria-label="main navigation">
<div class="pytorch-left-menu-search">
<div class="version">
<a href='http://pytorch.org/docs/versions.html'>1.7.0 ▼</a>
</div>
<div role="search">
<form id="rtd-search-form" class="wy-form" action="../search.html" method="get">
<input type="text" name="q" placeholder="Search Docs" />
<input type="hidden" name="check_keywords" value="yes" />
<input type="hidden" name="area" value="default" />
</form>
</div>
</div>
<p class="caption"><span class="caption-text">Notes</span></p>
<ul class="current">
<li class="toctree-l1"><a class="reference internal" href="amp_examples.html">Automatic Mixed Precision examples</a></li>
<li class="toctree-l1"><a class="reference internal" href="autograd.html">Autograd mechanics</a></li>
<li class="toctree-l1"><a class="reference internal" href="broadcasting.html">Broadcasting semantics</a></li>
<li class="toctree-l1"><a class="reference internal" href="cpu_threading_torchscript_inference.html">CPU threading and TorchScript inference</a></li>
<li class="toctree-l1"><a class="reference internal" href="cuda.html">CUDA semantics</a></li>
<li class="toctree-l1"><a class="reference internal" href="ddp.html">Distributed Data Parallel</a></li>
<li class="toctree-l1"><a class="reference internal" href="extending.html">Extending PyTorch</a></li>
<li class="toctree-l1 current"><a class="current reference internal" href="#">Frequently Asked Questions</a></li>
<li class="toctree-l1"><a class="reference internal" href="large_scale_deployments.html">Features for large-scale deployments</a></li>
<li class="toctree-l1"><a class="reference internal" href="multiprocessing.html">Multiprocessing best practices</a></li>
<li class="toctree-l1"><a class="reference internal" href="randomness.html">Reproducibility</a></li>
<li class="toctree-l1"><a class="reference internal" href="serialization.html">Serialization semantics</a></li>
<li class="toctree-l1"><a class="reference internal" href="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>Frequently Asked Questions</li>
<li class="pytorch-breadcrumbs-aside">
<a href="../_sources/notes/faq.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="frequently-asked-questions">
<h1>Frequently Asked Questions<a class="headerlink" href="#frequently-asked-questions" title="Permalink to this headline">¶</a></h1>
<div class="section" id="my-model-reports-cuda-runtime-error-2-out-of-memory">
<h2>My model reports “cuda runtime error(2): out of memory”<a class="headerlink" href="#my-model-reports-cuda-runtime-error-2-out-of-memory" title="Permalink to this headline">¶</a></h2>
<p>As the error message suggests, you have run out of memory on your
GPU. Since we often deal with large amounts of data in PyTorch,
small mistakes can rapidly cause your program to use up all of your
GPU; fortunately, the fixes in these cases are often simple.
Here are a few common things to check:</p>
<p><strong>Don’t accumulate history across your training loop.</strong>
By default, computations involving variables that require gradients
will keep history. This means that you should avoid using such
variables in computations which will live beyond your training loops,
e.g., when tracking statistics. Instead, you should detach the variable
or access its underlying data.</p>
<p>Sometimes, it can be non-obvious when differentiable variables can
occur. Consider the following training loop (abridged from <a class="reference external" href="https://discuss.pytorch.org/t/high-memory-usage-while-training/162">source</a>):</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="n">total_loss</span> <span class="o">=</span> <span class="mi">0</span>
<span class="k">for</span> <span class="n">i</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="mi">10000</span><span class="p">):</span>
<span class="n">optimizer</span><span class="o">.</span><span class="n">zero_grad</span><span class="p">()</span>
<span class="n">output</span> <span class="o">=</span> <span class="n">model</span><span class="p">(</span><span class="nb">input</span><span class="p">)</span>
<span class="n">loss</span> <span class="o">=</span> <span class="n">criterion</span><span class="p">(</span><span class="n">output</span><span class="p">)</span>
<span class="n">loss</span><span class="o">.</span><span class="n">backward</span><span class="p">()</span>
<span class="n">optimizer</span><span class="o">.</span><span class="n">step</span><span class="p">()</span>
<span class="n">total_loss</span> <span class="o">+=</span> <span class="n">loss</span>
</pre></div>
</div>
<p>Here, <code class="docutils literal notranslate"><span class="pre">total_loss</span></code> is accumulating history across your training loop, since
<code class="docutils literal notranslate"><span class="pre">loss</span></code> is a differentiable variable with autograd history. You can fix this by
writing <cite>total_loss += float(loss)</cite> instead.</p>
<p>Other instances of this problem:
<a class="reference external" href="https://discuss.pytorch.org/t/resolved-gpu-out-of-memory-error-with-batch-size-1/3719">1</a>.</p>
<p><strong>Don’t hold onto tensors and variables you don’t need.</strong>
If you assign a Tensor or Variable to a local, Python will not
deallocate until the local goes out of scope. You can free
this reference by using <code class="docutils literal notranslate"><span class="pre">del</span> <span class="pre">x</span></code>. Similarly, if you assign
a Tensor or Variable to a member variable of an object, it will
not deallocate until the object goes out of scope. You will
get the best memory usage if you don’t hold onto temporaries
you don’t need.</p>
<p>The scopes of locals can be larger than you expect. For example:</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="k">for</span> <span class="n">i</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="mi">5</span><span class="p">):</span>
<span class="n">intermediate</span> <span class="o">=</span> <span class="n">f</span><span class="p">(</span><span class="nb">input</span><span class="p">[</span><span class="n">i</span><span class="p">])</span>
<span class="n">result</span> <span class="o">+=</span> <span class="n">g</span><span class="p">(</span><span class="n">intermediate</span><span class="p">)</span>
<span class="n">output</span> <span class="o">=</span> <span class="n">h</span><span class="p">(</span><span class="n">result</span><span class="p">)</span>
<span class="k">return</span> <span class="n">output</span>
</pre></div>
</div>
<p>Here, <code class="docutils literal notranslate"><span class="pre">intermediate</span></code> remains live even while <code class="docutils literal notranslate"><span class="pre">h</span></code> is executing,
because its scope extrudes past the end of the loop. To free it
earlier, you should <code class="docutils literal notranslate"><span class="pre">del</span> <span class="pre">intermediate</span></code> when you are done with it.</p>
<p><strong>Don’t run RNNs on sequences that are too large.</strong>
The amount of memory required to backpropagate through an RNN scales
linearly with the length of the RNN input; thus, you will run out of memory
if you try to feed an RNN a sequence that is too long.</p>
<p>The technical term for this phenomenon is <a class="reference external" href="https://en.wikipedia.org/wiki/Backpropagation_through_time">backpropagation through time</a>,
and there are plenty of references for how to implement truncated
BPTT, including in the <a class="reference external" href="https://github.com/pytorch/examples/tree/master/word_language_model">word language model</a> example; truncation is handled by the
<code class="docutils literal notranslate"><span class="pre">repackage</span></code> function as described in
<a class="reference external" href="https://discuss.pytorch.org/t/help-clarifying-repackage-hidden-in-word-language-model/226">this forum post</a>.</p>
<p><strong>Don’t use linear layers that are too large.</strong>
A linear layer <code class="docutils literal notranslate"><span class="pre">nn.Linear(m,</span> <span class="pre">n)</span></code> uses <span class="math"><span class="katex"><span class="katex-mathml"><math xmlns="http://www.w3.org/1998/Math/MathML"><semantics><mrow><mi>O</mi><mo stretchy="false">(</mo><mi>n</mi><mi>m</mi><mo stretchy="false">)</mo></mrow><annotation encoding="application/x-tex">O(nm)</annotation></semantics></math></span><span class="katex-html" aria-hidden="true"><span class="base"><span class="strut" style="height:1em;vertical-align:-0.25em;"></span><span class="mord mathnormal" style="margin-right:0.02778em;">O</span><span class="mopen">(</span><span class="mord mathnormal">n</span><span class="mord mathnormal">m</span><span class="mclose">)</span></span></span></span>
</span> memory: that is to say,
the memory requirements of the weights
scales quadratically with the number of features. It is very easy
to <a class="reference external" href="https://github.com/pytorch/pytorch/issues/958">blow through your memory</a>
this way (and remember that you will need at least twice the size of the
weights, since you also need to store the gradients.)</p>
</div>
<div class="section" id="my-gpu-memory-isn-t-freed-properly">
<h2>My GPU memory isn’t freed properly<a class="headerlink" href="#my-gpu-memory-isn-t-freed-properly" title="Permalink to this headline">¶</a></h2>
<p>PyTorch uses a caching memory allocator to speed up memory allocations. As a
result, the values shown in <code class="docutils literal notranslate"><span class="pre">nvidia-smi</span></code> usually don’t reflect the true
memory usage. See <a class="reference internal" href="cuda.html#cuda-memory-management"><span class="std std-ref">Memory management</span></a> for more details about GPU
memory management.</p>
<p>If your GPU memory isn’t freed even after Python quits, it is very likely that
some Python subprocesses are still alive. You may find them via
<code class="docutils literal notranslate"><span class="pre">ps</span> <span class="pre">-elf</span> <span class="pre">|</span> <span class="pre">grep</span> <span class="pre">python</span></code> and manually kill them with <code class="docutils literal notranslate"><span class="pre">kill</span> <span class="pre">-9</span> <span class="pre">[pid]</span></code>.</p>
</div>
<div class="section" id="my-out-of-memory-exception-handler-can-t-allocate-memory">
<h2>My out of memory exception handler can’t allocate memory<a class="headerlink" href="#my-out-of-memory-exception-handler-can-t-allocate-memory" title="Permalink to this headline">¶</a></h2>
<p>You may have some code that tries to recover from out of memory errors.</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="k">try</span><span class="p">:</span>
<span class="n">run_model</span><span class="p">(</span><span class="n">batch_size</span><span class="p">)</span>
<span class="k">except</span> <span class="ne">RuntimeError</span><span class="p">:</span> <span class="c1"># Out of memory</span>
<span class="k">for</span> <span class="n">_</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="n">batch_size</span><span class="p">):</span>
<span class="n">run_model</span><span class="p">(</span><span class="mi">1</span><span class="p">)</span>
</pre></div>
</div>
<p>But find that when you do run out of memory, your recovery code can’t allocate
either. That’s because the python exception object holds a reference to the
stack frame where the error was raised. Which prevents the original tensor
objects from being freed. The solution is to move you OOM recovery code outside
of the <code class="docutils literal notranslate"><span class="pre">except</span></code> clause.</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="n">oom</span> <span class="o">=</span> <span class="kc">False</span>
<span class="k">try</span><span class="p">:</span>
<span class="n">run_model</span><span class="p">(</span><span class="n">batch_size</span><span class="p">)</span>
<span class="k">except</span> <span class="ne">RuntimeError</span><span class="p">:</span> <span class="c1"># Out of memory</span>
<span class="n">oom</span> <span class="o">=</span> <span class="kc">True</span>
<span class="k">if</span> <span class="n">oom</span><span class="p">:</span>
<span class="k">for</span> <span class="n">_</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="n">batch_size</span><span class="p">):</span>
<span class="n">run_model</span><span class="p">(</span><span class="mi">1</span><span class="p">)</span>
</pre></div>
</div>
</div>
<div class="section" id="my-data-loader-workers-return-identical-random-numbers">
<span id="dataloader-workers-random-seed"></span><h2>My data loader workers return identical random numbers<a class="headerlink" href="#my-data-loader-workers-return-identical-random-numbers" title="Permalink to this headline">¶</a></h2>
<p>You are likely using other libraries to generate random numbers in the dataset.
For example, NumPy’s RNG is duplicated when worker subprocesses are started via
<code class="docutils literal notranslate"><span class="pre">fork</span></code>. See <a class="reference internal" href="../data.html#torch.utils.data.DataLoader" title="torch.utils.data.DataLoader"><code class="xref py py-class docutils literal notranslate"><span class="pre">torch.utils.data.DataLoader</span></code></a>’s documentation for how to
properly set up random seeds in workers with its <code class="xref py py-attr docutils literal notranslate"><span class="pre">worker_init_fn</span></code> option.</p>
</div>
<div class="section" id="my-recurrent-network-doesn-t-work-with-data-parallelism">
<span id="pack-rnn-unpack-with-data-parallelism"></span><h2>My recurrent network doesn’t work with data parallelism<a class="headerlink" href="#my-recurrent-network-doesn-t-work-with-data-parallelism" title="Permalink to this headline">¶</a></h2>
<p>There is a subtlety in using the
<code class="docutils literal notranslate"><span class="pre">pack</span> <span class="pre">sequence</span> <span class="pre">-></span> <span class="pre">recurrent</span> <span class="pre">network</span> <span class="pre">-></span> <span class="pre">unpack</span> <span class="pre">sequence</span></code> pattern in a
<a class="reference internal" href="../generated/torch.nn.Module.html#torch.nn.Module" title="torch.nn.Module"><code class="xref py py-class docutils literal notranslate"><span class="pre">Module</span></code></a> with <a class="reference internal" href="../generated/torch.nn.DataParallel.html#torch.nn.DataParallel" title="torch.nn.DataParallel"><code class="xref py py-class docutils literal notranslate"><span class="pre">DataParallel</span></code></a> or
<a class="reference internal" href="../nn.functional.html#torch.nn.parallel.data_parallel" title="torch.nn.parallel.data_parallel"><code class="xref py py-func docutils literal notranslate"><span class="pre">data_parallel()</span></code></a>. Input to each the <code class="xref py py-meth docutils literal notranslate"><span class="pre">forward()</span></code> on
each device will only be part of the entire input. Because the unpack operation
<a class="reference internal" href="../generated/torch.nn.utils.rnn.pad_packed_sequence.html#torch.nn.utils.rnn.pad_packed_sequence" title="torch.nn.utils.rnn.pad_packed_sequence"><code class="xref py py-func docutils literal notranslate"><span class="pre">torch.nn.utils.rnn.pad_packed_sequence()</span></code></a> by default only pads up to the
longest input it sees, i.e., the longest on that particular device, size
mismatches will happen when results are gathered together. Therefore, you can
instead take advantage of the <code class="xref py py-attr docutils literal notranslate"><span class="pre">total_length</span></code> argument of
<a class="reference internal" href="../generated/torch.nn.utils.rnn.pad_packed_sequence.html#torch.nn.utils.rnn.pad_packed_sequence" title="torch.nn.utils.rnn.pad_packed_sequence"><code class="xref py py-func docutils literal notranslate"><span class="pre">pad_packed_sequence()</span></code></a> to make sure that the
<code class="xref py py-meth docutils literal notranslate"><span class="pre">forward()</span></code> calls return sequences of same length. For example, you can
write:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="kn">from</span> <span class="nn">torch.nn.utils.rnn</span> <span class="kn">import</span> <span class="n">pack_padded_sequence</span><span class="p">,</span> <span class="n">pad_packed_sequence</span>
<span class="k">class</span> <span class="nc">MyModule</span><span class="p">(</span><span class="n">nn</span><span class="o">.</span><span class="n">Module</span><span class="p">):</span>
<span class="c1"># ... __init__, other methods, etc.</span>
<span class="c1"># padded_input is of shape [B x T x *] (batch_first mode) and contains</span>
<span class="c1"># the sequences sorted by lengths</span>
<span class="c1"># B is the batch size</span>
<span class="c1"># T is max sequence length</span>
<span class="k">def</span> <span class="nf">forward</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">padded_input</span><span class="p">,</span> <span class="n">input_lengths</span><span class="p">):</span>
<span class="n">total_length</span> <span class="o">=</span> <span class="n">padded_input</span><span class="o">.</span><span class="n">size</span><span class="p">(</span><span class="mi">1</span><span class="p">)</span> <span class="c1"># get the max sequence length</span>
<span class="n">packed_input</span> <span class="o">=</span> <span class="n">pack_padded_sequence</span><span class="p">(</span><span class="n">padded_input</span><span class="p">,</span> <span class="n">input_lengths</span><span class="p">,</span>
<span class="n">batch_first</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span>
<span class="n">packed_output</span><span class="p">,</span> <span class="n">_</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">my_lstm</span><span class="p">(</span><span class="n">packed_input</span><span class="p">)</span>
<span class="n">output</span><span class="p">,</span> <span class="n">_</span> <span class="o">=</span> <span class="n">pad_packed_sequence</span><span class="p">(</span><span class="n">packed_output</span><span class="p">,</span> <span class="n">batch_first</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span>
<span class="n">total_length</span><span class="o">=</span><span class="n">total_length</span><span class="p">)</span>
<span class="k">return</span> <span class="n">output</span>
<span class="n">m</span> <span class="o">=</span> <span class="n">MyModule</span><span class="p">()</span><span class="o">.</span><span class="n">cuda</span><span class="p">()</span>
<span class="n">dp_m</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">DataParallel</span><span class="p">(</span><span class="n">m</span><span class="p">)</span>
</pre></div>
</div>
<p>Additionally, extra care needs to be taken when batch dimension is dim <code class="docutils literal notranslate"><span class="pre">1</span></code>
(i.e., <code class="docutils literal notranslate"><span class="pre">batch_first=False</span></code>) with data parallelism. In this case, the first
argument of pack_padded_sequence <code class="docutils literal notranslate"><span class="pre">padding_input</span></code> will be of shape
<code class="docutils literal notranslate"><span class="pre">[T</span> <span class="pre">x</span> <span class="pre">B</span> <span class="pre">x</span> <span class="pre">*]</span></code> and should be scattered along dim <code class="docutils literal notranslate"><span class="pre">1</span></code>, but the second argument
<code class="docutils literal notranslate"><span class="pre">input_lengths</span></code> will be of shape <code class="docutils literal notranslate"><span class="pre">[B]</span></code> and should be scattered along dim
<code class="docutils literal notranslate"><span class="pre">0</span></code>. Extra code to manipulate the tensor shapes will be needed.</p>
</div>
</div>
</article>
</div>
<footer>
<div class="rst-footer-buttons" role="navigation" aria-label="footer navigation">
<a href="large_scale_deployments.html" class="btn btn-neutral float-right" title="Features for large-scale deployments" accesskey="n" rel="next">Next <img src="../_static/images/chevron-right-orange.svg" class="next-page"></a>
<a href="extending.html" class="btn btn-neutral" title="Extending PyTorch" accesskey="p" rel="prev"><img src="../_static/images/chevron-right-orange.svg" class="previous-page"> Previous</a>
</div>
<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="#">Frequently Asked Questions</a><ul>
<li><a class="reference internal" href="#my-model-reports-cuda-runtime-error-2-out-of-memory">My model reports “cuda runtime error(2): out of memory”</a></li>
<li><a class="reference internal" href="#my-gpu-memory-isn-t-freed-properly">My GPU memory isn’t freed properly</a></li>
<li><a class="reference internal" href="#my-out-of-memory-exception-handler-can-t-allocate-memory">My out of memory exception handler can’t allocate memory</a></li>
<li><a class="reference internal" href="#my-data-loader-workers-return-identical-random-numbers">My data loader workers return identical random numbers</a></li>
<li><a class="reference internal" href="#my-recurrent-network-doesn-t-work-with-data-parallelism">My recurrent network doesn’t work with data parallelism</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/stable/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/stable/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/features">Features</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="active">
<a href="https://pytorch.org/docs/stable/index.html">Docs</a>
</li>
<li>
<a href="https://pytorch.org/resources">Resources</a>
</li>
<li>
<a href="https://github.com/pytorch/pytorch">Github</a>
</li>
</ul>
</div>
</div>
</div>
<!-- End Mobile Menu -->
<script type="text/javascript" src="../_static/js/vendor/anchor.min.js"></script>
<script type="text/javascript">
$(document).ready(function() {
mobileMenu.bind();
mobileTOC.bind();
pytorchAnchors.bind();
sideMenus.bind();
scrollToAnchor.bind();
highlightNavigation.bind();
mainMenuDropdown.bind();
filterTags.bind();
// Add class to links that have code blocks, since we cannot create links in code blocks
$("article.pytorch-article a span.pre").each(function(e) {
$(this).closest("a").addClass("has-code");
});
})
</script>
</body>
</html>