Register
Login
Resources
Docs Blog Datasets Glossary Case Studies Tutorials & Webinars
Product
Data Engine LLMs Platform Enterprise
Pricing Explore
Connect to our Discord channel
Resources  
Docs Blog Datasets Glossary Case Studies Tutorials & Webinars
Explore Pricing
Product  
Data Engine LLMs Platform Enterprise
Connect to Dagshub Discord channel
Register
Login

Deci-AI
/
super-gradients
connected to https://github.com/Deci-AI/super-gradients.git

Watch 4
Issues 62 Pull Requests 44 Discussions 0

#274 Remove all elasticsearch references

Merged
Ofri Masad merged 1 commits into Deci-AI:master from deci-ai:feature/LAB-0000_remove_elasticsearch_references
Conversation 0 Files changed 4 Commits 1
File Compare
super-gradients
/
docs
/
super_gradients.training.sg_model.html
View left file View right file
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
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405 
  1. <!DOCTYPE html>
    2. 

  2. <html class="writer-html5" lang="en" >
    3. 

  3. <head>
    4. 

  4.   <meta charset="utf-8" /><meta name="generator" content="Docutils 0.17.1: http://docutils.sourceforge.net/" />
    5. 

  5. 

  6.   <meta name="viewport" content="width=device-width, initial-scale=1.0" />
    7. 

  7.   <title>super_gradients.training.sg_model package &mdash; SuperGradients 1.0 documentation</title>
    8. 

  8.       <link rel="stylesheet" href="_static/pygments.css" type="text/css" />
    9. 

  9.       <link rel="stylesheet" href="_static/css/theme.css" type="text/css" />
    10. 

  10.       <link rel="stylesheet" href="_static/graphviz.css" type="text/css" />
    11. 

  11.   <!--[if lt IE 9]>
    12. 

  12.     <script src="_static/js/html5shiv.min.js"></script>
    13. 

  13.   <![endif]-->
    14. 

  14.   
    15. 

  15.         <script data-url_root="./" id="documentation_options" src="_static/documentation_options.js"></script>
    16. 

  16.         <script src="_static/jquery.js"></script>
    17. 

  17.         <script src="_static/underscore.js"></script>
    18. 

  18.         <script src="_static/doctools.js"></script>
    19. 

  19.     <script src="_static/js/theme.js"></script>
    20. 

  20.     <link rel="index" title="Index" href="genindex.html" />
    21. 

  21.     <link rel="search" title="Search" href="search.html" /> 
    22. 

  22. </head>
    23. 

  23. 

  24. <body class="wy-body-for-nav"> 
    25. 

  25.   <div class="wy-grid-for-nav">
    26. 

  26.     <nav data-toggle="wy-nav-shift" class="wy-nav-side">
    27. 

  27.       <div class="wy-side-scroll">
    28. 

  28.         <div class="wy-side-nav-search" >
    29. 

  29.             <a href="index.html" class="icon icon-home"> SuperGradients
    30. 

  30.           </a>
    31. 

  31. <div role="search">
    32. 

  32.   <form id="rtd-search-form" class="wy-form" action="search.html" method="get">
    33. 

  33.     <input type="text" name="q" placeholder="Search docs" />
    34. 

  34.     <input type="hidden" name="check_keywords" value="yes" />
    35. 

  35.     <input type="hidden" name="area" value="default" />
    36. 

  36.   </form>
    37. 

  37. </div>
    38. 

  38.         </div><div class="wy-menu wy-menu-vertical" data-spy="affix" role="navigation" aria-label="Navigation menu">
    39. 

  39.               <p class="caption"><span class="caption-text">Welcome To SuperGradients</span></p>
    40. 

  40. <ul>
    41. 

  41. <li class="toctree-l1"><a class="reference internal" href="welcome.html#supergradients">SuperGradients</a></li>
    42. 

  42. </ul>
    43. 

  43. <p class="caption"><span class="caption-text">Technical Documentation</span></p>
    44. 

  44. <ul>
    45. 

  45. <li class="toctree-l1"><a class="reference internal" href="super_gradients.common.html">Common package</a></li>
    46. 

  46. <li class="toctree-l1"><a class="reference internal" href="super_gradients.training.html">Training package</a></li>
    47. 

  47. </ul>
    48. 

  48. <p class="caption"><span class="caption-text">User Guide</span></p>
    49. 

  49. <ul>
    50. 

  50. <li class="toctree-l1"><a class="reference internal" href="user_guide.html">What is SuperGradients?</a></li>
    51. 

  51. <li class="toctree-l1"><a class="reference internal" href="user_guide.html#introducing-the-supergradients-library">Introducing the SuperGradients library</a></li>
    52. 

  52. <li class="toctree-l1"><a class="reference internal" href="user_guide.html#installation">Installation</a></li>
    53. 

  53. <li class="toctree-l1"><a class="reference internal" href="user_guide.html#integrating-your-training-code-complete-walkthrough">Integrating Your Training Code - Complete Walkthrough</a></li>
    54. 

  54. <li class="toctree-l1"><a class="reference internal" href="user_guide.html#training-parameters">Training Parameters</a></li>
    55. 

  55. <li class="toctree-l1"><a class="reference internal" href="user_guide.html#logs-and-checkpoints">Logs and Checkpoints</a></li>
    56. 

  56. <li class="toctree-l1"><a class="reference internal" href="user_guide.html#dataset-parameters">Dataset Parameters</a></li>
    57. 

  57. <li class="toctree-l1"><a class="reference internal" href="user_guide.html#network-architectures">Network Architectures</a></li>
    58. 

  58. <li class="toctree-l1"><a class="reference internal" href="user_guide.html#pretrained-models">Pretrained Models</a></li>
    59. 

  59. <li class="toctree-l1"><a class="reference internal" href="user_guide.html#how-to-reproduce-our-training-recipes">How To Reproduce Our Training Recipes</a></li>
    60. 

  60. <li class="toctree-l1"><a class="reference internal" href="user_guide.html#professional-tools-integration">Professional Tools Integration</a></li>
    61. 

  61. <li class="toctree-l1"><a class="reference internal" href="user_guide.html#supergradients-faq">SuperGradients FAQ</a></li>
    62. 

  62. </ul>
    63. 

  63. 

  64.         </div>
    65. 

  65.       </div>
    66. 

  66.     </nav>
    67. 

  67. 

  68.     <section data-toggle="wy-nav-shift" class="wy-nav-content-wrap"><nav class="wy-nav-top" aria-label="Mobile navigation menu" >
    69. 

  69.           <i data-toggle="wy-nav-top" class="fa fa-bars"></i>
    70. 

  70.           <a href="index.html">SuperGradients</a>
    71. 

  71.       </nav>
    72. 

  72. 

  73.       <div class="wy-nav-content">
    74. 

  74.         <div class="rst-content">
    75. 

  75.           <div role="navigation" aria-label="Page navigation">
    76. 

  76.   <ul class="wy-breadcrumbs">
    77. 

  77.       <li><a href="index.html" class="icon icon-home"></a> &raquo;</li>
    78. 

  78.       <li>super_gradients.training.sg_model package</li>
    79. 

  79.       <li class="wy-breadcrumbs-aside">
    80. 

  80.             <a href="_sources/super_gradients.training.sg_model.rst.txt" rel="nofollow"> View page source</a>
    81. 

  81.       </li>
    82. 

  82.   </ul>
    83. 

  83.   <hr/>
    84. 

  84. </div>
    85. 

  85.           <div role="main" class="document" itemscope="itemscope" itemtype="http://schema.org/Article">
    86. 

  86.            <div itemprop="articleBody">
    87. 

  87.              
    88. 

  88.   <section id="super-gradients-training-sg-model-package">
    89. 

  89. <h1>super_gradients.training.sg_model package<a class="headerlink" href="#super-gradients-training-sg-model-package" title="Permalink to this headline"></a></h1>
    90. 

  90. <section id="submodules">
    91. 

  91. <h2>Submodules<a class="headerlink" href="#submodules" title="Permalink to this headline"></a></h2>
    92. 

  92. </section>
    93. 

  93. <section id="module-super_gradients.training.sg_model.sg_model">
    94. 

  94. <span id="super-gradients-training-sg-model-sg-model-module"></span><h2>super_gradients.training.sg_model.sg_model module<a class="headerlink" href="#module-super_gradients.training.sg_model.sg_model" title="Permalink to this headline"></a></h2>
    95. 

  95. <dl class="py class">
    96. 

  96. <dt class="sig sig-object py" id="super_gradients.training.sg_model.sg_model.StrictLoad">
    97. 

  97. <em class="property"><span class="pre">class</span> </em><span class="sig-prename descclassname"><span class="pre">super_gradients.training.sg_model.sg_model.</span></span><span class="sig-name descname"><span class="pre">StrictLoad</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">value</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/sg_model/sg_model.html#StrictLoad"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.sg_model.sg_model.StrictLoad" title="Permalink to this definition"></a></dt>
    98. 

  98. <dd><p>Bases: <code class="xref py py-class docutils literal notranslate"><span class="pre">enum.Enum</span></code></p>
    99. 

  99. <dl>
    100. 

  100. <dt>Wrapper for adding more functionality to torch’s strict_load parameter in load_state_dict().</dt><dd><dl>
    101. 

  101. <dt>Attributes:</dt><dd><p>OFF              - Native torch “strict_load = off” behaviour. See nn.Module.load_state_dict() documentation for more details.
    102. 

  102. ON               - Native torch “strict_load = on” behaviour. See nn.Module.load_state_dict() documentation for more details.
    103. 

  103. NO_KEY_MATCHING  - Allows the usage of SuperGradient’s adapt_checkpoint function, which loads a checkpoint by matching each</p>
    104. 

  104. <blockquote>
    105. 

  105. <div><p>layer’s shapes (and bypasses the strict matching of the names of each layer (ie: disregards the state_dict key matching)).</p>
    106. 

  106. </div></blockquote>
    107. 

  107. </dd>
    108. 

  108. </dl>
    109. 

  109. </dd>
    110. 

  110. </dl>
    111. 

  111. <dl class="py attribute">
    112. 

  112. <dt class="sig sig-object py" id="super_gradients.training.sg_model.sg_model.StrictLoad.OFF">
    113. 

  113. <span class="sig-name descname"><span class="pre">OFF</span></span><em class="property"> <span class="pre">=</span> <span class="pre">False</span></em><a class="headerlink" href="#super_gradients.training.sg_model.sg_model.StrictLoad.OFF" title="Permalink to this definition"></a></dt>
    114. 

  114. <dd></dd></dl>
    115. 

  115. 

  116. <dl class="py attribute">
    117. 

  117. <dt class="sig sig-object py" id="super_gradients.training.sg_model.sg_model.StrictLoad.ON">
    118. 

  118. <span class="sig-name descname"><span class="pre">ON</span></span><em class="property"> <span class="pre">=</span> <span class="pre">True</span></em><a class="headerlink" href="#super_gradients.training.sg_model.sg_model.StrictLoad.ON" title="Permalink to this definition"></a></dt>
    119. 

  119. <dd></dd></dl>
    120. 

  120. 

  121. <dl class="py attribute">
    122. 

  122. <dt class="sig sig-object py" id="super_gradients.training.sg_model.sg_model.StrictLoad.NO_KEY_MATCHING">
    123. 

  123. <span class="sig-name descname"><span class="pre">NO_KEY_MATCHING</span></span><em class="property"> <span class="pre">=</span> <span class="pre">'no_key_matching'</span></em><a class="headerlink" href="#super_gradients.training.sg_model.sg_model.StrictLoad.NO_KEY_MATCHING" title="Permalink to this definition"></a></dt>
    124. 

  124. <dd></dd></dl>
    125. 

  125. 

  126. </dd></dl>
    127. 

  127. 

  128. <dl class="py class">
    129. 

  129. <dt class="sig sig-object py" id="super_gradients.training.sg_model.sg_model.MultiGPUMode">
    130. 

  130. <em class="property"><span class="pre">class</span> </em><span class="sig-prename descclassname"><span class="pre">super_gradients.training.sg_model.sg_model.</span></span><span class="sig-name descname"><span class="pre">MultiGPUMode</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">value</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/sg_model/sg_model.html#MultiGPUMode"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.sg_model.sg_model.MultiGPUMode" title="Permalink to this definition"></a></dt>
    131. 

  131. <dd><p>Bases: <code class="xref py py-class docutils literal notranslate"><span class="pre">str</span></code>, <code class="xref py py-class docutils literal notranslate"><span class="pre">enum.Enum</span></code></p>
    132. 

  132. <dl class="py attribute">
    133. 

  133. <dt class="sig sig-object py">
    134. 

  134. <span class="sig-name descname"><span class="pre">OFF</span>&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160; <span class="pre">-</span> <span class="pre">Single</span> <span class="pre">GPU</span> <span class="pre">Mode</span> <span class="pre">/</span> <span class="pre">CPU</span> <span class="pre">Mode</span></span></dt>
    135. 

  135. <dd></dd></dl>
    136. 

  136. 

  137. <dl class="py attribute">
    138. 

  138. <dt class="sig sig-object py">
    139. 

  139. <span class="sig-name descname"><span class="pre">DATA_PARALLEL</span>&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160; <span class="pre">-</span> <span class="pre">Multiple</span> <span class="pre">GPUs,</span> <span class="pre">Synchronous</span></span></dt>
    140. 

  140. <dd></dd></dl>
    141. 

  141. 

  142. <dl class="py attribute">
    143. 

  143. <dt class="sig sig-object py">
    144. 

  144. <span class="sig-name descname"><span class="pre">DISTRIBUTED_DATA_PARALLEL</span> <span class="pre">-</span> <span class="pre">Multiple</span> <span class="pre">GPUs,</span> <span class="pre">Asynchronous</span></span></dt>
    145. 

  145. <dd></dd></dl>
    146. 

  146. 

  147. <dl class="py attribute">
    148. 

  148. <dt class="sig sig-object py" id="super_gradients.training.sg_model.sg_model.MultiGPUMode.OFF">
    149. 

  149. <span class="sig-name descname"><span class="pre">OFF</span></span><em class="property"> <span class="pre">=</span> <span class="pre">'Off'</span></em><a class="headerlink" href="#super_gradients.training.sg_model.sg_model.MultiGPUMode.OFF" title="Permalink to this definition"></a></dt>
    150. 

  150. <dd></dd></dl>
    151. 

  151. 

  152. <dl class="py attribute">
    153. 

  153. <dt class="sig sig-object py" id="super_gradients.training.sg_model.sg_model.MultiGPUMode.DATA_PARALLEL">
    154. 

  154. <span class="sig-name descname"><span class="pre">DATA_PARALLEL</span></span><em class="property"> <span class="pre">=</span> <span class="pre">'DP'</span></em><a class="headerlink" href="#super_gradients.training.sg_model.sg_model.MultiGPUMode.DATA_PARALLEL" title="Permalink to this definition"></a></dt>
    155. 

  155. <dd></dd></dl>
    156. 

  156. 

  157. <dl class="py attribute">
    158. 

  158. <dt class="sig sig-object py" id="super_gradients.training.sg_model.sg_model.MultiGPUMode.DISTRIBUTED_DATA_PARALLEL">
    159. 

  159. <span class="sig-name descname"><span class="pre">DISTRIBUTED_DATA_PARALLEL</span></span><em class="property"> <span class="pre">=</span> <span class="pre">'DDP'</span></em><a class="headerlink" href="#super_gradients.training.sg_model.sg_model.MultiGPUMode.DISTRIBUTED_DATA_PARALLEL" title="Permalink to this definition"></a></dt>
    160. 

  160. <dd></dd></dl>
    161. 

  161. 

  162. <dl class="py attribute">
    163. 

  163. <dt class="sig sig-object py" id="super_gradients.training.sg_model.sg_model.MultiGPUMode.AUTO">
    164. 

  164. <span class="sig-name descname"><span class="pre">AUTO</span></span><em class="property"> <span class="pre">=</span> <span class="pre">'AUTO'</span></em><a class="headerlink" href="#super_gradients.training.sg_model.sg_model.MultiGPUMode.AUTO" title="Permalink to this definition"></a></dt>
    165. 

  165. <dd></dd></dl>
    166. 

  166. 

  167. </dd></dl>
    168. 

  168. 

  169. <dl class="py class">
    170. 

  170. <dt class="sig sig-object py" id="super_gradients.training.sg_model.sg_model.EvaluationType">
    171. 

  171. <em class="property"><span class="pre">class</span> </em><span class="sig-prename descclassname"><span class="pre">super_gradients.training.sg_model.sg_model.</span></span><span class="sig-name descname"><span class="pre">EvaluationType</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">value</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/sg_model/sg_model.html#EvaluationType"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.sg_model.sg_model.EvaluationType" title="Permalink to this definition"></a></dt>
    172. 

  172. <dd><p>Bases: <code class="xref py py-class docutils literal notranslate"><span class="pre">str</span></code>, <code class="xref py py-class docutils literal notranslate"><span class="pre">enum.Enum</span></code></p>
    173. 

  173. <p>Passed to SgModel.evaluate(..), and controls which phase callbacks should be triggered (if at all).</p>
    174. 

  174. <blockquote>
    175. 

  175. <div><dl class="simple">
    176. 

  176. <dt>Attributes:</dt><dd><p>TEST
    177. 

  177. VALIDATION</p>
    178. 

  178. </dd>
    179. 

  179. </dl>
    180. 

  180. </div></blockquote>
    181. 

  181. <dl class="py attribute">
    182. 

  182. <dt class="sig sig-object py" id="super_gradients.training.sg_model.sg_model.EvaluationType.TEST">
    183. 

  183. <span class="sig-name descname"><span class="pre">TEST</span></span><em class="property"> <span class="pre">=</span> <span class="pre">'TEST'</span></em><a class="headerlink" href="#super_gradients.training.sg_model.sg_model.EvaluationType.TEST" title="Permalink to this definition"></a></dt>
    184. 

  184. <dd></dd></dl>
    185. 

  185. 

  186. <dl class="py attribute">
    187. 

  187. <dt class="sig sig-object py" id="super_gradients.training.sg_model.sg_model.EvaluationType.VALIDATION">
    188. 

  188. <span class="sig-name descname"><span class="pre">VALIDATION</span></span><em class="property"> <span class="pre">=</span> <span class="pre">'VALIDATION'</span></em><a class="headerlink" href="#super_gradients.training.sg_model.sg_model.EvaluationType.VALIDATION" title="Permalink to this definition"></a></dt>
    189. 

  189. <dd></dd></dl>
    190. 

  190. 

  191. </dd></dl>
    192. 

  192. 

  193. <dl class="py class">
    194. 

  194. <dt class="sig sig-object py" id="super_gradients.training.sg_model.sg_model.SgModel">
    195. 

  195. <em class="property"><span class="pre">class</span> </em><span class="sig-prename descclassname"><span class="pre">super_gradients.training.sg_model.sg_model.</span></span><span class="sig-name descname"><span class="pre">SgModel</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="pre">experiment_name:</span> <span class="pre">str</span></em>, <em class="sig-param"><span class="pre">device:</span> <span class="pre">Optional[str]</span> <span class="pre">=</span> <span class="pre">None</span></em>, <em class="sig-param"><span class="pre">multi_gpu:</span> <span class="pre">Union[super_gradients.training.sg_model.sg_model.MultiGPUMode</span></em>, <em class="sig-param"><span class="pre">str]</span> <span class="pre">=</span> <span class="pre">&lt;MultiGPUMode.OFF:</span> <span class="pre">'Off'&gt;</span></em>, <em class="sig-param"><span class="pre">model_checkpoints_location:</span> <span class="pre">str</span> <span class="pre">=</span> <span class="pre">'local'</span></em>, <em class="sig-param"><span class="pre">overwrite_local_checkpoint:</span> <span class="pre">bool</span> <span class="pre">=</span> <span class="pre">True</span></em>, <em class="sig-param"><span class="pre">ckpt_name:</span> <span class="pre">str</span> <span class="pre">=</span> <span class="pre">'ckpt_latest.pth'</span></em>, <em class="sig-param"><span class="pre">post_prediction_callback:</span> <span class="pre">Optional[super_gradients.training.utils.detection_utils.DetectionPostPredictionCallback]</span> <span class="pre">=</span> <span class="pre">None</span></em>, <em class="sig-param"><span class="pre">ckpt_root_dir=None</span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/sg_model/sg_model.html#SgModel"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.sg_model.sg_model.SgModel" title="Permalink to this definition"></a></dt>
    196. 

  196. <dd><p>Bases: <code class="xref py py-class docutils literal notranslate"><span class="pre">object</span></code></p>
    197. 

  197. <p>SuperGradient Model - Base Class for Sg Models</p>
    198. 

  198. <dl class="py method">
    199. 

  199. <dt class="sig sig-object py" id="super_gradients.training.sg_model.sg_model.SgModel.train">
    200. 

  200. <span class="sig-name descname"><span class="pre">train</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">max_epochs</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">int</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">initial_epoch</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">int</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">save_model</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">bool</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/sg_model/sg_model.html#SgModel.train"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.sg_model.sg_model.SgModel.train" title="Permalink to this definition"></a></dt>
    201. 

  201. <dd><p>the main function used for the training, h.p. updating, logging etc.</p>
    202. 

  202. </dd></dl>
    203. 

  203. 

  204. <dl class="py method">
    205. 

  205. <dt class="sig sig-object py" id="super_gradients.training.sg_model.sg_model.SgModel.predict">
    206. 

  206. <span class="sig-name descname"><span class="pre">predict</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">idx</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">int</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/sg_model/sg_model.html#SgModel.predict"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.sg_model.sg_model.SgModel.predict" title="Permalink to this definition"></a></dt>
    207. 

  207. <dd><p>returns the predictions and label of the current inputs</p>
    208. 

  208. </dd></dl>
    209. 

  209. 

  210. <dl class="py method">
    211. 

  211. <dt class="sig sig-object py">
    212. 

  212. <span class="sig-name descname"><span class="pre">test(epoch</span> <span class="pre">:</span> <span class="pre">int,</span> <span class="pre">idx</span> <span class="pre">:</span> <span class="pre">int,</span> <span class="pre">save</span> <span class="pre">:</span> <span class="pre">bool):</span></span></dt>
    213. 

  213. <dd><p>returns the test loss, accuracy and runtime</p>
    214. 

  214. </dd></dl>
    215. 

  215. 

  216. <dl class="py method">
    217. 

  217. <dt class="sig sig-object py" id="super_gradients.training.sg_model.sg_model.SgModel.connect_dataset_interface">
    218. 

  218. <span class="sig-name descname"><span class="pre">connect_dataset_interface</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">dataset_interface</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><a class="reference internal" href="super_gradients.training.datasets.dataset_interfaces.html#super_gradients.training.datasets.dataset_interfaces.dataset_interface.DatasetInterface" title="super_gradients.training.datasets.dataset_interfaces.dataset_interface.DatasetInterface"><span class="pre">super_gradients.training.datasets.dataset_interfaces.dataset_interface.DatasetInterface</span></a></span></em>, <em class="sig-param"><span class="n"><span class="pre">data_loader_num_workers</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">int</span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">8</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/sg_model/sg_model.html#SgModel.connect_dataset_interface"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.sg_model.sg_model.SgModel.connect_dataset_interface" title="Permalink to this definition"></a></dt>
    219. 

  219. <dd><dl class="field-list simple">
    220. 

  220. <dt class="field-odd">Parameters</dt>
    221. 

  221. <dd class="field-odd"><ul class="simple">
    222. 

  222. <li><p><strong>dataset_interface</strong> – DatasetInterface object</p></li>
    223. 

  223. <li><p><strong>data_loader_num_workers</strong> – The number of threads to initialize the Data Loaders with
    224. 

  224. The dataset to be connected</p></li>
    225. 

  225. </ul>
    226. 

  226. </dd>
    227. 

  227. </dl>
    228. 

  228. </dd></dl>
    229. 

  229. 

  230. <dl class="py method">
    231. 

  231. <dt class="sig sig-object py" id="super_gradients.training.sg_model.sg_model.SgModel.build_model">
    232. 

  232. <span class="sig-name descname"><span class="pre">build_model</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">architecture</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">Union</span><span class="p"><span class="pre">[</span></span><span class="pre">str</span><span class="p"><span class="pre">,</span> </span><span class="pre">torch.nn.modules.module.Module</span><span class="p"><span class="pre">]</span></span></span></em>, <em class="sig-param"><span class="n"><span class="pre">arch_params</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">{}</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">checkpoint_params</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">{}</span></span></em>, <em class="sig-param"><span class="o"><span class="pre">*</span></span><span class="n"><span class="pre">args</span></span></em>, <em class="sig-param"><span class="o"><span class="pre">**</span></span><span class="n"><span class="pre">kwargs</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/sg_model/sg_model.html#SgModel.build_model"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.sg_model.sg_model.SgModel.build_model" title="Permalink to this definition"></a></dt>
    233. 

  233. <dd><dl class="field-list simple">
    234. 

  234. <dt class="field-odd">Parameters</dt>
    235. 

  235. <dd class="field-odd"><ul class="simple">
    236. 

  236. <li><p><strong>architecture</strong> – Defines the network’s architecture from models/ALL_ARCHITECTURES</p></li>
    237. 

  237. <li><p><strong>arch_params</strong> – Architecture H.P. e.g.: block, num_blocks, num_classes, etc.</p></li>
    238. 

  238. <li><p><strong>checkpoint_params</strong> – <p>Dictionary like object with the following key:values:</p>
    239. 

  239. <p>load_checkpoint:            Load a pre-trained checkpoint
    240. 

  240. strict_load:                See StrictLoad class documentation for details.
    241. 

  241. source_ckpt_folder_name:    folder name to load the checkpoint from (self.experiment_name if none is given)
    242. 

  242. load_weights_only:          loads only the weight from the checkpoint and zeroize the training params
    243. 

  243. load_backbone:              loads the provided checkpoint to self.net.backbone instead of self.net
    244. 

  244. external_checkpoint_path:   The path to the external checkpoint to be loaded. Can be absolute or relative</p>
    245. 

  245. <blockquote>
    246. 

  246. <div><p>(ie: path/to/checkpoint.pth). If provided, will automatically attempt to
    247. 

  247. load the checkpoint even if the load_checkpoint flag is not provided.</p>
    248. 

  248. </div></blockquote>
    249. 

  249. </p></li>
    250. 

  250. </ul>
    251. 

  251. </dd>
    252. 

  252. </dl>
    253. 

  253. </dd></dl>
    254. 

  254. 

  255. <dl class="py method">
    256. 

  256. <dt class="sig sig-object py" id="super_gradients.training.sg_model.sg_model.SgModel.backward_step">
    257. 

  257. <span class="sig-name descname"><span class="pre">backward_step</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">loss</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">torch.Tensor</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">epoch</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">int</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">batch_idx</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">int</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">context</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><a class="reference internal" href="super_gradients.training.utils.html#super_gradients.training.utils.callbacks.PhaseContext" title="super_gradients.training.utils.callbacks.PhaseContext"><span class="pre">super_gradients.training.utils.callbacks.PhaseContext</span></a></span></em>, <em class="sig-param"><span class="o"><span class="pre">*</span></span><span class="n"><span class="pre">args</span></span></em>, <em class="sig-param"><span class="o"><span class="pre">**</span></span><span class="n"><span class="pre">kwargs</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/sg_model/sg_model.html#SgModel.backward_step"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.sg_model.sg_model.SgModel.backward_step" title="Permalink to this definition"></a></dt>
    258. 

  258. <dd><p>Run backprop on the loss and perform a step
    259. 

  259. :param loss: The value computed by the loss function
    260. 

  260. :param optimizer: An object that can perform a gradient step and zeroize model gradient
    261. 

  261. :param epoch: number of epoch the training is on
    262. 

  262. :param batch_idx: number of iteration inside the current epoch
    263. 

  263. :param context: current phase context
    264. 

  264. :return:</p>
    265. 

  265. </dd></dl>
    266. 

  266. 

  267. <dl class="py method">
    268. 

  268. <dt class="sig sig-object py" id="super_gradients.training.sg_model.sg_model.SgModel.save_checkpoint">
    269. 

  269. <span class="sig-name descname"><span class="pre">save_checkpoint</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">optimizer</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">epoch</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">Optional</span><span class="p"><span class="pre">[</span></span><span class="pre">int</span><span class="p"><span class="pre">]</span></span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">None</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">validation_results_tuple</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">Optional</span><span class="p"><span class="pre">[</span></span><span class="pre">tuple</span><span class="p"><span class="pre">]</span></span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">None</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">context</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">Optional</span><span class="p"><span class="pre">[</span></span><a class="reference internal" href="super_gradients.training.utils.html#super_gradients.training.utils.callbacks.PhaseContext" title="super_gradients.training.utils.callbacks.PhaseContext"><span class="pre">super_gradients.training.utils.callbacks.PhaseContext</span></a><span class="p"><span class="pre">]</span></span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">None</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/sg_model/sg_model.html#SgModel.save_checkpoint"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.sg_model.sg_model.SgModel.save_checkpoint" title="Permalink to this definition"></a></dt>
    270. 

  270. <dd><p>Save the current state dict as latest (always), best (if metric was improved), epoch# (if determined in training
    271. 

  271. params)</p>
    272. 

  272. </dd></dl>
    273. 

  273. 

  274. <dl class="py method">
    275. 

  275. <dt class="sig sig-object py" id="id0">
    276. 

  276. <span class="sig-name descname"><span class="pre">train</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">training_params</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">dict</span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">{}</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/sg_model/sg_model.html#SgModel.train"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#id0" title="Permalink to this definition"></a></dt>
    277. 

  277. <dd><p>train - Trains the Model</p>
    278. 

  278. <dl>
    279. 

  279. <dt>IMPORTANT NOTE: Additional batch parameters can be added as a third item (optional) if a tuple is returned by</dt><dd><p>the data loaders, as dictionary. The phase context will hold the additional items, under an attribute with
    280. 

  280. the same name as the key in this dictionary. Then such items can be accessed through phase callbacks.</p>
    281. 

  281. <blockquote>
    282. 

  282. <div><dl class="field-list">
    283. 

  283. <dt class="field-odd">param training_params</dt>
    284. 

  284. <dd class="field-odd"><ul>
    285. 

  285. <li><p><cite>max_epochs</cite> : int</p>
    286. 

  286. <blockquote>
    287. 

  287. <div><p>Number of epochs to run training.</p>
    288. 

  288. </div></blockquote>
    289. 

  289. </li>
    290. 

  290. <li><p><cite>lr_updates</cite> : list(int)</p>
    291. 

  291. <blockquote>
    292. 

  292. <div><p>List of fixed epoch numbers to perform learning rate updates when <cite>lr_mode=’step’</cite>.</p>
    293. 

  293. </div></blockquote>
    294. 

  294. </li>
    295. 

  295. <li><p><cite>lr_decay_factor</cite> : float</p>
    296. 

  296. <blockquote>
    297. 

  297. <div><p>Decay factor to apply to the learning rate at each update when <cite>lr_mode=’step’</cite>.</p>
    298. 

  298. </div></blockquote>
    299. 

  299. </li>
    300. 

  300. <li><p><cite>lr_mode</cite> : str</p>
    301. 

  301. <blockquote>
    302. 

  302. <div><p>Learning rate scheduling policy, one of [‘step’,’poly’,’cosine’,’function’]. ‘step’ refers to
    303. 

  303. constant updates at epoch numbers passed through <cite>lr_updates</cite>. ‘cosine’ refers to Cosine Anealing
    304. 

  304. policy as mentioned in <a class="reference external" href="https://arxiv.org/abs/1608.03983">https://arxiv.org/abs/1608.03983</a>. ‘poly’ refers to polynomial decrease i.e
    305. 

  305. in each epoch iteration <cite>self.lr = self.initial_lr * pow((1.0 - (current_iter / max_iter)),
    306. 

  306. 0.9)</cite> ‘function’ refers to user defined learning rate scheduling function, that is passed through
    307. 

  307. <cite>lr_schedule_function</cite>.</p>
    308. 

  308. </div></blockquote>
    309. 

  309. </li>
    310. 

  310. <li><p><cite>lr_schedule_function</cite> : Union[callable,None]</p>
    311. 

  311. <blockquote>
    312. 

  312. <div><p>Learning rate scheduling function to be used when <cite>lr_mode</cite> is ‘function’.</p>
    313. 

  313. </div></blockquote>
    314. 

  314. </li>
    315. 

  315. <li><p><cite>lr_warmup_epochs</cite> : int (default=0)</p>
    316. 

  316. <blockquote>
    317. 

  317. <div><p>Number of epochs for learning rate warm up - see <a class="reference external" href="https://arxiv.org/pdf/1706.02677.pdf">https://arxiv.org/pdf/1706.02677.pdf</a> (Section 2.2).</p>
    318. 

  318. </div></blockquote>
    319. 

  319. </li>
    320. 

  320. <li><dl class="simple">
    321. 

  321. <dt><cite>cosine_final_lr_ratio</cite><span class="classifier">float (default=0.01)</span></dt><dd><dl class="simple">
    322. 

  322. <dt>Final learning rate ratio (only relevant when <a href="#id1"><span class="problematic" id="id2">`</span></a>lr_mode`=’cosine’). The cosine starts from initial_lr and reaches</dt><dd><p>initial_lr * cosine_final_lr_ratio in last epoch</p>
    323. 

  323. </dd>
    324. 

  324. </dl>
    325. 

  325. </dd>
    326. 

  326. </dl>
    327. 

  327. </li>
    328. 

  328. <li><p><cite>inital_lr</cite> : float</p>
    329. 

  329. <blockquote>
    330. 

  330. <div><p>Initial learning rate.</p>
    331. 

  331. </div></blockquote>
    332. 

  332. </li>
    333. 

  333. <li><p><cite>loss</cite> : Union[nn.module, str]</p>
    334. 

  334. <blockquote>
    335. 

  335. <div><p>Loss function for training.
    336. 

  336. One of SuperGradient’s built in options:</p>
    337. 

  337. <blockquote>
    338. 

  338. <div><p>“cross_entropy”: LabelSmoothingCrossEntropyLoss,
    339. 

  339. “mse”: MSELoss,
    340. 

  340. “r_squared_loss”: RSquaredLoss,
    341. 

  341. “detection_loss”: YoLoV3DetectionLoss,
    342. 

  342. “shelfnet_ohem_loss”: ShelfNetOHEMLoss,
    343. 

  343. “shelfnet_se_loss”: ShelfNetSemanticEncodingLoss,
    344. 

  344. “yolo_v5_loss”: YoLoV5DetectionLoss,
    345. 

  345. “ssd_loss”: SSDLoss,</p>
    346. 

  346. </div></blockquote>
    347. 

  347. <p>or user defined nn.module loss function.</p>
    348. 

  348. <p>IMPORTANT: forward(…) should return a (loss, loss_items) tuple where loss is the tensor used
    349. 

  349. for backprop (i.e what your original loss function returns), and loss_items should be a tensor of
    350. 

  350. shape (n_items), of values computed during the forward pass which we desire to log over the
    351. 

  351. entire epoch. For example- the loss itself should always be logged. Another example is a scenario
    352. 

  352. where the computed loss is the sum of a few components we would like to log- these entries in
    353. 

  353. loss_items).</p>
    354. 

  354. <p>When training, set the loss_logging_items_names parameter in train_params to be a list of
    355. 

  355. strings, of length n_items who’s ith element is the name of the ith entry in loss_items. Then
    356. 

  356. each item will be logged, rendered on tensorboard and “watched” (i.e saving model checkpoints
    357. 

  357. according to it).</p>
    358. 

  358. <p>Since running logs will save the loss_items in some internal state, it is recommended that
    359. 

  359. loss_items are detached from their computational graph for memory efficiency.</p>
    360. 

  360. </div></blockquote>
    361. 

  361. </li>
    362. 

  362. <li><p><cite>optimizer</cite> : Union[str, torch.optim.Optimizer]</p>
    363. 

  363. <blockquote>
    364. 

  364. <div><p>Optimization algorithm. One of [‘Adam’,’SGD’,’RMSProp’] corresponding to the torch.optim
    365. 

  365. optimzers implementations, or any object that implements torch.optim.Optimizer.</p>
    366. 

  366. </div></blockquote>
    367. 

  367. </li>
    368. 

  368. <li><p><cite>criterion_params</cite> : dict</p>
    369. 

  369. <blockquote>
    370. 

  370. <div><p>Loss function parameters.</p>
    371. 

  371. </div></blockquote>
    372. 

  372. </li>
    373. 

  373. <li><dl>
    374. 

  374. <dt><cite>optimizer_params</cite><span class="classifier">dict</span></dt><dd><p>When <cite>optimizer</cite> is one of [‘Adam’,’SGD’,’RMSProp’], it will be initialized with optimizer_params.</p>
    375. 

  375. <p>(see <a class="reference external" href="https://pytorch.org/docs/stable/optim.html">https://pytorch.org/docs/stable/optim.html</a> for the full list of
    376. 

  376. parameters for each optimizer).</p>
    377. 

  377. </dd>
    378. 

  378. </dl>
    379. 

  379. </li>
    380. 

  380. <li><p><cite>train_metrics_list</cite> : list(torchmetrics.Metric)</p>
    381. 

  381. <blockquote>
    382. 

  382. <div><p>Metrics to log during training. For more information on torchmetrics see
    383. 

  383. <a class="reference external" href="https://torchmetrics.rtfd.io/en/latest/">https://torchmetrics.rtfd.io/en/latest/</a>.</p>
    384. 

  384. </div></blockquote>
    385. 

  385. </li>
    386. 

  386. <li><p><cite>valid_metrics_list</cite> : list(torchmetrics.Metric)</p>
    387. 

  387. <blockquote>
    388. 

  388. <div><p>Metrics to log during validation/testing. For more information on torchmetrics see
    389. 

  389. <a class="reference external" href="https://torchmetrics.rtfd.io/en/latest/">https://torchmetrics.rtfd.io/en/latest/</a>.</p>
    390. 

  390. </div></blockquote>
    391. 

  391. </li>
    392. 

  392. <li><p><cite>loss_logging_items_names</cite> : list(str)</p>
    393. 

  393. <blockquote>
    394. 

  394. <div><p>The list of names/titles for the outputs returned from the loss functions forward pass (reminder-
    395. 

  395. the loss function should return the tuple (loss, loss_items)). These names will be used for
    396. 

  396. logging their values.</p>
    397. 

  397. </div></blockquote>
    398. 

  398. </li>
    399. 

  399. <li><p><cite>metric_to_watch</cite> : str (default=”Accuracy”)</p>
    400. 

  400. <blockquote>
    401. 

  401. <div><p>will be the metric which the model checkpoint will be saved according to, and can be set to any
    402. 

  402. of the following:</p>
    403. 

  403. <blockquote>
    404. 

  404. <div><p>a metric name (str) of one of the metric objects from the valid_metrics_list</p>
    405. 

  405. <p>a “metric_name” if some metric in valid_metrics_list has an attribute component_names which
    406. 

  406. is a list referring to the names of each entry in the output metric (torch tensor of size n)</p>
    407. 

  407. <p>one of “loss_logging_items_names” i.e which will correspond to an item returned during the
    408. 

  408. loss function’s forward pass.</p>
    409. 

  409. </div></blockquote>
    410. 

  410. <p>At the end of each epoch, if a new best metric_to_watch value is achieved, the models checkpoint
    411. 

  411. is saved in YOUR_PYTHON_PATH/checkpoints/ckpt_best.pth</p>
    412. 

  412. </div></blockquote>
    413. 

  413. </li>
    414. 

  414. <li><p><cite>greater_metric_to_watch_is_better</cite> : bool</p>
    415. 

  415. <blockquote>
    416. 

  416. <div><dl class="simple">
    417. 

  417. <dt>When choosing a model’s checkpoint to be saved, the best achieved model is the one that maximizes the</dt><dd><p>metric_to_watch when this parameter is set to True, and a one that minimizes it otherwise.</p>
    418. 

  418. </dd>
    419. 

  419. </dl>
    420. 

  420. </div></blockquote>
    421. 

  421. </li>
    422. 

  422. <li><p><cite>ema</cite> : bool (default=False)</p>
    423. 

  423. <blockquote>
    424. 

  424. <div><p>Whether to use Model Exponential Moving Average (see
    425. 

  425. <a class="reference external" href="https://github.com/rwightman/pytorch-image-models">https://github.com/rwightman/pytorch-image-models</a> ema implementation)</p>
    426. 

  426. </div></blockquote>
    427. 

  427. </li>
    428. 

  428. <li><p><cite>batch_accumulate</cite> : int (default=1)</p>
    429. 

  429. <blockquote>
    430. 

  430. <div><p>Number of batches to accumulate before every backward pass.</p>
    431. 

  431. </div></blockquote>
    432. 

  432. </li>
    433. 

  433. <li><p><cite>ema_params</cite> : dict</p>
    434. 

  434. <blockquote>
    435. 

  435. <div><p>Parameters for the ema model.</p>
    436. 

  436. </div></blockquote>
    437. 

  437. </li>
    438. 

  438. <li><p><cite>zero_weight_decay_on_bias_and_bn</cite> : bool (default=False)</p>
    439. 

  439. <blockquote>
    440. 

  440. <div><p>Whether to apply weight decay on batch normalization parameters or not (ignored when the passed
    441. 

  441. optimizer has already been initialized).</p>
    442. 

  442. </div></blockquote>
    443. 

  443. </li>
    444. 

  444. <li><p><cite>load_opt_params</cite> : bool (default=True)</p>
    445. 

  445. <blockquote>
    446. 

  446. <div><p>Whether to load the optimizers parameters as well when loading a model’s checkpoint.</p>
    447. 

  447. </div></blockquote>
    448. 

  448. </li>
    449. 

  449. <li><p><cite>run_validation_freq</cite> : int (default=1)</p>
    450. 

  450. <blockquote>
    451. 

  451. <div><dl class="simple">
    452. 

  452. <dt>The frequency in which validation is performed during training (i.e the validation is ran every</dt><dd><p><cite>run_validation_freq</cite> epochs.</p>
    453. 

  453. </dd>
    454. 

  454. </dl>
    455. 

  455. </div></blockquote>
    456. 

  456. </li>
    457. 

  457. <li><p><cite>save_model</cite> : bool (default=True)</p>
    458. 

  458. <blockquote>
    459. 

  459. <div><p>Whether to save the model checkpoints.</p>
    460. 

  460. </div></blockquote>
    461. 

  461. </li>
    462. 

  462. <li><p><cite>silent_mode</cite> : bool</p>
    463. 

  463. <blockquote>
    464. 

  464. <div><p>Silents the print outs.</p>
    465. 

  465. </div></blockquote>
    466. 

  466. </li>
    467. 

  467. <li><p><cite>mixed_precision</cite> : bool</p>
    468. 

  468. <blockquote>
    469. 

  469. <div><p>Whether to use mixed precision or not.</p>
    470. 

  470. </div></blockquote>
    471. 

  471. </li>
    472. 

  472. <li><p><cite>save_ckpt_epoch_list</cite> : list(int) (default=[])</p>
    473. 

  473. <blockquote>
    474. 

  474. <div><p>List of fixed epoch indices the user wishes to save checkpoints in.</p>
    475. 

  475. </div></blockquote>
    476. 

  476. </li>
    477. 

  477. <li><p><cite>average_best_models</cite> : bool (default=False)</p>
    478. 

  478. <blockquote>
    479. 

  479. <div><p>If set, a snapshot dictionary file and the average model will be saved / updated at every epoch
    480. 

  480. and evaluated only when training is completed. The snapshot file will only be deleted upon
    481. 

  481. completing the training. The snapshot dict will be managed on cpu.</p>
    482. 

  482. </div></blockquote>
    483. 

  483. </li>
    484. 

  484. <li><p><cite>precise_bn</cite> : bool (default=False)</p>
    485. 

  485. <blockquote>
    486. 

  486. <div><p>Whether to use precise_bn calculation during the training.</p>
    487. 

  487. </div></blockquote>
    488. 

  488. </li>
    489. 

  489. <li><p><cite>precise_bn_batch_size</cite> : int (default=None)</p>
    490. 

  490. <blockquote>
    491. 

  491. <div><p>The effective batch size we want to calculate the batchnorm on. For example, if we are training a model
    492. 

  492. on 8 gpus, with a batch of 128 on each gpu, a good rule of thumb would be to give it 8192
    493. 

  493. (ie: effective_batch_size * num_gpus = batch_per_gpu * num_gpus * num_gpus).
    494. 

  494. If precise_bn_batch_size is not provided in the training_params, the latter heuristic will be taken.</p>
    495. 

  495. </div></blockquote>
    496. 

  496. </li>
    497. 

  497. <li><p><cite>seed</cite> : int (default=42)</p>
    498. 

  498. <blockquote>
    499. 

  499. <div><p>Random seed to be set for torch, numpy, and random. When using DDP each process will have it’s seed
    500. 

  500. set to seed + rank.</p>
    501. 

  501. </div></blockquote>
    502. 

  502. </li>
    503. 

  503. <li><p><cite>log_installed_packages</cite> : bool (default=False)</p>
    504. 

  504. <blockquote>
    505. 

  505. <div><dl class="simple">
    506. 

  506. <dt>When set, the list of all installed packages (and their versions) will be written to the tensorboard</dt><dd><p>and logfile (useful when trying to reproduce results).</p>
    507. 

  507. </dd>
    508. 

  508. </dl>
    509. 

  509. </div></blockquote>
    510. 

  510. </li>
    511. 

  511. <li><p><cite>dataset_statistics</cite> : bool (default=False)</p>
    512. 

  512. <blockquote>
    513. 

  513. <div><p>Enable a statistic analysis of the dataset. If set to True the dataset will be analyzed and a report
    514. 

  514. will be added to the tensorboard along with some sample images from the dataset. Currently only
    515. 

  515. detection datasets are supported for analysis.</p>
    516. 

  516. </div></blockquote>
    517. 

  517. </li>
    518. 

  518. <li><p><cite>save_full_train_log</cite> : bool (default=False)</p>
    519. 

  519. <blockquote>
    520. 

  520. <div><dl class="simple">
    521. 

  521. <dt>When set, a full log (of all super_gradients modules, including uncaught exceptions from any other</dt><dd><p>module) of the training will be saved in the checkpoint directory under full_train_log.log</p>
    522. 

  522. </dd>
    523. 

  523. </dl>
    524. 

  524. </div></blockquote>
    525. 

  525. </li>
    526. 

  526. <li><p><cite>sg_logger</cite> : Union[AbstractSGLogger, str] (defauls=base_sg_logger)</p>
    527. 

  527. <blockquote>
    528. 

  528. <div><p>Define the SGLogger object for this training process. The SGLogger handles all disk writes, logs, TensorBoard, remote logging
    529. 

  529. and remote storage. By overriding the default base_sg_logger, you can change the storage location, support external monitoring and logging
    530. 

  530. or support remote storage.</p>
    531. 

  531. </div></blockquote>
    532. 

  532. </li>
    533. 

  533. <li><p><cite>sg_logger_params</cite> : dict</p>
    534. 

  534. <p>SGLogger parameters</p>
    535. 

  535. </li>
    536. 

  536. <li><p><cite>clip_grad_norm</cite> : float</p>
    537. 

  537. <p>Defines a maximal L2 norm of the gradients. Values which exceed the given value will be clipped</p>
    538. 

  538. </li>
    539. 

  539. </ul>
    540. 

  540. </dd>
    541. 

  541. </dl>
    542. 

  542. </div></blockquote>
    543. 

  543. </dd>
    544. 

  544. </dl>
    545. 

  545. <dl class="field-list simple">
    546. 

  546. <dt class="field-odd">Returns</dt>
    547. 

  547. <dd class="field-odd"><p></p>
    548. 

  548. </dd>
    549. 

  549. </dl>
    550. 

  550. </dd></dl>
    551. 

  551. 

  552. <dl class="py method">
    553. 

  553. <dt class="sig sig-object py" id="id3">
    554. 

  554. <span class="sig-name descname"><span class="pre">predict</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">inputs</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">targets</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">half</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">False</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">normalize</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">False</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">verbose</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">False</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">move_outputs_to_cpu</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">True</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/sg_model/sg_model.html#SgModel.predict"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#id3" title="Permalink to this definition"></a></dt>
    555. 

  555. <dd><p>A fast predictor for a batch of inputs
    556. 

  556. :param inputs: torch.tensor or numpy.array</p>
    557. 

  557. <blockquote>
    558. 

  558. <div><p>a batch of inputs</p>
    559. 

  559. </div></blockquote>
    560. 

  560. <dl class="field-list simple">
    561. 

  561. <dt class="field-odd">Parameters</dt>
    562. 

  562. <dd class="field-odd"><ul class="simple">
    563. 

  563. <li><p><strong>targets</strong> – torch.tensor()
    564. 

  564. corresponding labels - if non are given - accuracy will not be computed</p></li>
    565. 

  565. <li><p><strong>verbose</strong> – bool
    566. 

  566. print the results to screen</p></li>
    567. 

  567. <li><p><strong>normalize</strong> – bool
    568. 

  568. If true, normalizes the tensor according to the dataloader’s normalization values</p></li>
    569. 

  569. <li><p><strong>half</strong> – Performs half precision evaluation</p></li>
    570. 

  570. <li><p><strong>move_outputs_to_cpu</strong> – Moves the results from the GPU to the CPU</p></li>
    571. 

  571. </ul>
    572. 

  572. </dd>
    573. 

  573. <dt class="field-even">Returns</dt>
    574. 

  574. <dd class="field-even"><p>outputs, acc, net_time, gross_time
    575. 

  575. networks predictions, accuracy calculation, forward pass net time, function gross time</p>
    576. 

  576. </dd>
    577. 

  577. </dl>
    578. 

  578. </dd></dl>
    579. 

  579. 

  580. <dl class="py method">
    581. 

  581. <dt class="sig sig-object py" id="super_gradients.training.sg_model.sg_model.SgModel.compute_model_runtime">
    582. 

  582. <span class="sig-name descname"><span class="pre">compute_model_runtime</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">input_dims</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">Optional</span><span class="p"><span class="pre">[</span></span><span class="pre">tuple</span><span class="p"><span class="pre">]</span></span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">None</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">batch_sizes</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">Union</span><span class="p"><span class="pre">[</span></span><span class="pre">tuple</span><span class="p"><span class="pre">,</span> </span><span class="pre">list</span><span class="p"><span class="pre">,</span> </span><span class="pre">int</span><span class="p"><span class="pre">]</span></span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">(1,</span> <span class="pre">8,</span> <span class="pre">16,</span> <span class="pre">32,</span> <span class="pre">64)</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">verbose</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">bool</span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">True</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/sg_model/sg_model.html#SgModel.compute_model_runtime"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.sg_model.sg_model.SgModel.compute_model_runtime" title="Permalink to this definition"></a></dt>
    583. 

  583. <dd><p>Compute the “atomic” inference time and throughput.
    584. 

  584. Atomic refers to calculating the forward pass independently, discarding effects such as data augmentation,
    585. 

  585. data upload to device, multi-gpu distribution etc.
    586. 

  586. :param input_dims: tuple</p>
    587. 

  587. <blockquote>
    588. 

  588. <div><p>shape of a basic input to the network (without the first index) e.g. (3, 224, 224)
    589. 

  589. if None uses an input from the test loader</p>
    590. 

  590. </div></blockquote>
    591. 

  591. <dl class="field-list simple">
    592. 

  592. <dt class="field-odd">Parameters</dt>
    593. 

  593. <dd class="field-odd"><ul class="simple">
    594. 

  594. <li><p><strong>batch_sizes</strong> – int or list
    595. 

  595. Batch sizes for latency calculation</p></li>
    596. 

  596. <li><p><strong>verbose</strong> – bool
    597. 

  597. Prints results to screen</p></li>
    598. 

  598. </ul>
    599. 

  599. </dd>
    600. 

  600. <dt class="field-even">Returns</dt>
    601. 

  601. <dd class="field-even"><p>log: dict
    602. 

  602. Latency and throughput for each tested batch size</p>
    603. 

  603. </dd>
    604. 

  604. </dl>
    605. 

  605. </dd></dl>
    606. 

  606. 

  607. <dl class="py method">
    608. 

  608. <dt class="sig sig-object py" id="super_gradients.training.sg_model.sg_model.SgModel.get_arch_params">
    609. 

  609. <span class="sig-name descname"><span class="pre">get_arch_params</span></span><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/sg_model/sg_model.html#SgModel.get_arch_params"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.sg_model.sg_model.SgModel.get_arch_params" title="Permalink to this definition"></a></dt>
    610. 

  610. <dd></dd></dl>
    611. 

  611. 

  612. <dl class="py method">
    613. 

  613. <dt class="sig sig-object py" id="super_gradients.training.sg_model.sg_model.SgModel.get_structure">
    614. 

  614. <span class="sig-name descname"><span class="pre">get_structure</span></span><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/sg_model/sg_model.html#SgModel.get_structure"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.sg_model.sg_model.SgModel.get_structure" title="Permalink to this definition"></a></dt>
    615. 

  615. <dd></dd></dl>
    616. 

  616. 

  617. <dl class="py method">
    618. 

  618. <dt class="sig sig-object py" id="super_gradients.training.sg_model.sg_model.SgModel.get_architecture">
    619. 

  619. <span class="sig-name descname"><span class="pre">get_architecture</span></span><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/sg_model/sg_model.html#SgModel.get_architecture"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.sg_model.sg_model.SgModel.get_architecture" title="Permalink to this definition"></a></dt>
    620. 

  620. <dd></dd></dl>
    621. 

  621. 

  622. <dl class="py method">
    623. 

  623. <dt class="sig sig-object py" id="super_gradients.training.sg_model.sg_model.SgModel.set_experiment_name">
    624. 

  624. <span class="sig-name descname"><span class="pre">set_experiment_name</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">experiment_name</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/sg_model/sg_model.html#SgModel.set_experiment_name"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.sg_model.sg_model.SgModel.set_experiment_name" title="Permalink to this definition"></a></dt>
    625. 

  625. <dd></dd></dl>
    626. 

  626. 

  627. <dl class="py method">
    628. 

  628. <dt class="sig sig-object py" id="super_gradients.training.sg_model.sg_model.SgModel.re_build_model">
    629. 

  629. <span class="sig-name descname"><span class="pre">re_build_model</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">arch_params</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">{}</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/sg_model/sg_model.html#SgModel.re_build_model"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.sg_model.sg_model.SgModel.re_build_model" title="Permalink to this definition"></a></dt>
    630. 

  630. <dd><dl class="simple">
    631. 

  631. <dt>arch_params<span class="classifier">dict</span></dt><dd><p>Architecture H.P. e.g.: block, num_blocks, num_classes, etc.</p>
    632. 

  632. </dd>
    633. 

  633. </dl>
    634. 

  634. <dl class="field-list simple">
    635. 

  635. <dt class="field-odd">Returns</dt>
    636. 

  636. <dd class="field-odd"><p></p>
    637. 

  637. </dd>
    638. 

  638. </dl>
    639. 

  639. </dd></dl>
    640. 

  640. 

  641. <dl class="py method">
    642. 

  642. <dt class="sig sig-object py" id="super_gradients.training.sg_model.sg_model.SgModel.update_architecture">
    643. 

  643. <span class="sig-name descname"><span class="pre">update_architecture</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">structure</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/sg_model/sg_model.html#SgModel.update_architecture"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.sg_model.sg_model.SgModel.update_architecture" title="Permalink to this definition"></a></dt>
    644. 

  644. <dd><dl class="simple">
    645. 

  645. <dt>architecture<span class="classifier">str</span></dt><dd><p>Defines the network’s architecture according to the options in models/all_architectures</p>
    646. 

  646. </dd>
    647. 

  647. <dt>load_checkpoint<span class="classifier">bool</span></dt><dd><p>Loads a checkpoint according to experiment_name</p>
    648. 

  648. </dd>
    649. 

  649. <dt>arch_params<span class="classifier">dict</span></dt><dd><p>Architecture H.P. e.g.: block, num_blocks, num_classes, etc.</p>
    650. 

  650. </dd>
    651. 

  651. </dl>
    652. 

  652. <dl class="field-list simple">
    653. 

  653. <dt class="field-odd">Returns</dt>
    654. 

  654. <dd class="field-odd"><p></p>
    655. 

  655. </dd>
    656. 

  656. </dl>
    657. 

  657. </dd></dl>
    658. 

  658. 

  659. <dl class="py method">
    660. 

  660. <dt class="sig sig-object py" id="super_gradients.training.sg_model.sg_model.SgModel.get_module">
    661. 

  661. <span class="sig-name descname"><span class="pre">get_module</span></span><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/sg_model/sg_model.html#SgModel.get_module"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.sg_model.sg_model.SgModel.get_module" title="Permalink to this definition"></a></dt>
    662. 

  662. <dd></dd></dl>
    663. 

  663. 

  664. <dl class="py method">
    665. 

  665. <dt class="sig sig-object py" id="super_gradients.training.sg_model.sg_model.SgModel.set_module">
    666. 

  666. <span class="sig-name descname"><span class="pre">set_module</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">module</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/sg_model/sg_model.html#SgModel.set_module"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.sg_model.sg_model.SgModel.set_module" title="Permalink to this definition"></a></dt>
    667. 

  667. <dd></dd></dl>
    668. 

  668. 

  669. <dl class="py method">
    670. 

  670. <dt class="sig sig-object py" id="super_gradients.training.sg_model.sg_model.SgModel.test">
    671. 

  671. <span class="sig-name descname"><span class="pre">test</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">test_loader</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">Optional</span><span class="p"><span class="pre">[</span></span><span class="pre">torch.utils.data.dataloader.DataLoader</span><span class="p"><span class="pre">]</span></span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">None</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">loss</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">Optional</span><span class="p"><span class="pre">[</span></span><span class="pre">torch.nn.modules.loss._Loss</span><span class="p"><span class="pre">]</span></span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">None</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">silent_mode</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">bool</span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">False</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">test_metrics_list</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">loss_logging_items_names</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">metrics_progress_verbose</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">False</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">test_phase_callbacks</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">use_ema_net</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">True</span></span></em><span class="sig-paren">)</span> &#x2192; <span class="pre">tuple</span><a class="reference internal" href="_modules/super_gradients/training/sg_model/sg_model.html#SgModel.test"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.sg_model.sg_model.SgModel.test" title="Permalink to this definition"></a></dt>
    672. 

  672. <dd><p>Evaluates the model on given dataloader and metrics.</p>
    673. 

  673. <dl class="field-list simple">
    674. 

  674. <dt class="field-odd">Parameters</dt>
    675. 

  675. <dd class="field-odd"><ul class="simple">
    676. 

  676. <li><p><strong>test_loader</strong> – dataloader to perform test on.</p></li>
    677. 

  677. <li><p><strong>test_metrics_list</strong> – (list(torchmetrics.Metric)) metrics list for evaluation.</p></li>
    678. 

  678. <li><p><strong>silent_mode</strong> – (bool) controls verbosity</p></li>
    679. 

  679. <li><p><strong>metrics_progress_verbose</strong> – (bool) controls the verbosity of metrics progress (default=False). Slows down the program.</p></li>
    680. 

  680. </ul>
    681. 

  681. </dd>
    682. 

  682. </dl>
    683. 

  683. <dl class="simple">
    684. 

  684. <dt>:param use_ema_net (bool) whether to perform test on self.ema_model.ema (when self.ema_model.ema exists,</dt><dd><p>otherwise self.net will be tested) (default=True)</p>
    685. 

  685. </dd>
    686. 

  686. </dl>
    687. 

  687. <dl class="field-list simple">
    688. 

  688. <dt class="field-odd">Returns</dt>
    689. 

  689. <dd class="field-odd"><p>results tuple (tuple) containing the loss items and metric values.</p>
    690. 

  690. </dd>
    691. 

  691. </dl>
    692. 

  692. <dl class="simple">
    693. 

  693. <dt>All of the above args will override SgModel’s corresponding attribute when not equal to None. Then evaluation</dt><dd><p>is ran on self.test_loader with self.test_metrics.</p>
    694. 

  694. </dd>
    695. 

  695. </dl>
    696. 

  696. </dd></dl>
    697. 

  697. 

  698. <dl class="py method">
    699. 

  699. <dt class="sig sig-object py" id="super_gradients.training.sg_model.sg_model.SgModel.evaluate">
    700. 

  700. <span class="sig-name descname"><span class="pre">evaluate</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">data_loader</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">torch.utils.data.dataloader.DataLoader</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">metrics</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">torchmetrics.collections.MetricCollection</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">evaluation_type</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><a class="reference internal" href="#super_gradients.training.sg_model.sg_model.EvaluationType" title="super_gradients.training.sg_model.sg_model.EvaluationType"><span class="pre">super_gradients.training.sg_model.sg_model.EvaluationType</span></a></span></em>, <em class="sig-param"><span class="n"><span class="pre">epoch</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">Optional</span><span class="p"><span class="pre">[</span></span><span class="pre">int</span><span class="p"><span class="pre">]</span></span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">None</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">silent_mode</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">bool</span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">False</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">metrics_progress_verbose</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">bool</span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">False</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/sg_model/sg_model.html#SgModel.evaluate"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.sg_model.sg_model.SgModel.evaluate" title="Permalink to this definition"></a></dt>
    701. 

  701. <dd><p>Evaluates the model on given dataloader and metrics.</p>
    702. 

  702. <dl class="field-list simple">
    703. 

  703. <dt class="field-odd">Parameters</dt>
    704. 

  704. <dd class="field-odd"><ul class="simple">
    705. 

  705. <li><p><strong>data_loader</strong> – dataloader to perform evaluataion on</p></li>
    706. 

  706. <li><p><strong>metrics</strong> – (MetricCollection) metrics for evaluation</p></li>
    707. 

  707. <li><p><strong>evaluation_type</strong> – (EvaluationType) controls which phase callbacks will be used (for example, on batch end,
    708. 

  708. when evaluation_type=EvaluationType.VALIDATION the Phase.VALIDATION_BATCH_END callbacks will be triggered)</p></li>
    709. 

  709. <li><p><strong>epoch</strong> – (int) epoch idx</p></li>
    710. 

  710. <li><p><strong>silent_mode</strong> – (bool) controls verbosity</p></li>
    711. 

  711. <li><p><strong>metrics_progress_verbose</strong> – (bool) controls the verbosity of metrics progress (default=False).
    712. 

  712. Slows down the program significantly.</p></li>
    713. 

  713. </ul>
    714. 

  714. </dd>
    715. 

  715. <dt class="field-even">Returns</dt>
    716. 

  716. <dd class="field-even"><p>results tuple (tuple) containing the loss items and metric values.</p>
    717. 

  717. </dd>
    718. 

  718. </dl>
    719. 

  719. </dd></dl>
    720. 

  720. 

  721. <dl class="py method">
    722. 

  722. <dt class="sig sig-object py" id="super_gradients.training.sg_model.sg_model.SgModel.instantiate_net">
    723. 

  723. <span class="sig-name descname"><span class="pre">instantiate_net</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">architecture</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">Union</span><span class="p"><span class="pre">[</span></span><span class="pre">torch.nn.modules.module.Module</span><span class="p"><span class="pre">,</span> </span><span class="pre">type</span><span class="p"><span class="pre">,</span> </span><span class="pre">str</span><span class="p"><span class="pre">]</span></span></span></em>, <em class="sig-param"><span class="n"><span class="pre">arch_params</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">dict</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">checkpoint_params</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">dict</span></span></em>, <em class="sig-param"><span class="o"><span class="pre">*</span></span><span class="n"><span class="pre">args</span></span></em>, <em class="sig-param"><span class="o"><span class="pre">**</span></span><span class="n"><span class="pre">kwargs</span></span></em><span class="sig-paren">)</span> &#x2192; <span class="pre">tuple</span><a class="reference internal" href="_modules/super_gradients/training/sg_model/sg_model.html#SgModel.instantiate_net"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.sg_model.sg_model.SgModel.instantiate_net" title="Permalink to this definition"></a></dt>
    724. 

  724. <dd><dl class="simple">
    725. 

  725. <dt>Instantiates nn.Module according to architecture and arch_params, and handles pretrained weights and the required</dt><dd><p>module manipulation (i.e head replacement).</p>
    726. 

  726. </dd>
    727. 

  727. </dl>
    728. 

  728. <dl class="field-list simple">
    729. 

  729. <dt class="field-odd">Parameters</dt>
    730. 

  730. <dd class="field-odd"><ul class="simple">
    731. 

  731. <li><p><strong>architecture</strong> – String, torch.nn.Module or uninstantiated SgModule class describing the netowrks architecture.</p></li>
    732. 

  732. <li><p><strong>arch_params</strong> – Architecture’s parameters passed to networks c’tor.</p></li>
    733. 

  733. <li><p><strong>checkpoint_params</strong> – checkpoint loading related parameters dictionary with ‘pretrained_weights’ key,
    734. 

  734. s.t it’s value is a string describing the dataset of the pretrained weights (for example “imagenent”).</p></li>
    735. 

  735. </ul>
    736. 

  736. </dd>
    737. 

  737. <dt class="field-even">Returns</dt>
    738. 

  738. <dd class="field-even"><p>instantiated netowrk i.e torch.nn.Module, architecture_class (will be none when architecture is not str)</p>
    739. 

  739. </dd>
    740. 

  740. </dl>
    741. 

  741. </dd></dl>
    742. 

  742. 

  743. </dd></dl>
    744. 

  744. 

  745. </section>
    746. 

  746. <section id="module-super_gradients.training.sg_model">
    747. 

  747. <span id="module-contents"></span><h2>Module contents<a class="headerlink" href="#module-super_gradients.training.sg_model" title="Permalink to this headline"></a></h2>
    748. 

  748. <dl class="py class">
    749. 

  749. <dt class="sig sig-object py" id="super_gradients.training.sg_model.SgModel">
    750. 

  750. <em class="property"><span class="pre">class</span> </em><span class="sig-prename descclassname"><span class="pre">super_gradients.training.sg_model.</span></span><span class="sig-name descname"><span class="pre">SgModel</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="pre">experiment_name:</span> <span class="pre">str</span></em>, <em class="sig-param"><span class="pre">device:</span> <span class="pre">Optional[str]</span> <span class="pre">=</span> <span class="pre">None</span></em>, <em class="sig-param"><span class="pre">multi_gpu:</span> <span class="pre">Union[super_gradients.training.sg_model.sg_model.MultiGPUMode</span></em>, <em class="sig-param"><span class="pre">str]</span> <span class="pre">=</span> <span class="pre">&lt;MultiGPUMode.OFF:</span> <span class="pre">'Off'&gt;</span></em>, <em class="sig-param"><span class="pre">model_checkpoints_location:</span> <span class="pre">str</span> <span class="pre">=</span> <span class="pre">'local'</span></em>, <em class="sig-param"><span class="pre">overwrite_local_checkpoint:</span> <span class="pre">bool</span> <span class="pre">=</span> <span class="pre">True</span></em>, <em class="sig-param"><span class="pre">ckpt_name:</span> <span class="pre">str</span> <span class="pre">=</span> <span class="pre">'ckpt_latest.pth'</span></em>, <em class="sig-param"><span class="pre">post_prediction_callback:</span> <span class="pre">Optional[super_gradients.training.utils.detection_utils.DetectionPostPredictionCallback]</span> <span class="pre">=</span> <span class="pre">None</span></em>, <em class="sig-param"><span class="pre">ckpt_root_dir=None</span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/sg_model/sg_model.html#SgModel"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.sg_model.SgModel" title="Permalink to this definition"></a></dt>
    751. 

  751. <dd><p>Bases: <code class="xref py py-class docutils literal notranslate"><span class="pre">object</span></code></p>
    752. 

  752. <p>SuperGradient Model - Base Class for Sg Models</p>
    753. 

  753. <dl class="py method">
    754. 

  754. <dt class="sig sig-object py" id="super_gradients.training.sg_model.SgModel.train">
    755. 

  755. <span class="sig-name descname"><span class="pre">train</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">max_epochs</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">int</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">initial_epoch</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">int</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">save_model</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">bool</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/sg_model/sg_model.html#SgModel.train"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.sg_model.SgModel.train" title="Permalink to this definition"></a></dt>
    756. 

  756. <dd><p>the main function used for the training, h.p. updating, logging etc.</p>
    757. 

  757. </dd></dl>
    758. 

  758. 

  759. <dl class="py method">
    760. 

  760. <dt class="sig sig-object py" id="super_gradients.training.sg_model.SgModel.predict">
    761. 

  761. <span class="sig-name descname"><span class="pre">predict</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">idx</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">int</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/sg_model/sg_model.html#SgModel.predict"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.sg_model.SgModel.predict" title="Permalink to this definition"></a></dt>
    762. 

  762. <dd><p>returns the predictions and label of the current inputs</p>
    763. 

  763. </dd></dl>
    764. 

  764. 

  765. <dl class="py method">
    766. 

  766. <dt class="sig sig-object py">
    767. 

  767. <span class="sig-name descname"><span class="pre">test(epoch</span> <span class="pre">:</span> <span class="pre">int,</span> <span class="pre">idx</span> <span class="pre">:</span> <span class="pre">int,</span> <span class="pre">save</span> <span class="pre">:</span> <span class="pre">bool):</span></span></dt>
    768. 

  768. <dd><p>returns the test loss, accuracy and runtime</p>
    769. 

  769. </dd></dl>
    770. 

  770. 

  771. <dl class="py method">
    772. 

  772. <dt class="sig sig-object py" id="super_gradients.training.sg_model.SgModel.connect_dataset_interface">
    773. 

  773. <span class="sig-name descname"><span class="pre">connect_dataset_interface</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">dataset_interface</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><a class="reference internal" href="super_gradients.training.datasets.dataset_interfaces.html#super_gradients.training.datasets.dataset_interfaces.dataset_interface.DatasetInterface" title="super_gradients.training.datasets.dataset_interfaces.dataset_interface.DatasetInterface"><span class="pre">super_gradients.training.datasets.dataset_interfaces.dataset_interface.DatasetInterface</span></a></span></em>, <em class="sig-param"><span class="n"><span class="pre">data_loader_num_workers</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">int</span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">8</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/sg_model/sg_model.html#SgModel.connect_dataset_interface"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.sg_model.SgModel.connect_dataset_interface" title="Permalink to this definition"></a></dt>
    774. 

  774. <dd><dl class="field-list simple">
    775. 

  775. <dt class="field-odd">Parameters</dt>
    776. 

  776. <dd class="field-odd"><ul class="simple">
    777. 

  777. <li><p><strong>dataset_interface</strong> – DatasetInterface object</p></li>
    778. 

  778. <li><p><strong>data_loader_num_workers</strong> – The number of threads to initialize the Data Loaders with
    779. 

  779. The dataset to be connected</p></li>
    780. 

  780. </ul>
    781. 

  781. </dd>
    782. 

  782. </dl>
    783. 

  783. </dd></dl>
    784. 

  784. 

  785. <dl class="py method">
    786. 

  786. <dt class="sig sig-object py" id="super_gradients.training.sg_model.SgModel.build_model">
    787. 

  787. <span class="sig-name descname"><span class="pre">build_model</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">architecture</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">Union</span><span class="p"><span class="pre">[</span></span><span class="pre">str</span><span class="p"><span class="pre">,</span> </span><span class="pre">torch.nn.modules.module.Module</span><span class="p"><span class="pre">]</span></span></span></em>, <em class="sig-param"><span class="n"><span class="pre">arch_params</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">{}</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">checkpoint_params</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">{}</span></span></em>, <em class="sig-param"><span class="o"><span class="pre">*</span></span><span class="n"><span class="pre">args</span></span></em>, <em class="sig-param"><span class="o"><span class="pre">**</span></span><span class="n"><span class="pre">kwargs</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/sg_model/sg_model.html#SgModel.build_model"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.sg_model.SgModel.build_model" title="Permalink to this definition"></a></dt>
    788. 

  788. <dd><dl class="field-list simple">
    789. 

  789. <dt class="field-odd">Parameters</dt>
    790. 

  790. <dd class="field-odd"><ul class="simple">
    791. 

  791. <li><p><strong>architecture</strong> – Defines the network’s architecture from models/ALL_ARCHITECTURES</p></li>
    792. 

  792. <li><p><strong>arch_params</strong> – Architecture H.P. e.g.: block, num_blocks, num_classes, etc.</p></li>
    793. 

  793. <li><p><strong>checkpoint_params</strong> – <p>Dictionary like object with the following key:values:</p>
    794. 

  794. <p>load_checkpoint:            Load a pre-trained checkpoint
    795. 

  795. strict_load:                See StrictLoad class documentation for details.
    796. 

  796. source_ckpt_folder_name:    folder name to load the checkpoint from (self.experiment_name if none is given)
    797. 

  797. load_weights_only:          loads only the weight from the checkpoint and zeroize the training params
    798. 

  798. load_backbone:              loads the provided checkpoint to self.net.backbone instead of self.net
    799. 

  799. external_checkpoint_path:   The path to the external checkpoint to be loaded. Can be absolute or relative</p>
    800. 

  800. <blockquote>
    801. 

  801. <div><p>(ie: path/to/checkpoint.pth). If provided, will automatically attempt to
    802. 

  802. load the checkpoint even if the load_checkpoint flag is not provided.</p>
    803. 

  803. </div></blockquote>
    804. 

  804. </p></li>
    805. 

  805. </ul>
    806. 

  806. </dd>
    807. 

  807. </dl>
    808. 

  808. </dd></dl>
    809. 

  809. 

  810. <dl class="py method">
    811. 

  811. <dt class="sig sig-object py" id="super_gradients.training.sg_model.SgModel.backward_step">
    812. 

  812. <span class="sig-name descname"><span class="pre">backward_step</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">loss</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">torch.Tensor</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">epoch</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">int</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">batch_idx</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">int</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">context</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><a class="reference internal" href="super_gradients.training.utils.html#super_gradients.training.utils.callbacks.PhaseContext" title="super_gradients.training.utils.callbacks.PhaseContext"><span class="pre">super_gradients.training.utils.callbacks.PhaseContext</span></a></span></em>, <em class="sig-param"><span class="o"><span class="pre">*</span></span><span class="n"><span class="pre">args</span></span></em>, <em class="sig-param"><span class="o"><span class="pre">**</span></span><span class="n"><span class="pre">kwargs</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/sg_model/sg_model.html#SgModel.backward_step"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.sg_model.SgModel.backward_step" title="Permalink to this definition"></a></dt>
    813. 

  813. <dd><p>Run backprop on the loss and perform a step
    814. 

  814. :param loss: The value computed by the loss function
    815. 

  815. :param optimizer: An object that can perform a gradient step and zeroize model gradient
    816. 

  816. :param epoch: number of epoch the training is on
    817. 

  817. :param batch_idx: number of iteration inside the current epoch
    818. 

  818. :param context: current phase context
    819. 

  819. :return:</p>
    820. 

  820. </dd></dl>
    821. 

  821. 

  822. <dl class="py method">
    823. 

  823. <dt class="sig sig-object py" id="super_gradients.training.sg_model.SgModel.save_checkpoint">
    824. 

  824. <span class="sig-name descname"><span class="pre">save_checkpoint</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">optimizer</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">epoch</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">Optional</span><span class="p"><span class="pre">[</span></span><span class="pre">int</span><span class="p"><span class="pre">]</span></span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">None</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">validation_results_tuple</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">Optional</span><span class="p"><span class="pre">[</span></span><span class="pre">tuple</span><span class="p"><span class="pre">]</span></span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">None</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">context</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">Optional</span><span class="p"><span class="pre">[</span></span><a class="reference internal" href="super_gradients.training.utils.html#super_gradients.training.utils.callbacks.PhaseContext" title="super_gradients.training.utils.callbacks.PhaseContext"><span class="pre">super_gradients.training.utils.callbacks.PhaseContext</span></a><span class="p"><span class="pre">]</span></span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">None</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/sg_model/sg_model.html#SgModel.save_checkpoint"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.sg_model.SgModel.save_checkpoint" title="Permalink to this definition"></a></dt>
    825. 

  825. <dd><p>Save the current state dict as latest (always), best (if metric was improved), epoch# (if determined in training
    826. 

  826. params)</p>
    827. 

  827. </dd></dl>
    828. 

  828. 

  829. <dl class="py method">
    830. 

  830. <dt class="sig sig-object py" id="id4">
    831. 

  831. <span class="sig-name descname"><span class="pre">train</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">training_params</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">dict</span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">{}</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/sg_model/sg_model.html#SgModel.train"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#id4" title="Permalink to this definition"></a></dt>
    832. 

  832. <dd><p>train - Trains the Model</p>
    833. 

  833. <dl>
    834. 

  834. <dt>IMPORTANT NOTE: Additional batch parameters can be added as a third item (optional) if a tuple is returned by</dt><dd><p>the data loaders, as dictionary. The phase context will hold the additional items, under an attribute with
    835. 

  835. the same name as the key in this dictionary. Then such items can be accessed through phase callbacks.</p>
    836. 

  836. <blockquote>
    837. 

  837. <div><dl class="field-list">
    838. 

  838. <dt class="field-odd">param training_params</dt>
    839. 

  839. <dd class="field-odd"><ul>
    840. 

  840. <li><p><cite>max_epochs</cite> : int</p>
    841. 

  841. <blockquote>
    842. 

  842. <div><p>Number of epochs to run training.</p>
    843. 

  843. </div></blockquote>
    844. 

  844. </li>
    845. 

  845. <li><p><cite>lr_updates</cite> : list(int)</p>
    846. 

  846. <blockquote>
    847. 

  847. <div><p>List of fixed epoch numbers to perform learning rate updates when <cite>lr_mode=’step’</cite>.</p>
    848. 

  848. </div></blockquote>
    849. 

  849. </li>
    850. 

  850. <li><p><cite>lr_decay_factor</cite> : float</p>
    851. 

  851. <blockquote>
    852. 

  852. <div><p>Decay factor to apply to the learning rate at each update when <cite>lr_mode=’step’</cite>.</p>
    853. 

  853. </div></blockquote>
    854. 

  854. </li>
    855. 

  855. <li><p><cite>lr_mode</cite> : str</p>
    856. 

  856. <blockquote>
    857. 

  857. <div><p>Learning rate scheduling policy, one of [‘step’,’poly’,’cosine’,’function’]. ‘step’ refers to
    858. 

  858. constant updates at epoch numbers passed through <cite>lr_updates</cite>. ‘cosine’ refers to Cosine Anealing
    859. 

  859. policy as mentioned in <a class="reference external" href="https://arxiv.org/abs/1608.03983">https://arxiv.org/abs/1608.03983</a>. ‘poly’ refers to polynomial decrease i.e
    860. 

  860. in each epoch iteration <cite>self.lr = self.initial_lr * pow((1.0 - (current_iter / max_iter)),
    861. 

  861. 0.9)</cite> ‘function’ refers to user defined learning rate scheduling function, that is passed through
    862. 

  862. <cite>lr_schedule_function</cite>.</p>
    863. 

  863. </div></blockquote>
    864. 

  864. </li>
    865. 

  865. <li><p><cite>lr_schedule_function</cite> : Union[callable,None]</p>
    866. 

  866. <blockquote>
    867. 

  867. <div><p>Learning rate scheduling function to be used when <cite>lr_mode</cite> is ‘function’.</p>
    868. 

  868. </div></blockquote>
    869. 

  869. </li>
    870. 

  870. <li><p><cite>lr_warmup_epochs</cite> : int (default=0)</p>
    871. 

  871. <blockquote>
    872. 

  872. <div><p>Number of epochs for learning rate warm up - see <a class="reference external" href="https://arxiv.org/pdf/1706.02677.pdf">https://arxiv.org/pdf/1706.02677.pdf</a> (Section 2.2).</p>
    873. 

  873. </div></blockquote>
    874. 

  874. </li>
    875. 

  875. <li><dl class="simple">
    876. 

  876. <dt><cite>cosine_final_lr_ratio</cite><span class="classifier">float (default=0.01)</span></dt><dd><dl class="simple">
    877. 

  877. <dt>Final learning rate ratio (only relevant when <a href="#id5"><span class="problematic" id="id6">`</span></a>lr_mode`=’cosine’). The cosine starts from initial_lr and reaches</dt><dd><p>initial_lr * cosine_final_lr_ratio in last epoch</p>
    878. 

  878. </dd>
    879. 

  879. </dl>
    880. 

  880. </dd>
    881. 

  881. </dl>
    882. 

  882. </li>
    883. 

  883. <li><p><cite>inital_lr</cite> : float</p>
    884. 

  884. <blockquote>
    885. 

  885. <div><p>Initial learning rate.</p>
    886. 

  886. </div></blockquote>
    887. 

  887. </li>
    888. 

  888. <li><p><cite>loss</cite> : Union[nn.module, str]</p>
    889. 

  889. <blockquote>
    890. 

  890. <div><p>Loss function for training.
    891. 

  891. One of SuperGradient’s built in options:</p>
    892. 

  892. <blockquote>
    893. 

  893. <div><p>“cross_entropy”: LabelSmoothingCrossEntropyLoss,
    894. 

  894. “mse”: MSELoss,
    895. 

  895. “r_squared_loss”: RSquaredLoss,
    896. 

  896. “detection_loss”: YoLoV3DetectionLoss,
    897. 

  897. “shelfnet_ohem_loss”: ShelfNetOHEMLoss,
    898. 

  898. “shelfnet_se_loss”: ShelfNetSemanticEncodingLoss,
    899. 

  899. “yolo_v5_loss”: YoLoV5DetectionLoss,
    900. 

  900. “ssd_loss”: SSDLoss,</p>
    901. 

  901. </div></blockquote>
    902. 

  902. <p>or user defined nn.module loss function.</p>
    903. 

  903. <p>IMPORTANT: forward(…) should return a (loss, loss_items) tuple where loss is the tensor used
    904. 

  904. for backprop (i.e what your original loss function returns), and loss_items should be a tensor of
    905. 

  905. shape (n_items), of values computed during the forward pass which we desire to log over the
    906. 

  906. entire epoch. For example- the loss itself should always be logged. Another example is a scenario
    907. 

  907. where the computed loss is the sum of a few components we would like to log- these entries in
    908. 

  908. loss_items).</p>
    909. 

  909. <p>When training, set the loss_logging_items_names parameter in train_params to be a list of
    910. 

  910. strings, of length n_items who’s ith element is the name of the ith entry in loss_items. Then
    911. 

  911. each item will be logged, rendered on tensorboard and “watched” (i.e saving model checkpoints
    912. 

  912. according to it).</p>
    913. 

  913. <p>Since running logs will save the loss_items in some internal state, it is recommended that
    914. 

  914. loss_items are detached from their computational graph for memory efficiency.</p>
    915. 

  915. </div></blockquote>
    916. 

  916. </li>
    917. 

  917. <li><p><cite>optimizer</cite> : Union[str, torch.optim.Optimizer]</p>
    918. 

  918. <blockquote>
    919. 

  919. <div><p>Optimization algorithm. One of [‘Adam’,’SGD’,’RMSProp’] corresponding to the torch.optim
    920. 

  920. optimzers implementations, or any object that implements torch.optim.Optimizer.</p>
    921. 

  921. </div></blockquote>
    922. 

  922. </li>
    923. 

  923. <li><p><cite>criterion_params</cite> : dict</p>
    924. 

  924. <blockquote>
    925. 

  925. <div><p>Loss function parameters.</p>
    926. 

  926. </div></blockquote>
    927. 

  927. </li>
    928. 

  928. <li><dl>
    929. 

  929. <dt><cite>optimizer_params</cite><span class="classifier">dict</span></dt><dd><p>When <cite>optimizer</cite> is one of [‘Adam’,’SGD’,’RMSProp’], it will be initialized with optimizer_params.</p>
    930. 

  930. <p>(see <a class="reference external" href="https://pytorch.org/docs/stable/optim.html">https://pytorch.org/docs/stable/optim.html</a> for the full list of
    931. 

  931. parameters for each optimizer).</p>
    932. 

  932. </dd>
    933. 

  933. </dl>
    934. 

  934. </li>
    935. 

  935. <li><p><cite>train_metrics_list</cite> : list(torchmetrics.Metric)</p>
    936. 

  936. <blockquote>
    937. 

  937. <div><p>Metrics to log during training. For more information on torchmetrics see
    938. 

  938. <a class="reference external" href="https://torchmetrics.rtfd.io/en/latest/">https://torchmetrics.rtfd.io/en/latest/</a>.</p>
    939. 

  939. </div></blockquote>
    940. 

  940. </li>
    941. 

  941. <li><p><cite>valid_metrics_list</cite> : list(torchmetrics.Metric)</p>
    942. 

  942. <blockquote>
    943. 

  943. <div><p>Metrics to log during validation/testing. For more information on torchmetrics see
    944. 

  944. <a class="reference external" href="https://torchmetrics.rtfd.io/en/latest/">https://torchmetrics.rtfd.io/en/latest/</a>.</p>
    945. 

  945. </div></blockquote>
    946. 

  946. </li>
    947. 

  947. <li><p><cite>loss_logging_items_names</cite> : list(str)</p>
    948. 

  948. <blockquote>
    949. 

  949. <div><p>The list of names/titles for the outputs returned from the loss functions forward pass (reminder-
    950. 

  950. the loss function should return the tuple (loss, loss_items)). These names will be used for
    951. 

  951. logging their values.</p>
    952. 

  952. </div></blockquote>
    953. 

  953. </li>
    954. 

  954. <li><p><cite>metric_to_watch</cite> : str (default=”Accuracy”)</p>
    955. 

  955. <blockquote>
    956. 

  956. <div><p>will be the metric which the model checkpoint will be saved according to, and can be set to any
    957. 

  957. of the following:</p>
    958. 

  958. <blockquote>
    959. 

  959. <div><p>a metric name (str) of one of the metric objects from the valid_metrics_list</p>
    960. 

  960. <p>a “metric_name” if some metric in valid_metrics_list has an attribute component_names which
    961. 

  961. is a list referring to the names of each entry in the output metric (torch tensor of size n)</p>
    962. 

  962. <p>one of “loss_logging_items_names” i.e which will correspond to an item returned during the
    963. 

  963. loss function’s forward pass.</p>
    964. 

  964. </div></blockquote>
    965. 

  965. <p>At the end of each epoch, if a new best metric_to_watch value is achieved, the models checkpoint
    966. 

  966. is saved in YOUR_PYTHON_PATH/checkpoints/ckpt_best.pth</p>
    967. 

  967. </div></blockquote>
    968. 

  968. </li>
    969. 

  969. <li><p><cite>greater_metric_to_watch_is_better</cite> : bool</p>
    970. 

  970. <blockquote>
    971. 

  971. <div><dl class="simple">
    972. 

  972. <dt>When choosing a model’s checkpoint to be saved, the best achieved model is the one that maximizes the</dt><dd><p>metric_to_watch when this parameter is set to True, and a one that minimizes it otherwise.</p>
    973. 

  973. </dd>
    974. 

  974. </dl>
    975. 

  975. </div></blockquote>
    976. 

  976. </li>
    977. 

  977. <li><p><cite>ema</cite> : bool (default=False)</p>
    978. 

  978. <blockquote>
    979. 

  979. <div><p>Whether to use Model Exponential Moving Average (see
    980. 

  980. <a class="reference external" href="https://github.com/rwightman/pytorch-image-models">https://github.com/rwightman/pytorch-image-models</a> ema implementation)</p>
    981. 

  981. </div></blockquote>
    982. 

  982. </li>
    983. 

  983. <li><p><cite>batch_accumulate</cite> : int (default=1)</p>
    984. 

  984. <blockquote>
    985. 

  985. <div><p>Number of batches to accumulate before every backward pass.</p>
    986. 

  986. </div></blockquote>
    987. 

  987. </li>
    988. 

  988. <li><p><cite>ema_params</cite> : dict</p>
    989. 

  989. <blockquote>
    990. 

  990. <div><p>Parameters for the ema model.</p>
    991. 

  991. </div></blockquote>
    992. 

  992. </li>
    993. 

  993. <li><p><cite>zero_weight_decay_on_bias_and_bn</cite> : bool (default=False)</p>
    994. 

  994. <blockquote>
    995. 

  995. <div><p>Whether to apply weight decay on batch normalization parameters or not (ignored when the passed
    996. 

  996. optimizer has already been initialized).</p>
    997. 

  997. </div></blockquote>
    998. 

  998. </li>
    999. 

  999. <li><p><cite>load_opt_params</cite> : bool (default=True)</p>
    1000. 

  1000. <blockquote>
    1001. 

  1001. <div><p>Whether to load the optimizers parameters as well when loading a model’s checkpoint.</p>
    1002. 

  1002. </div></blockquote>
    1003. 

  1003. </li>
    1004. 

  1004. <li><p><cite>run_validation_freq</cite> : int (default=1)</p>
    1005. 

  1005. <blockquote>
    1006. 

  1006. <div><dl class="simple">
    1007. 

  1007. <dt>The frequency in which validation is performed during training (i.e the validation is ran every</dt><dd><p><cite>run_validation_freq</cite> epochs.</p>
    1008. 

  1008. </dd>
    1009. 

  1009. </dl>
    1010. 

  1010. </div></blockquote>
    1011. 

  1011. </li>
    1012. 

  1012. <li><p><cite>save_model</cite> : bool (default=True)</p>
    1013. 

  1013. <blockquote>
    1014. 

  1014. <div><p>Whether to save the model checkpoints.</p>
    1015. 

  1015. </div></blockquote>
    1016. 

  1016. </li>
    1017. 

  1017. <li><p><cite>silent_mode</cite> : bool</p>
    1018. 

  1018. <blockquote>
    1019. 

  1019. <div><p>Silents the print outs.</p>
    1020. 

  1020. </div></blockquote>
    1021. 

  1021. </li>
    1022. 

  1022. <li><p><cite>mixed_precision</cite> : bool</p>
    1023. 

  1023. <blockquote>
    1024. 

  1024. <div><p>Whether to use mixed precision or not.</p>
    1025. 

  1025. </div></blockquote>
    1026. 

  1026. </li>
    1027. 

  1027. <li><p><cite>save_ckpt_epoch_list</cite> : list(int) (default=[])</p>
    1028. 

  1028. <blockquote>
    1029. 

  1029. <div><p>List of fixed epoch indices the user wishes to save checkpoints in.</p>
    1030. 

  1030. </div></blockquote>
    1031. 

  1031. </li>
    1032. 

  1032. <li><p><cite>average_best_models</cite> : bool (default=False)</p>
    1033. 

  1033. <blockquote>
    1034. 

  1034. <div><p>If set, a snapshot dictionary file and the average model will be saved / updated at every epoch
    1035. 

  1035. and evaluated only when training is completed. The snapshot file will only be deleted upon
    1036. 

  1036. completing the training. The snapshot dict will be managed on cpu.</p>
    1037. 

  1037. </div></blockquote>
    1038. 

  1038. </li>
    1039. 

  1039. <li><p><cite>precise_bn</cite> : bool (default=False)</p>
    1040. 

  1040. <blockquote>
    1041. 

  1041. <div><p>Whether to use precise_bn calculation during the training.</p>
    1042. 

  1042. </div></blockquote>
    1043. 

  1043. </li>
    1044. 

  1044. <li><p><cite>precise_bn_batch_size</cite> : int (default=None)</p>
    1045. 

  1045. <blockquote>
    1046. 

  1046. <div><p>The effective batch size we want to calculate the batchnorm on. For example, if we are training a model
    1047. 

  1047. on 8 gpus, with a batch of 128 on each gpu, a good rule of thumb would be to give it 8192
    1048. 

  1048. (ie: effective_batch_size * num_gpus = batch_per_gpu * num_gpus * num_gpus).
    1049. 

  1049. If precise_bn_batch_size is not provided in the training_params, the latter heuristic will be taken.</p>
    1050. 

  1050. </div></blockquote>
    1051. 

  1051. </li>
    1052. 

  1052. <li><p><cite>seed</cite> : int (default=42)</p>
    1053. 

  1053. <blockquote>
    1054. 

  1054. <div><p>Random seed to be set for torch, numpy, and random. When using DDP each process will have it’s seed
    1055. 

  1055. set to seed + rank.</p>
    1056. 

  1056. </div></blockquote>
    1057. 

  1057. </li>
    1058. 

  1058. <li><p><cite>log_installed_packages</cite> : bool (default=False)</p>
    1059. 

  1059. <blockquote>
    1060. 

  1060. <div><dl class="simple">
    1061. 

  1061. <dt>When set, the list of all installed packages (and their versions) will be written to the tensorboard</dt><dd><p>and logfile (useful when trying to reproduce results).</p>
    1062. 

  1062. </dd>
    1063. 

  1063. </dl>
    1064. 

  1064. </div></blockquote>
    1065. 

  1065. </li>
    1066. 

  1066. <li><p><cite>dataset_statistics</cite> : bool (default=False)</p>
    1067. 

  1067. <blockquote>
    1068. 

  1068. <div><p>Enable a statistic analysis of the dataset. If set to True the dataset will be analyzed and a report
    1069. 

  1069. will be added to the tensorboard along with some sample images from the dataset. Currently only
    1070. 

  1070. detection datasets are supported for analysis.</p>
    1071. 

  1071. </div></blockquote>
    1072. 

  1072. </li>
    1073. 

  1073. <li><p><cite>save_full_train_log</cite> : bool (default=False)</p>
    1074. 

  1074. <blockquote>
    1075. 

  1075. <div><dl class="simple">
    1076. 

  1076. <dt>When set, a full log (of all super_gradients modules, including uncaught exceptions from any other</dt><dd><p>module) of the training will be saved in the checkpoint directory under full_train_log.log</p>
    1077. 

  1077. </dd>
    1078. 

  1078. </dl>
    1079. 

  1079. </div></blockquote>
    1080. 

  1080. </li>
    1081. 

  1081. <li><p><cite>sg_logger</cite> : Union[AbstractSGLogger, str] (defauls=base_sg_logger)</p>
    1082. 

  1082. <blockquote>
    1083. 

  1083. <div><p>Define the SGLogger object for this training process. The SGLogger handles all disk writes, logs, TensorBoard, remote logging
    1084. 

  1084. and remote storage. By overriding the default base_sg_logger, you can change the storage location, support external monitoring and logging
    1085. 

  1085. or support remote storage.</p>
    1086. 

  1086. </div></blockquote>
    1087. 

  1087. </li>
    1088. 

  1088. <li><p><cite>sg_logger_params</cite> : dict</p>
    1089. 

  1089. <p>SGLogger parameters</p>
    1090. 

  1090. </li>
    1091. 

  1091. <li><p><cite>clip_grad_norm</cite> : float</p>
    1092. 

  1092. <p>Defines a maximal L2 norm of the gradients. Values which exceed the given value will be clipped</p>
    1093. 

  1093. </li>
    1094. 

  1094. </ul>
    1095. 

  1095. </dd>
    1096. 

  1096. </dl>
    1097. 

  1097. </div></blockquote>
    1098. 

  1098. </dd>
    1099. 

  1099. </dl>
    1100. 

  1100. <dl class="field-list simple">
    1101. 

  1101. <dt class="field-odd">Returns</dt>
    1102. 

  1102. <dd class="field-odd"><p></p>
    1103. 

  1103. </dd>
    1104. 

  1104. </dl>
    1105. 

  1105. </dd></dl>
    1106. 

  1106. 

  1107. <dl class="py method">
    1108. 

  1108. <dt class="sig sig-object py" id="id7">
    1109. 

  1109. <span class="sig-name descname"><span class="pre">predict</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">inputs</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">targets</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">half</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">False</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">normalize</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">False</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">verbose</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">False</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">move_outputs_to_cpu</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">True</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/sg_model/sg_model.html#SgModel.predict"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#id7" title="Permalink to this definition"></a></dt>
    1110. 

  1110. <dd><p>A fast predictor for a batch of inputs
    1111. 

  1111. :param inputs: torch.tensor or numpy.array</p>
    1112. 

  1112. <blockquote>
    1113. 

  1113. <div><p>a batch of inputs</p>
    1114. 

  1114. </div></blockquote>
    1115. 

  1115. <dl class="field-list simple">
    1116. 

  1116. <dt class="field-odd">Parameters</dt>
    1117. 

  1117. <dd class="field-odd"><ul class="simple">
    1118. 

  1118. <li><p><strong>targets</strong> – torch.tensor()
    1119. 

  1119. corresponding labels - if non are given - accuracy will not be computed</p></li>
    1120. 

  1120. <li><p><strong>verbose</strong> – bool
    1121. 

  1121. print the results to screen</p></li>
    1122. 

  1122. <li><p><strong>normalize</strong> – bool
    1123. 

  1123. If true, normalizes the tensor according to the dataloader’s normalization values</p></li>
    1124. 

  1124. <li><p><strong>half</strong> – Performs half precision evaluation</p></li>
    1125. 

  1125. <li><p><strong>move_outputs_to_cpu</strong> – Moves the results from the GPU to the CPU</p></li>
    1126. 

  1126. </ul>
    1127. 

  1127. </dd>
    1128. 

  1128. <dt class="field-even">Returns</dt>
    1129. 

  1129. <dd class="field-even"><p>outputs, acc, net_time, gross_time
    1130. 

  1130. networks predictions, accuracy calculation, forward pass net time, function gross time</p>
    1131. 

  1131. </dd>
    1132. 

  1132. </dl>
    1133. 

  1133. </dd></dl>
    1134. 

  1134. 

  1135. <dl class="py method">
    1136. 

  1136. <dt class="sig sig-object py" id="super_gradients.training.sg_model.SgModel.compute_model_runtime">
    1137. 

  1137. <span class="sig-name descname"><span class="pre">compute_model_runtime</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">input_dims</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">Optional</span><span class="p"><span class="pre">[</span></span><span class="pre">tuple</span><span class="p"><span class="pre">]</span></span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">None</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">batch_sizes</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">Union</span><span class="p"><span class="pre">[</span></span><span class="pre">tuple</span><span class="p"><span class="pre">,</span> </span><span class="pre">list</span><span class="p"><span class="pre">,</span> </span><span class="pre">int</span><span class="p"><span class="pre">]</span></span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">(1,</span> <span class="pre">8,</span> <span class="pre">16,</span> <span class="pre">32,</span> <span class="pre">64)</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">verbose</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">bool</span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">True</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/sg_model/sg_model.html#SgModel.compute_model_runtime"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.sg_model.SgModel.compute_model_runtime" title="Permalink to this definition"></a></dt>
    1138. 

  1138. <dd><p>Compute the “atomic” inference time and throughput.
    1139. 

  1139. Atomic refers to calculating the forward pass independently, discarding effects such as data augmentation,
    1140. 

  1140. data upload to device, multi-gpu distribution etc.
    1141. 

  1141. :param input_dims: tuple</p>
    1142. 

  1142. <blockquote>
    1143. 

  1143. <div><p>shape of a basic input to the network (without the first index) e.g. (3, 224, 224)
    1144. 

  1144. if None uses an input from the test loader</p>
    1145. 

  1145. </div></blockquote>
    1146. 

  1146. <dl class="field-list simple">
    1147. 

  1147. <dt class="field-odd">Parameters</dt>
    1148. 

  1148. <dd class="field-odd"><ul class="simple">
    1149. 

  1149. <li><p><strong>batch_sizes</strong> – int or list
    1150. 

  1150. Batch sizes for latency calculation</p></li>
    1151. 

  1151. <li><p><strong>verbose</strong> – bool
    1152. 

  1152. Prints results to screen</p></li>
    1153. 

  1153. </ul>
    1154. 

  1154. </dd>
    1155. 

  1155. <dt class="field-even">Returns</dt>
    1156. 

  1156. <dd class="field-even"><p>log: dict
    1157. 

  1157. Latency and throughput for each tested batch size</p>
    1158. 

  1158. </dd>
    1159. 

  1159. </dl>
    1160. 

  1160. </dd></dl>
    1161. 

  1161. 

  1162. <dl class="py method">
    1163. 

  1163. <dt class="sig sig-object py" id="super_gradients.training.sg_model.SgModel.get_arch_params">
    1164. 

  1164. <span class="sig-name descname"><span class="pre">get_arch_params</span></span><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/sg_model/sg_model.html#SgModel.get_arch_params"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.sg_model.SgModel.get_arch_params" title="Permalink to this definition"></a></dt>
    1165. 

  1165. <dd></dd></dl>
    1166. 

  1166. 

  1167. <dl class="py method">
    1168. 

  1168. <dt class="sig sig-object py" id="super_gradients.training.sg_model.SgModel.get_structure">
    1169. 

  1169. <span class="sig-name descname"><span class="pre">get_structure</span></span><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/sg_model/sg_model.html#SgModel.get_structure"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.sg_model.SgModel.get_structure" title="Permalink to this definition"></a></dt>
    1170. 

  1170. <dd></dd></dl>
    1171. 

  1171. 

  1172. <dl class="py method">
    1173. 

  1173. <dt class="sig sig-object py" id="super_gradients.training.sg_model.SgModel.get_architecture">
    1174. 

  1174. <span class="sig-name descname"><span class="pre">get_architecture</span></span><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/sg_model/sg_model.html#SgModel.get_architecture"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.sg_model.SgModel.get_architecture" title="Permalink to this definition"></a></dt>
    1175. 

  1175. <dd></dd></dl>
    1176. 

  1176. 

  1177. <dl class="py method">
    1178. 

  1178. <dt class="sig sig-object py" id="super_gradients.training.sg_model.SgModel.set_experiment_name">
    1179. 

  1179. <span class="sig-name descname"><span class="pre">set_experiment_name</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">experiment_name</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/sg_model/sg_model.html#SgModel.set_experiment_name"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.sg_model.SgModel.set_experiment_name" title="Permalink to this definition"></a></dt>
    1180. 

  1180. <dd></dd></dl>
    1181. 

  1181. 

  1182. <dl class="py method">
    1183. 

  1183. <dt class="sig sig-object py" id="super_gradients.training.sg_model.SgModel.re_build_model">
    1184. 

  1184. <span class="sig-name descname"><span class="pre">re_build_model</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">arch_params</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">{}</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/sg_model/sg_model.html#SgModel.re_build_model"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.sg_model.SgModel.re_build_model" title="Permalink to this definition"></a></dt>
    1185. 

  1185. <dd><dl class="simple">
    1186. 

  1186. <dt>arch_params<span class="classifier">dict</span></dt><dd><p>Architecture H.P. e.g.: block, num_blocks, num_classes, etc.</p>
    1187. 

  1187. </dd>
    1188. 

  1188. </dl>
    1189. 

  1189. <dl class="field-list simple">
    1190. 

  1190. <dt class="field-odd">Returns</dt>
    1191. 

  1191. <dd class="field-odd"><p></p>
    1192. 

  1192. </dd>
    1193. 

  1193. </dl>
    1194. 

  1194. </dd></dl>
    1195. 

  1195. 

  1196. <dl class="py method">
    1197. 

  1197. <dt class="sig sig-object py" id="super_gradients.training.sg_model.SgModel.update_architecture">
    1198. 

  1198. <span class="sig-name descname"><span class="pre">update_architecture</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">structure</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/sg_model/sg_model.html#SgModel.update_architecture"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.sg_model.SgModel.update_architecture" title="Permalink to this definition"></a></dt>
    1199. 

  1199. <dd><dl class="simple">
    1200. 

  1200. <dt>architecture<span class="classifier">str</span></dt><dd><p>Defines the network’s architecture according to the options in models/all_architectures</p>
    1201. 

  1201. </dd>
    1202. 

  1202. <dt>load_checkpoint<span class="classifier">bool</span></dt><dd><p>Loads a checkpoint according to experiment_name</p>
    1203. 

  1203. </dd>
    1204. 

  1204. <dt>arch_params<span class="classifier">dict</span></dt><dd><p>Architecture H.P. e.g.: block, num_blocks, num_classes, etc.</p>
    1205. 

  1205. </dd>
    1206. 

  1206. </dl>
    1207. 

  1207. <dl class="field-list simple">
    1208. 

  1208. <dt class="field-odd">Returns</dt>
    1209. 

  1209. <dd class="field-odd"><p></p>
    1210. 

  1210. </dd>
    1211. 

  1211. </dl>
    1212. 

  1212. </dd></dl>
    1213. 

  1213. 

  1214. <dl class="py method">
    1215. 

  1215. <dt class="sig sig-object py" id="super_gradients.training.sg_model.SgModel.get_module">
    1216. 

  1216. <span class="sig-name descname"><span class="pre">get_module</span></span><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/sg_model/sg_model.html#SgModel.get_module"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.sg_model.SgModel.get_module" title="Permalink to this definition"></a></dt>
    1217. 

  1217. <dd></dd></dl>
    1218. 

  1218. 

  1219. <dl class="py method">
    1220. 

  1220. <dt class="sig sig-object py" id="super_gradients.training.sg_model.SgModel.set_module">
    1221. 

  1221. <span class="sig-name descname"><span class="pre">set_module</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">module</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/sg_model/sg_model.html#SgModel.set_module"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.sg_model.SgModel.set_module" title="Permalink to this definition"></a></dt>
    1222. 

  1222. <dd></dd></dl>
    1223. 

  1223. 

  1224. <dl class="py method">
    1225. 

  1225. <dt class="sig sig-object py" id="super_gradients.training.sg_model.SgModel.test">
    1226. 

  1226. <span class="sig-name descname"><span class="pre">test</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">test_loader</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">Optional</span><span class="p"><span class="pre">[</span></span><span class="pre">torch.utils.data.dataloader.DataLoader</span><span class="p"><span class="pre">]</span></span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">None</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">loss</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">Optional</span><span class="p"><span class="pre">[</span></span><span class="pre">torch.nn.modules.loss._Loss</span><span class="p"><span class="pre">]</span></span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">None</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">silent_mode</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">bool</span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">False</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">test_metrics_list</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">loss_logging_items_names</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">metrics_progress_verbose</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">False</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">test_phase_callbacks</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">use_ema_net</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">True</span></span></em><span class="sig-paren">)</span> &#x2192; <span class="pre">tuple</span><a class="reference internal" href="_modules/super_gradients/training/sg_model/sg_model.html#SgModel.test"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.sg_model.SgModel.test" title="Permalink to this definition"></a></dt>
    1227. 

  1227. <dd><p>Evaluates the model on given dataloader and metrics.</p>
    1228. 

  1228. <dl class="field-list simple">
    1229. 

  1229. <dt class="field-odd">Parameters</dt>
    1230. 

  1230. <dd class="field-odd"><ul class="simple">
    1231. 

  1231. <li><p><strong>test_loader</strong> – dataloader to perform test on.</p></li>
    1232. 

  1232. <li><p><strong>test_metrics_list</strong> – (list(torchmetrics.Metric)) metrics list for evaluation.</p></li>
    1233. 

  1233. <li><p><strong>silent_mode</strong> – (bool) controls verbosity</p></li>
    1234. 

  1234. <li><p><strong>metrics_progress_verbose</strong> – (bool) controls the verbosity of metrics progress (default=False). Slows down the program.</p></li>
    1235. 

  1235. </ul>
    1236. 

  1236. </dd>
    1237. 

  1237. </dl>
    1238. 

  1238. <dl class="simple">
    1239. 

  1239. <dt>:param use_ema_net (bool) whether to perform test on self.ema_model.ema (when self.ema_model.ema exists,</dt><dd><p>otherwise self.net will be tested) (default=True)</p>
    1240. 

  1240. </dd>
    1241. 

  1241. </dl>
    1242. 

  1242. <dl class="field-list simple">
    1243. 

  1243. <dt class="field-odd">Returns</dt>
    1244. 

  1244. <dd class="field-odd"><p>results tuple (tuple) containing the loss items and metric values.</p>
    1245. 

  1245. </dd>
    1246. 

  1246. </dl>
    1247. 

  1247. <dl class="simple">
    1248. 

  1248. <dt>All of the above args will override SgModel’s corresponding attribute when not equal to None. Then evaluation</dt><dd><p>is ran on self.test_loader with self.test_metrics.</p>
    1249. 

  1249. </dd>
    1250. 

  1250. </dl>
    1251. 

  1251. </dd></dl>
    1252. 

  1252. 

  1253. <dl class="py method">
    1254. 

  1254. <dt class="sig sig-object py" id="super_gradients.training.sg_model.SgModel.evaluate">
    1255. 

  1255. <span class="sig-name descname"><span class="pre">evaluate</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">data_loader</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">torch.utils.data.dataloader.DataLoader</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">metrics</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">torchmetrics.collections.MetricCollection</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">evaluation_type</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><a class="reference internal" href="#super_gradients.training.sg_model.sg_model.EvaluationType" title="super_gradients.training.sg_model.sg_model.EvaluationType"><span class="pre">super_gradients.training.sg_model.sg_model.EvaluationType</span></a></span></em>, <em class="sig-param"><span class="n"><span class="pre">epoch</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">Optional</span><span class="p"><span class="pre">[</span></span><span class="pre">int</span><span class="p"><span class="pre">]</span></span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">None</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">silent_mode</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">bool</span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">False</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">metrics_progress_verbose</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">bool</span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">False</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/sg_model/sg_model.html#SgModel.evaluate"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.sg_model.SgModel.evaluate" title="Permalink to this definition"></a></dt>
    1256. 

  1256. <dd><p>Evaluates the model on given dataloader and metrics.</p>
    1257. 

  1257. <dl class="field-list simple">
    1258. 

  1258. <dt class="field-odd">Parameters</dt>
    1259. 

  1259. <dd class="field-odd"><ul class="simple">
    1260. 

  1260. <li><p><strong>data_loader</strong> – dataloader to perform evaluataion on</p></li>
    1261. 

  1261. <li><p><strong>metrics</strong> – (MetricCollection) metrics for evaluation</p></li>
    1262. 

  1262. <li><p><strong>evaluation_type</strong> – (EvaluationType) controls which phase callbacks will be used (for example, on batch end,
    1263. 

  1263. when evaluation_type=EvaluationType.VALIDATION the Phase.VALIDATION_BATCH_END callbacks will be triggered)</p></li>
    1264. 

  1264. <li><p><strong>epoch</strong> – (int) epoch idx</p></li>
    1265. 

  1265. <li><p><strong>silent_mode</strong> – (bool) controls verbosity</p></li>
    1266. 

  1266. <li><p><strong>metrics_progress_verbose</strong> – (bool) controls the verbosity of metrics progress (default=False).
    1267. 

  1267. Slows down the program significantly.</p></li>
    1268. 

  1268. </ul>
    1269. 

  1269. </dd>
    1270. 

  1270. <dt class="field-even">Returns</dt>
    1271. 

  1271. <dd class="field-even"><p>results tuple (tuple) containing the loss items and metric values.</p>
    1272. 

  1272. </dd>
    1273. 

  1273. </dl>
    1274. 

  1274. </dd></dl>
    1275. 

  1275. 

  1276. <dl class="py method">
    1277. 

  1277. <dt class="sig sig-object py" id="super_gradients.training.sg_model.SgModel.instantiate_net">
    1278. 

  1278. <span class="sig-name descname"><span class="pre">instantiate_net</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">architecture</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">Union</span><span class="p"><span class="pre">[</span></span><span class="pre">torch.nn.modules.module.Module</span><span class="p"><span class="pre">,</span> </span><span class="pre">type</span><span class="p"><span class="pre">,</span> </span><span class="pre">str</span><span class="p"><span class="pre">]</span></span></span></em>, <em class="sig-param"><span class="n"><span class="pre">arch_params</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">dict</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">checkpoint_params</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">dict</span></span></em>, <em class="sig-param"><span class="o"><span class="pre">*</span></span><span class="n"><span class="pre">args</span></span></em>, <em class="sig-param"><span class="o"><span class="pre">**</span></span><span class="n"><span class="pre">kwargs</span></span></em><span class="sig-paren">)</span> &#x2192; <span class="pre">tuple</span><a class="reference internal" href="_modules/super_gradients/training/sg_model/sg_model.html#SgModel.instantiate_net"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.sg_model.SgModel.instantiate_net" title="Permalink to this definition"></a></dt>
    1279. 

  1279. <dd><dl class="simple">
    1280. 

  1280. <dt>Instantiates nn.Module according to architecture and arch_params, and handles pretrained weights and the required</dt><dd><p>module manipulation (i.e head replacement).</p>
    1281. 

  1281. </dd>
    1282. 

  1282. </dl>
    1283. 

  1283. <dl class="field-list simple">
    1284. 

  1284. <dt class="field-odd">Parameters</dt>
    1285. 

  1285. <dd class="field-odd"><ul class="simple">
    1286. 

  1286. <li><p><strong>architecture</strong> – String, torch.nn.Module or uninstantiated SgModule class describing the netowrks architecture.</p></li>
    1287. 

  1287. <li><p><strong>arch_params</strong> – Architecture’s parameters passed to networks c’tor.</p></li>
    1288. 

  1288. <li><p><strong>checkpoint_params</strong> – checkpoint loading related parameters dictionary with ‘pretrained_weights’ key,
    1289. 

  1289. s.t it’s value is a string describing the dataset of the pretrained weights (for example “imagenent”).</p></li>
    1290. 

  1290. </ul>
    1291. 

  1291. </dd>
    1292. 

  1292. <dt class="field-even">Returns</dt>
    1293. 

  1293. <dd class="field-even"><p>instantiated netowrk i.e torch.nn.Module, architecture_class (will be none when architecture is not str)</p>
    1294. 

  1294. </dd>
    1295. 

  1295. </dl>
    1296. 

  1296. </dd></dl>
    1297. 

  1297. 

  1298. </dd></dl>
    1299. 

  1299. 

  1300. <dl class="py class">
    1301. 

  1301. <dt class="sig sig-object py" id="super_gradients.training.sg_model.MultiGPUMode">
    1302. 

  1302. <em class="property"><span class="pre">class</span> </em><span class="sig-prename descclassname"><span class="pre">super_gradients.training.sg_model.</span></span><span class="sig-name descname"><span class="pre">MultiGPUMode</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">value</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/sg_model/sg_model.html#MultiGPUMode"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.sg_model.MultiGPUMode" title="Permalink to this definition"></a></dt>
    1303. 

  1303. <dd><p>Bases: <code class="xref py py-class docutils literal notranslate"><span class="pre">str</span></code>, <code class="xref py py-class docutils literal notranslate"><span class="pre">enum.Enum</span></code></p>
    1304. 

  1304. <dl class="py attribute">
    1305. 

  1305. <dt class="sig sig-object py">
    1306. 

  1306. <span class="sig-name descname"><span class="pre">OFF</span>&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160; <span class="pre">-</span> <span class="pre">Single</span> <span class="pre">GPU</span> <span class="pre">Mode</span> <span class="pre">/</span> <span class="pre">CPU</span> <span class="pre">Mode</span></span></dt>
    1307. 

  1307. <dd></dd></dl>
    1308. 

  1308. 

  1309. <dl class="py attribute">
    1310. 

  1310. <dt class="sig sig-object py">
    1311. 

  1311. <span class="sig-name descname"><span class="pre">DATA_PARALLEL</span>&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160; <span class="pre">-</span> <span class="pre">Multiple</span> <span class="pre">GPUs,</span> <span class="pre">Synchronous</span></span></dt>
    1312. 

  1312. <dd></dd></dl>
    1313. 

  1313. 

  1314. <dl class="py attribute">
    1315. 

  1315. <dt class="sig sig-object py">
    1316. 

  1316. <span class="sig-name descname"><span class="pre">DISTRIBUTED_DATA_PARALLEL</span> <span class="pre">-</span> <span class="pre">Multiple</span> <span class="pre">GPUs,</span> <span class="pre">Asynchronous</span></span></dt>
    1317. 

  1317. <dd></dd></dl>
    1318. 

  1318. 

  1319. <dl class="py attribute">
    1320. 

  1320. <dt class="sig sig-object py" id="super_gradients.training.sg_model.MultiGPUMode.OFF">
    1321. 

  1321. <span class="sig-name descname"><span class="pre">OFF</span></span><em class="property"> <span class="pre">=</span> <span class="pre">'Off'</span></em><a class="headerlink" href="#super_gradients.training.sg_model.MultiGPUMode.OFF" title="Permalink to this definition"></a></dt>
    1322. 

  1322. <dd></dd></dl>
    1323. 

  1323. 

  1324. <dl class="py attribute">
    1325. 

  1325. <dt class="sig sig-object py" id="super_gradients.training.sg_model.MultiGPUMode.DATA_PARALLEL">
    1326. 

  1326. <span class="sig-name descname"><span class="pre">DATA_PARALLEL</span></span><em class="property"> <span class="pre">=</span> <span class="pre">'DP'</span></em><a class="headerlink" href="#super_gradients.training.sg_model.MultiGPUMode.DATA_PARALLEL" title="Permalink to this definition"></a></dt>
    1327. 

  1327. <dd></dd></dl>
    1328. 

  1328. 

  1329. <dl class="py attribute">
    1330. 

  1330. <dt class="sig sig-object py" id="super_gradients.training.sg_model.MultiGPUMode.DISTRIBUTED_DATA_PARALLEL">
    1331. 

  1331. <span class="sig-name descname"><span class="pre">DISTRIBUTED_DATA_PARALLEL</span></span><em class="property"> <span class="pre">=</span> <span class="pre">'DDP'</span></em><a class="headerlink" href="#super_gradients.training.sg_model.MultiGPUMode.DISTRIBUTED_DATA_PARALLEL" title="Permalink to this definition"></a></dt>
    1332. 

  1332. <dd></dd></dl>
    1333. 

  1333. 

  1334. <dl class="py attribute">
    1335. 

  1335. <dt class="sig sig-object py" id="super_gradients.training.sg_model.MultiGPUMode.AUTO">
    1336. 

  1336. <span class="sig-name descname"><span class="pre">AUTO</span></span><em class="property"> <span class="pre">=</span> <span class="pre">'AUTO'</span></em><a class="headerlink" href="#super_gradients.training.sg_model.MultiGPUMode.AUTO" title="Permalink to this definition"></a></dt>
    1337. 

  1337. <dd></dd></dl>
    1338. 

  1338. 

  1339. </dd></dl>
    1340. 

  1340. 

  1341. <dl class="py class">
    1342. 

  1342. <dt class="sig sig-object py" id="super_gradients.training.sg_model.StrictLoad">
    1343. 

  1343. <em class="property"><span class="pre">class</span> </em><span class="sig-prename descclassname"><span class="pre">super_gradients.training.sg_model.</span></span><span class="sig-name descname"><span class="pre">StrictLoad</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">value</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/sg_model/sg_model.html#StrictLoad"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.sg_model.StrictLoad" title="Permalink to this definition"></a></dt>
    1344. 

  1344. <dd><p>Bases: <code class="xref py py-class docutils literal notranslate"><span class="pre">enum.Enum</span></code></p>
    1345. 

  1345. <dl>
    1346. 

  1346. <dt>Wrapper for adding more functionality to torch’s strict_load parameter in load_state_dict().</dt><dd><dl>
    1347. 

  1347. <dt>Attributes:</dt><dd><p>OFF              - Native torch “strict_load = off” behaviour. See nn.Module.load_state_dict() documentation for more details.
    1348. 

  1348. ON               - Native torch “strict_load = on” behaviour. See nn.Module.load_state_dict() documentation for more details.
    1349. 

  1349. NO_KEY_MATCHING  - Allows the usage of SuperGradient’s adapt_checkpoint function, which loads a checkpoint by matching each</p>
    1350. 

  1350. <blockquote>
    1351. 

  1351. <div><p>layer’s shapes (and bypasses the strict matching of the names of each layer (ie: disregards the state_dict key matching)).</p>
    1352. 

  1352. </div></blockquote>
    1353. 

  1353. </dd>
    1354. 

  1354. </dl>
    1355. 

  1355. </dd>
    1356. 

  1356. </dl>
    1357. 

  1357. <dl class="py attribute">
    1358. 

  1358. <dt class="sig sig-object py" id="super_gradients.training.sg_model.StrictLoad.OFF">
    1359. 

  1359. <span class="sig-name descname"><span class="pre">OFF</span></span><em class="property"> <span class="pre">=</span> <span class="pre">False</span></em><a class="headerlink" href="#super_gradients.training.sg_model.StrictLoad.OFF" title="Permalink to this definition"></a></dt>
    1360. 

  1360. <dd></dd></dl>
    1361. 

  1361. 

  1362. <dl class="py attribute">
    1363. 

  1363. <dt class="sig sig-object py" id="super_gradients.training.sg_model.StrictLoad.ON">
    1364. 

  1364. <span class="sig-name descname"><span class="pre">ON</span></span><em class="property"> <span class="pre">=</span> <span class="pre">True</span></em><a class="headerlink" href="#super_gradients.training.sg_model.StrictLoad.ON" title="Permalink to this definition"></a></dt>
    1365. 

  1365. <dd></dd></dl>
    1366. 

  1366. 

  1367. <dl class="py attribute">
    1368. 

  1368. <dt class="sig sig-object py" id="super_gradients.training.sg_model.StrictLoad.NO_KEY_MATCHING">
    1369. 

  1369. <span class="sig-name descname"><span class="pre">NO_KEY_MATCHING</span></span><em class="property"> <span class="pre">=</span> <span class="pre">'no_key_matching'</span></em><a class="headerlink" href="#super_gradients.training.sg_model.StrictLoad.NO_KEY_MATCHING" title="Permalink to this definition"></a></dt>
    1370. 

  1370. <dd></dd></dl>
    1371. 

  1371. 

  1372. </dd></dl>
    1373. 

  1373. 

  1374. </section>
    1375. 

  1375. </section>
    1376. 

  1376. 

  1377. 

  1378.            </div>
    1379. 

  1379.           </div>
    1380. 

  1380.           <footer>
    1381. 

  1381. 

  1382.   <hr/>
    1383. 

  1383. 

  1384.   <div role="contentinfo">
    1385. 

  1385.     <p>&#169; Copyright 2021, SuperGradients team.</p>
    1386. 

  1386.   </div>
    1387. 

  1387. 

  1388.   Built with <a href="https://www.sphinx-doc.org/">Sphinx</a> using a
    1389. 

  1389.     <a href="https://github.com/readthedocs/sphinx_rtd_theme">theme</a>
    1390. 

  1390.     provided by <a href="https://readthedocs.org">Read the Docs</a>.
    1391. 

  1391.    
    1392. 

  1392. 

  1393. </footer>
    1394. 

  1394.         </div>
    1395. 

  1395.       </div>
    1396. 

  1396.     </section>
    1397. 

  1397.   </div>
    1398. 

  1398.   <script>
    1399. 

  1399.       jQuery(function () {
    1400. 

  1400.           SphinxRtdTheme.Navigation.enable(true);
    1401. 

  1401.       });
    1402. 

  1402.   </script> 
    1403. 

  1403. 

  1404. </body>
    1405. 

  1405. </html>
    1406.
Discard
Prev
Next
Tip!

Press p or to see the previous file or, n or to see the next file