forked from pytorch/pytorch.github.io
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathmultiprocessing.html
691 lines (515 loc) · 36.5 KB
/
multiprocessing.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
<!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>Multiprocessing best practices — PyTorch master documentation</title>
<link rel="canonical" href="https://pytorch.org/docs/stable/notes/multiprocessing.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/katex@0.10.0-beta/dist/katex.min.css" type="text/css" />
<link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/katex@0.10.2/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="Reproducibility" href="randomness.html" />
<link rel="prev" title="Features for large-scale deployments" href="large_scale_deployments.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/katex@0.10.0/dist/fonts/KaTeX_Math-Italic.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/katex@0.10.0/dist/fonts/KaTeX_Main-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/katex@0.10.0/dist/fonts/KaTeX_Main-Bold.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/katex@0.10.0/dist/fonts/KaTeX_Size1-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/katex@0.10.0/dist/fonts/KaTeX_Size4-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/katex@0.10.0/dist/fonts/KaTeX_Size2-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/katex@0.10.0/dist/fonts/KaTeX_Size3-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/katex@0.10.0/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/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://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>
<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'>1.3.0 ▼</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="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="extending.html">Extending PyTorch</a></li>
<li class="toctree-l1"><a class="reference internal" href="faq.html">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 current"><a class="current reference internal" href="#">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>
<li class="toctree-l1"><a class="reference external" href="* `PyTorch on XLA Devices <http://pytorch.org/xla/>`_">* `PyTorch on XLA Devices <http://pytorch.org/xla/>`_</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>
<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="../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="../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="../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="../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="../quantization.html">Quantization</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="../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">torchaudio Reference</span></p>
<ul>
<li class="toctree-l1"><a class="reference external" href="* `torchaudio <https://pytorch.org/audio>`_">* `torchaudio <https://pytorch.org/audio>`_</a></li>
</ul>
<p class="caption"><span class="caption-text">torchtext Reference</span></p>
<ul>
<li class="toctree-l1"><a class="reference external" href="* `torchtext <https://pytorch.org/text>`_">* `torchtext <https://pytorch.org/text>`_</a></li>
</ul>
<p class="caption"><span class="caption-text">Other Languages</span></p>
<ul>
<li class="toctree-l1"><a class="reference external" href="* `C++ API <https://pytorch.org/cppdocs/>`_">* `C++ API <https://pytorch.org/cppdocs/>`_</a></li>
<li class="toctree-l1"><a class="reference internal" href="../packages.html">Javadoc</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>Multiprocessing best practices</li>
<li class="pytorch-breadcrumbs-aside">
<a href="../_sources/notes/multiprocessing.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="multiprocessing-best-practices">
<span id="id1"></span><h1>Multiprocessing best practices<a class="headerlink" href="#multiprocessing-best-practices" title="Permalink to this headline">¶</a></h1>
<p><a class="reference internal" href="../multiprocessing.html#module-torch.multiprocessing" title="torch.multiprocessing"><code class="xref py py-mod docutils literal notranslate"><span class="pre">torch.multiprocessing</span></code></a> is a drop in replacement for Python’s
<a class="reference external" href="https://docs.python.org/3/library/multiprocessing.html#module-multiprocessing" title="(in Python v3.7)"><code class="docutils literal notranslate"><span class="pre">multiprocessing</span></code></a> module. It supports the exact same operations,
but extends it, so that all tensors sent through a
<a class="reference external" href="https://docs.python.org/3/library/multiprocessing.html#multiprocessing.Queue" title="(in Python v3.7)"><code class="docutils literal notranslate"><span class="pre">multiprocessing.Queue</span></code></a>, will have their data moved into shared
memory and will only send a handle to another process.</p>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>When a <a class="reference internal" href="../tensors.html#torch.Tensor" title="torch.Tensor"><code class="xref py py-class docutils literal notranslate"><span class="pre">Tensor</span></code></a> is sent to another process, the
<a class="reference internal" href="../tensors.html#torch.Tensor" title="torch.Tensor"><code class="xref py py-class docutils literal notranslate"><span class="pre">Tensor</span></code></a> data is shared. If <a class="reference internal" href="../tensors.html#torch.Tensor.grad" title="torch.Tensor.grad"><code class="xref py py-attr docutils literal notranslate"><span class="pre">torch.Tensor.grad</span></code></a> is
not <code class="docutils literal notranslate"><span class="pre">None</span></code>, it is also shared. After a <a class="reference internal" href="../tensors.html#torch.Tensor" title="torch.Tensor"><code class="xref py py-class docutils literal notranslate"><span class="pre">Tensor</span></code></a> without
a <a class="reference internal" href="../tensors.html#torch.Tensor.grad" title="torch.Tensor.grad"><code class="xref py py-attr docutils literal notranslate"><span class="pre">torch.Tensor.grad</span></code></a> field is sent to the other process, it
creates a standard process-specific <code class="docutils literal notranslate"><span class="pre">.grad</span></code> <a class="reference internal" href="../tensors.html#torch.Tensor" title="torch.Tensor"><code class="xref py py-class docutils literal notranslate"><span class="pre">Tensor</span></code></a> that
is not automatically shared across all processes, unlike how the
<a class="reference internal" href="../tensors.html#torch.Tensor" title="torch.Tensor"><code class="xref py py-class docutils literal notranslate"><span class="pre">Tensor</span></code></a>’s data has been shared.</p>
</div>
<p>This allows to implement various training methods, like Hogwild, A3C, or any
others that require asynchronous operation.</p>
<div class="section" id="cuda-in-multiprocessing">
<span id="multiprocessing-cuda-note"></span><h2>CUDA in multiprocessing<a class="headerlink" href="#cuda-in-multiprocessing" title="Permalink to this headline">¶</a></h2>
<p>The CUDA runtime does not support the <code class="docutils literal notranslate"><span class="pre">fork</span></code> start method. However,
<a class="reference external" href="https://docs.python.org/3/library/multiprocessing.html#module-multiprocessing" title="(in Python v3.7)"><code class="docutils literal notranslate"><span class="pre">multiprocessing</span></code></a> in Python 2 can only create subprocesses using
<code class="docutils literal notranslate"><span class="pre">fork</span></code>. So Python 3 and either <code class="docutils literal notranslate"><span class="pre">spawn</span></code> or <code class="docutils literal notranslate"><span class="pre">forkserver</span></code> start method are
required to use CUDA in subprocesses.</p>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>The start method can be set via either creating a context with
<code class="docutils literal notranslate"><span class="pre">multiprocessing.get_context(...)</span></code> or directly using
<code class="docutils literal notranslate"><span class="pre">multiprocessing.set_start_method(...)</span></code>.</p>
</div>
<p>Unlike CPU tensors, the sending process is required to keep the original tensor
as long as the receiving process retains a copy of the tensor. It is implemented
under the hood but requires users to follow the best practices for the program
to run correctly. For example, the sending process must stay alive as long as
the consumer process has references to the tensor, and the refcounting can not
save you if the consumer process exits abnormally via a fatal signal. See
<a class="reference internal" href="../multiprocessing.html#multiprocessing-cuda-sharing-details"><span class="std std-ref">this section</span></a>.</p>
<p>See also: <a class="reference internal" href="cuda.html#cuda-nn-dataparallel-instead"><span class="std std-ref">Use nn.DataParallel instead of multiprocessing</span></a></p>
</div>
<div class="section" id="best-practices-and-tips">
<h2>Best practices and tips<a class="headerlink" href="#best-practices-and-tips" title="Permalink to this headline">¶</a></h2>
<div class="section" id="avoiding-and-fighting-deadlocks">
<h3>Avoiding and fighting deadlocks<a class="headerlink" href="#avoiding-and-fighting-deadlocks" title="Permalink to this headline">¶</a></h3>
<p>There are a lot of things that can go wrong when a new process is spawned, with
the most common cause of deadlocks being background threads. If there’s any
thread that holds a lock or imports a module, and <code class="docutils literal notranslate"><span class="pre">fork</span></code> is called, it’s very
likely that the subprocess will be in a corrupted state and will deadlock or
fail in a different way. Note that even if you don’t, Python built in
libraries do - no need to look further than <a class="reference external" href="https://docs.python.org/3/library/multiprocessing.html#module-multiprocessing" title="(in Python v3.7)"><code class="docutils literal notranslate"><span class="pre">multiprocessing</span></code></a>.
<a class="reference external" href="https://docs.python.org/3/library/multiprocessing.html#multiprocessing.Queue" title="(in Python v3.7)"><code class="docutils literal notranslate"><span class="pre">multiprocessing.Queue</span></code></a> is actually a very complex class, that
spawns multiple threads used to serialize, send and receive objects, and they
can cause aforementioned problems too. If you find yourself in such situation
try using a <code class="xref py py-class docutils literal notranslate"><span class="pre">multiprocessing.queues.SimpleQueue</span></code>, that doesn’t
use any additional threads.</p>
<p>We’re trying our best to make it easy for you and ensure these deadlocks don’t
happen but some things are out of our control. If you have any issues you can’t
cope with for a while, try reaching out on forums, and we’ll see if it’s an
issue we can fix.</p>
</div>
<div class="section" id="reuse-buffers-passed-through-a-queue">
<h3>Reuse buffers passed through a Queue<a class="headerlink" href="#reuse-buffers-passed-through-a-queue" title="Permalink to this headline">¶</a></h3>
<p>Remember that each time you put a <a class="reference internal" href="../tensors.html#torch.Tensor" title="torch.Tensor"><code class="xref py py-class docutils literal notranslate"><span class="pre">Tensor</span></code></a> into a
<a class="reference external" href="https://docs.python.org/3/library/multiprocessing.html#multiprocessing.Queue" title="(in Python v3.7)"><code class="docutils literal notranslate"><span class="pre">multiprocessing.Queue</span></code></a>, it has to be moved into shared memory.
If it’s already shared, it is a no-op, otherwise it will incur an additional
memory copy that can slow down the whole process. Even if you have a pool of
processes sending data to a single one, make it send the buffers back - this
is nearly free and will let you avoid a copy when sending next batch.</p>
</div>
<div class="section" id="asynchronous-multiprocess-training-e-g-hogwild">
<h3>Asynchronous multiprocess training (e.g. Hogwild)<a class="headerlink" href="#asynchronous-multiprocess-training-e-g-hogwild" title="Permalink to this headline">¶</a></h3>
<p>Using <a class="reference internal" href="../multiprocessing.html#module-torch.multiprocessing" title="torch.multiprocessing"><code class="xref py py-mod docutils literal notranslate"><span class="pre">torch.multiprocessing</span></code></a>, it is possible to train a model
asynchronously, with parameters either shared all the time, or being
periodically synchronized. In the first case, we recommend sending over the whole
model object, while in the latter, we advise to only send the
<a class="reference internal" href="../nn.html#torch.nn.Module.state_dict" title="torch.nn.Module.state_dict"><code class="xref py py-meth docutils literal notranslate"><span class="pre">state_dict()</span></code></a>.</p>
<p>We recommend using <a class="reference external" href="https://docs.python.org/3/library/multiprocessing.html#multiprocessing.Queue" title="(in Python v3.7)"><code class="docutils literal notranslate"><span class="pre">multiprocessing.Queue</span></code></a> for passing all kinds
of PyTorch objects between processes. It is possible to e.g. inherit the tensors
and storages already in shared memory, when using the <code class="docutils literal notranslate"><span class="pre">fork</span></code> start method,
however it is very bug prone and should be used with care, and only by advanced
users. Queues, even though they’re sometimes a less elegant solution, will work
properly in all cases.</p>
<div class="admonition warning">
<p class="admonition-title">Warning</p>
<p>You should be careful about having global statements, that are not guarded
with an <code class="docutils literal notranslate"><span class="pre">if</span> <span class="pre">__name__</span> <span class="pre">==</span> <span class="pre">'__main__'</span></code>. If a different start method than
<code class="docutils literal notranslate"><span class="pre">fork</span></code> is used, they will be executed in all subprocesses.</p>
</div>
<div class="section" id="hogwild">
<h4>Hogwild<a class="headerlink" href="#hogwild" title="Permalink to this headline">¶</a></h4>
<p>A concrete Hogwild implementation can be found in the <a class="reference external" href="https://github.com/pytorch/examples/tree/master/mnist_hogwild">examples repository</a>,
but to showcase the overall structure of the code, there’s also a minimal
example below as well:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="kn">import</span> <span class="nn">torch.multiprocessing</span> <span class="k">as</span> <span class="nn">mp</span>
<span class="kn">from</span> <span class="nn">model</span> <span class="k">import</span> <span class="n">MyModel</span>
<span class="k">def</span> <span class="nf">train</span><span class="p">(</span><span class="n">model</span><span class="p">):</span>
<span class="c1"># Construct data_loader, optimizer, etc.</span>
<span class="k">for</span> <span class="n">data</span><span class="p">,</span> <span class="n">labels</span> <span class="ow">in</span> <span class="n">data_loader</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">loss_fn</span><span class="p">(</span><span class="n">model</span><span class="p">(</span><span class="n">data</span><span class="p">),</span> <span class="n">labels</span><span class="p">)</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="c1"># This will update the shared parameters</span>
<span class="k">if</span> <span class="vm">__name__</span> <span class="o">==</span> <span class="s1">'__main__'</span><span class="p">:</span>
<span class="n">num_processes</span> <span class="o">=</span> <span class="mi">4</span>
<span class="n">model</span> <span class="o">=</span> <span class="n">MyModel</span><span class="p">()</span>
<span class="c1"># NOTE: this is required for the ``fork`` method to work</span>
<span class="n">model</span><span class="o">.</span><span class="n">share_memory</span><span class="p">()</span>
<span class="n">processes</span> <span class="o">=</span> <span class="p">[]</span>
<span class="k">for</span> <span class="n">rank</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="n">num_processes</span><span class="p">):</span>
<span class="n">p</span> <span class="o">=</span> <span class="n">mp</span><span class="o">.</span><span class="n">Process</span><span class="p">(</span><span class="n">target</span><span class="o">=</span><span class="n">train</span><span class="p">,</span> <span class="n">args</span><span class="o">=</span><span class="p">(</span><span class="n">model</span><span class="p">,))</span>
<span class="n">p</span><span class="o">.</span><span class="n">start</span><span class="p">()</span>
<span class="n">processes</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="n">p</span><span class="p">)</span>
<span class="k">for</span> <span class="n">p</span> <span class="ow">in</span> <span class="n">processes</span><span class="p">:</span>
<span class="n">p</span><span class="o">.</span><span class="n">join</span><span class="p">()</span>
</pre></div>
</div>
</div>
</div>
</div>
</div>
</article>
</div>
<footer>
<div class="rst-footer-buttons" role="navigation" aria-label="footer navigation">
<a href="randomness.html" class="btn btn-neutral float-right" title="Reproducibility" accesskey="n" rel="next">Next <img src="../_static/images/chevron-right-orange.svg" class="next-page"></a>
<a href="large_scale_deployments.html" class="btn btn-neutral" title="Features for large-scale deployments" 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="#">Multiprocessing best practices</a><ul>
<li><a class="reference internal" href="#cuda-in-multiprocessing">CUDA in multiprocessing</a></li>
<li><a class="reference internal" href="#best-practices-and-tips">Best practices and tips</a><ul>
<li><a class="reference internal" href="#avoiding-and-fighting-deadlocks">Avoiding and fighting deadlocks</a></li>
<li><a class="reference internal" href="#reuse-buffers-passed-through-a-queue">Reuse buffers passed through a Queue</a></li>
<li><a class="reference internal" href="#asynchronous-multiprocess-training-e-g-hogwild">Asynchronous multiprocess training (e.g. Hogwild)</a><ul>
<li><a class="reference internal" href="#hogwild">Hogwild</a></li>
</ul>
</li>
</ul>
</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 type="text/javascript" src="../_static/jquery.js"></script>
<script type="text/javascript" src="../_static/underscore.js"></script>
<script type="text/javascript" src="../_static/doctools.js"></script>
<script type="text/javascript" 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 type="text/javascript" src="../_static/js/theme.js"></script>
<script type="text/javascript">
jQuery(function () {
SphinxRtdTheme.Navigation.enable(true);
});
</script>
<!-- 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://pytorch.org/resources">Resources</a></li>
</ul>
</div>
<div class="footer-links-col">
<ul>
<li class="list-title"><a href="https://pytorch.org/support">Support</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.slack.com" target="_blank">Slack</a></li>
<li><a href="https://github.com/pytorch/pytorch/blob/master/CONTRIBUTING.md" target="_blank">Contributing</a></li>
</ul>
</div>
<div class="footer-links-col follow-us-col">
<ul>
<li class="list-title">Follow Us</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>
</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="#">Get Started</a>
</li>
<li>
<a href="#">Features</a>
</li>
<li>
<a href="#">Ecosystem</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();
// 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>