|
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
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
|
- <!DOCTYPE html>
- <html class="writer-html5" lang="en" >
- <head>
- <meta charset="utf-8" />
- <meta name="viewport" content="width=device-width, initial-scale=1.0" />
- <title>super_gradients.training.sg_model.sg_model — SuperGradients 1.0 documentation</title>
- <link rel="stylesheet" href="../../../../_static/pygments.css" type="text/css" />
- <link rel="stylesheet" href="../../../../_static/css/theme.css" type="text/css" />
- <link rel="stylesheet" href="../../../../_static/graphviz.css" type="text/css" />
- <!--[if lt IE 9]>
- <script src="../../../../_static/js/html5shiv.min.js"></script>
- <![endif]-->
-
- <script data-url_root="../../../../" id="documentation_options" src="../../../../_static/documentation_options.js"></script>
- <script src="../../../../_static/jquery.js"></script>
- <script src="../../../../_static/underscore.js"></script>
- <script src="../../../../_static/doctools.js"></script>
- <script src="../../../../_static/js/theme.js"></script>
- <link rel="index" title="Index" href="../../../../genindex.html" />
- <link rel="search" title="Search" href="../../../../search.html" />
- </head>
- <body class="wy-body-for-nav">
- <div class="wy-grid-for-nav">
- <nav data-toggle="wy-nav-shift" class="wy-nav-side">
- <div class="wy-side-scroll">
- <div class="wy-side-nav-search" >
- <a href="../../../../index.html" class="icon icon-home"> SuperGradients
- </a>
- <div role="search">
- <form id="rtd-search-form" class="wy-form" action="../../../../search.html" method="get">
- <input type="text" name="q" placeholder="Search docs" />
- <input type="hidden" name="check_keywords" value="yes" />
- <input type="hidden" name="area" value="default" />
- </form>
- </div>
- </div><div class="wy-menu wy-menu-vertical" data-spy="affix" role="navigation" aria-label="Navigation menu">
- <p class="caption"><span class="caption-text">Welcome To SuperGradients</span></p>
- <ul>
- <li class="toctree-l1"><a class="reference internal" href="../../../../welcome.html">SuperGradients</a></li>
- </ul>
- <p class="caption"><span class="caption-text">Technical Documentation</span></p>
- <ul>
- <li class="toctree-l1"><a class="reference internal" href="../../../../super_gradients.common.html">Common package</a></li>
- <li class="toctree-l1"><a class="reference internal" href="../../../../super_gradients.training.html">Training package</a></li>
- </ul>
- <p class="caption"><span class="caption-text">User Guide</span></p>
- <ul>
- <li class="toctree-l1"><a class="reference internal" href="../../../../user_guide.html">What is SuperGradients?</a></li>
- <li class="toctree-l1"><a class="reference internal" href="../../../../user_guide.html#introducing-the-supergradients-library">Introducing the SuperGradients library</a></li>
- <li class="toctree-l1"><a class="reference internal" href="../../../../user_guide.html#installation">Installation</a></li>
- <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>
- <li class="toctree-l1"><a class="reference internal" href="../../../../user_guide.html#training-parameters">Training Parameters</a></li>
- <li class="toctree-l1"><a class="reference internal" href="../../../../user_guide.html#logs-and-checkpoints">Logs and Checkpoints</a></li>
- <li class="toctree-l1"><a class="reference internal" href="../../../../user_guide.html#dataset-parameters">Dataset Parameters</a></li>
- <li class="toctree-l1"><a class="reference internal" href="../../../../user_guide.html#network-architectures">Network Architectures</a></li>
- <li class="toctree-l1"><a class="reference internal" href="../../../../user_guide.html#pretrained-models">Pretrained Models</a></li>
- <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>
- <li class="toctree-l1"><a class="reference internal" href="../../../../user_guide.html#supergradients-faq">SuperGradients FAQ</a></li>
- </ul>
- </div>
- </div>
- </nav>
- <section data-toggle="wy-nav-shift" class="wy-nav-content-wrap"><nav class="wy-nav-top" aria-label="Mobile navigation menu" >
- <i data-toggle="wy-nav-top" class="fa fa-bars"></i>
- <a href="../../../../index.html">SuperGradients</a>
- </nav>
- <div class="wy-nav-content">
- <div class="rst-content">
- <div role="navigation" aria-label="Page navigation">
- <ul class="wy-breadcrumbs">
- <li><a href="../../../../index.html" class="icon icon-home"></a> »</li>
- <li><a href="../../../index.html">Module code</a> »</li>
- <li>super_gradients.training.sg_model.sg_model</li>
- <li class="wy-breadcrumbs-aside">
- </li>
- </ul>
- <hr/>
- </div>
- <div role="main" class="document" itemscope="itemscope" itemtype="http://schema.org/Article">
- <div itemprop="articleBody">
-
- <h1>Source code for super_gradients.training.sg_model.sg_model</h1><div class="highlight"><pre>
- <span></span><span class="kn">import</span> <span class="nn">os</span>
- <span class="kn">import</span> <span class="nn">sys</span>
- <span class="kn">from</span> <span class="nn">copy</span> <span class="kn">import</span> <span class="n">deepcopy</span>
- <span class="kn">from</span> <span class="nn">enum</span> <span class="kn">import</span> <span class="n">Enum</span>
- <span class="kn">from</span> <span class="nn">typing</span> <span class="kn">import</span> <span class="n">Union</span><span class="p">,</span> <span class="n">Tuple</span><span class="p">,</span> <span class="n">Mapping</span>
- <span class="kn">import</span> <span class="nn">numpy</span> <span class="k">as</span> <span class="nn">np</span>
- <span class="kn">import</span> <span class="nn">pkg_resources</span>
- <span class="kn">import</span> <span class="nn">torch</span>
- <span class="kn">import</span> <span class="nn">torchvision.transforms</span> <span class="k">as</span> <span class="nn">transforms</span>
- <span class="kn">from</span> <span class="nn">deprecated</span> <span class="kn">import</span> <span class="n">deprecated</span>
- <span class="kn">from</span> <span class="nn">torch</span> <span class="kn">import</span> <span class="n">nn</span>
- <span class="kn">from</span> <span class="nn">torch.cuda.amp</span> <span class="kn">import</span> <span class="n">GradScaler</span><span class="p">,</span> <span class="n">autocast</span>
- <span class="kn">from</span> <span class="nn">torchmetrics</span> <span class="kn">import</span> <span class="n">MetricCollection</span>
- <span class="kn">from</span> <span class="nn">tqdm</span> <span class="kn">import</span> <span class="n">tqdm</span>
- <span class="kn">from</span> <span class="nn">piptools.scripts.sync</span> <span class="kn">import</span> <span class="n">_get_installed_distributions</span>
- <span class="kn">from</span> <span class="nn">super_gradients.common.decorators.factory_decorator</span> <span class="kn">import</span> <span class="n">resolve_param</span>
- <span class="kn">from</span> <span class="nn">super_gradients.common.environment</span> <span class="kn">import</span> <span class="n">env_helpers</span>
- <span class="kn">from</span> <span class="nn">super_gradients.common.abstractions.abstract_logger</span> <span class="kn">import</span> <span class="n">get_logger</span>
- <span class="kn">from</span> <span class="nn">super_gradients.common.factories.datasets_factory</span> <span class="kn">import</span> <span class="n">DatasetsFactory</span>
- <span class="kn">from</span> <span class="nn">super_gradients.common.factories.list_factory</span> <span class="kn">import</span> <span class="n">ListFactory</span>
- <span class="kn">from</span> <span class="nn">super_gradients.common.factories.losses_factory</span> <span class="kn">import</span> <span class="n">LossesFactory</span>
- <span class="kn">from</span> <span class="nn">super_gradients.common.factories.metrics_factory</span> <span class="kn">import</span> <span class="n">MetricsFactory</span>
- <span class="kn">from</span> <span class="nn">super_gradients.common.sg_loggers</span> <span class="kn">import</span> <span class="n">SG_LOGGERS</span>
- <span class="kn">from</span> <span class="nn">super_gradients.common.sg_loggers.abstract_sg_logger</span> <span class="kn">import</span> <span class="n">AbstractSGLogger</span>
- <span class="kn">from</span> <span class="nn">super_gradients.common.sg_loggers.base_sg_logger</span> <span class="kn">import</span> <span class="n">BaseSGLogger</span>
- <span class="kn">from</span> <span class="nn">super_gradients.training</span> <span class="kn">import</span> <span class="n">ARCHITECTURES</span><span class="p">,</span> <span class="n">utils</span> <span class="k">as</span> <span class="n">core_utils</span>
- <span class="kn">from</span> <span class="nn">super_gradients.training.utils</span> <span class="kn">import</span> <span class="n">sg_model_utils</span>
- <span class="kn">from</span> <span class="nn">super_gradients.training</span> <span class="kn">import</span> <span class="n">metrics</span>
- <span class="kn">from</span> <span class="nn">super_gradients.training.exceptions.sg_model_exceptions</span> <span class="kn">import</span> <span class="n">UnsupportedOptimizerFormat</span>
- <span class="kn">from</span> <span class="nn">super_gradients.training.datasets</span> <span class="kn">import</span> <span class="n">DatasetInterface</span>
- <span class="kn">from</span> <span class="nn">super_gradients.training.losses</span> <span class="kn">import</span> <span class="n">LOSSES</span>
- <span class="kn">from</span> <span class="nn">super_gradients.training.metrics.metric_utils</span> <span class="kn">import</span> <span class="n">get_metrics_titles</span><span class="p">,</span> <span class="n">get_metrics_results_tuple</span><span class="p">,</span> \
- <span class="n">get_logging_values</span><span class="p">,</span> \
- <span class="n">get_metrics_dict</span><span class="p">,</span> <span class="n">get_train_loop_description_dict</span>
- <span class="kn">from</span> <span class="nn">super_gradients.training.models</span> <span class="kn">import</span> <span class="n">SgModule</span>
- <span class="kn">from</span> <span class="nn">super_gradients.training.params</span> <span class="kn">import</span> <span class="n">TrainingParams</span>
- <span class="kn">from</span> <span class="nn">super_gradients.training.utils.detection_utils</span> <span class="kn">import</span> <span class="n">DetectionPostPredictionCallback</span>
- <span class="kn">from</span> <span class="nn">super_gradients.training.utils.distributed_training_utils</span> <span class="kn">import</span> <span class="n">MultiGPUModeAutocastWrapper</span><span class="p">,</span> \
- <span class="n">reduce_results_tuple_for_ddp</span><span class="p">,</span> <span class="n">compute_precise_bn_stats</span>
- <span class="kn">from</span> <span class="nn">super_gradients.training.utils.ema</span> <span class="kn">import</span> <span class="n">ModelEMA</span>
- <span class="kn">from</span> <span class="nn">super_gradients.training.utils.optimizer_utils</span> <span class="kn">import</span> <span class="n">build_optimizer</span>
- <span class="kn">from</span> <span class="nn">super_gradients.training.utils.weight_averaging_utils</span> <span class="kn">import</span> <span class="n">ModelWeightAveraging</span>
- <span class="kn">from</span> <span class="nn">super_gradients.training.metrics</span> <span class="kn">import</span> <span class="n">Accuracy</span><span class="p">,</span> <span class="n">Top5</span>
- <span class="kn">from</span> <span class="nn">super_gradients.training.utils</span> <span class="kn">import</span> <span class="n">random_seed</span>
- <span class="kn">from</span> <span class="nn">super_gradients.training.utils.checkpoint_utils</span> <span class="kn">import</span> <span class="n">get_ckpt_local_path</span><span class="p">,</span> <span class="n">read_ckpt_state_dict</span><span class="p">,</span> \
- <span class="n">load_checkpoint_to_model</span><span class="p">,</span> <span class="n">load_pretrained_weights</span>
- <span class="kn">from</span> <span class="nn">super_gradients.training.datasets.datasets_utils</span> <span class="kn">import</span> <span class="n">DatasetStatisticsTensorboardLogger</span>
- <span class="kn">from</span> <span class="nn">super_gradients.training.utils.callbacks</span> <span class="kn">import</span> <span class="n">CallbackHandler</span><span class="p">,</span> <span class="n">Phase</span><span class="p">,</span> <span class="n">LR_SCHEDULERS_CLS_DICT</span><span class="p">,</span> <span class="n">PhaseContext</span><span class="p">,</span> \
- <span class="n">MetricsUpdateCallback</span><span class="p">,</span> <span class="n">LR_WARMUP_CLS_DICT</span>
- <span class="kn">from</span> <span class="nn">super_gradients.common.environment</span> <span class="kn">import</span> <span class="n">environment_config</span>
- <span class="kn">from</span> <span class="nn">super_gradients.training.pretrained_models</span> <span class="kn">import</span> <span class="n">PRETRAINED_NUM_CLASSES</span>
- <span class="n">logger</span> <span class="o">=</span> <span class="n">get_logger</span><span class="p">(</span><span class="vm">__name__</span><span class="p">)</span>
- <div class="viewcode-block" id="StrictLoad"><a class="viewcode-back" href="../../../../super_gradients.training.sg_model.html#super_gradients.training.StrictLoad">[docs]</a><span class="k">class</span> <span class="nc">StrictLoad</span><span class="p">(</span><span class="n">Enum</span><span class="p">):</span>
- <span class="sd">"""</span>
- <span class="sd"> Wrapper for adding more functionality to torch's strict_load parameter in load_state_dict().</span>
- <span class="sd"> Attributes:</span>
- <span class="sd"> OFF - Native torch "strict_load = off" behaviour. See nn.Module.load_state_dict() documentation for more details.</span>
- <span class="sd"> ON - Native torch "strict_load = on" behaviour. See nn.Module.load_state_dict() documentation for more details.</span>
- <span class="sd"> NO_KEY_MATCHING - Allows the usage of SuperGradient's adapt_checkpoint function, which loads a checkpoint by matching each</span>
- <span class="sd"> layer's shapes (and bypasses the strict matching of the names of each layer (ie: disregards the state_dict key matching)).</span>
- <span class="sd"> """</span>
- <span class="n">OFF</span> <span class="o">=</span> <span class="kc">False</span>
- <span class="n">ON</span> <span class="o">=</span> <span class="kc">True</span>
- <span class="n">NO_KEY_MATCHING</span> <span class="o">=</span> <span class="s1">'no_key_matching'</span></div>
- <div class="viewcode-block" id="MultiGPUMode"><a class="viewcode-back" href="../../../../super_gradients.training.sg_model.html#super_gradients.training.MultiGPUMode">[docs]</a><span class="k">class</span> <span class="nc">MultiGPUMode</span><span class="p">(</span><span class="nb">str</span><span class="p">,</span> <span class="n">Enum</span><span class="p">):</span>
- <span class="sd">"""</span>
- <span class="sd"> MultiGPUMode</span>
- <span class="sd"> Attributes:</span>
- <span class="sd"> OFF - Single GPU Mode / CPU Mode</span>
- <span class="sd"> DATA_PARALLEL - Multiple GPUs, Synchronous</span>
- <span class="sd"> DISTRIBUTED_DATA_PARALLEL - Multiple GPUs, Asynchronous</span>
- <span class="sd"> """</span>
- <span class="n">OFF</span> <span class="o">=</span> <span class="s1">'Off'</span>
- <span class="n">DATA_PARALLEL</span> <span class="o">=</span> <span class="s1">'DP'</span>
- <span class="n">DISTRIBUTED_DATA_PARALLEL</span> <span class="o">=</span> <span class="s1">'DDP'</span>
- <span class="n">AUTO</span> <span class="o">=</span> <span class="s2">"AUTO"</span></div>
- <div class="viewcode-block" id="EvaluationType"><a class="viewcode-back" href="../../../../super_gradients.training.sg_model.html#super_gradients.training.EvaluationType">[docs]</a><span class="k">class</span> <span class="nc">EvaluationType</span><span class="p">(</span><span class="nb">str</span><span class="p">,</span> <span class="n">Enum</span><span class="p">):</span>
- <span class="sd">"""</span>
- <span class="sd"> EvaluationType</span>
- <span class="sd"> Passed to SgModel.evaluate(..), and controls which phase callbacks should be triggered (if at all).</span>
- <span class="sd"> Attributes:</span>
- <span class="sd"> TEST</span>
- <span class="sd"> VALIDATION</span>
- <span class="sd"> """</span>
- <span class="n">TEST</span> <span class="o">=</span> <span class="s1">'TEST'</span>
- <span class="n">VALIDATION</span> <span class="o">=</span> <span class="s1">'VALIDATION'</span></div>
- <div class="viewcode-block" id="SgModel"><a class="viewcode-back" href="../../../../super_gradients.training.sg_model.html#super_gradients.training.SgModel">[docs]</a><span class="k">class</span> <span class="nc">SgModel</span><span class="p">:</span>
- <span class="sd">"""</span>
- <span class="sd"> SuperGradient Model - Base Class for Sg Models</span>
- <span class="sd"> Methods</span>
- <span class="sd"> -------</span>
- <span class="sd"> train(max_epochs : int, initial_epoch : int, save_model : bool)</span>
- <span class="sd"> the main function used for the training, h.p. updating, logging etc.</span>
- <span class="sd"> predict(idx : int)</span>
- <span class="sd"> returns the predictions and label of the current inputs</span>
- <span class="sd"> test(epoch : int, idx : int, save : bool):</span>
- <span class="sd"> returns the test loss, accuracy and runtime</span>
- <span class="sd"> """</span>
- <span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">experiment_name</span><span class="p">:</span> <span class="nb">str</span><span class="p">,</span> <span class="n">device</span><span class="p">:</span> <span class="nb">str</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span> <span class="n">multi_gpu</span><span class="p">:</span> <span class="n">Union</span><span class="p">[</span><span class="n">MultiGPUMode</span><span class="p">,</span> <span class="nb">str</span><span class="p">]</span> <span class="o">=</span> <span class="n">MultiGPUMode</span><span class="o">.</span><span class="n">AUTO</span><span class="p">,</span>
- <span class="n">model_checkpoints_location</span><span class="p">:</span> <span class="nb">str</span> <span class="o">=</span> <span class="s1">'local'</span><span class="p">,</span>
- <span class="n">overwrite_local_checkpoint</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="kc">True</span><span class="p">,</span> <span class="n">ckpt_name</span><span class="p">:</span> <span class="nb">str</span> <span class="o">=</span> <span class="s1">'ckpt_latest.pth'</span><span class="p">,</span>
- <span class="n">post_prediction_callback</span><span class="p">:</span> <span class="n">DetectionPostPredictionCallback</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span> <span class="n">ckpt_root_dir</span><span class="o">=</span><span class="kc">None</span><span class="p">):</span>
- <span class="sd">"""</span>
- <span class="sd"> :param experiment_name: Used for logging and loading purposes</span>
- <span class="sd"> :param device: If equal to 'cpu' runs on the CPU otherwise on GPU</span>
- <span class="sd"> :param multi_gpu: If True, runs on all available devices</span>
- <span class="sd"> :param model_checkpoints_location: If set to 's3' saves the Checkpoints in AWS S3</span>
- <span class="sd"> otherwise saves the Checkpoints Locally</span>
- <span class="sd"> :param overwrite_local_checkpoint: If set to False keeps the current local checkpoint when importing</span>
- <span class="sd"> checkpoint from cloud service, otherwise overwrites the local checkpoints file</span>
- <span class="sd"> :param ckpt_name: The Checkpoint to Load</span>
- <span class="sd"> :ckpt_root_dir: Local root directory path where all experiment logging directories will</span>
- <span class="sd"> reside. When none is give, it is assumed that</span>
- <span class="sd"> pkg_resources.resource_filename('checkpoints', "") exists and will be used.</span>
- <span class="sd"> """</span>
- <span class="c1"># SET THE EMPTY PROPERTIES</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">net</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">architecture</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">arch_params</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">dataset_interface</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span> <span class="kc">None</span><span class="p">,</span> <span class="kc">None</span><span class="p">,</span> <span class="kc">None</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">architecture_cls</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">device</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">multi_gpu</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span> <span class="kc">None</span><span class="p">,</span> <span class="kc">None</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">dataset_params</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">train_loader</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">valid_loader</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">test_loader</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">classes</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span> <span class="kc">None</span><span class="p">,</span> <span class="kc">None</span><span class="p">,</span> <span class="kc">None</span><span class="p">,</span> <span class="kc">None</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">ema</span> <span class="o">=</span> <span class="kc">None</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">ema_model</span> <span class="o">=</span> <span class="kc">None</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">sg_logger</span> <span class="o">=</span> <span class="kc">None</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">update_param_groups</span> <span class="o">=</span> <span class="kc">None</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">post_prediction_callback</span> <span class="o">=</span> <span class="kc">None</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">criterion</span> <span class="o">=</span> <span class="kc">None</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">training_params</span> <span class="o">=</span> <span class="kc">None</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">scaler</span> <span class="o">=</span> <span class="kc">None</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">phase_callbacks</span> <span class="o">=</span> <span class="kc">None</span>
- <span class="c1"># SET THE DEFAULT PROPERTIES</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">half_precision</span> <span class="o">=</span> <span class="kc">False</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">load_checkpoint</span> <span class="o">=</span> <span class="kc">False</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">load_backbone</span> <span class="o">=</span> <span class="kc">False</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">load_weights_only</span> <span class="o">=</span> <span class="kc">False</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">ddp_silent_mode</span> <span class="o">=</span> <span class="kc">False</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">source_ckpt_folder_name</span> <span class="o">=</span> <span class="kc">None</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">model_weight_averaging</span> <span class="o">=</span> <span class="kc">None</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">average_model_checkpoint_filename</span> <span class="o">=</span> <span class="s1">'average_model.pth'</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">start_epoch</span> <span class="o">=</span> <span class="mi">0</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">best_metric</span> <span class="o">=</span> <span class="n">np</span><span class="o">.</span><span class="n">inf</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">external_checkpoint_path</span> <span class="o">=</span> <span class="kc">None</span>
- <span class="c1"># DETERMINE THE LOCATION OF THE LOSS AND ACCURACY IN THE RESULTS TUPLE OUTPUTED BY THE TEST</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">loss_idx_in_results_tuple</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">acc_idx_in_results_tuple</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span> <span class="kc">None</span>
- <span class="c1"># METRICS</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">loss_logging_items_names</span> <span class="o">=</span> <span class="kc">None</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">train_metrics</span> <span class="o">=</span> <span class="kc">None</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">valid_metrics</span> <span class="o">=</span> <span class="kc">None</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">greater_metric_to_watch_is_better</span> <span class="o">=</span> <span class="kc">None</span>
- <span class="c1"># SETTING THE PROPERTIES FROM THE CONSTRUCTOR</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">experiment_name</span> <span class="o">=</span> <span class="n">experiment_name</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">ckpt_name</span> <span class="o">=</span> <span class="n">ckpt_name</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">overwrite_local_checkpoint</span> <span class="o">=</span> <span class="n">overwrite_local_checkpoint</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">model_checkpoints_location</span> <span class="o">=</span> <span class="n">model_checkpoints_location</span>
- <span class="c1"># CREATING THE LOGGING DIR BASED ON THE INPUT PARAMS TO PREVENT OVERWRITE OF LOCAL VERSION</span>
- <span class="k">if</span> <span class="n">ckpt_root_dir</span><span class="p">:</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">checkpoints_dir_path</span> <span class="o">=</span> <span class="n">os</span><span class="o">.</span><span class="n">path</span><span class="o">.</span><span class="n">join</span><span class="p">(</span><span class="n">ckpt_root_dir</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">experiment_name</span><span class="p">)</span>
- <span class="k">elif</span> <span class="n">pkg_resources</span><span class="o">.</span><span class="n">resource_exists</span><span class="p">(</span><span class="s2">"checkpoints"</span><span class="p">,</span> <span class="s2">""</span><span class="p">):</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">checkpoints_dir_path</span> <span class="o">=</span> <span class="n">pkg_resources</span><span class="o">.</span><span class="n">resource_filename</span><span class="p">(</span><span class="s1">'checkpoints'</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">experiment_name</span><span class="p">)</span>
- <span class="k">else</span><span class="p">:</span>
- <span class="k">raise</span> <span class="ne">ValueError</span><span class="p">(</span><span class="s2">"Illegal checkpoints directory: pass ckpt_root_dir that exists, or add 'checkpoints' to"</span>
- <span class="s2">"resources."</span><span class="p">)</span>
- <span class="c1"># INITIALIZE THE DEVICE FOR THE MODEL</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">_initialize_device</span><span class="p">(</span><span class="n">requested_device</span><span class="o">=</span><span class="n">device</span><span class="p">,</span> <span class="n">requested_multi_gpu</span><span class="o">=</span><span class="n">multi_gpu</span><span class="p">)</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">post_prediction_callback</span> <span class="o">=</span> <span class="n">post_prediction_callback</span>
- <span class="c1"># SET THE DEFAULTS</span>
- <span class="c1"># TODO: SET DEFAULT TRAINING PARAMS FOR EACH TASK</span>
- <span class="n">default_results_titles</span> <span class="o">=</span> <span class="p">[</span><span class="s1">'Train Loss'</span><span class="p">,</span> <span class="s1">'Train Acc'</span><span class="p">,</span> <span class="s1">'Train Top5'</span><span class="p">,</span> <span class="s1">'Valid Loss'</span><span class="p">,</span> <span class="s1">'Valid Acc'</span><span class="p">,</span> <span class="s1">'Valid Top5'</span><span class="p">]</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">results_titles</span> <span class="o">=</span> <span class="n">default_results_titles</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">loss_idx_in_results_tuple</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">acc_idx_in_results_tuple</span> <span class="o">=</span> <span class="mi">0</span><span class="p">,</span> <span class="mi">1</span>
- <span class="n">default_train_metrics</span><span class="p">,</span> <span class="n">default_valid_metrics</span> <span class="o">=</span> <span class="n">MetricCollection</span><span class="p">([</span><span class="n">Accuracy</span><span class="p">(),</span> <span class="n">Top5</span><span class="p">()]),</span> <span class="n">MetricCollection</span><span class="p">(</span>
- <span class="p">[</span><span class="n">Accuracy</span><span class="p">(),</span> <span class="n">Top5</span><span class="p">()])</span>
- <span class="n">default_loss_logging_items_names</span> <span class="o">=</span> <span class="p">[</span><span class="s2">"Loss"</span><span class="p">]</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">train_metrics</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">valid_metrics</span> <span class="o">=</span> <span class="n">default_train_metrics</span><span class="p">,</span> <span class="n">default_valid_metrics</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">loss_logging_items_names</span> <span class="o">=</span> <span class="n">default_loss_logging_items_names</span>
- <div class="viewcode-block" id="SgModel.connect_dataset_interface"><a class="viewcode-back" href="../../../../super_gradients.training.sg_model.html#super_gradients.training.SgModel.connect_dataset_interface">[docs]</a> <span class="nd">@resolve_param</span><span class="p">(</span><span class="s1">'dataset_interface'</span><span class="p">,</span> <span class="n">DatasetsFactory</span><span class="p">())</span>
- <span class="k">def</span> <span class="nf">connect_dataset_interface</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">dataset_interface</span><span class="p">:</span> <span class="n">DatasetInterface</span><span class="p">,</span> <span class="n">data_loader_num_workers</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="mi">8</span><span class="p">):</span>
- <span class="sd">"""</span>
- <span class="sd"> :param dataset_interface: DatasetInterface object</span>
- <span class="sd"> :param data_loader_num_workers: The number of threads to initialize the Data Loaders with</span>
- <span class="sd"> The dataset to be connected</span>
- <span class="sd"> """</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">dataset_interface</span> <span class="o">=</span> <span class="n">dataset_interface</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">train_loader</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">valid_loader</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">test_loader</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">classes</span> <span class="o">=</span> \
- <span class="bp">self</span><span class="o">.</span><span class="n">dataset_interface</span><span class="o">.</span><span class="n">get_data_loaders</span><span class="p">(</span><span class="n">batch_size_factor</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">num_devices</span><span class="p">,</span>
- <span class="n">num_workers</span><span class="o">=</span><span class="n">data_loader_num_workers</span><span class="p">,</span>
- <span class="n">distributed_sampler</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">multi_gpu</span> <span class="o">==</span> <span class="n">MultiGPUMode</span><span class="o">.</span><span class="n">DISTRIBUTED_DATA_PARALLEL</span><span class="p">)</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">dataset_params</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">dataset_interface</span><span class="o">.</span><span class="n">get_dataset_params</span><span class="p">()</span></div>
- <span class="c1"># FIXME - we need to resolve flake8's 'function is too complex' for this function</span>
- <div class="viewcode-block" id="SgModel.build_model"><a class="viewcode-back" href="../../../../super_gradients.training.sg_model.html#super_gradients.training.SgModel.build_model">[docs]</a> <span class="k">def</span> <span class="nf">build_model</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="c1"># noqa: C901 - too complex</span>
- <span class="n">architecture</span><span class="p">:</span> <span class="n">Union</span><span class="p">[</span><span class="nb">str</span><span class="p">,</span> <span class="n">nn</span><span class="o">.</span><span class="n">Module</span><span class="p">],</span>
- <span class="n">arch_params</span><span class="o">=</span><span class="p">{},</span>
- <span class="n">load_checkpoint</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="kc">False</span><span class="p">,</span>
- <span class="n">strict_load</span><span class="p">:</span> <span class="n">StrictLoad</span> <span class="o">=</span> <span class="n">StrictLoad</span><span class="o">.</span><span class="n">ON</span><span class="p">,</span>
- <span class="n">source_ckpt_folder_name</span><span class="p">:</span> <span class="nb">str</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
- <span class="n">load_weights_only</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="kc">False</span><span class="p">,</span>
- <span class="n">load_backbone</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="kc">False</span><span class="p">,</span>
- <span class="n">external_checkpoint_path</span><span class="p">:</span> <span class="nb">str</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
- <span class="n">load_ema_as_net</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="kc">False</span><span class="p">):</span>
- <span class="sd">"""</span>
- <span class="sd"> :param architecture: Defines the network's architecture from models/ALL_ARCHITECTURES</span>
- <span class="sd"> :param arch_params: Architecture H.P. e.g.: block, num_blocks, num_classes, etc.</span>
- <span class="sd"> :param load_checkpoint: Load a pre-trained checkpoint</span>
- <span class="sd"> :param strict_load: See StrictLoad class documentation for details.</span>
- <span class="sd"> :param source_ckpt_folder_name: folder name to load the checkpoint from (self.experiment_name if none is given)</span>
- <span class="sd"> :param load_weights_only: loads only the weight from the checkpoint and zeroize the training params</span>
- <span class="sd"> :param load_backbone: loads the provided checkpoint to self.net.backbone instead of self.net</span>
- <span class="sd"> :param external_checkpoint_path: The path to the external checkpoint to be loaded. Can be absolute or relative</span>
- <span class="sd"> (ie: path/to/checkpoint.pth). If provided, will automatically attempt to</span>
- <span class="sd"> load the checkpoint even if the load_checkpoint flag is not provided.</span>
- <span class="sd"> """</span>
- <span class="k">if</span> <span class="s1">'num_classes'</span> <span class="ow">not</span> <span class="ow">in</span> <span class="n">arch_params</span><span class="o">.</span><span class="n">keys</span><span class="p">():</span>
- <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">dataset_interface</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
- <span class="k">raise</span> <span class="ne">Exception</span><span class="p">(</span><span class="s1">'Error'</span><span class="p">,</span> <span class="s1">'Number of classes not defined in arch params and dataset is not defined'</span><span class="p">)</span>
- <span class="k">else</span><span class="p">:</span>
- <span class="n">arch_params</span><span class="p">[</span><span class="s1">'num_classes'</span><span class="p">]</span> <span class="o">=</span> <span class="nb">len</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">classes</span><span class="p">)</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">arch_params</span> <span class="o">=</span> <span class="n">core_utils</span><span class="o">.</span><span class="n">HpmStruct</span><span class="p">(</span><span class="o">**</span><span class="n">arch_params</span><span class="p">)</span>
- <span class="n">pretrained_weights</span> <span class="o">=</span> <span class="n">core_utils</span><span class="o">.</span><span class="n">get_param</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">arch_params</span><span class="p">,</span> <span class="s1">'pretrained_weights'</span><span class="p">,</span> <span class="n">default_val</span><span class="o">=</span><span class="kc">None</span><span class="p">)</span>
- <span class="k">if</span> <span class="n">pretrained_weights</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
- <span class="n">num_classes_new_head</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">arch_params</span><span class="o">.</span><span class="n">num_classes</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">arch_params</span><span class="o">.</span><span class="n">num_classes</span> <span class="o">=</span> <span class="n">PRETRAINED_NUM_CLASSES</span><span class="p">[</span><span class="n">pretrained_weights</span><span class="p">]</span>
- <span class="c1"># OVERRIDE THE INPUT PARAMS WITH THE arch_params VALUES</span>
- <span class="n">load_weights_only</span> <span class="o">=</span> <span class="n">core_utils</span><span class="o">.</span><span class="n">get_param</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">arch_params</span><span class="p">,</span> <span class="s1">'load_weights_only'</span><span class="p">,</span> <span class="n">default_val</span><span class="o">=</span><span class="n">load_weights_only</span><span class="p">)</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">source_ckpt_folder_name</span> <span class="o">=</span> <span class="n">core_utils</span><span class="o">.</span><span class="n">get_param</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">arch_params</span><span class="p">,</span> <span class="s1">'source_ckpt_folder_name'</span><span class="p">,</span>
- <span class="n">default_val</span><span class="o">=</span><span class="n">source_ckpt_folder_name</span><span class="p">)</span>
- <span class="n">strict_load</span> <span class="o">=</span> <span class="n">core_utils</span><span class="o">.</span><span class="n">get_param</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">arch_params</span><span class="p">,</span> <span class="s1">'strict_load'</span><span class="p">,</span> <span class="n">default_val</span><span class="o">=</span><span class="n">strict_load</span><span class="p">)</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">arch_params</span><span class="o">.</span><span class="n">sync_bn</span> <span class="o">=</span> <span class="n">core_utils</span><span class="o">.</span><span class="n">get_param</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">arch_params</span><span class="p">,</span> <span class="s1">'sync_bn'</span><span class="p">,</span> <span class="n">default_val</span><span class="o">=</span><span class="kc">False</span><span class="p">)</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">load_checkpoint</span> <span class="o">=</span> <span class="n">load_checkpoint</span> <span class="ow">or</span> <span class="n">core_utils</span><span class="o">.</span><span class="n">get_param</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">arch_params</span><span class="p">,</span> <span class="s1">'load_checkpoint'</span><span class="p">,</span>
- <span class="n">default_val</span><span class="o">=</span><span class="kc">False</span><span class="p">)</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">load_backbone</span> <span class="o">=</span> <span class="n">core_utils</span><span class="o">.</span><span class="n">get_param</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">arch_params</span><span class="p">,</span> <span class="s1">'load_backbone'</span><span class="p">,</span> <span class="n">default_val</span><span class="o">=</span><span class="n">load_backbone</span><span class="p">)</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">external_checkpoint_path</span> <span class="o">=</span> <span class="n">core_utils</span><span class="o">.</span><span class="n">get_param</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">arch_params</span><span class="p">,</span> <span class="s1">'external_checkpoint_path'</span><span class="p">,</span>
- <span class="n">default_val</span><span class="o">=</span><span class="n">external_checkpoint_path</span><span class="p">)</span>
- <span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">architecture</span><span class="p">,</span> <span class="nb">str</span><span class="p">):</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">architecture_cls</span> <span class="o">=</span> <span class="n">ARCHITECTURES</span><span class="p">[</span><span class="n">architecture</span><span class="p">]</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">net</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">architecture_cls</span><span class="p">(</span><span class="n">arch_params</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">arch_params</span><span class="p">)</span>
- <span class="k">elif</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">architecture</span><span class="p">,</span> <span class="n">SgModule</span><span class="o">.</span><span class="vm">__class__</span><span class="p">):</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">net</span> <span class="o">=</span> <span class="n">architecture</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">arch_params</span><span class="p">)</span>
- <span class="k">else</span><span class="p">:</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">net</span> <span class="o">=</span> <span class="n">architecture</span>
- <span class="c1"># SAVE THE ARCHITECTURE FOR NEURAL ARCHITECTURE SEARCH</span>
- <span class="k">if</span> <span class="nb">hasattr</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">net</span><span class="p">,</span> <span class="s1">'structure'</span><span class="p">):</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">architecture</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">net</span><span class="o">.</span><span class="n">structure</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">net</span><span class="o">.</span><span class="n">to</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">device</span><span class="p">)</span>
- <span class="c1"># FOR MULTI-GPU TRAINING (not distributed)</span>
- <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">multi_gpu</span> <span class="o">==</span> <span class="n">MultiGPUMode</span><span class="o">.</span><span class="n">DATA_PARALLEL</span><span class="p">:</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">net</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">nn</span><span class="o">.</span><span class="n">DataParallel</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">net</span><span class="p">,</span> <span class="n">device_ids</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">device_ids</span><span class="p">)</span>
- <span class="k">elif</span> <span class="bp">self</span><span class="o">.</span><span class="n">multi_gpu</span> <span class="o">==</span> <span class="n">MultiGPUMode</span><span class="o">.</span><span class="n">DISTRIBUTED_DATA_PARALLEL</span><span class="p">:</span>
- <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">arch_params</span><span class="o">.</span><span class="n">sync_bn</span><span class="p">:</span>
- <span class="k">if</span> <span class="ow">not</span> <span class="bp">self</span><span class="o">.</span><span class="n">ddp_silent_mode</span><span class="p">:</span>
- <span class="n">logger</span><span class="o">.</span><span class="n">info</span><span class="p">(</span><span class="s1">'DDP - Using Sync Batch Norm... Training time will be affected accordingly'</span><span class="p">)</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">net</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">nn</span><span class="o">.</span><span class="n">SyncBatchNorm</span><span class="o">.</span><span class="n">convert_sync_batchnorm</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">net</span><span class="p">)</span><span class="o">.</span><span class="n">to</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">device</span><span class="p">)</span>
- <span class="n">local_rank</span> <span class="o">=</span> <span class="nb">int</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">device</span><span class="o">.</span><span class="n">split</span><span class="p">(</span><span class="s1">':'</span><span class="p">)[</span><span class="mi">1</span><span class="p">])</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">net</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">nn</span><span class="o">.</span><span class="n">parallel</span><span class="o">.</span><span class="n">DistributedDataParallel</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">net</span><span class="p">,</span>
- <span class="n">device_ids</span><span class="o">=</span><span class="p">[</span><span class="n">local_rank</span><span class="p">],</span>
- <span class="n">output_device</span><span class="o">=</span><span class="n">local_rank</span><span class="p">,</span>
- <span class="n">find_unused_parameters</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span>
- <span class="k">else</span><span class="p">:</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">net</span> <span class="o">=</span> <span class="n">core_utils</span><span class="o">.</span><span class="n">WrappedModel</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">net</span><span class="p">)</span>
- <span class="c1"># SET THE FLAG FOR DIFFERENT PARAMETER GROUP OPTIMIZER UPDATE</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">update_param_groups</span> <span class="o">=</span> <span class="nb">hasattr</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">net</span><span class="o">.</span><span class="n">module</span><span class="p">,</span> <span class="s1">'update_param_groups'</span><span class="p">)</span>
- <span class="c1"># LOAD AN EXISTING CHECKPOINT IF INDICATED</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">checkpoint</span> <span class="o">=</span> <span class="p">{}</span>
- <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">load_checkpoint</span> <span class="ow">or</span> <span class="bp">self</span><span class="o">.</span><span class="n">external_checkpoint_path</span><span class="p">:</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">load_weights_only</span> <span class="o">=</span> <span class="n">load_weights_only</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">_load_checkpoint_to_model</span><span class="p">(</span><span class="n">strict</span><span class="o">=</span><span class="n">strict_load</span><span class="p">,</span> <span class="n">load_backbone</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">load_backbone</span><span class="p">,</span>
- <span class="n">source_ckpt_folder_name</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">source_ckpt_folder_name</span><span class="p">,</span>
- <span class="n">load_ema_as_net</span><span class="o">=</span><span class="n">load_ema_as_net</span><span class="p">)</span>
- <span class="k">if</span> <span class="n">pretrained_weights</span><span class="p">:</span>
- <span class="n">load_pretrained_weights</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">net</span><span class="p">,</span> <span class="n">architecture</span><span class="p">,</span> <span class="n">pretrained_weights</span><span class="p">)</span>
- <span class="k">if</span> <span class="n">num_classes_new_head</span> <span class="o">!=</span> <span class="bp">self</span><span class="o">.</span><span class="n">arch_params</span><span class="o">.</span><span class="n">num_classes</span><span class="p">:</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">net</span><span class="o">.</span><span class="n">module</span><span class="o">.</span><span class="n">replace_head</span><span class="p">(</span><span class="n">new_num_classes</span><span class="o">=</span><span class="n">num_classes_new_head</span><span class="p">)</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">arch_params</span><span class="o">.</span><span class="n">num_classes</span> <span class="o">=</span> <span class="n">num_classes_new_head</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">net</span><span class="o">.</span><span class="n">to</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">device</span><span class="p">)</span></div>
- <span class="k">def</span> <span class="nf">_train_epoch</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">epoch</span><span class="p">:</span> <span class="nb">int</span><span class="p">,</span> <span class="n">silent_mode</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="kc">False</span><span class="p">)</span> <span class="o">-></span> <span class="nb">tuple</span><span class="p">:</span>
- <span class="sd">"""</span>
- <span class="sd"> train_epoch - A single epoch training procedure</span>
- <span class="sd"> :param optimizer: The optimizer for the network</span>
- <span class="sd"> :param epoch: The current epoch</span>
- <span class="sd"> :param silent_mode: No verbosity</span>
- <span class="sd"> """</span>
- <span class="c1"># SET THE MODEL IN training STATE</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">net</span><span class="o">.</span><span class="n">train</span><span class="p">()</span>
- <span class="c1"># THE DISABLE FLAG CONTROLS WHETHER THE PROGRESS BAR IS SILENT OR PRINTS THE LOGS</span>
- <span class="n">progress_bar_train_loader</span> <span class="o">=</span> <span class="n">tqdm</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">train_loader</span><span class="p">,</span> <span class="n">bar_format</span><span class="o">=</span><span class="s2">"</span><span class="si">{l_bar}{bar:10}{r_bar}</span><span class="s2">"</span><span class="p">,</span> <span class="n">dynamic_ncols</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span>
- <span class="n">disable</span><span class="o">=</span><span class="n">silent_mode</span><span class="p">)</span>
- <span class="n">progress_bar_train_loader</span><span class="o">.</span><span class="n">set_description</span><span class="p">(</span><span class="sa">f</span><span class="s2">"Train epoch </span><span class="si">{</span><span class="n">epoch</span><span class="si">}</span><span class="s2">"</span><span class="p">)</span>
- <span class="c1"># RESET/INIT THE METRIC LOGGERS</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">train_metrics</span><span class="o">.</span><span class="n">reset</span><span class="p">()</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">train_metrics</span><span class="o">.</span><span class="n">to</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">device</span><span class="p">)</span>
- <span class="n">loss_avg_meter</span> <span class="o">=</span> <span class="n">core_utils</span><span class="o">.</span><span class="n">utils</span><span class="o">.</span><span class="n">AverageMeter</span><span class="p">()</span>
- <span class="n">context</span> <span class="o">=</span> <span class="n">PhaseContext</span><span class="p">(</span><span class="n">epoch</span><span class="o">=</span><span class="n">epoch</span><span class="p">,</span>
- <span class="n">optimizer</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">optimizer</span><span class="p">,</span>
- <span class="n">metrics_compute_fn</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">train_metrics</span><span class="p">,</span>
- <span class="n">loss_avg_meter</span><span class="o">=</span><span class="n">loss_avg_meter</span><span class="p">,</span>
- <span class="n">criterion</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">criterion</span><span class="p">,</span>
- <span class="n">device</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">device</span><span class="p">,</span>
- <span class="n">lr_warmup_epochs</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">training_params</span><span class="o">.</span><span class="n">lr_warmup_epochs</span><span class="p">,</span>
- <span class="n">sg_logger</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">sg_logger</span><span class="p">)</span>
- <span class="k">for</span> <span class="n">batch_idx</span><span class="p">,</span> <span class="n">batch_items</span> <span class="ow">in</span> <span class="nb">enumerate</span><span class="p">(</span><span class="n">progress_bar_train_loader</span><span class="p">):</span>
- <span class="n">batch_items</span> <span class="o">=</span> <span class="n">core_utils</span><span class="o">.</span><span class="n">tensor_container_to_device</span><span class="p">(</span><span class="n">batch_items</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">device</span><span class="p">,</span> <span class="n">non_blocking</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span>
- <span class="n">inputs</span><span class="p">,</span> <span class="n">targets</span><span class="p">,</span> <span class="n">additional_batch_items</span> <span class="o">=</span> <span class="n">sg_model_utils</span><span class="o">.</span><span class="n">unpack_batch_items</span><span class="p">(</span><span class="n">batch_items</span><span class="p">)</span>
- <span class="c1"># AUTOCAST IS ENABLED ONLY IF self.training_params.mixed_precision - IF enabled=False AUTOCAST HAS NO EFFECT</span>
- <span class="k">with</span> <span class="n">autocast</span><span class="p">(</span><span class="n">enabled</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">training_params</span><span class="o">.</span><span class="n">mixed_precision</span><span class="p">):</span>
- <span class="c1"># FORWARD PASS TO GET NETWORK'S PREDICTIONS</span>
- <span class="n">outputs</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">net</span><span class="p">(</span><span class="n">inputs</span><span class="p">)</span>
- <span class="c1"># COMPUTE THE LOSS FOR BACK PROP + EXTRA METRICS COMPUTED DURING THE LOSS FORWARD PASS</span>
- <span class="n">loss</span><span class="p">,</span> <span class="n">loss_log_items</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_get_losses</span><span class="p">(</span><span class="n">outputs</span><span class="p">,</span> <span class="n">targets</span><span class="p">)</span>
- <span class="n">context</span><span class="o">.</span><span class="n">update_context</span><span class="p">(</span><span class="n">batch_idx</span><span class="o">=</span><span class="n">batch_idx</span><span class="p">,</span>
- <span class="n">inputs</span><span class="o">=</span><span class="n">inputs</span><span class="p">,</span>
- <span class="n">preds</span><span class="o">=</span><span class="n">outputs</span><span class="p">,</span>
- <span class="n">target</span><span class="o">=</span><span class="n">targets</span><span class="p">,</span>
- <span class="n">loss_log_items</span><span class="o">=</span><span class="n">loss_log_items</span><span class="p">,</span>
- <span class="o">**</span><span class="n">additional_batch_items</span><span class="p">)</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">phase_callback_handler</span><span class="p">(</span><span class="n">Phase</span><span class="o">.</span><span class="n">TRAIN_BATCH_END</span><span class="p">,</span> <span class="n">context</span><span class="p">)</span>
- <span class="c1"># LOG LR THAT WILL BE USED IN CURRENT EPOCH AND AFTER FIRST WARMUP/LR_SCHEDULER UPDATE BEFORE WEIGHT UPDATE</span>
- <span class="k">if</span> <span class="ow">not</span> <span class="bp">self</span><span class="o">.</span><span class="n">ddp_silent_mode</span> <span class="ow">and</span> <span class="n">batch_idx</span> <span class="o">==</span> <span class="mi">0</span><span class="p">:</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">_write_lrs</span><span class="p">(</span><span class="n">epoch</span><span class="p">)</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">backward_step</span><span class="p">(</span><span class="n">loss</span><span class="p">,</span> <span class="n">epoch</span><span class="p">,</span> <span class="n">batch_idx</span><span class="p">,</span> <span class="n">context</span><span class="p">)</span>
- <span class="c1"># COMPUTE THE RUNNING USER METRICS AND LOSS RUNNING ITEMS. RESULT TUPLE IS THEIR CONCATENATION.</span>
- <span class="n">logging_values</span> <span class="o">=</span> <span class="n">loss_avg_meter</span><span class="o">.</span><span class="n">average</span> <span class="o">+</span> <span class="n">get_metrics_results_tuple</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">train_metrics</span><span class="p">)</span>
- <span class="n">gpu_memory_utilization</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">cuda</span><span class="o">.</span><span class="n">memory_cached</span><span class="p">()</span> <span class="o">/</span> <span class="mf">1E9</span> <span class="k">if</span> <span class="n">torch</span><span class="o">.</span><span class="n">cuda</span><span class="o">.</span><span class="n">is_available</span><span class="p">()</span> <span class="k">else</span> <span class="mi">0</span>
- <span class="c1"># RENDER METRICS PROGRESS</span>
- <span class="n">pbar_message_dict</span> <span class="o">=</span> <span class="n">get_train_loop_description_dict</span><span class="p">(</span><span class="n">logging_values</span><span class="p">,</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">train_metrics</span><span class="p">,</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">loss_logging_items_names</span><span class="p">,</span>
- <span class="n">gpu_mem</span><span class="o">=</span><span class="n">gpu_memory_utilization</span><span class="p">)</span>
- <span class="n">progress_bar_train_loader</span><span class="o">.</span><span class="n">set_postfix</span><span class="p">(</span><span class="o">**</span><span class="n">pbar_message_dict</span><span class="p">)</span>
- <span class="k">if</span> <span class="ow">not</span> <span class="bp">self</span><span class="o">.</span><span class="n">ddp_silent_mode</span><span class="p">:</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">sg_logger</span><span class="o">.</span><span class="n">upload</span><span class="p">()</span>
- <span class="k">return</span> <span class="n">logging_values</span>
- <span class="k">def</span> <span class="nf">_get_losses</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">outputs</span><span class="p">:</span> <span class="n">torch</span><span class="o">.</span><span class="n">Tensor</span><span class="p">,</span> <span class="n">targets</span><span class="p">:</span> <span class="n">torch</span><span class="o">.</span><span class="n">Tensor</span><span class="p">)</span> <span class="o">-></span> <span class="n">Tuple</span><span class="p">[</span><span class="n">torch</span><span class="o">.</span><span class="n">Tensor</span><span class="p">,</span> <span class="nb">tuple</span><span class="p">]:</span>
- <span class="c1"># GET THE OUTPUT OF THE LOSS FUNCTION</span>
- <span class="n">loss</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">criterion</span><span class="p">(</span><span class="n">outputs</span><span class="p">,</span> <span class="n">targets</span><span class="p">)</span>
- <span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">loss</span><span class="p">,</span> <span class="nb">tuple</span><span class="p">):</span>
- <span class="n">loss</span><span class="p">,</span> <span class="n">loss_logging_items</span> <span class="o">=</span> <span class="n">loss</span>
- <span class="c1"># IF ITS NOT A TUPLE THE LOGGING ITEMS CONTAIN ONLY THE LOSS FOR BACKPROP (USER DEFINED LOSS RETURNS SCALAR)</span>
- <span class="k">else</span><span class="p">:</span>
- <span class="n">loss_logging_items</span> <span class="o">=</span> <span class="n">loss</span><span class="o">.</span><span class="n">unsqueeze</span><span class="p">(</span><span class="mi">0</span><span class="p">)</span><span class="o">.</span><span class="n">detach</span><span class="p">()</span>
- <span class="k">if</span> <span class="nb">len</span><span class="p">(</span><span class="n">loss_logging_items</span><span class="p">)</span> <span class="o">!=</span> <span class="nb">len</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">loss_logging_items_names</span><span class="p">):</span>
- <span class="k">raise</span> <span class="ne">ValueError</span><span class="p">(</span><span class="s2">"Loss output length must match loss_logging_items_names. Got "</span> <span class="o">+</span> <span class="nb">str</span><span class="p">(</span>
- <span class="nb">len</span><span class="p">(</span><span class="n">loss_logging_items</span><span class="p">))</span> <span class="o">+</span> <span class="s1">', and '</span> <span class="o">+</span> <span class="nb">str</span><span class="p">(</span><span class="nb">len</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">loss_logging_items_names</span><span class="p">)))</span>
- <span class="c1"># RETURN AND THE LOSS LOGGING ITEMS COMPUTED DURING LOSS FORWARD PASS</span>
- <span class="k">return</span> <span class="n">loss</span><span class="p">,</span> <span class="n">loss_logging_items</span>
- <div class="viewcode-block" id="SgModel.backward_step"><a class="viewcode-back" href="../../../../super_gradients.training.sg_model.html#super_gradients.training.SgModel.backward_step">[docs]</a> <span class="k">def</span> <span class="nf">backward_step</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">loss</span><span class="p">:</span> <span class="n">torch</span><span class="o">.</span><span class="n">Tensor</span><span class="p">,</span> <span class="n">epoch</span><span class="p">:</span> <span class="nb">int</span><span class="p">,</span> <span class="n">batch_idx</span><span class="p">:</span> <span class="nb">int</span><span class="p">,</span> <span class="n">context</span><span class="p">:</span> <span class="n">PhaseContext</span><span class="p">):</span>
- <span class="sd">"""</span>
- <span class="sd"> Run backprop on the loss and perform a step</span>
- <span class="sd"> :param loss: The value computed by the loss function</span>
- <span class="sd"> :param optimizer: An object that can perform a gradient step and zeroize model gradient</span>
- <span class="sd"> :param epoch: number of epoch the training is on</span>
- <span class="sd"> :param batch_idx: number of iteration inside the current epoch</span>
- <span class="sd"> :param context: current phase context</span>
- <span class="sd"> :return:</span>
- <span class="sd"> """</span>
- <span class="c1"># SCALER IS ENABLED ONLY IF self.training_params.mixed_precision=True</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">scaler</span><span class="o">.</span><span class="n">scale</span><span class="p">(</span><span class="n">loss</span><span class="p">)</span><span class="o">.</span><span class="n">backward</span><span class="p">()</span>
- <span class="c1"># ACCUMULATE GRADIENT FOR X BATCHES BEFORE OPTIMIZING</span>
- <span class="n">integrated_batches_num</span> <span class="o">=</span> <span class="n">batch_idx</span> <span class="o">+</span> <span class="nb">len</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">train_loader</span><span class="p">)</span> <span class="o">*</span> <span class="n">epoch</span> <span class="o">+</span> <span class="mi">1</span>
- <span class="k">if</span> <span class="n">integrated_batches_num</span> <span class="o">%</span> <span class="bp">self</span><span class="o">.</span><span class="n">batch_accumulate</span> <span class="o">==</span> <span class="mi">0</span><span class="p">:</span>
- <span class="c1"># SCALER IS ENABLED ONLY IF self.training_params.mixed_precision=True</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">scaler</span><span class="o">.</span><span class="n">step</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">optimizer</span><span class="p">)</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">scaler</span><span class="o">.</span><span class="n">update</span><span class="p">()</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">optimizer</span><span class="o">.</span><span class="n">zero_grad</span><span class="p">()</span>
- <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">ema</span><span class="p">:</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">ema_model</span><span class="o">.</span><span class="n">update</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">net</span><span class="p">,</span> <span class="n">integrated_batches_num</span> <span class="o">/</span> <span class="p">(</span><span class="nb">len</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">train_loader</span><span class="p">)</span> <span class="o">*</span> <span class="bp">self</span><span class="o">.</span><span class="n">max_epochs</span><span class="p">))</span>
- <span class="c1"># RUN PHASE CALLBACKS</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">phase_callback_handler</span><span class="p">(</span><span class="n">Phase</span><span class="o">.</span><span class="n">TRAIN_BATCH_STEP</span><span class="p">,</span> <span class="n">context</span><span class="p">)</span></div>
- <div class="viewcode-block" id="SgModel.save_checkpoint"><a class="viewcode-back" href="../../../../super_gradients.training.sg_model.html#super_gradients.training.SgModel.save_checkpoint">[docs]</a> <span class="k">def</span> <span class="nf">save_checkpoint</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">optimizer</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span> <span class="n">epoch</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span> <span class="n">validation_results_tuple</span><span class="p">:</span> <span class="nb">tuple</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span> <span class="n">context</span><span class="p">:</span> <span class="n">PhaseContext</span> <span class="o">=</span> <span class="kc">None</span><span class="p">):</span>
- <span class="sd">"""</span>
- <span class="sd"> Save the current state dict as latest (always), best (if metric was improved), epoch# (if determined in training</span>
- <span class="sd"> params)</span>
- <span class="sd"> """</span>
- <span class="c1"># WHEN THE validation_results_tuple IS NONE WE SIMPLY SAVE THE state_dict AS LATEST AND Return</span>
- <span class="k">if</span> <span class="n">validation_results_tuple</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">sg_logger</span><span class="o">.</span><span class="n">add_checkpoint</span><span class="p">(</span><span class="n">tag</span><span class="o">=</span><span class="s1">'ckpt_latest_weights_only.pth'</span><span class="p">,</span> <span class="n">state_dict</span><span class="o">=</span><span class="p">{</span><span class="s1">'net'</span><span class="p">:</span> <span class="bp">self</span><span class="o">.</span><span class="n">net</span><span class="o">.</span><span class="n">state_dict</span><span class="p">()},</span>
- <span class="n">global_step</span><span class="o">=</span><span class="n">epoch</span><span class="p">)</span>
- <span class="k">return</span>
- <span class="c1"># COMPUTE THE CURRENT metric</span>
- <span class="c1"># IF idx IS A LIST - SUM ALL THE VALUES STORED IN THE LIST'S INDICES</span>
- <span class="n">metric</span> <span class="o">=</span> <span class="n">validation_results_tuple</span><span class="p">[</span><span class="bp">self</span><span class="o">.</span><span class="n">metric_idx_in_results_tuple</span><span class="p">]</span> <span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">metric_idx_in_results_tuple</span><span class="p">,</span> <span class="nb">int</span><span class="p">)</span> <span class="k">else</span> \
- <span class="nb">sum</span><span class="p">([</span><span class="n">validation_results_tuple</span><span class="p">[</span><span class="n">idx</span><span class="p">]</span> <span class="k">for</span> <span class="n">idx</span> <span class="ow">in</span> <span class="bp">self</span><span class="o">.</span><span class="n">metric_idx_in_results_tuple</span><span class="p">])</span>
- <span class="c1"># BUILD THE state_dict</span>
- <span class="n">state</span> <span class="o">=</span> <span class="p">{</span><span class="s1">'net'</span><span class="p">:</span> <span class="bp">self</span><span class="o">.</span><span class="n">net</span><span class="o">.</span><span class="n">state_dict</span><span class="p">(),</span> <span class="s1">'acc'</span><span class="p">:</span> <span class="n">metric</span><span class="p">,</span> <span class="s1">'epoch'</span><span class="p">:</span> <span class="n">epoch</span><span class="p">}</span>
- <span class="k">if</span> <span class="n">optimizer</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
- <span class="n">state</span><span class="p">[</span><span class="s1">'optimizer_state_dict'</span><span class="p">]</span> <span class="o">=</span> <span class="n">optimizer</span><span class="o">.</span><span class="n">state_dict</span><span class="p">()</span>
- <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">scaler</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
- <span class="n">state</span><span class="p">[</span><span class="s1">'scaler_state_dict'</span><span class="p">]</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">scaler</span><span class="o">.</span><span class="n">state_dict</span><span class="p">()</span>
- <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">ema</span><span class="p">:</span>
- <span class="n">state</span><span class="p">[</span><span class="s1">'ema_net'</span><span class="p">]</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">ema_model</span><span class="o">.</span><span class="n">ema</span><span class="o">.</span><span class="n">state_dict</span><span class="p">()</span>
- <span class="c1"># SAVES CURRENT MODEL AS ckpt_latest</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">sg_logger</span><span class="o">.</span><span class="n">add_checkpoint</span><span class="p">(</span><span class="n">tag</span><span class="o">=</span><span class="s1">'ckpt_latest.pth'</span><span class="p">,</span> <span class="n">state_dict</span><span class="o">=</span><span class="n">state</span><span class="p">,</span> <span class="n">global_step</span><span class="o">=</span><span class="n">epoch</span><span class="p">)</span>
- <span class="c1"># SAVE MODEL AT SPECIFIC EPOCHS DETERMINED BY save_ckpt_epoch_list</span>
- <span class="k">if</span> <span class="n">epoch</span> <span class="ow">in</span> <span class="bp">self</span><span class="o">.</span><span class="n">training_params</span><span class="o">.</span><span class="n">save_ckpt_epoch_list</span><span class="p">:</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">sg_logger</span><span class="o">.</span><span class="n">add_checkpoint</span><span class="p">(</span><span class="n">tag</span><span class="o">=</span><span class="sa">f</span><span class="s1">'ckpt_epoch_</span><span class="si">{</span><span class="n">epoch</span><span class="si">}</span><span class="s1">.pth'</span><span class="p">,</span> <span class="n">state_dict</span><span class="o">=</span><span class="n">state</span><span class="p">,</span> <span class="n">global_step</span><span class="o">=</span><span class="n">epoch</span><span class="p">)</span>
- <span class="c1"># OVERRIDE THE BEST CHECKPOINT AND best_metric IF metric GOT BETTER THAN THE PREVIOUS BEST</span>
- <span class="k">if</span> <span class="p">(</span><span class="n">metric</span> <span class="o">></span> <span class="bp">self</span><span class="o">.</span><span class="n">best_metric</span> <span class="ow">and</span> <span class="bp">self</span><span class="o">.</span><span class="n">greater_metric_to_watch_is_better</span><span class="p">)</span> <span class="ow">or</span> <span class="p">(</span>
- <span class="n">metric</span> <span class="o"><</span> <span class="bp">self</span><span class="o">.</span><span class="n">best_metric</span> <span class="ow">and</span> <span class="ow">not</span> <span class="bp">self</span><span class="o">.</span><span class="n">greater_metric_to_watch_is_better</span><span class="p">):</span>
- <span class="c1"># STORE THE CURRENT metric AS BEST</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">best_metric</span> <span class="o">=</span> <span class="n">metric</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">sg_logger</span><span class="o">.</span><span class="n">add_checkpoint</span><span class="p">(</span><span class="n">tag</span><span class="o">=</span><span class="s1">'ckpt_best.pth'</span><span class="p">,</span> <span class="n">state_dict</span><span class="o">=</span><span class="n">state</span><span class="p">,</span> <span class="n">global_step</span><span class="o">=</span><span class="n">epoch</span><span class="p">)</span>
- <span class="c1"># RUN PHASE CALLBACKS</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">phase_callback_handler</span><span class="p">(</span><span class="n">Phase</span><span class="o">.</span><span class="n">VALIDATION_END_BEST_EPOCH</span><span class="p">,</span> <span class="n">context</span><span class="p">)</span>
- <span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">metric</span><span class="p">,</span> <span class="n">torch</span><span class="o">.</span><span class="n">Tensor</span><span class="p">):</span>
- <span class="n">metric</span> <span class="o">=</span> <span class="n">metric</span><span class="o">.</span><span class="n">item</span><span class="p">()</span>
- <span class="n">logger</span><span class="o">.</span><span class="n">info</span><span class="p">(</span><span class="s2">"Best checkpoint overriden: validation "</span> <span class="o">+</span> <span class="bp">self</span><span class="o">.</span><span class="n">metric_to_watch</span> <span class="o">+</span> <span class="s2">": "</span> <span class="o">+</span> <span class="nb">str</span><span class="p">(</span><span class="n">metric</span><span class="p">))</span>
- <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">training_params</span><span class="o">.</span><span class="n">average_best_models</span><span class="p">:</span>
- <span class="n">net_for_averaging</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">ema_model</span><span class="o">.</span><span class="n">ema</span> <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">ema</span> <span class="k">else</span> <span class="bp">self</span><span class="o">.</span><span class="n">net</span>
- <span class="n">averaged_model_sd</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">model_weight_averaging</span><span class="o">.</span><span class="n">get_average_model</span><span class="p">(</span><span class="n">net_for_averaging</span><span class="p">,</span>
- <span class="n">validation_results_tuple</span><span class="o">=</span><span class="n">validation_results_tuple</span><span class="p">)</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">sg_logger</span><span class="o">.</span><span class="n">add_checkpoint</span><span class="p">(</span><span class="n">tag</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">average_model_checkpoint_filename</span><span class="p">,</span>
- <span class="n">state_dict</span><span class="o">=</span><span class="p">{</span><span class="s1">'net'</span><span class="p">:</span> <span class="n">averaged_model_sd</span><span class="p">},</span> <span class="n">global_step</span><span class="o">=</span><span class="n">epoch</span><span class="p">)</span></div>
- <span class="c1"># FIXME - we need to resolve flake8's 'function is too complex' for this function</span>
- <div class="viewcode-block" id="SgModel.train"><a class="viewcode-back" href="../../../../super_gradients.training.sg_model.html#super_gradients.training.SgModel.train">[docs]</a> <span class="k">def</span> <span class="nf">train</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">training_params</span><span class="p">:</span> <span class="nb">dict</span> <span class="o">=</span> <span class="nb">dict</span><span class="p">()):</span> <span class="c1"># noqa: C901</span>
- <span class="sd">"""</span>
- <span class="sd"> train - Trains the Model</span>
- <span class="sd"> IMPORTANT NOTE: Additional batch parameters can be added as a third item (optional) if a tuple is returned by</span>
- <span class="sd"> the data loaders, as dictionary. The phase context will hold the additional items, under an attribute with</span>
- <span class="sd"> the same name as the key in this dictionary. Then such items can be accessed through phase callbacks.</span>
- <span class="sd"> :param training_params:</span>
- <span class="sd"> - `max_epochs` : int</span>
- <span class="sd"> Number of epochs to run training.</span>
- <span class="sd"> - `lr_updates` : list(int)</span>
- <span class="sd"> List of fixed epoch numbers to perform learning rate updates when `lr_mode='step'`.</span>
- <span class="sd"> - `lr_decay_factor` : float</span>
- <span class="sd"> Decay factor to apply to the learning rate at each update when `lr_mode='step'`.</span>
- <span class="sd"> - `lr_mode` : str</span>
- <span class="sd"> Learning rate scheduling policy, one of ['step','poly','cosine','function']. 'step' refers to</span>
- <span class="sd"> constant updates at epoch numbers passed through `lr_updates`. 'cosine' refers to Cosine Anealing</span>
- <span class="sd"> policy as mentioned in https://arxiv.org/abs/1608.03983. 'poly' refers to polynomial decrease i.e</span>
- <span class="sd"> in each epoch iteration `self.lr = self.initial_lr * pow((1.0 - (current_iter / max_iter)),</span>
- <span class="sd"> 0.9)` 'function' refers to user defined learning rate scheduling function, that is passed through</span>
- <span class="sd"> `lr_schedule_function`.</span>
- <span class="sd"> - `lr_schedule_function` : Union[callable,None]</span>
- <span class="sd"> Learning rate scheduling function to be used when `lr_mode` is 'function'.</span>
- <span class="sd"> - `lr_warmup_epochs` : int (default=0)</span>
- <span class="sd"> Number of epochs for learning rate warm up - see https://arxiv.org/pdf/1706.02677.pdf (Section 2.2).</span>
- <span class="sd"> - `cosine_final_lr_ratio` : float (default=0.01)</span>
- <span class="sd"> Final learning rate ratio (only relevant when `lr_mode`='cosine'). The cosine starts from initial_lr and reaches</span>
- <span class="sd"> initial_lr * cosine_final_lr_ratio in last epoch</span>
- <span class="sd"> - `inital_lr` : float</span>
- <span class="sd"> Initial learning rate.</span>
- <span class="sd"> - `loss` : Union[nn.module, str]</span>
- <span class="sd"> Loss function for training.</span>
- <span class="sd"> One of SuperGradient's built in options:</span>
- <span class="sd"> "cross_entropy": LabelSmoothingCrossEntropyLoss,</span>
- <span class="sd"> "mse": MSELoss,</span>
- <span class="sd"> "r_squared_loss": RSquaredLoss,</span>
- <span class="sd"> "detection_loss": YoLoV3DetectionLoss,</span>
- <span class="sd"> "shelfnet_ohem_loss": ShelfNetOHEMLoss,</span>
- <span class="sd"> "shelfnet_se_loss": ShelfNetSemanticEncodingLoss,</span>
- <span class="sd"> "yolo_v5_loss": YoLoV5DetectionLoss,</span>
- <span class="sd"> "ssd_loss": SSDLoss,</span>
- <span class="sd"> or user defined nn.module loss function.</span>
- <span class="sd"> IMPORTANT: forward(...) should return a (loss, loss_items) tuple where loss is the tensor used</span>
- <span class="sd"> for backprop (i.e what your original loss function returns), and loss_items should be a tensor of</span>
- <span class="sd"> shape (n_items), of values computed during the forward pass which we desire to log over the</span>
- <span class="sd"> entire epoch. For example- the loss itself should always be logged. Another example is a scenario</span>
- <span class="sd"> where the computed loss is the sum of a few components we would like to log- these entries in</span>
- <span class="sd"> loss_items).</span>
- <span class="sd"> When training, set the loss_logging_items_names parameter in train_params to be a list of</span>
- <span class="sd"> strings, of length n_items who's ith element is the name of the ith entry in loss_items. Then</span>
- <span class="sd"> each item will be logged, rendered on tensorboard and "watched" (i.e saving model checkpoints</span>
- <span class="sd"> according to it).</span>
- <span class="sd"> Since running logs will save the loss_items in some internal state, it is recommended that</span>
- <span class="sd"> loss_items are detached from their computational graph for memory efficiency.</span>
- <span class="sd"> - `optimizer` : Union[str, torch.optim.Optimizer]</span>
- <span class="sd"> Optimization algorithm. One of ['Adam','SGD','RMSProp'] corresponding to the torch.optim</span>
- <span class="sd"> optimzers implementations, or any object that implements torch.optim.Optimizer.</span>
- <span class="sd"> - `criterion_params` : dict</span>
- <span class="sd"> Loss function parameters.</span>
- <span class="sd"> - `optimizer_params` : dict</span>
- <span class="sd"> When `optimizer` is one of ['Adam','SGD','RMSProp'], it will be initialized with optimizer_params.</span>
- <span class="sd"> (see https://pytorch.org/docs/stable/optim.html for the full list of</span>
- <span class="sd"> parameters for each optimizer).</span>
- <span class="sd"> - `train_metrics_list` : list(torchmetrics.Metric)</span>
- <span class="sd"> Metrics to log during training. For more information on torchmetrics see</span>
- <span class="sd"> https://torchmetrics.rtfd.io/en/latest/.</span>
- <span class="sd"> - `valid_metrics_list` : list(torchmetrics.Metric)</span>
- <span class="sd"> Metrics to log during validation/testing. For more information on torchmetrics see</span>
- <span class="sd"> https://torchmetrics.rtfd.io/en/latest/.</span>
- <span class="sd"> - `loss_logging_items_names` : list(str)</span>
- <span class="sd"> The list of names/titles for the outputs returned from the loss functions forward pass (reminder-</span>
- <span class="sd"> the loss function should return the tuple (loss, loss_items)). These names will be used for</span>
- <span class="sd"> logging their values.</span>
- <span class="sd"> - `metric_to_watch` : str (default="Accuracy")</span>
- <span class="sd"> will be the metric which the model checkpoint will be saved according to, and can be set to any</span>
- <span class="sd"> of the following:</span>
- <span class="sd"> a metric name (str) of one of the metric objects from the valid_metrics_list</span>
- <span class="sd"> a "metric_name" if some metric in valid_metrics_list has an attribute component_names which</span>
- <span class="sd"> is a list referring to the names of each entry in the output metric (torch tensor of size n)</span>
- <span class="sd"> one of "loss_logging_items_names" i.e which will correspond to an item returned during the</span>
- <span class="sd"> loss function's forward pass.</span>
- <span class="sd"> At the end of each epoch, if a new best metric_to_watch value is achieved, the models checkpoint</span>
- <span class="sd"> is saved in YOUR_PYTHON_PATH/checkpoints/ckpt_best.pth</span>
- <span class="sd"> - `greater_metric_to_watch_is_better` : bool</span>
- <span class="sd"> When choosing a model's checkpoint to be saved, the best achieved model is the one that maximizes the</span>
- <span class="sd"> metric_to_watch when this parameter is set to True, and a one that minimizes it otherwise.</span>
- <span class="sd"> - `ema` : bool (default=False)</span>
- <span class="sd"> Whether to use Model Exponential Moving Average (see</span>
- <span class="sd"> https://github.com/rwightman/pytorch-image-models ema implementation)</span>
- <span class="sd"> - `batch_accumulate` : int (default=1)</span>
- <span class="sd"> Number of batches to accumulate before every backward pass.</span>
- <span class="sd"> - `ema_params` : dict</span>
- <span class="sd"> Parameters for the ema model.</span>
- <span class="sd"> - `zero_weight_decay_on_bias_and_bn` : bool (default=False)</span>
- <span class="sd"> Whether to apply weight decay on batch normalization parameters or not (ignored when the passed</span>
- <span class="sd"> optimizer has already been initialized).</span>
- <span class="sd"> - `load_opt_params` : bool (default=True)</span>
- <span class="sd"> Whether to load the optimizers parameters as well when loading a model's checkpoint.</span>
- <span class="sd"> - `run_validation_freq` : int (default=1)</span>
- <span class="sd"> The frequency in which validation is performed during training (i.e the validation is ran every</span>
- <span class="sd"> `run_validation_freq` epochs.</span>
- <span class="sd"> - `save_model` : bool (default=True)</span>
- <span class="sd"> Whether to save the model checkpoints.</span>
- <span class="sd"> - `silent_mode` : bool</span>
- <span class="sd"> Silents the print outs.</span>
- <span class="sd"> - `mixed_precision` : bool</span>
- <span class="sd"> Whether to use mixed precision or not.</span>
- <span class="sd"> - `save_ckpt_epoch_list` : list(int) (default=[])</span>
- <span class="sd"> List of fixed epoch indices the user wishes to save checkpoints in.</span>
- <span class="sd"> - `average_best_models` : bool (default=False)</span>
- <span class="sd"> If set, a snapshot dictionary file and the average model will be saved / updated at every epoch</span>
- <span class="sd"> and evaluated only when training is completed. The snapshot file will only be deleted upon</span>
- <span class="sd"> completing the training. The snapshot dict will be managed on cpu.</span>
- <span class="sd"> - `precise_bn` : bool (default=False)</span>
- <span class="sd"> Whether to use precise_bn calculation during the training.</span>
- <span class="sd"> - `precise_bn_batch_size` : int (default=None)</span>
- <span class="sd"> The effective batch size we want to calculate the batchnorm on. For example, if we are training a model</span>
- <span class="sd"> on 8 gpus, with a batch of 128 on each gpu, a good rule of thumb would be to give it 8192</span>
- <span class="sd"> (ie: effective_batch_size * num_gpus = batch_per_gpu * num_gpus * num_gpus).</span>
- <span class="sd"> If precise_bn_batch_size is not provided in the training_params, the latter heuristic will be taken.</span>
- <span class="sd"> - `seed` : int (default=42)</span>
- <span class="sd"> Random seed to be set for torch, numpy, and random. When using DDP each process will have it's seed</span>
- <span class="sd"> set to seed + rank.</span>
- <span class="sd"> - `log_installed_packages` : bool (default=False)</span>
- <span class="sd"> When set, the list of all installed packages (and their versions) will be written to the tensorboard</span>
- <span class="sd"> and logfile (useful when trying to reproduce results).</span>
- <span class="sd"> - `dataset_statistics` : bool (default=False)</span>
- <span class="sd"> Enable a statistic analysis of the dataset. If set to True the dataset will be analyzed and a report</span>
- <span class="sd"> will be added to the tensorboard along with some sample images from the dataset. Currently only</span>
- <span class="sd"> detection datasets are supported for analysis.</span>
- <span class="sd"> - `save_full_train_log` : bool (default=False)</span>
- <span class="sd"> When set, a full log (of all super_gradients modules, including uncaught exceptions from any other</span>
- <span class="sd"> module) of the training will be saved in the checkpoint directory under full_train_log.log</span>
- <span class="sd"> - `sg_logger` : Union[AbstractSGLogger, str] (defauls=base_sg_logger)</span>
- <span class="sd"> Define the SGLogger object for this training process. The SGLogger handles all disk writes, logs, TensorBoard, remote logging</span>
- <span class="sd"> and remote storage. By overriding the default base_sg_logger, you can change the storage location, support external monitoring and logging</span>
- <span class="sd"> or support remote storage.</span>
- <span class="sd"> - `sg_logger_params` : dict</span>
- <span class="sd"> SGLogger parameters</span>
- <span class="sd"> :return:</span>
- <span class="sd"> """</span>
- <span class="k">global</span> <span class="n">logger</span>
- <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">net</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
- <span class="k">raise</span> <span class="ne">Exception</span><span class="p">(</span><span class="s1">'Model'</span><span class="p">,</span> <span class="s1">'No model found'</span><span class="p">)</span>
- <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">dataset_interface</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
- <span class="k">raise</span> <span class="ne">Exception</span><span class="p">(</span><span class="s1">'Data'</span><span class="p">,</span> <span class="s1">'No dataset found'</span><span class="p">)</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">training_params</span> <span class="o">=</span> <span class="n">TrainingParams</span><span class="p">()</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">training_params</span><span class="o">.</span><span class="n">override</span><span class="p">(</span><span class="o">**</span><span class="n">training_params</span><span class="p">)</span>
- <span class="c1"># SET RANDOM SEED</span>
- <span class="n">random_seed</span><span class="p">(</span><span class="n">is_ddp</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">multi_gpu</span> <span class="o">==</span> <span class="n">MultiGPUMode</span><span class="o">.</span><span class="n">DISTRIBUTED_DATA_PARALLEL</span><span class="p">,</span>
- <span class="n">device</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">device</span><span class="p">,</span> <span class="n">seed</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">training_params</span><span class="o">.</span><span class="n">seed</span><span class="p">)</span>
- <span class="n">silent_mode</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">training_params</span><span class="o">.</span><span class="n">silent_mode</span> <span class="ow">or</span> <span class="bp">self</span><span class="o">.</span><span class="n">ddp_silent_mode</span>
- <span class="c1"># METRICS</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">_set_train_metrics</span><span class="p">(</span><span class="n">train_metrics_list</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">training_params</span><span class="o">.</span><span class="n">train_metrics_list</span><span class="p">)</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">_set_valid_metrics</span><span class="p">(</span><span class="n">valid_metrics_list</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">training_params</span><span class="o">.</span><span class="n">valid_metrics_list</span><span class="p">)</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">loss_logging_items_names</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">training_params</span><span class="o">.</span><span class="n">loss_logging_items_names</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">results_titles</span> <span class="o">=</span> <span class="p">[</span><span class="s2">"Train_"</span> <span class="o">+</span> <span class="n">t</span> <span class="k">for</span> <span class="n">t</span> <span class="ow">in</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">loss_logging_items_names</span> <span class="o">+</span> <span class="n">get_metrics_titles</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">train_metrics</span><span class="p">)]</span> <span class="o">+</span> \
- <span class="p">[</span><span class="s2">"Valid_"</span> <span class="o">+</span> <span class="n">t</span> <span class="k">for</span> <span class="n">t</span> <span class="ow">in</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">loss_logging_items_names</span> <span class="o">+</span> <span class="n">get_metrics_titles</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">valid_metrics</span><span class="p">)]</span>
- <span class="c1"># Store the metric to follow (loss\accuracy) and initialize as the worst value</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">metric_to_watch</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">training_params</span><span class="o">.</span><span class="n">metric_to_watch</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">greater_metric_to_watch_is_better</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">training_params</span><span class="o">.</span><span class="n">greater_metric_to_watch_is_better</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">metric_idx_in_results_tuple</span> <span class="o">=</span> <span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">loss_logging_items_names</span> <span class="o">+</span> <span class="n">get_metrics_titles</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">valid_metrics</span><span class="p">))</span><span class="o">.</span><span class="n">index</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">metric_to_watch</span><span class="p">)</span>
- <span class="c1"># Allowing loading instantiated loss or string</span>
- <span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">training_params</span><span class="o">.</span><span class="n">loss</span><span class="p">,</span> <span class="nb">str</span><span class="p">):</span>
- <span class="n">criterion_cls</span> <span class="o">=</span> <span class="n">LOSSES</span><span class="p">[</span><span class="bp">self</span><span class="o">.</span><span class="n">training_params</span><span class="o">.</span><span class="n">loss</span><span class="p">]</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">criterion</span> <span class="o">=</span> <span class="n">criterion_cls</span><span class="p">(</span><span class="o">**</span><span class="bp">self</span><span class="o">.</span><span class="n">training_params</span><span class="o">.</span><span class="n">criterion_params</span><span class="p">)</span>
- <span class="k">elif</span> <span class="nb">isinstance</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">training_params</span><span class="o">.</span><span class="n">loss</span><span class="p">,</span> <span class="n">Mapping</span><span class="p">):</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">criterion</span> <span class="o">=</span> <span class="n">LossesFactory</span><span class="p">()</span><span class="o">.</span><span class="n">get</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">training_params</span><span class="o">.</span><span class="n">loss</span><span class="p">)</span>
- <span class="k">elif</span> <span class="nb">isinstance</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">training_params</span><span class="o">.</span><span class="n">loss</span><span class="p">,</span> <span class="n">nn</span><span class="o">.</span><span class="n">Module</span><span class="p">):</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">criterion</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">training_params</span><span class="o">.</span><span class="n">loss</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">criterion</span><span class="o">.</span><span class="n">to</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">device</span><span class="p">)</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">max_epochs</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">training_params</span><span class="o">.</span><span class="n">max_epochs</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">ema</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">training_params</span><span class="o">.</span><span class="n">ema</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">precise_bn</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">training_params</span><span class="o">.</span><span class="n">precise_bn</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">precise_bn_batch_size</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">training_params</span><span class="o">.</span><span class="n">precise_bn_batch_size</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">batch_accumulate</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">training_params</span><span class="o">.</span><span class="n">batch_accumulate</span>
- <span class="n">num_batches</span> <span class="o">=</span> <span class="nb">len</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">train_loader</span><span class="p">)</span>
- <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">ema</span><span class="p">:</span>
- <span class="n">ema_params</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">training_params</span><span class="o">.</span><span class="n">ema_params</span>
- <span class="n">logger</span><span class="o">.</span><span class="n">info</span><span class="p">(</span><span class="sa">f</span><span class="s1">'Using EMA with params </span><span class="si">{</span><span class="n">ema_params</span><span class="si">}</span><span class="s1">'</span><span class="p">)</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">ema_model</span> <span class="o">=</span> <span class="n">ModelEMA</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">net</span><span class="p">,</span> <span class="o">**</span><span class="n">ema_params</span><span class="p">)</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">ema_model</span><span class="o">.</span><span class="n">updates</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">start_epoch</span> <span class="o">*</span> <span class="n">num_batches</span> <span class="o">//</span> <span class="bp">self</span><span class="o">.</span><span class="n">batch_accumulate</span>
- <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">load_checkpoint</span><span class="p">:</span>
- <span class="k">if</span> <span class="s1">'ema_net'</span> <span class="ow">in</span> <span class="bp">self</span><span class="o">.</span><span class="n">checkpoint</span><span class="o">.</span><span class="n">keys</span><span class="p">():</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">ema_model</span><span class="o">.</span><span class="n">ema</span><span class="o">.</span><span class="n">load_state_dict</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">checkpoint</span><span class="p">[</span><span class="s1">'ema_net'</span><span class="p">])</span>
- <span class="k">else</span><span class="p">:</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">ema</span> <span class="o">=</span> <span class="kc">False</span>
- <span class="n">logger</span><span class="o">.</span><span class="n">warning</span><span class="p">(</span>
- <span class="s2">"[Warning] Checkpoint does not include EMA weights, continuing training without EMA."</span><span class="p">)</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">run_validation_freq</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">training_params</span><span class="o">.</span><span class="n">run_validation_freq</span>
- <span class="n">validation_results_tuple</span> <span class="o">=</span> <span class="p">(</span><span class="mi">0</span><span class="p">,</span> <span class="mi">0</span><span class="p">)</span>
- <span class="n">inf_time</span> <span class="o">=</span> <span class="mi">0</span>
- <span class="n">timer</span> <span class="o">=</span> <span class="n">core_utils</span><span class="o">.</span><span class="n">Timer</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">device</span><span class="p">)</span>
- <span class="c1"># IF THE LR MODE IS NOT DEFAULT TAKE IT FROM THE TRAINING PARAMS</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">lr_mode</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">training_params</span><span class="o">.</span><span class="n">lr_mode</span>
- <span class="n">load_opt_params</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">training_params</span><span class="o">.</span><span class="n">load_opt_params</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">phase_callbacks</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">training_params</span><span class="o">.</span><span class="n">phase_callbacks</span>
- <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">lr_mode</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
- <span class="n">sg_lr_callback_cls</span> <span class="o">=</span> <span class="n">LR_SCHEDULERS_CLS_DICT</span><span class="p">[</span><span class="bp">self</span><span class="o">.</span><span class="n">lr_mode</span><span class="p">]</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">phase_callbacks</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="n">sg_lr_callback_cls</span><span class="p">(</span><span class="n">train_loader_len</span><span class="o">=</span><span class="nb">len</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">train_loader</span><span class="p">),</span>
- <span class="n">net</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">net</span><span class="p">,</span>
- <span class="n">training_params</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">training_params</span><span class="p">,</span>
- <span class="n">update_param_groups</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">update_param_groups</span><span class="p">,</span>
- <span class="o">**</span><span class="bp">self</span><span class="o">.</span><span class="n">training_params</span><span class="o">.</span><span class="n">to_dict</span><span class="p">()))</span>
- <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">training_params</span><span class="o">.</span><span class="n">lr_warmup_epochs</span> <span class="o">></span> <span class="mi">0</span><span class="p">:</span>
- <span class="n">warmup_callback_cls</span> <span class="o">=</span> <span class="n">LR_WARMUP_CLS_DICT</span><span class="p">[</span><span class="bp">self</span><span class="o">.</span><span class="n">training_params</span><span class="o">.</span><span class="n">warmup_mode</span><span class="p">]</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">phase_callbacks</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="n">warmup_callback_cls</span><span class="p">(</span><span class="n">train_loader_len</span><span class="o">=</span><span class="nb">len</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">train_loader</span><span class="p">),</span>
- <span class="n">net</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">net</span><span class="p">,</span>
- <span class="n">training_params</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">training_params</span><span class="p">,</span>
- <span class="n">update_param_groups</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">update_param_groups</span><span class="p">,</span>
- <span class="o">**</span><span class="bp">self</span><span class="o">.</span><span class="n">training_params</span><span class="o">.</span><span class="n">to_dict</span><span class="p">()))</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">phase_callbacks</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="n">MetricsUpdateCallback</span><span class="p">(</span><span class="n">Phase</span><span class="o">.</span><span class="n">TRAIN_BATCH_END</span><span class="p">))</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">phase_callbacks</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="n">MetricsUpdateCallback</span><span class="p">(</span><span class="n">Phase</span><span class="o">.</span><span class="n">VALIDATION_BATCH_END</span><span class="p">))</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">phase_callback_handler</span> <span class="o">=</span> <span class="n">CallbackHandler</span><span class="p">(</span><span class="n">callbacks</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">phase_callbacks</span><span class="p">)</span>
- <span class="k">if</span> <span class="ow">not</span> <span class="bp">self</span><span class="o">.</span><span class="n">ddp_silent_mode</span><span class="p">:</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">_initialize_sg_logger_objects</span><span class="p">()</span>
- <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">training_params</span><span class="o">.</span><span class="n">dataset_statistics</span><span class="p">:</span>
- <span class="n">dataset_statistics_logger</span> <span class="o">=</span> <span class="n">DatasetStatisticsTensorboardLogger</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">sg_logger</span><span class="p">)</span>
- <span class="n">dataset_statistics_logger</span><span class="o">.</span><span class="n">analyze</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">train_loader</span><span class="p">,</span> <span class="n">dataset_params</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">dataset_params</span><span class="p">,</span>
- <span class="n">title</span><span class="o">=</span><span class="s2">"Train-set"</span><span class="p">,</span> <span class="n">anchors</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">net</span><span class="o">.</span><span class="n">module</span><span class="o">.</span><span class="n">arch_params</span><span class="o">.</span><span class="n">anchors</span><span class="p">)</span>
- <span class="n">dataset_statistics_logger</span><span class="o">.</span><span class="n">analyze</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">valid_loader</span><span class="p">,</span> <span class="n">dataset_params</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">dataset_params</span><span class="p">,</span>
- <span class="n">title</span><span class="o">=</span><span class="s2">"val-set"</span><span class="p">)</span>
- <span class="c1"># AVERAGE BEST 10 MODELS PARAMS</span>
- <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">training_params</span><span class="o">.</span><span class="n">average_best_models</span><span class="p">:</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">model_weight_averaging</span> <span class="o">=</span> <span class="n">ModelWeightAveraging</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">checkpoints_dir_path</span><span class="p">,</span>
- <span class="n">greater_is_better</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">greater_metric_to_watch_is_better</span><span class="p">,</span>
- <span class="n">source_ckpt_folder_name</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">source_ckpt_folder_name</span><span class="p">,</span>
- <span class="n">metric_to_watch</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">metric_to_watch</span><span class="p">,</span>
- <span class="n">metric_idx</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">metric_idx_in_results_tuple</span><span class="p">,</span>
- <span class="n">load_checkpoint</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">load_checkpoint</span><span class="p">,</span>
- <span class="n">model_checkpoints_location</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">model_checkpoints_location</span><span class="p">)</span>
- <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">training_params</span><span class="o">.</span><span class="n">save_full_train_log</span> <span class="ow">and</span> <span class="ow">not</span> <span class="bp">self</span><span class="o">.</span><span class="n">ddp_silent_mode</span><span class="p">:</span>
- <span class="n">logger</span> <span class="o">=</span> <span class="n">get_logger</span><span class="p">(</span><span class="vm">__name__</span><span class="p">,</span>
- <span class="n">training_log_path</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">sg_logger</span><span class="o">.</span><span class="n">log_file_path</span><span class="o">.</span><span class="n">replace</span><span class="p">(</span><span class="s1">'.txt'</span><span class="p">,</span> <span class="s1">'full_train_log.log'</span><span class="p">))</span>
- <span class="n">sg_model_utils</span><span class="o">.</span><span class="n">log_uncaught_exceptions</span><span class="p">(</span><span class="n">logger</span><span class="p">)</span>
- <span class="k">if</span> <span class="ow">not</span> <span class="bp">self</span><span class="o">.</span><span class="n">load_checkpoint</span> <span class="ow">or</span> <span class="bp">self</span><span class="o">.</span><span class="n">load_weights_only</span><span class="p">:</span>
- <span class="c1"># WHEN STARTING TRAINING FROM SCRATCH, DO NOT LOAD OPTIMIZER PARAMS (EVEN IF LOADING BACKBONE)</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">start_epoch</span> <span class="o">=</span> <span class="mi">0</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">best_metric</span> <span class="o">=</span> <span class="o">-</span><span class="mi">1</span> <span class="o">*</span> <span class="n">np</span><span class="o">.</span><span class="n">inf</span> <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">greater_metric_to_watch_is_better</span> <span class="k">else</span> <span class="n">np</span><span class="o">.</span><span class="n">inf</span>
- <span class="n">load_opt_params</span> <span class="o">=</span> <span class="kc">False</span>
- <span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">training_params</span><span class="o">.</span><span class="n">optimizer</span><span class="p">,</span> <span class="nb">str</span><span class="p">):</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">optimizer</span> <span class="o">=</span> <span class="n">build_optimizer</span><span class="p">(</span><span class="n">net</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">net</span><span class="p">,</span> <span class="n">lr</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">training_params</span><span class="o">.</span><span class="n">initial_lr</span><span class="p">,</span>
- <span class="n">training_params</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">training_params</span><span class="p">)</span>
- <span class="k">elif</span> <span class="nb">isinstance</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">training_params</span><span class="o">.</span><span class="n">optimizer</span><span class="p">,</span> <span class="n">torch</span><span class="o">.</span><span class="n">optim</span><span class="o">.</span><span class="n">Optimizer</span><span class="p">):</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">optimizer</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">training_params</span><span class="o">.</span><span class="n">optimizer</span>
- <span class="k">else</span><span class="p">:</span>
- <span class="k">raise</span> <span class="n">UnsupportedOptimizerFormat</span><span class="p">()</span>
- <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">load_checkpoint</span> <span class="ow">and</span> <span class="n">load_opt_params</span><span class="p">:</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">optimizer</span><span class="o">.</span><span class="n">load_state_dict</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">checkpoint</span><span class="p">[</span><span class="s1">'optimizer_state_dict'</span><span class="p">])</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">_initialize_mixed_precision</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">training_params</span><span class="o">.</span><span class="n">mixed_precision</span><span class="p">)</span>
- <span class="n">context</span> <span class="o">=</span> <span class="n">PhaseContext</span><span class="p">(</span><span class="n">optimizer</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">optimizer</span><span class="p">,</span> <span class="n">net</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">net</span><span class="p">,</span> <span class="n">experiment_name</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">experiment_name</span><span class="p">,</span>
- <span class="n">ckpt_dir</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">checkpoints_dir_path</span><span class="p">,</span>
- <span class="n">lr_warmup_epochs</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">training_params</span><span class="o">.</span><span class="n">lr_warmup_epochs</span><span class="p">,</span> <span class="n">sg_logger</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">sg_logger</span><span class="p">)</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">phase_callback_handler</span><span class="p">(</span><span class="n">Phase</span><span class="o">.</span><span class="n">PRE_TRAINING</span><span class="p">,</span> <span class="n">context</span><span class="p">)</span>
- <span class="k">try</span><span class="p">:</span>
- <span class="c1"># HEADERS OF THE TRAINING PROGRESS</span>
- <span class="k">if</span> <span class="ow">not</span> <span class="n">silent_mode</span><span class="p">:</span>
- <span class="n">logger</span><span class="o">.</span><span class="n">info</span><span class="p">(</span>
- <span class="sa">f</span><span class="s1">'Started training for </span><span class="si">{</span><span class="bp">self</span><span class="o">.</span><span class="n">max_epochs</span> <span class="o">-</span> <span class="bp">self</span><span class="o">.</span><span class="n">start_epoch</span><span class="si">}</span><span class="s1"> epochs (</span><span class="si">{</span><span class="bp">self</span><span class="o">.</span><span class="n">start_epoch</span><span class="si">}</span><span class="s1">/'</span><span class="sa">f</span><span class="s1">'</span><span class="si">{</span><span class="bp">self</span><span class="o">.</span><span class="n">max_epochs</span> <span class="o">-</span> <span class="mi">1</span><span class="si">}</span><span class="s1">)</span><span class="se">\n</span><span class="s1">'</span><span class="p">)</span>
- <span class="k">for</span> <span class="n">epoch</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">start_epoch</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">max_epochs</span><span class="p">):</span>
- <span class="k">if</span> <span class="n">context</span><span class="o">.</span><span class="n">stop_training</span><span class="p">:</span>
- <span class="n">logger</span><span class="o">.</span><span class="n">info</span><span class="p">(</span><span class="s2">"Request to stop training has been received, stopping training"</span><span class="p">)</span>
- <span class="k">break</span>
- <span class="c1"># Phase.TRAIN_EPOCH_START</span>
- <span class="c1"># RUN PHASE CALLBACKS</span>
- <span class="n">context</span><span class="o">.</span><span class="n">update_context</span><span class="p">(</span><span class="n">epoch</span><span class="o">=</span><span class="n">epoch</span><span class="p">)</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">phase_callback_handler</span><span class="p">(</span><span class="n">Phase</span><span class="o">.</span><span class="n">TRAIN_EPOCH_START</span><span class="p">,</span> <span class="n">context</span><span class="p">)</span>
- <span class="c1"># IN DDP- SET_EPOCH WILL CAUSE EVERY PROCESS TO BE EXPOSED TO THE ENTIRE DATASET BY SHUFFLING WITH A</span>
- <span class="c1"># DIFFERENT SEED EACH EPOCH START</span>
- <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">multi_gpu</span> <span class="o">==</span> <span class="n">MultiGPUMode</span><span class="o">.</span><span class="n">DISTRIBUTED_DATA_PARALLEL</span><span class="p">:</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">train_loader</span><span class="o">.</span><span class="n">sampler</span><span class="o">.</span><span class="n">set_epoch</span><span class="p">(</span><span class="n">epoch</span><span class="p">)</span>
- <span class="n">train_metrics_tuple</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_train_epoch</span><span class="p">(</span><span class="n">epoch</span><span class="o">=</span><span class="n">epoch</span><span class="p">,</span> <span class="n">silent_mode</span><span class="o">=</span><span class="n">silent_mode</span><span class="p">)</span>
- <span class="c1"># Phase.TRAIN_EPOCH_END</span>
- <span class="c1"># RUN PHASE CALLBACKS</span>
- <span class="n">train_metrics_dict</span> <span class="o">=</span> <span class="n">get_metrics_dict</span><span class="p">(</span><span class="n">train_metrics_tuple</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">train_metrics</span><span class="p">,</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">loss_logging_items_names</span><span class="p">)</span>
- <span class="n">context</span><span class="o">.</span><span class="n">update_context</span><span class="p">(</span><span class="n">metrics_dict</span><span class="o">=</span><span class="n">train_metrics_dict</span><span class="p">)</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">phase_callback_handler</span><span class="p">(</span><span class="n">Phase</span><span class="o">.</span><span class="n">TRAIN_EPOCH_END</span><span class="p">,</span> <span class="n">context</span><span class="p">)</span>
- <span class="c1"># CALCULATE PRECISE BATCHNORM STATS</span>
- <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">precise_bn</span><span class="p">:</span>
- <span class="n">compute_precise_bn_stats</span><span class="p">(</span><span class="n">model</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">net</span><span class="p">,</span> <span class="n">loader</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">train_loader</span><span class="p">,</span>
- <span class="n">precise_bn_batch_size</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">precise_bn_batch_size</span><span class="p">,</span>
- <span class="n">num_gpus</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">num_devices</span><span class="p">)</span>
- <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">ema</span><span class="p">:</span>
- <span class="n">compute_precise_bn_stats</span><span class="p">(</span><span class="n">model</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">ema_model</span><span class="o">.</span><span class="n">ema</span><span class="p">,</span> <span class="n">loader</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">train_loader</span><span class="p">,</span>
- <span class="n">precise_bn_batch_size</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">precise_bn_batch_size</span><span class="p">,</span>
- <span class="n">num_gpus</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">num_devices</span><span class="p">)</span>
- <span class="c1"># model switch - we replace self.net.module with the ema model for the testing and saving part</span>
- <span class="c1"># and then switch it back before the next training epoch</span>
- <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">ema</span><span class="p">:</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">ema_model</span><span class="o">.</span><span class="n">update_attr</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">net</span><span class="p">)</span>
- <span class="n">keep_model</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">net</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">net</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">ema_model</span><span class="o">.</span><span class="n">ema</span>
- <span class="c1"># RUN TEST ON VALIDATION SET EVERY self.run_validation_freq EPOCHS</span>
- <span class="k">if</span> <span class="p">(</span><span class="n">epoch</span> <span class="o">+</span> <span class="mi">1</span><span class="p">)</span> <span class="o">%</span> <span class="bp">self</span><span class="o">.</span><span class="n">run_validation_freq</span> <span class="o">==</span> <span class="mi">0</span><span class="p">:</span>
- <span class="n">timer</span><span class="o">.</span><span class="n">start</span><span class="p">()</span>
- <span class="n">validation_results_tuple</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_validate_epoch</span><span class="p">(</span><span class="n">epoch</span><span class="o">=</span><span class="n">epoch</span><span class="p">,</span> <span class="n">silent_mode</span><span class="o">=</span><span class="n">silent_mode</span><span class="p">)</span>
- <span class="n">inf_time</span> <span class="o">=</span> <span class="n">timer</span><span class="o">.</span><span class="n">stop</span><span class="p">()</span>
- <span class="c1"># Phase.VALIDATION_EPOCH_END</span>
- <span class="c1"># RUN PHASE CALLBACKS</span>
- <span class="n">valid_metrics_dict</span> <span class="o">=</span> <span class="n">get_metrics_dict</span><span class="p">(</span><span class="n">validation_results_tuple</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">valid_metrics</span><span class="p">,</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">loss_logging_items_names</span><span class="p">)</span>
- <span class="n">context</span><span class="o">.</span><span class="n">update_context</span><span class="p">(</span><span class="n">metrics_dict</span><span class="o">=</span><span class="n">valid_metrics_dict</span><span class="p">)</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">phase_callback_handler</span><span class="p">(</span><span class="n">Phase</span><span class="o">.</span><span class="n">VALIDATION_EPOCH_END</span><span class="p">,</span> <span class="n">context</span><span class="p">)</span>
- <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">ema</span><span class="p">:</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">net</span> <span class="o">=</span> <span class="n">keep_model</span>
- <span class="k">if</span> <span class="ow">not</span> <span class="bp">self</span><span class="o">.</span><span class="n">ddp_silent_mode</span><span class="p">:</span>
- <span class="c1"># SAVING AND LOGGING OCCURS ONLY IN THE MAIN PROCESS (IN CASES THERE ARE SEVERAL PROCESSES - DDP)</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">_write_to_disk_operations</span><span class="p">(</span><span class="n">train_metrics_tuple</span><span class="p">,</span> <span class="n">validation_results_tuple</span><span class="p">,</span> <span class="n">inf_time</span><span class="p">,</span> <span class="n">epoch</span><span class="p">,</span> <span class="n">context</span><span class="p">)</span>
- <span class="c1"># Evaluating the average model and removing snapshot averaging file if training is completed</span>
- <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">training_params</span><span class="o">.</span><span class="n">average_best_models</span><span class="p">:</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">_validate_final_average_model</span><span class="p">(</span><span class="n">cleanup_snapshots_pkl_file</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span>
- <span class="k">except</span> <span class="ne">KeyboardInterrupt</span><span class="p">:</span>
- <span class="n">logger</span><span class="o">.</span><span class="n">info</span><span class="p">(</span>
- <span class="s1">'</span><span class="se">\n</span><span class="s1">[MODEL TRAINING EXECUTION HAS BEEN INTERRUPTED]... Please wait until SOFT-TERMINATION process '</span>
- <span class="s1">'finishes and saves all of the Model Checkpoints and log files before terminating...'</span><span class="p">)</span>
- <span class="n">logger</span><span class="o">.</span><span class="n">info</span><span class="p">(</span><span class="s1">'For HARD Termination - Stop the process again'</span><span class="p">)</span>
- <span class="k">finally</span><span class="p">:</span>
- <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">multi_gpu</span> <span class="o">==</span> <span class="n">MultiGPUMode</span><span class="o">.</span><span class="n">DISTRIBUTED_DATA_PARALLEL</span><span class="p">:</span>
- <span class="c1"># CLEAN UP THE MULTI-GPU PROCESS GROUP WHEN DONE</span>
- <span class="k">if</span> <span class="n">torch</span><span class="o">.</span><span class="n">distributed</span><span class="o">.</span><span class="n">is_initialized</span><span class="p">():</span>
- <span class="n">torch</span><span class="o">.</span><span class="n">distributed</span><span class="o">.</span><span class="n">destroy_process_group</span><span class="p">()</span>
- <span class="c1"># PHASE.TRAIN_END</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">phase_callback_handler</span><span class="p">(</span><span class="n">Phase</span><span class="o">.</span><span class="n">POST_TRAINING</span><span class="p">,</span> <span class="n">context</span><span class="p">)</span>
- <span class="k">if</span> <span class="ow">not</span> <span class="bp">self</span><span class="o">.</span><span class="n">ddp_silent_mode</span><span class="p">:</span>
- <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">model_checkpoints_location</span> <span class="o">!=</span> <span class="s1">'local'</span><span class="p">:</span>
- <span class="n">logger</span><span class="o">.</span><span class="n">info</span><span class="p">(</span><span class="s1">'[CLEANUP] - Saving Checkpoint files'</span><span class="p">)</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">sg_logger</span><span class="o">.</span><span class="n">upload</span><span class="p">()</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">sg_logger</span><span class="o">.</span><span class="n">close</span><span class="p">()</span></div>
- <span class="nd">@resolve_param</span><span class="p">(</span><span class="s1">'train_metrics_list'</span><span class="p">,</span> <span class="n">ListFactory</span><span class="p">(</span><span class="n">MetricsFactory</span><span class="p">()))</span>
- <span class="k">def</span> <span class="nf">_set_train_metrics</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">train_metrics_list</span><span class="p">):</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">train_metrics</span> <span class="o">=</span> <span class="n">MetricCollection</span><span class="p">(</span><span class="n">train_metrics_list</span><span class="p">)</span>
- <span class="nd">@resolve_param</span><span class="p">(</span><span class="s1">'valid_metrics_list'</span><span class="p">,</span> <span class="n">ListFactory</span><span class="p">(</span><span class="n">MetricsFactory</span><span class="p">()))</span>
- <span class="k">def</span> <span class="nf">_set_valid_metrics</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">valid_metrics_list</span><span class="p">):</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">valid_metrics</span> <span class="o">=</span> <span class="n">MetricCollection</span><span class="p">(</span><span class="n">valid_metrics_list</span><span class="p">)</span>
- <span class="k">def</span> <span class="nf">_initialize_mixed_precision</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">mixed_precision_enabled</span><span class="p">:</span> <span class="nb">bool</span><span class="p">):</span>
- <span class="c1"># SCALER IS ALWAYS INITIALIZED BUT IS DISABLED IF MIXED PRECISION WAS NOT SET</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">scaler</span> <span class="o">=</span> <span class="n">GradScaler</span><span class="p">(</span><span class="n">enabled</span><span class="o">=</span><span class="n">mixed_precision_enabled</span><span class="p">)</span>
- <span class="k">if</span> <span class="n">mixed_precision_enabled</span><span class="p">:</span>
- <span class="k">assert</span> <span class="bp">self</span><span class="o">.</span><span class="n">device</span><span class="o">.</span><span class="n">startswith</span><span class="p">(</span><span class="s1">'cuda'</span><span class="p">),</span> <span class="s2">"mixed precision is not available for CPU"</span>
- <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">multi_gpu</span> <span class="o">==</span> <span class="n">MultiGPUMode</span><span class="o">.</span><span class="n">DATA_PARALLEL</span><span class="p">:</span>
- <span class="c1"># IN DATAPARALLEL MODE WE NEED TO WRAP THE FORWARD FUNCTION OF OUR MODEL SO IT WILL RUN WITH AUTOCAST.</span>
- <span class="c1"># BUT SINCE THE MODULE IS CLONED TO THE DEVICES ON EACH FORWARD CALL OF A DATAPARALLEL MODEL,</span>
- <span class="c1"># WE HAVE TO REGISTER THE WRAPPER BEFORE EVERY FORWARD CALL</span>
- <span class="k">def</span> <span class="nf">hook</span><span class="p">(</span><span class="n">module</span><span class="p">,</span> <span class="n">_</span><span class="p">):</span>
- <span class="n">module</span><span class="o">.</span><span class="n">forward</span> <span class="o">=</span> <span class="n">MultiGPUModeAutocastWrapper</span><span class="p">(</span><span class="n">module</span><span class="o">.</span><span class="n">forward</span><span class="p">)</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">net</span><span class="o">.</span><span class="n">module</span><span class="o">.</span><span class="n">register_forward_pre_hook</span><span class="p">(</span><span class="n">hook</span><span class="o">=</span><span class="n">hook</span><span class="p">)</span>
- <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">load_checkpoint</span><span class="p">:</span>
- <span class="n">scaler_state_dict</span> <span class="o">=</span> <span class="n">core_utils</span><span class="o">.</span><span class="n">get_param</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">checkpoint</span><span class="p">,</span> <span class="s1">'scaler_state_dict'</span><span class="p">)</span>
- <span class="k">if</span> <span class="n">scaler_state_dict</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
- <span class="n">logger</span><span class="o">.</span><span class="n">warning</span><span class="p">(</span>
- <span class="s1">'Mixed Precision - scaler state_dict not found in loaded model. This may case issues '</span>
- <span class="s1">'with loss scaling'</span><span class="p">)</span>
- <span class="k">else</span><span class="p">:</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">scaler</span><span class="o">.</span><span class="n">load_state_dict</span><span class="p">(</span><span class="n">scaler_state_dict</span><span class="p">)</span>
- <span class="k">def</span> <span class="nf">_validate_final_average_model</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">cleanup_snapshots_pkl_file</span><span class="o">=</span><span class="kc">False</span><span class="p">):</span>
- <span class="sd">"""</span>
- <span class="sd"> Testing the averaged model by loading the last saved average checkpoint and running test.</span>
- <span class="sd"> Will be loaded to each of DDP processes</span>
- <span class="sd"> :param cleanup_pkl_file: a flag for deleting the 10 best snapshots dictionary</span>
- <span class="sd"> """</span>
- <span class="n">logger</span><span class="o">.</span><span class="n">info</span><span class="p">(</span><span class="s1">'RUNNING ADDITIONAL TEST ON THE AVERAGED MODEL...'</span><span class="p">)</span>
- <span class="n">keep_state_dict</span> <span class="o">=</span> <span class="n">deepcopy</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">net</span><span class="o">.</span><span class="n">state_dict</span><span class="p">())</span>
- <span class="c1"># SETTING STATE DICT TO THE AVERAGE MODEL FOR EVALUATION</span>
- <span class="n">average_model_ckpt_path</span> <span class="o">=</span> <span class="n">os</span><span class="o">.</span><span class="n">path</span><span class="o">.</span><span class="n">join</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">checkpoints_dir_path</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">average_model_checkpoint_filename</span><span class="p">)</span>
- <span class="n">average_model_sd</span> <span class="o">=</span> <span class="n">read_ckpt_state_dict</span><span class="p">(</span><span class="n">average_model_ckpt_path</span><span class="p">)[</span><span class="s1">'net'</span><span class="p">]</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">net</span><span class="o">.</span><span class="n">load_state_dict</span><span class="p">(</span><span class="n">average_model_sd</span><span class="p">)</span>
- <span class="c1"># testing the averaged model and save instead of best model if needed</span>
- <span class="n">averaged_model_results_tuple</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_validate_epoch</span><span class="p">(</span><span class="n">epoch</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">max_epochs</span><span class="p">)</span>
- <span class="c1"># Reverting the current model</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">net</span><span class="o">.</span><span class="n">load_state_dict</span><span class="p">(</span><span class="n">keep_state_dict</span><span class="p">)</span>
- <span class="k">if</span> <span class="ow">not</span> <span class="bp">self</span><span class="o">.</span><span class="n">ddp_silent_mode</span><span class="p">:</span>
- <span class="c1"># Adding values to sg_logger</span>
- <span class="c1"># looping over last titles which corresponds to validation (and average model) metrics.</span>
- <span class="n">all_titles</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">results_titles</span><span class="p">[</span><span class="o">-</span><span class="mi">1</span> <span class="o">*</span> <span class="nb">len</span><span class="p">(</span><span class="n">averaged_model_results_tuple</span><span class="p">):]</span>
- <span class="n">result_dict</span> <span class="o">=</span> <span class="p">{</span><span class="n">all_titles</span><span class="p">[</span><span class="n">i</span><span class="p">]:</span> <span class="n">averaged_model_results_tuple</span><span class="p">[</span><span class="n">i</span><span class="p">]</span> <span class="k">for</span> <span class="n">i</span> <span class="ow">in</span>
- <span class="nb">range</span><span class="p">(</span><span class="nb">len</span><span class="p">(</span><span class="n">averaged_model_results_tuple</span><span class="p">))}</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">sg_logger</span><span class="o">.</span><span class="n">add_scalars</span><span class="p">(</span><span class="n">tag_scalar_dict</span><span class="o">=</span><span class="n">result_dict</span><span class="p">,</span> <span class="n">global_step</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">max_epochs</span><span class="p">)</span>
- <span class="n">average_model_tb_titles</span> <span class="o">=</span> <span class="p">[</span><span class="s1">'Averaged Model '</span> <span class="o">+</span> <span class="n">x</span> <span class="k">for</span> <span class="n">x</span> <span class="ow">in</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">results_titles</span><span class="p">[</span><span class="o">-</span><span class="mi">1</span> <span class="o">*</span> <span class="nb">len</span><span class="p">(</span><span class="n">averaged_model_results_tuple</span><span class="p">):]]</span>
- <span class="n">write_struct</span> <span class="o">=</span> <span class="s1">''</span>
- <span class="k">for</span> <span class="n">ind</span><span class="p">,</span> <span class="n">title</span> <span class="ow">in</span> <span class="nb">enumerate</span><span class="p">(</span><span class="n">average_model_tb_titles</span><span class="p">):</span>
- <span class="n">write_struct</span> <span class="o">+=</span> <span class="s1">'</span><span class="si">%s</span><span class="s1">: </span><span class="si">%.3f</span><span class="s1"> </span><span class="se">\n</span><span class="s1"> '</span> <span class="o">%</span> <span class="p">(</span><span class="n">title</span><span class="p">,</span> <span class="n">averaged_model_results_tuple</span><span class="p">[</span><span class="n">ind</span><span class="p">])</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">sg_logger</span><span class="o">.</span><span class="n">add_scalar</span><span class="p">(</span><span class="n">title</span><span class="p">,</span> <span class="n">averaged_model_results_tuple</span><span class="p">[</span><span class="n">ind</span><span class="p">],</span> <span class="n">global_step</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">max_epochs</span><span class="p">)</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">sg_logger</span><span class="o">.</span><span class="n">add_text</span><span class="p">(</span><span class="s2">"Averaged_Model_Performance"</span><span class="p">,</span> <span class="n">write_struct</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">max_epochs</span><span class="p">)</span>
- <span class="k">if</span> <span class="n">cleanup_snapshots_pkl_file</span><span class="p">:</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">model_weight_averaging</span><span class="o">.</span><span class="n">cleanup</span><span class="p">()</span>
- <span class="c1"># FIXME - we need to resolve flake8's 'function is too complex' for this function</span>
- <div class="viewcode-block" id="SgModel.predict"><a class="viewcode-back" href="../../../../super_gradients.training.sg_model.html#super_gradients.training.SgModel.predict">[docs]</a> <span class="nd">@deprecated</span><span class="p">(</span><span class="n">version</span><span class="o">=</span><span class="s1">'0.1'</span><span class="p">,</span> <span class="n">reason</span><span class="o">=</span><span class="s2">"directly predict using the nn_module"</span><span class="p">)</span> <span class="c1"># noqa: C901</span>
- <span class="k">def</span> <span class="nf">predict</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">inputs</span><span class="p">,</span> <span class="n">targets</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span> <span class="n">half</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span> <span class="n">normalize</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span> <span class="n">verbose</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
- <span class="n">move_outputs_to_cpu</span><span class="o">=</span><span class="kc">True</span><span class="p">):</span>
- <span class="sd">"""</span>
- <span class="sd"> A fast predictor for a batch of inputs</span>
- <span class="sd"> :param inputs: torch.tensor or numpy.array</span>
- <span class="sd"> a batch of inputs</span>
- <span class="sd"> :param targets: torch.tensor()</span>
- <span class="sd"> corresponding labels - if non are given - accuracy will not be computed</span>
- <span class="sd"> :param verbose: bool</span>
- <span class="sd"> print the results to screen</span>
- <span class="sd"> :param normalize: bool</span>
- <span class="sd"> If true, normalizes the tensor according to the dataloader's normalization values</span>
- <span class="sd"> :param half:</span>
- <span class="sd"> Performs half precision evaluation</span>
- <span class="sd"> :param move_outputs_to_cpu:</span>
- <span class="sd"> Moves the results from the GPU to the CPU</span>
- <span class="sd"> :return: outputs, acc, net_time, gross_time</span>
- <span class="sd"> networks predictions, accuracy calculation, forward pass net time, function gross time</span>
- <span class="sd"> """</span>
- <span class="n">transform_list</span> <span class="o">=</span> <span class="p">[]</span>
- <span class="c1"># Create a 'to_tensor' transformation and a place holder of input_t</span>
- <span class="k">if</span> <span class="nb">type</span><span class="p">(</span><span class="n">inputs</span><span class="p">)</span> <span class="o">==</span> <span class="n">torch</span><span class="o">.</span><span class="n">Tensor</span><span class="p">:</span>
- <span class="n">inputs_t</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">zeros_like</span><span class="p">(</span><span class="n">inputs</span><span class="p">)</span>
- <span class="k">else</span><span class="p">:</span>
- <span class="n">transform_list</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="n">transforms</span><span class="o">.</span><span class="n">ToTensor</span><span class="p">())</span>
- <span class="n">inputs_t</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">zeros</span><span class="p">(</span><span class="n">size</span><span class="o">=</span><span class="p">(</span><span class="n">inputs</span><span class="o">.</span><span class="n">shape</span><span class="p">[</span><span class="mi">0</span><span class="p">],</span> <span class="n">inputs</span><span class="o">.</span><span class="n">shape</span><span class="p">[</span><span class="mi">3</span><span class="p">],</span> <span class="n">inputs</span><span class="o">.</span><span class="n">shape</span><span class="p">[</span><span class="mi">1</span><span class="p">],</span> <span class="n">inputs</span><span class="o">.</span><span class="n">shape</span><span class="p">[</span><span class="mi">2</span><span class="p">]))</span>
- <span class="c1"># Create a normalization transformation</span>
- <span class="k">if</span> <span class="n">normalize</span><span class="p">:</span>
- <span class="k">try</span><span class="p">:</span>
- <span class="n">mean</span><span class="p">,</span> <span class="n">std</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">dataset_interface</span><span class="o">.</span><span class="n">lib_dataset_params</span><span class="p">[</span><span class="s1">'mean'</span><span class="p">],</span> <span class="bp">self</span><span class="o">.</span><span class="n">dataset_interface</span><span class="o">.</span><span class="n">lib_dataset_params</span><span class="p">[</span><span class="s1">'std'</span><span class="p">]</span>
- <span class="k">except</span> <span class="ne">AttributeError</span><span class="p">:</span>
- <span class="k">raise</span> <span class="ne">AttributeError</span><span class="p">(</span><span class="s1">'In </span><span class="se">\'</span><span class="s1">predict()</span><span class="se">\'</span><span class="s1">, Normalization is set to True while the dataset has no default '</span>
- <span class="s1">'mean & std => deactivate normalization or inject it to the datasets library.'</span><span class="p">)</span>
- <span class="n">transform_list</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="n">transforms</span><span class="o">.</span><span class="n">Normalize</span><span class="p">(</span><span class="n">mean</span><span class="p">,</span> <span class="n">std</span><span class="p">))</span>
- <span class="c1"># Compose all transformations into one</span>
- <span class="n">transformation</span> <span class="o">=</span> <span class="n">transforms</span><span class="o">.</span><span class="n">Compose</span><span class="p">(</span><span class="n">transform_list</span><span class="p">)</span>
- <span class="c1"># Transform the input</span>
- <span class="k">for</span> <span class="n">idx</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="nb">len</span><span class="p">(</span><span class="n">inputs_t</span><span class="p">)):</span>
- <span class="n">inputs_t</span><span class="p">[</span><span class="n">idx</span><span class="p">]</span> <span class="o">=</span> <span class="n">transformation</span><span class="p">(</span><span class="n">inputs</span><span class="p">[</span><span class="n">idx</span><span class="p">])</span>
- <span class="c1"># Timer instances</span>
- <span class="n">gross_timer</span> <span class="o">=</span> <span class="n">core_utils</span><span class="o">.</span><span class="n">Timer</span><span class="p">(</span><span class="s1">'cpu'</span><span class="p">)</span>
- <span class="n">gross_timer</span><span class="o">.</span><span class="n">start</span><span class="p">()</span>
- <span class="n">net_timer</span> <span class="o">=</span> <span class="n">core_utils</span><span class="o">.</span><span class="n">Timer</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">device</span><span class="p">)</span>
- <span class="c1"># Set network in eval mode</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">net</span><span class="o">.</span><span class="n">eval</span><span class="p">()</span>
- <span class="c1"># Half is not supported on CPU</span>
- <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">device</span> <span class="o">!=</span> <span class="s1">'cuda'</span> <span class="ow">and</span> <span class="n">half</span><span class="p">:</span>
- <span class="n">half</span> <span class="o">=</span> <span class="kc">False</span>
- <span class="n">logger</span><span class="o">.</span><span class="n">warning</span><span class="p">(</span><span class="s1">'NOTICE: half is set to True but is not supported on CPU ==> using full precision'</span><span class="p">)</span>
- <span class="c1"># Apply half precision to network and input</span>
- <span class="k">if</span> <span class="n">half</span><span class="p">:</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">net</span><span class="o">.</span><span class="n">half</span><span class="p">()</span>
- <span class="n">inputs_t</span> <span class="o">=</span> <span class="n">inputs_t</span><span class="o">.</span><span class="n">half</span><span class="p">()</span>
- <span class="k">with</span> <span class="n">torch</span><span class="o">.</span><span class="n">no_grad</span><span class="p">():</span>
- <span class="c1"># Move input to compute device</span>
- <span class="n">inputs_t</span> <span class="o">=</span> <span class="n">inputs_t</span><span class="o">.</span><span class="n">to</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">device</span><span class="p">)</span>
- <span class="c1"># Forward pass (timed...)</span>
- <span class="n">net_timer</span><span class="o">.</span><span class="n">start</span><span class="p">()</span>
- <span class="n">outputs</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">net</span><span class="p">(</span><span class="n">inputs_t</span><span class="p">)</span>
- <span class="n">net_time</span> <span class="o">=</span> <span class="n">net_timer</span><span class="o">.</span><span class="n">stop</span><span class="p">()</span>
- <span class="k">if</span> <span class="n">move_outputs_to_cpu</span><span class="p">:</span>
- <span class="n">outputs</span> <span class="o">=</span> <span class="n">outputs</span><span class="o">.</span><span class="n">cpu</span><span class="p">()</span>
- <span class="n">gross_time</span> <span class="o">=</span> <span class="n">gross_timer</span><span class="o">.</span><span class="n">stop</span><span class="p">()</span>
- <span class="c1"># Convert targets to tensor</span>
- <span class="n">targets</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">tensor</span><span class="p">(</span><span class="n">targets</span><span class="p">)</span> <span class="k">if</span> <span class="p">(</span><span class="nb">type</span><span class="p">(</span><span class="n">targets</span><span class="p">)</span> <span class="o">!=</span> <span class="n">torch</span><span class="o">.</span><span class="n">Tensor</span> <span class="ow">and</span> <span class="n">targets</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">)</span> <span class="k">else</span> <span class="n">targets</span>
- <span class="c1"># Compute accuracy</span>
- <span class="n">acc</span> <span class="o">=</span> <span class="n">metrics</span><span class="o">.</span><span class="n">accuracy</span><span class="p">(</span><span class="n">outputs</span><span class="o">.</span><span class="n">float</span><span class="p">(),</span> <span class="n">targets</span><span class="o">.</span><span class="n">cpu</span><span class="p">())[</span><span class="mi">0</span><span class="p">]</span> <span class="k">if</span> <span class="n">targets</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span> <span class="k">else</span> <span class="kc">None</span>
- <span class="n">acc_str</span> <span class="o">=</span> <span class="s1">'</span><span class="si">%.2f</span><span class="s1">'</span> <span class="o">%</span> <span class="n">acc</span> <span class="k">if</span> <span class="n">targets</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span> <span class="k">else</span> <span class="s1">'N/A'</span>
- <span class="k">if</span> <span class="n">verbose</span><span class="p">:</span>
- <span class="n">logger</span><span class="o">.</span><span class="n">info</span><span class="p">(</span><span class="s1">'</span><span class="si">%s</span><span class="se">\n</span><span class="s1">Predicted </span><span class="si">%d</span><span class="s1"> examples: </span><span class="se">\n\t</span><span class="si">%.2f</span><span class="s1"> ms (gross) --> </span><span class="si">%.2f</span><span class="s1"> ms (net)</span><span class="se">\n\t</span><span class="s1">With accuracy </span><span class="si">%s</span><span class="se">\n</span><span class="si">%s</span><span class="s1">'</span> <span class="o">%</span>
- <span class="p">(</span><span class="s1">'-'</span> <span class="o">*</span> <span class="mi">50</span><span class="p">,</span> <span class="n">inputs_t</span><span class="o">.</span><span class="n">shape</span><span class="p">[</span><span class="mi">0</span><span class="p">],</span> <span class="n">gross_time</span><span class="p">,</span> <span class="n">net_time</span><span class="p">,</span> <span class="n">acc_str</span><span class="p">,</span> <span class="s1">'-'</span> <span class="o">*</span> <span class="mi">50</span><span class="p">))</span>
- <span class="c1"># Undo the half precision</span>
- <span class="k">if</span> <span class="n">half</span> <span class="ow">and</span> <span class="ow">not</span> <span class="bp">self</span><span class="o">.</span><span class="n">half_precision</span><span class="p">:</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">net</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">net</span><span class="o">.</span><span class="n">float</span><span class="p">()</span>
- <span class="k">return</span> <span class="n">outputs</span><span class="p">,</span> <span class="n">acc</span><span class="p">,</span> <span class="n">net_time</span><span class="p">,</span> <span class="n">gross_time</span></div>
- <div class="viewcode-block" id="SgModel.compute_model_runtime"><a class="viewcode-back" href="../../../../super_gradients.training.sg_model.html#super_gradients.training.SgModel.compute_model_runtime">[docs]</a> <span class="k">def</span> <span class="nf">compute_model_runtime</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">input_dims</span><span class="p">:</span> <span class="nb">tuple</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
- <span class="n">batch_sizes</span><span class="p">:</span> <span class="n">Union</span><span class="p">[</span><span class="nb">tuple</span><span class="p">,</span> <span class="nb">list</span><span class="p">,</span> <span class="nb">int</span><span class="p">]</span> <span class="o">=</span> <span class="p">(</span><span class="mi">1</span><span class="p">,</span> <span class="mi">8</span><span class="p">,</span> <span class="mi">16</span><span class="p">,</span> <span class="mi">32</span><span class="p">,</span> <span class="mi">64</span><span class="p">),</span>
- <span class="n">verbose</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="kc">True</span><span class="p">):</span>
- <span class="sd">"""</span>
- <span class="sd"> Compute the "atomic" inference time and throughput.</span>
- <span class="sd"> Atomic refers to calculating the forward pass independently, discarding effects such as data augmentation,</span>
- <span class="sd"> data upload to device, multi-gpu distribution etc.</span>
- <span class="sd"> :param input_dims: tuple</span>
- <span class="sd"> shape of a basic input to the network (without the first index) e.g. (3, 224, 224)</span>
- <span class="sd"> if None uses an input from the test loader</span>
- <span class="sd"> :param batch_sizes: int or list</span>
- <span class="sd"> Batch sizes for latency calculation</span>
- <span class="sd"> :param verbose: bool</span>
- <span class="sd"> Prints results to screen</span>
- <span class="sd"> :return: log: dict</span>
- <span class="sd"> Latency and throughput for each tested batch size</span>
- <span class="sd"> """</span>
- <span class="k">assert</span> <span class="n">input_dims</span> <span class="ow">or</span> <span class="bp">self</span><span class="o">.</span><span class="n">test_loader</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">,</span> <span class="s1">'Must get </span><span class="se">\'</span><span class="s1">input_dims</span><span class="se">\'</span><span class="s1"> or connect a dataset interface'</span>
- <span class="k">assert</span> <span class="bp">self</span><span class="o">.</span><span class="n">multi_gpu</span> <span class="ow">not</span> <span class="ow">in</span> <span class="p">(</span><span class="n">MultiGPUMode</span><span class="o">.</span><span class="n">DATA_PARALLEL</span><span class="p">,</span> <span class="n">MultiGPUMode</span><span class="o">.</span><span class="n">DISTRIBUTED_DATA_PARALLEL</span><span class="p">),</span> \
- <span class="s1">'The model is on multiple GPUs, move it to a single GPU is order to compute runtime'</span>
- <span class="c1"># TRANSFER THE MODEL TO EVALUATION MODE BUT REMEMBER THE MODE TO RETURN TO</span>
- <span class="n">was_in_training_mode</span> <span class="o">=</span> <span class="kc">True</span> <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">net</span><span class="o">.</span><span class="n">training</span> <span class="k">else</span> <span class="kc">False</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">net</span><span class="o">.</span><span class="n">eval</span><span class="p">()</span>
- <span class="c1"># INITIALIZE LOGS AND PRINTS</span>
- <span class="n">timer</span> <span class="o">=</span> <span class="n">core_utils</span><span class="o">.</span><span class="n">Timer</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">device</span><span class="p">)</span>
- <span class="n">logs</span> <span class="o">=</span> <span class="p">{}</span>
- <span class="n">log_print</span> <span class="o">=</span> <span class="sa">f</span><span class="s2">"</span><span class="si">{</span><span class="s1">'-'</span> <span class="o">*</span> <span class="mi">35</span><span class="si">}</span><span class="se">\n</span><span class="s2">"</span> \
- <span class="sa">f</span><span class="s2">"Batch Time per Batch Throughput</span><span class="se">\n</span><span class="s2">"</span> \
- <span class="sa">f</span><span class="s2">"size (ms) (im/s)</span><span class="se">\n</span><span class="s2">"</span> \
- <span class="sa">f</span><span class="s2">"</span><span class="si">{</span><span class="s1">'-'</span> <span class="o">*</span> <span class="mi">35</span><span class="si">}</span><span class="se">\n</span><span class="s2">"</span>
- <span class="c1"># GET THE INPUT SHAPE FROM THE DATA LOADER IF NOT PROVIDED EXPLICITLY</span>
- <span class="n">input_dims</span> <span class="o">=</span> <span class="n">input_dims</span> <span class="ow">or</span> <span class="nb">next</span><span class="p">(</span><span class="nb">iter</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">test_loader</span><span class="p">))[</span><span class="mi">0</span><span class="p">]</span><span class="o">.</span><span class="n">shape</span><span class="p">[</span><span class="mi">1</span><span class="p">:]</span>
- <span class="c1"># DEFINE NUMBER ACCORDING TO DEVICE</span>
- <span class="n">repetitions</span> <span class="o">=</span> <span class="mi">200</span> <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">device</span> <span class="o">==</span> <span class="s1">'cuda'</span> <span class="k">else</span> <span class="mi">20</span>
- <span class="c1"># CREATE A LIST OF BATCH SIZES</span>
- <span class="n">batch_sizes</span> <span class="o">=</span> <span class="p">[</span><span class="n">batch_sizes</span><span class="p">]</span> <span class="k">if</span> <span class="nb">type</span><span class="p">(</span><span class="n">batch_sizes</span><span class="p">)</span> <span class="o">==</span> <span class="nb">int</span> <span class="k">else</span> <span class="n">batch_sizes</span>
- <span class="k">for</span> <span class="n">batch_size</span> <span class="ow">in</span> <span class="nb">sorted</span><span class="p">(</span><span class="n">batch_sizes</span><span class="p">):</span>
- <span class="k">try</span><span class="p">:</span>
- <span class="c1"># CREATE A RANDOM TENSOR AS INPUT</span>
- <span class="n">dummy_batch</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">randn</span><span class="p">((</span><span class="n">batch_size</span><span class="p">,</span> <span class="o">*</span><span class="n">input_dims</span><span class="p">),</span> <span class="n">device</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">device</span><span class="p">)</span>
- <span class="c1"># WARM UP</span>
- <span class="k">for</span> <span class="n">_</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="mi">10</span><span class="p">):</span>
- <span class="n">_</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">net</span><span class="p">(</span><span class="n">dummy_batch</span><span class="p">)</span>
- <span class="c1"># RUN & TIME</span>
- <span class="n">accumulated_time</span> <span class="o">=</span> <span class="mi">0</span>
- <span class="k">with</span> <span class="n">torch</span><span class="o">.</span><span class="n">no_grad</span><span class="p">():</span>
- <span class="k">for</span> <span class="n">_</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="n">repetitions</span><span class="p">):</span>
- <span class="n">timer</span><span class="o">.</span><span class="n">start</span><span class="p">()</span>
- <span class="n">_</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">net</span><span class="p">(</span><span class="n">dummy_batch</span><span class="p">)</span>
- <span class="n">accumulated_time</span> <span class="o">+=</span> <span class="n">timer</span><span class="o">.</span><span class="n">stop</span><span class="p">()</span>
- <span class="c1"># PERFORMANCE CALCULATION</span>
- <span class="n">time_per_batch</span> <span class="o">=</span> <span class="n">accumulated_time</span> <span class="o">/</span> <span class="n">repetitions</span>
- <span class="n">throughput</span> <span class="o">=</span> <span class="n">batch_size</span> <span class="o">*</span> <span class="mi">1000</span> <span class="o">/</span> <span class="n">time_per_batch</span>
- <span class="n">logs</span><span class="p">[</span><span class="n">batch_size</span><span class="p">]</span> <span class="o">=</span> <span class="p">{</span><span class="s1">'time_per_batch'</span><span class="p">:</span> <span class="n">time_per_batch</span><span class="p">,</span> <span class="s1">'throughput'</span><span class="p">:</span> <span class="n">throughput</span><span class="p">}</span>
- <span class="n">log_print</span> <span class="o">+=</span> <span class="sa">f</span><span class="s2">"</span><span class="si">{</span><span class="n">batch_size</span><span class="si">:</span><span class="s2">4.0f</span><span class="si">}</span><span class="s2"> </span><span class="si">{</span><span class="n">time_per_batch</span><span class="si">:</span><span class="s2">12.1f</span><span class="si">}</span><span class="s2"> </span><span class="si">{</span><span class="n">throughput</span><span class="si">:</span><span class="s2">12.0f</span><span class="si">}</span><span class="se">\n</span><span class="s2">"</span>
- <span class="k">except</span> <span class="ne">RuntimeError</span> <span class="k">as</span> <span class="n">e</span><span class="p">:</span>
- <span class="c1"># ONLY FOR THE CASE OF CUDA OUT OF MEMORY WE CATCH THE EXCEPTION AND CONTINUE THE FUNCTION</span>
- <span class="k">if</span> <span class="s1">'CUDA out of memory'</span> <span class="ow">in</span> <span class="nb">str</span><span class="p">(</span><span class="n">e</span><span class="p">):</span>
- <span class="n">log_print</span> <span class="o">+=</span> <span class="sa">f</span><span class="s2">"</span><span class="si">{</span><span class="n">batch_size</span><span class="si">:</span><span class="s2">4d</span><span class="si">}</span><span class="se">\t</span><span class="si">{</span><span class="s1">'CUDA out of memory'</span><span class="si">:</span><span class="s2">13s</span><span class="si">}</span><span class="se">\n</span><span class="s2">"</span>
- <span class="k">else</span><span class="p">:</span>
- <span class="k">raise</span>
- <span class="c1"># PRINT RESULTS</span>
- <span class="k">if</span> <span class="n">verbose</span><span class="p">:</span>
- <span class="n">logger</span><span class="o">.</span><span class="n">info</span><span class="p">(</span><span class="n">log_print</span><span class="p">)</span>
- <span class="c1"># RETURN THE MODEL TO THE PREVIOUS MODE</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">net</span><span class="o">.</span><span class="n">train</span><span class="p">(</span><span class="n">was_in_training_mode</span><span class="p">)</span>
- <span class="k">return</span> <span class="n">logs</span></div>
- <div class="viewcode-block" id="SgModel.get_arch_params"><a class="viewcode-back" href="../../../../super_gradients.training.sg_model.html#super_gradients.training.SgModel.get_arch_params">[docs]</a> <span class="k">def</span> <span class="nf">get_arch_params</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
- <span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">arch_params</span><span class="o">.</span><span class="n">to_dict</span><span class="p">()</span></div>
- <div class="viewcode-block" id="SgModel.get_structure"><a class="viewcode-back" href="../../../../super_gradients.training.sg_model.html#super_gradients.training.SgModel.get_structure">[docs]</a> <span class="k">def</span> <span class="nf">get_structure</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
- <span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">net</span><span class="o">.</span><span class="n">module</span><span class="o">.</span><span class="n">structure</span></div>
- <div class="viewcode-block" id="SgModel.get_architecture"><a class="viewcode-back" href="../../../../super_gradients.training.sg_model.html#super_gradients.training.SgModel.get_architecture">[docs]</a> <span class="k">def</span> <span class="nf">get_architecture</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
- <span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">architecture</span></div>
- <div class="viewcode-block" id="SgModel.set_experiment_name"><a class="viewcode-back" href="../../../../super_gradients.training.sg_model.html#super_gradients.training.SgModel.set_experiment_name">[docs]</a> <span class="k">def</span> <span class="nf">set_experiment_name</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">experiment_name</span><span class="p">):</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">experiment_name</span> <span class="o">=</span> <span class="n">experiment_name</span></div>
- <div class="viewcode-block" id="SgModel.re_build_model"><a class="viewcode-back" href="../../../../super_gradients.training.sg_model.html#super_gradients.training.SgModel.re_build_model">[docs]</a> <span class="k">def</span> <span class="nf">re_build_model</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">arch_params</span><span class="o">=</span><span class="p">{}):</span>
- <span class="sd">"""</span>
- <span class="sd"> arch_params : dict</span>
- <span class="sd"> Architecture H.P. e.g.: block, num_blocks, num_classes, etc.</span>
- <span class="sd"> :return:</span>
- <span class="sd"> """</span>
- <span class="k">if</span> <span class="s1">'num_classes'</span> <span class="ow">not</span> <span class="ow">in</span> <span class="n">arch_params</span><span class="o">.</span><span class="n">keys</span><span class="p">():</span>
- <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">dataset_interface</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
- <span class="k">raise</span> <span class="ne">Exception</span><span class="p">(</span><span class="s1">'Error'</span><span class="p">,</span> <span class="s1">'Number of classes not defined in arch params and dataset is not defined'</span><span class="p">)</span>
- <span class="k">else</span><span class="p">:</span>
- <span class="n">arch_params</span><span class="p">[</span><span class="s1">'num_classes'</span><span class="p">]</span> <span class="o">=</span> <span class="nb">len</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">classes</span><span class="p">)</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">arch_params</span> <span class="o">=</span> <span class="n">core_utils</span><span class="o">.</span><span class="n">HpmStruct</span><span class="p">(</span><span class="o">**</span><span class="n">arch_params</span><span class="p">)</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">classes</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">arch_params</span><span class="o">.</span><span class="n">num_classes</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">net</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">architecture_cls</span><span class="p">(</span><span class="n">arch_params</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">arch_params</span><span class="p">)</span>
- <span class="c1"># save the architecture for neural architecture search</span>
- <span class="k">if</span> <span class="nb">hasattr</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">net</span><span class="p">,</span> <span class="s1">'structure'</span><span class="p">):</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">architecture</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">net</span><span class="o">.</span><span class="n">structure</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">net</span><span class="o">.</span><span class="n">to</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">device</span><span class="p">)</span>
- <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">multi_gpu</span> <span class="o">==</span> <span class="n">MultiGPUMode</span><span class="o">.</span><span class="n">DISTRIBUTED_DATA_PARALLEL</span><span class="p">:</span>
- <span class="n">logger</span><span class="o">.</span><span class="n">warning</span><span class="p">(</span><span class="s2">"Warning: distributed training is not supported in re_build_model()"</span><span class="p">)</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">net</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">nn</span><span class="o">.</span><span class="n">DataParallel</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">net</span><span class="p">,</span>
- <span class="n">device_ids</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">device_ids</span><span class="p">)</span> <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">multi_gpu</span> <span class="k">else</span> <span class="n">core_utils</span><span class="o">.</span><span class="n">WrappedModel</span><span class="p">(</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">net</span><span class="p">)</span></div>
- <div class="viewcode-block" id="SgModel.update_architecture"><a class="viewcode-back" href="../../../../super_gradients.training.sg_model.html#super_gradients.training.SgModel.update_architecture">[docs]</a> <span class="k">def</span> <span class="nf">update_architecture</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">structure</span><span class="p">):</span>
- <span class="sd">'''</span>
- <span class="sd"> architecture : str</span>
- <span class="sd"> Defines the network's architecture according to the options in models/all_architectures</span>
- <span class="sd"> load_checkpoint : bool</span>
- <span class="sd"> Loads a checkpoint according to experiment_name</span>
- <span class="sd"> arch_params : dict</span>
- <span class="sd"> Architecture H.P. e.g.: block, num_blocks, num_classes, etc.</span>
- <span class="sd"> :return:</span>
- <span class="sd"> '''</span>
- <span class="k">if</span> <span class="nb">hasattr</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">net</span><span class="o">.</span><span class="n">module</span><span class="p">,</span> <span class="s1">'update_structure'</span><span class="p">):</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">net</span><span class="o">.</span><span class="n">module</span><span class="o">.</span><span class="n">update_structure</span><span class="p">(</span><span class="n">structure</span><span class="p">)</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">net</span><span class="o">.</span><span class="n">to</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">device</span><span class="p">)</span>
- <span class="k">else</span><span class="p">:</span>
- <span class="k">raise</span> <span class="ne">Exception</span><span class="p">(</span><span class="s2">"architecture is not valid for NAS"</span><span class="p">)</span></div>
- <div class="viewcode-block" id="SgModel.get_module"><a class="viewcode-back" href="../../../../super_gradients.training.sg_model.html#super_gradients.training.SgModel.get_module">[docs]</a> <span class="k">def</span> <span class="nf">get_module</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
- <span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">net</span></div>
- <div class="viewcode-block" id="SgModel.set_module"><a class="viewcode-back" href="../../../../super_gradients.training.sg_model.html#super_gradients.training.SgModel.set_module">[docs]</a> <span class="k">def</span> <span class="nf">set_module</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">module</span><span class="p">):</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">net</span> <span class="o">=</span> <span class="n">module</span></div>
- <span class="k">def</span> <span class="nf">_initialize_device</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">requested_device</span><span class="p">:</span> <span class="nb">str</span><span class="p">,</span> <span class="n">requested_multi_gpu</span><span class="p">:</span> <span class="n">Union</span><span class="p">[</span><span class="n">MultiGPUMode</span><span class="p">,</span> <span class="nb">str</span><span class="p">]):</span>
- <span class="sd">"""</span>
- <span class="sd"> _initialize_device - Initializes the device for the model - Default is CUDA</span>
- <span class="sd"> :param requested_device: Device to initialize ('cuda' / 'cpu')</span>
- <span class="sd"> :param requested_multi_gpu: Get Multiple GPU</span>
- <span class="sd"> """</span>
- <span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">requested_multi_gpu</span><span class="p">,</span> <span class="nb">str</span><span class="p">):</span>
- <span class="n">requested_multi_gpu</span> <span class="o">=</span> <span class="n">MultiGPUMode</span><span class="p">(</span><span class="n">requested_multi_gpu</span><span class="p">)</span>
- <span class="c1"># SELECT CUDA DEVICE</span>
- <span class="k">if</span> <span class="n">requested_device</span> <span class="o">==</span> <span class="s1">'cuda'</span><span class="p">:</span>
- <span class="k">if</span> <span class="n">torch</span><span class="o">.</span><span class="n">cuda</span><span class="o">.</span><span class="n">is_available</span><span class="p">():</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">device</span> <span class="o">=</span> <span class="s1">'cuda'</span> <span class="c1"># TODO - we may want to set the device number as well i.e. 'cuda:1'</span>
- <span class="k">else</span><span class="p">:</span>
- <span class="k">raise</span> <span class="ne">RuntimeError</span><span class="p">(</span><span class="s1">'CUDA DEVICE NOT FOUND... EXITING'</span><span class="p">)</span>
- <span class="c1"># SELECT CPU DEVICE</span>
- <span class="k">elif</span> <span class="n">requested_device</span> <span class="o">==</span> <span class="s1">'cpu'</span><span class="p">:</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">device</span> <span class="o">=</span> <span class="s1">'cpu'</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">multi_gpu</span> <span class="o">=</span> <span class="kc">False</span>
- <span class="k">else</span><span class="p">:</span>
- <span class="c1"># SELECT CUDA DEVICE BY DEFAULT IF AVAILABLE</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">device</span> <span class="o">=</span> <span class="s1">'cuda'</span> <span class="k">if</span> <span class="n">torch</span><span class="o">.</span><span class="n">cuda</span><span class="o">.</span><span class="n">is_available</span><span class="p">()</span> <span class="k">else</span> <span class="s1">'cpu'</span>
- <span class="c1"># DEFUALT IS SET TO 1 - IT IS CHANGED IF MULTI-GPU IS USED</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">num_devices</span> <span class="o">=</span> <span class="mi">1</span>
- <span class="c1"># IN CASE OF MULTIPLE GPUS UPDATE THE LEARNING AND DATA PARAMETERS</span>
- <span class="c1"># FIXME - CREATE A DISCUSSION ON THESE PARAMETERS - WE MIGHT WANT TO CHANGE THE WAY WE USE THE LR AND</span>
- <span class="k">if</span> <span class="n">requested_multi_gpu</span> <span class="o">!=</span> <span class="n">MultiGPUMode</span><span class="o">.</span><span class="n">OFF</span><span class="p">:</span>
- <span class="k">if</span> <span class="s1">'cuda'</span> <span class="ow">in</span> <span class="bp">self</span><span class="o">.</span><span class="n">device</span><span class="p">:</span>
- <span class="c1"># COLLECT THE AVAILABLE GPU AND COUNT THE AVAILABLE GPUS AMOUNT</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">device_ids</span> <span class="o">=</span> <span class="nb">list</span><span class="p">(</span><span class="nb">range</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">cuda</span><span class="o">.</span><span class="n">device_count</span><span class="p">()))</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">num_devices</span> <span class="o">=</span> <span class="nb">len</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">device_ids</span><span class="p">)</span>
- <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">num_devices</span> <span class="o">==</span> <span class="mi">1</span><span class="p">:</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">multi_gpu</span> <span class="o">=</span> <span class="n">MultiGPUMode</span><span class="o">.</span><span class="n">OFF</span>
- <span class="k">if</span> <span class="n">requested_multi_gpu</span> <span class="o">!=</span> <span class="n">MultiGPUMode</span><span class="o">.</span><span class="n">AUTO</span><span class="p">:</span>
- <span class="c1"># if AUTO mode was set - do not log a warning</span>
- <span class="n">logger</span><span class="o">.</span><span class="n">warning</span><span class="p">(</span><span class="s1">'</span><span class="se">\n</span><span class="s1">[WARNING] - Tried running on multiple GPU but only a single GPU is available</span><span class="se">\n</span><span class="s1">'</span><span class="p">)</span>
- <span class="k">else</span><span class="p">:</span>
- <span class="k">if</span> <span class="n">requested_multi_gpu</span> <span class="o">==</span> <span class="n">MultiGPUMode</span><span class="o">.</span><span class="n">AUTO</span><span class="p">:</span>
- <span class="k">if</span> <span class="n">env_helpers</span><span class="o">.</span><span class="n">is_distributed</span><span class="p">():</span>
- <span class="n">requested_multi_gpu</span> <span class="o">=</span> <span class="n">MultiGPUMode</span><span class="o">.</span><span class="n">DISTRIBUTED_DATA_PARALLEL</span>
- <span class="k">else</span><span class="p">:</span>
- <span class="n">requested_multi_gpu</span> <span class="o">=</span> <span class="n">MultiGPUMode</span><span class="o">.</span><span class="n">DATA_PARALLEL</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">multi_gpu</span> <span class="o">=</span> <span class="n">requested_multi_gpu</span>
- <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">multi_gpu</span> <span class="o">==</span> <span class="n">MultiGPUMode</span><span class="o">.</span><span class="n">DISTRIBUTED_DATA_PARALLEL</span><span class="p">:</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">_initialize_ddp</span><span class="p">()</span>
- <span class="k">else</span><span class="p">:</span>
- <span class="c1"># MULTIPLE GPUS CAN BE ACTIVE ONLY IF A GPU IS AVAILABLE</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">multi_gpu</span> <span class="o">=</span> <span class="n">MultiGPUMode</span><span class="o">.</span><span class="n">OFF</span>
- <span class="n">logger</span><span class="o">.</span><span class="n">warning</span><span class="p">(</span><span class="s1">'</span><span class="se">\n</span><span class="s1">[WARNING] - Tried running on multiple GPU but none are available => running on CPU</span><span class="se">\n</span><span class="s1">'</span><span class="p">)</span>
- <span class="k">def</span> <span class="nf">_initialize_ddp</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
- <span class="sd">"""</span>
- <span class="sd"> Initializes Distributed Data Parallel</span>
- <span class="sd"> Usage:</span>
- <span class="sd"> python -m torch.distributed.launch --nproc_per_node=n YOUR_TRAINING_SCRIPT.py</span>
- <span class="sd"> where n is the number of GPUs required, e.g., n=8</span>
- <span class="sd"> Important note: (1) in distributed training it is customary to specify learning rates and batch sizes per GPU.</span>
- <span class="sd"> Whatever learning rate and schedule you specify will be applied to the each GPU individually.</span>
- <span class="sd"> Since gradients are passed and summed (reduced) from all to all GPUs, the effective batch size is the</span>
- <span class="sd"> batch you specify times the number of GPUs. In the literature there are several "best practices" to set</span>
- <span class="sd"> learning rates and schedules for large batch sizes.</span>
- <span class="sd"> """</span>
- <span class="n">logger</span><span class="o">.</span><span class="n">info</span><span class="p">(</span><span class="s2">"Distributed training starting..."</span><span class="p">)</span>
- <span class="n">local_rank</span> <span class="o">=</span> <span class="n">environment_config</span><span class="o">.</span><span class="n">DDP_LOCAL_RANK</span>
- <span class="k">if</span> <span class="ow">not</span> <span class="n">torch</span><span class="o">.</span><span class="n">distributed</span><span class="o">.</span><span class="n">is_initialized</span><span class="p">():</span>
- <span class="n">torch</span><span class="o">.</span><span class="n">distributed</span><span class="o">.</span><span class="n">init_process_group</span><span class="p">(</span><span class="n">backend</span><span class="o">=</span><span class="s1">'nccl'</span><span class="p">,</span> <span class="n">init_method</span><span class="o">=</span><span class="s1">'env://'</span><span class="p">)</span>
- <span class="k">if</span> <span class="n">local_rank</span> <span class="o">></span> <span class="mi">0</span><span class="p">:</span>
- <span class="n">f</span> <span class="o">=</span> <span class="nb">open</span><span class="p">(</span><span class="n">os</span><span class="o">.</span><span class="n">devnull</span><span class="p">,</span> <span class="s1">'w'</span><span class="p">)</span>
- <span class="n">sys</span><span class="o">.</span><span class="n">stdout</span> <span class="o">=</span> <span class="n">f</span> <span class="c1"># silent all printing for non master process</span>
- <span class="n">torch</span><span class="o">.</span><span class="n">cuda</span><span class="o">.</span><span class="n">set_device</span><span class="p">(</span><span class="n">local_rank</span><span class="p">)</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">device</span> <span class="o">=</span> <span class="s1">'cuda:</span><span class="si">%d</span><span class="s1">'</span> <span class="o">%</span> <span class="n">local_rank</span>
- <span class="c1"># MAKE ALL HIGHER-RANK GPUS SILENT (DISTRIBUTED MODE)</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">ddp_silent_mode</span> <span class="o">=</span> <span class="n">local_rank</span> <span class="o">></span> <span class="mi">0</span>
- <span class="k">if</span> <span class="n">torch</span><span class="o">.</span><span class="n">distributed</span><span class="o">.</span><span class="n">get_rank</span><span class="p">()</span> <span class="o">==</span> <span class="mi">0</span><span class="p">:</span>
- <span class="n">logger</span><span class="o">.</span><span class="n">info</span><span class="p">(</span><span class="sa">f</span><span class="s2">"Training in distributed mode... with </span><span class="si">{</span><span class="nb">str</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">distributed</span><span class="o">.</span><span class="n">get_world_size</span><span class="p">())</span><span class="si">}</span><span class="s2"> GPUs"</span><span class="p">)</span>
- <span class="k">def</span> <span class="nf">_switch_device</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">new_device</span><span class="p">):</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">device</span> <span class="o">=</span> <span class="n">new_device</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">net</span><span class="o">.</span><span class="n">to</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">device</span><span class="p">)</span>
- <span class="c1"># FIXME - we need to resolve flake8's 'function is too complex' for this function</span>
- <span class="k">def</span> <span class="nf">_load_checkpoint_to_model</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">strict</span><span class="p">:</span> <span class="n">StrictLoad</span><span class="p">,</span> <span class="n">load_backbone</span><span class="p">:</span> <span class="nb">bool</span><span class="p">,</span> <span class="n">source_ckpt_folder_name</span><span class="p">:</span> <span class="nb">str</span><span class="p">,</span>
- <span class="n">load_ema_as_net</span><span class="p">:</span> <span class="nb">bool</span><span class="p">):</span> <span class="c1"># noqa: C901 - too complex</span>
- <span class="sd">"""</span>
- <span class="sd"> Copies the source checkpoint to a local folder and loads the checkpoint's data to the model</span>
- <span class="sd"> :param strict: See StrictLoad class documentation for details.</span>
- <span class="sd"> :param load_backbone: loads the provided checkpoint to self.net.backbone instead of self.net</span>
- <span class="sd"> :param source_ckpt_folder_name: The folder where the checkpoint is saved. By default uses the self.experiment_name</span>
- <span class="sd"> NOTE: 'acc', 'epoch', 'optimizer_state_dict' and the logs are NOT loaded if self.zeroize_prev_train_params is True</span>
- <span class="sd"> """</span>
- <span class="c1"># GET LOCAL PATH TO THE CHECKPOINT FILE FIRST</span>
- <span class="n">ckpt_local_path</span> <span class="o">=</span> <span class="n">get_ckpt_local_path</span><span class="p">(</span><span class="n">source_ckpt_folder_name</span><span class="o">=</span><span class="n">source_ckpt_folder_name</span><span class="p">,</span>
- <span class="n">experiment_name</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">experiment_name</span><span class="p">,</span>
- <span class="n">ckpt_name</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">ckpt_name</span><span class="p">,</span>
- <span class="n">model_checkpoints_location</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">model_checkpoints_location</span><span class="p">,</span>
- <span class="n">external_checkpoint_path</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">external_checkpoint_path</span><span class="p">,</span>
- <span class="n">overwrite_local_checkpoint</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">overwrite_local_checkpoint</span><span class="p">,</span>
- <span class="n">load_weights_only</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">load_weights_only</span><span class="p">)</span>
- <span class="c1"># LOAD CHECKPOINT TO MODEL</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">checkpoint</span> <span class="o">=</span> <span class="n">load_checkpoint_to_model</span><span class="p">(</span><span class="n">ckpt_local_path</span><span class="o">=</span><span class="n">ckpt_local_path</span><span class="p">,</span>
- <span class="n">load_backbone</span><span class="o">=</span><span class="n">load_backbone</span><span class="p">,</span>
- <span class="n">net</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">net</span><span class="p">,</span>
- <span class="n">strict</span><span class="o">=</span><span class="n">strict</span><span class="o">.</span><span class="n">value</span> <span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">strict</span><span class="p">,</span> <span class="n">StrictLoad</span><span class="p">)</span> <span class="k">else</span> <span class="n">strict</span><span class="p">,</span>
- <span class="n">load_weights_only</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">load_weights_only</span><span class="p">,</span>
- <span class="n">load_ema_as_net</span><span class="o">=</span><span class="n">load_ema_as_net</span><span class="p">)</span>
- <span class="k">if</span> <span class="s1">'ema_net'</span> <span class="ow">in</span> <span class="bp">self</span><span class="o">.</span><span class="n">checkpoint</span><span class="o">.</span><span class="n">keys</span><span class="p">():</span>
- <span class="n">logger</span><span class="o">.</span><span class="n">warning</span><span class="p">(</span><span class="s2">"[WARNING] Main network has been loaded from checkpoint but EMA network exists as well. It "</span>
- <span class="s2">" will only be loaded during validation when training with ema=True. "</span><span class="p">)</span>
- <span class="c1"># UPDATE TRAINING PARAMS IF THEY EXIST & WE ARE NOT LOADING AN EXTERNAL MODEL's WEIGHTS</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">best_metric</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">checkpoint</span><span class="p">[</span><span class="s1">'acc'</span><span class="p">]</span> <span class="k">if</span> <span class="s1">'acc'</span> <span class="ow">in</span> <span class="bp">self</span><span class="o">.</span><span class="n">checkpoint</span><span class="o">.</span><span class="n">keys</span><span class="p">()</span> <span class="k">else</span> <span class="o">-</span><span class="mi">1</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">start_epoch</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">checkpoint</span><span class="p">[</span><span class="s1">'epoch'</span><span class="p">]</span> <span class="k">if</span> <span class="s1">'epoch'</span> <span class="ow">in</span> <span class="bp">self</span><span class="o">.</span><span class="n">checkpoint</span><span class="o">.</span><span class="n">keys</span><span class="p">()</span> <span class="k">else</span> <span class="mi">0</span>
- <span class="k">def</span> <span class="nf">_prep_for_test</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">test_loader</span><span class="p">:</span> <span class="n">torch</span><span class="o">.</span><span class="n">utils</span><span class="o">.</span><span class="n">data</span><span class="o">.</span><span class="n">DataLoader</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span> <span class="n">loss</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span> <span class="n">post_prediction_callback</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
- <span class="n">test_metrics_list</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
- <span class="n">loss_logging_items_names</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span> <span class="n">test_phase_callbacks</span><span class="o">=</span><span class="kc">None</span><span class="p">):</span>
- <span class="sd">"""Run commands that are common to all SgModels"""</span>
- <span class="c1"># SET THE MODEL IN evaluation STATE</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">net</span><span class="o">.</span><span class="n">eval</span><span class="p">()</span>
- <span class="c1"># IF SPECIFIED IN THE FUNCTION CALL - OVERRIDE THE self ARGUMENTS</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">test_loader</span> <span class="o">=</span> <span class="n">test_loader</span> <span class="ow">or</span> <span class="bp">self</span><span class="o">.</span><span class="n">test_loader</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">criterion</span> <span class="o">=</span> <span class="n">loss</span> <span class="ow">or</span> <span class="bp">self</span><span class="o">.</span><span class="n">criterion</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">post_prediction_callback</span> <span class="o">=</span> <span class="n">post_prediction_callback</span> <span class="ow">or</span> <span class="bp">self</span><span class="o">.</span><span class="n">post_prediction_callback</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">loss_logging_items_names</span> <span class="o">=</span> <span class="n">loss_logging_items_names</span> <span class="ow">or</span> <span class="bp">self</span><span class="o">.</span><span class="n">loss_logging_items_names</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">phase_callbacks</span> <span class="o">=</span> <span class="n">test_phase_callbacks</span> <span class="ow">or</span> <span class="bp">self</span><span class="o">.</span><span class="n">phase_callbacks</span>
- <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">phase_callbacks</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">phase_callbacks</span> <span class="o">=</span> <span class="p">[]</span>
- <span class="k">if</span> <span class="n">test_metrics_list</span><span class="p">:</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">test_metrics</span> <span class="o">=</span> <span class="n">MetricCollection</span><span class="p">(</span><span class="n">test_metrics_list</span><span class="p">)</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">phase_callbacks</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="n">MetricsUpdateCallback</span><span class="p">(</span><span class="n">Phase</span><span class="o">.</span><span class="n">TEST_BATCH_END</span><span class="p">))</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">phase_callback_handler</span> <span class="o">=</span> <span class="n">CallbackHandler</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">phase_callbacks</span><span class="p">)</span>
- <span class="c1"># WHEN TESTING WITHOUT A LOSS FUNCTION- CREATE EPOCH HEADERS FOR PRINTS</span>
- <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">criterion</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">loss_logging_items_names</span> <span class="o">=</span> <span class="p">[]</span>
- <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">test_metrics</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
- <span class="k">raise</span> <span class="ne">ValueError</span><span class="p">(</span><span class="s2">"Metrics are required to perform test. Pass them through test_metrics_list arg when "</span>
- <span class="s2">"calling test or through training_params when calling train(...)"</span><span class="p">)</span>
- <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">test_loader</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
- <span class="k">raise</span> <span class="ne">ValueError</span><span class="p">(</span><span class="s2">"Test dataloader is required to perform test. Make sure to either pass it through "</span>
- <span class="s2">"test_loader arg or calling connect_dataset_interface upon a DatasetInterface instance "</span>
- <span class="s2">"with a non empty testset attribute."</span><span class="p">)</span>
- <span class="c1"># RESET METRIC RUNNERS</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">test_metrics</span><span class="o">.</span><span class="n">reset</span><span class="p">()</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">test_metrics</span><span class="o">.</span><span class="n">to</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">device</span><span class="p">)</span>
- <span class="k">def</span> <span class="nf">_initialize_sg_logger_objects</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
- <span class="sd">"""Initialize object that collect, write to disk, monitor and store remotely all training outputs"""</span>
- <span class="n">sg_logger</span> <span class="o">=</span> <span class="n">core_utils</span><span class="o">.</span><span class="n">get_param</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">training_params</span><span class="p">,</span> <span class="s1">'sg_logger'</span><span class="p">)</span>
- <span class="c1"># OVERRIDE SOME PARAMETERS TO MAKE SURE THEY MATCH THE TRAINING PARAMETERS</span>
- <span class="n">general_sg_logger_params</span> <span class="o">=</span> <span class="p">{</span><span class="s1">'experiment_name'</span><span class="p">:</span> <span class="bp">self</span><span class="o">.</span><span class="n">experiment_name</span><span class="p">,</span>
- <span class="s1">'storage_location'</span><span class="p">:</span> <span class="bp">self</span><span class="o">.</span><span class="n">model_checkpoints_location</span><span class="p">,</span>
- <span class="s1">'resumed'</span><span class="p">:</span> <span class="bp">self</span><span class="o">.</span><span class="n">load_checkpoint</span><span class="p">,</span>
- <span class="s1">'training_params'</span><span class="p">:</span> <span class="bp">self</span><span class="o">.</span><span class="n">training_params</span><span class="p">,</span>
- <span class="s1">'checkpoints_dir_path'</span><span class="p">:</span> <span class="bp">self</span><span class="o">.</span><span class="n">checkpoints_dir_path</span><span class="p">}</span>
- <span class="k">if</span> <span class="n">sg_logger</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
- <span class="k">raise</span> <span class="ne">RuntimeError</span><span class="p">(</span><span class="s1">'sg_logger must be defined in training params (see default_training_params)'</span><span class="p">)</span>
- <span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">sg_logger</span><span class="p">,</span> <span class="n">AbstractSGLogger</span><span class="p">):</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">sg_logger</span> <span class="o">=</span> <span class="n">sg_logger</span>
- <span class="k">elif</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">sg_logger</span><span class="p">,</span> <span class="nb">str</span><span class="p">):</span>
- <span class="n">sg_logger_params</span> <span class="o">=</span> <span class="n">core_utils</span><span class="o">.</span><span class="n">get_param</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">training_params</span><span class="p">,</span> <span class="s1">'sg_logger_params'</span><span class="p">,</span> <span class="p">{})</span>
- <span class="k">if</span> <span class="nb">issubclass</span><span class="p">(</span><span class="n">SG_LOGGERS</span><span class="p">[</span><span class="n">sg_logger</span><span class="p">],</span> <span class="n">BaseSGLogger</span><span class="p">):</span>
- <span class="n">sg_logger_params</span> <span class="o">=</span> <span class="p">{</span><span class="o">**</span><span class="n">sg_logger_params</span><span class="p">,</span> <span class="o">**</span><span class="n">general_sg_logger_params</span><span class="p">}</span>
- <span class="k">if</span> <span class="n">sg_logger</span> <span class="ow">not</span> <span class="ow">in</span> <span class="n">SG_LOGGERS</span><span class="p">:</span>
- <span class="k">raise</span> <span class="ne">RuntimeError</span><span class="p">(</span><span class="s1">'sg_logger not defined in SG_LOGGERS'</span><span class="p">)</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">sg_logger</span> <span class="o">=</span> <span class="n">SG_LOGGERS</span><span class="p">[</span><span class="n">sg_logger</span><span class="p">](</span><span class="o">**</span><span class="n">sg_logger_params</span><span class="p">)</span>
- <span class="k">else</span><span class="p">:</span>
- <span class="k">raise</span> <span class="ne">RuntimeError</span><span class="p">(</span><span class="s1">'sg_logger can be either an sg_logger name (str) or a subcalss of AbstractSGLogger'</span><span class="p">)</span>
- <span class="k">if</span> <span class="ow">not</span> <span class="nb">isinstance</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">sg_logger</span><span class="p">,</span> <span class="n">BaseSGLogger</span><span class="p">):</span>
- <span class="n">logger</span><span class="o">.</span><span class="n">warning</span><span class="p">(</span><span class="s2">"WARNING! Using a user-defined sg_logger: files will not be automatically written to disk!</span><span class="se">\n</span><span class="s2">"</span>
- <span class="s2">"Please make sure the provided sg_logger writes to disk or compose your sg_logger to BaseSGLogger"</span><span class="p">)</span>
- <span class="c1"># IN CASE SG_LOGGER UPDATED THE DIR PATH</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">checkpoints_dir_path</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">sg_logger</span><span class="o">.</span><span class="n">local_dir</span><span class="p">()</span>
- <span class="n">additional_log_items</span> <span class="o">=</span> <span class="p">{</span><span class="s1">'initial_LR'</span><span class="p">:</span> <span class="bp">self</span><span class="o">.</span><span class="n">training_params</span><span class="o">.</span><span class="n">initial_lr</span><span class="p">,</span>
- <span class="s1">'num_devices'</span><span class="p">:</span> <span class="bp">self</span><span class="o">.</span><span class="n">num_devices</span><span class="p">,</span>
- <span class="s1">'multi_gpu'</span><span class="p">:</span> <span class="nb">str</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">multi_gpu</span><span class="p">),</span>
- <span class="s1">'device_type'</span><span class="p">:</span> <span class="n">torch</span><span class="o">.</span><span class="n">cuda</span><span class="o">.</span><span class="n">get_device_name</span><span class="p">(</span><span class="mi">0</span><span class="p">)</span> <span class="k">if</span> <span class="n">torch</span><span class="o">.</span><span class="n">cuda</span><span class="o">.</span><span class="n">is_available</span><span class="p">()</span> <span class="k">else</span> <span class="s1">'cpu'</span><span class="p">}</span>
- <span class="c1"># ADD INSTALLED PACKAGE LIST + THEIR VERSIONS</span>
- <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">training_params</span><span class="o">.</span><span class="n">log_installed_packages</span><span class="p">:</span>
- <span class="n">pkg_list</span> <span class="o">=</span> <span class="nb">list</span><span class="p">(</span><span class="nb">map</span><span class="p">(</span><span class="k">lambda</span> <span class="n">pkg</span><span class="p">:</span> <span class="nb">str</span><span class="p">(</span><span class="n">pkg</span><span class="p">),</span> <span class="n">_get_installed_distributions</span><span class="p">()))</span>
- <span class="n">additional_log_items</span><span class="p">[</span><span class="s1">'installed_packages'</span><span class="p">]</span> <span class="o">=</span> <span class="n">pkg_list</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">sg_logger</span><span class="o">.</span><span class="n">add_config</span><span class="p">(</span><span class="s2">"hyper_params"</span><span class="p">,</span> <span class="p">{</span><span class="s2">"arch_params"</span><span class="p">:</span> <span class="bp">self</span><span class="o">.</span><span class="n">arch_params</span><span class="o">.</span><span class="vm">__dict__</span><span class="p">,</span>
- <span class="s2">"training_hyperparams"</span><span class="p">:</span> <span class="bp">self</span><span class="o">.</span><span class="n">training_params</span><span class="o">.</span><span class="vm">__dict__</span><span class="p">,</span>
- <span class="s2">"dataset_params"</span><span class="p">:</span> <span class="bp">self</span><span class="o">.</span><span class="n">dataset_params</span><span class="o">.</span><span class="vm">__dict__</span><span class="p">,</span>
- <span class="s2">"additional_log_items"</span><span class="p">:</span> <span class="n">additional_log_items</span><span class="p">})</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">sg_logger</span><span class="o">.</span><span class="n">flush</span><span class="p">()</span>
- <span class="k">def</span> <span class="nf">_write_to_disk_operations</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">train_metrics</span><span class="p">:</span> <span class="nb">tuple</span><span class="p">,</span> <span class="n">validation_results</span><span class="p">:</span> <span class="nb">tuple</span><span class="p">,</span> <span class="n">inf_time</span><span class="p">:</span> <span class="nb">float</span><span class="p">,</span> <span class="n">epoch</span><span class="p">:</span> <span class="nb">int</span><span class="p">,</span> <span class="n">context</span><span class="p">:</span> <span class="n">PhaseContext</span><span class="p">):</span>
- <span class="sd">"""Run the various logging operations, e.g.: log file, Tensorboard, save checkpoint etc."""</span>
- <span class="c1"># STORE VALUES IN A TENSORBOARD FILE</span>
- <span class="n">train_results</span> <span class="o">=</span> <span class="nb">list</span><span class="p">(</span><span class="n">train_metrics</span><span class="p">)</span> <span class="o">+</span> <span class="nb">list</span><span class="p">(</span><span class="n">validation_results</span><span class="p">)</span> <span class="o">+</span> <span class="p">[</span><span class="n">inf_time</span><span class="p">]</span>
- <span class="n">all_titles</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">results_titles</span> <span class="o">+</span> <span class="p">[</span><span class="s1">'Inference Time'</span><span class="p">]</span>
- <span class="n">result_dict</span> <span class="o">=</span> <span class="p">{</span><span class="n">all_titles</span><span class="p">[</span><span class="n">i</span><span class="p">]:</span> <span class="n">train_results</span><span class="p">[</span><span class="n">i</span><span class="p">]</span> <span class="k">for</span> <span class="n">i</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="nb">len</span><span class="p">(</span><span class="n">train_results</span><span class="p">))}</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">sg_logger</span><span class="o">.</span><span class="n">add_scalars</span><span class="p">(</span><span class="n">tag_scalar_dict</span><span class="o">=</span><span class="n">result_dict</span><span class="p">,</span> <span class="n">global_step</span><span class="o">=</span><span class="n">epoch</span><span class="p">)</span>
- <span class="c1"># SAVE THE CHECKPOINT</span>
- <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">training_params</span><span class="o">.</span><span class="n">save_model</span><span class="p">:</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">save_checkpoint</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">optimizer</span><span class="p">,</span> <span class="n">epoch</span> <span class="o">+</span> <span class="mi">1</span><span class="p">,</span> <span class="n">validation_results</span><span class="p">,</span> <span class="n">context</span><span class="p">)</span>
- <span class="k">def</span> <span class="nf">_write_lrs</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">epoch</span><span class="p">):</span>
- <span class="n">lrs</span> <span class="o">=</span> <span class="p">[</span><span class="bp">self</span><span class="o">.</span><span class="n">optimizer</span><span class="o">.</span><span class="n">param_groups</span><span class="p">[</span><span class="n">i</span><span class="p">][</span><span class="s1">'lr'</span><span class="p">]</span> <span class="k">for</span> <span class="n">i</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="nb">len</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">optimizer</span><span class="o">.</span><span class="n">param_groups</span><span class="p">))]</span>
- <span class="n">lr_titles</span> <span class="o">=</span> <span class="p">[</span><span class="s1">'LR/Param_group_'</span> <span class="o">+</span> <span class="nb">str</span><span class="p">(</span><span class="n">i</span><span class="p">)</span> <span class="k">for</span> <span class="n">i</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="nb">len</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">optimizer</span><span class="o">.</span><span class="n">param_groups</span><span class="p">))]</span> <span class="k">if</span> <span class="nb">len</span><span class="p">(</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">optimizer</span><span class="o">.</span><span class="n">param_groups</span><span class="p">)</span> <span class="o">></span> <span class="mi">1</span> <span class="k">else</span> <span class="p">[</span><span class="s1">'LR'</span><span class="p">]</span>
- <span class="n">lr_dict</span> <span class="o">=</span> <span class="p">{</span><span class="n">lr_titles</span><span class="p">[</span><span class="n">i</span><span class="p">]:</span> <span class="n">lrs</span><span class="p">[</span><span class="n">i</span><span class="p">]</span> <span class="k">for</span> <span class="n">i</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="nb">len</span><span class="p">(</span><span class="n">lrs</span><span class="p">))}</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">sg_logger</span><span class="o">.</span><span class="n">add_scalars</span><span class="p">(</span><span class="n">tag_scalar_dict</span><span class="o">=</span><span class="n">lr_dict</span><span class="p">,</span> <span class="n">global_step</span><span class="o">=</span><span class="n">epoch</span><span class="p">)</span>
- <div class="viewcode-block" id="SgModel.test"><a class="viewcode-back" href="../../../../super_gradients.training.sg_model.html#super_gradients.training.SgModel.test">[docs]</a> <span class="k">def</span> <span class="nf">test</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="c1"># noqa: C901</span>
- <span class="n">test_loader</span><span class="p">:</span> <span class="n">torch</span><span class="o">.</span><span class="n">utils</span><span class="o">.</span><span class="n">data</span><span class="o">.</span><span class="n">DataLoader</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
- <span class="n">loss</span><span class="p">:</span> <span class="n">torch</span><span class="o">.</span><span class="n">nn</span><span class="o">.</span><span class="n">modules</span><span class="o">.</span><span class="n">loss</span><span class="o">.</span><span class="n">_Loss</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
- <span class="n">silent_mode</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="kc">False</span><span class="p">,</span>
- <span class="n">test_metrics_list</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
- <span class="n">loss_logging_items_names</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span> <span class="n">metrics_progress_verbose</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span> <span class="n">test_phase_callbacks</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
- <span class="n">use_ema_net</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span> <span class="o">-></span> <span class="nb">tuple</span><span class="p">:</span>
- <span class="sd">"""</span>
- <span class="sd"> Evaluates the model on given dataloader and metrics.</span>
- <span class="sd"> :param test_loader: dataloader to perform test on.</span>
- <span class="sd"> :param test_metrics_list: (list(torchmetrics.Metric)) metrics list for evaluation.</span>
- <span class="sd"> :param silent_mode: (bool) controls verbosity</span>
- <span class="sd"> :param metrics_progress_verbose: (bool) controls the verbosity of metrics progress (default=False). Slows down the program.</span>
- <span class="sd"> :param use_ema_net (bool) whether to perform test on self.ema_model.ema (when self.ema_model.ema exists,</span>
- <span class="sd"> otherwise self.net will be tested) (default=True)</span>
- <span class="sd"> :return: results tuple (tuple) containing the loss items and metric values.</span>
- <span class="sd"> All of the above args will override SgModel's corresponding attribute when not equal to None. Then evaluation</span>
- <span class="sd"> is ran on self.test_loader with self.test_metrics.</span>
- <span class="sd"> """</span>
- <span class="c1"># IN CASE TRAINING WAS PERFROMED BEFORE TEST- MAKE SURE TO TEST THE EMA MODEL (UNLESS SPECIFIED OTHERWISE BY</span>
- <span class="c1"># use_ema_net)</span>
- <span class="k">if</span> <span class="n">use_ema_net</span> <span class="ow">and</span> <span class="bp">self</span><span class="o">.</span><span class="n">ema_model</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
- <span class="n">keep_model</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">net</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">net</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">ema_model</span><span class="o">.</span><span class="n">ema</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">_prep_for_test</span><span class="p">(</span><span class="n">test_loader</span><span class="o">=</span><span class="n">test_loader</span><span class="p">,</span>
- <span class="n">loss</span><span class="o">=</span><span class="n">loss</span><span class="p">,</span>
- <span class="n">test_metrics_list</span><span class="o">=</span><span class="n">test_metrics_list</span><span class="p">,</span>
- <span class="n">loss_logging_items_names</span><span class="o">=</span><span class="n">loss_logging_items_names</span><span class="p">,</span>
- <span class="n">test_phase_callbacks</span><span class="o">=</span><span class="n">test_phase_callbacks</span><span class="p">,</span>
- <span class="p">)</span>
- <span class="n">test_results</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">evaluate</span><span class="p">(</span><span class="n">data_loader</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">test_loader</span><span class="p">,</span>
- <span class="n">metrics</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">test_metrics</span><span class="p">,</span>
- <span class="n">evaluation_type</span><span class="o">=</span><span class="n">EvaluationType</span><span class="o">.</span><span class="n">TEST</span><span class="p">,</span>
- <span class="n">silent_mode</span><span class="o">=</span><span class="n">silent_mode</span><span class="p">,</span>
- <span class="n">metrics_progress_verbose</span><span class="o">=</span><span class="n">metrics_progress_verbose</span><span class="p">)</span>
- <span class="c1"># SWITCH BACK BETWEEN NETS SO AN ADDITIONAL TRAINING CAN BE DONE AFTER TEST</span>
- <span class="k">if</span> <span class="n">use_ema_net</span> <span class="ow">and</span> <span class="bp">self</span><span class="o">.</span><span class="n">ema_model</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">net</span> <span class="o">=</span> <span class="n">keep_model</span>
- <span class="k">return</span> <span class="n">test_results</span></div>
- <span class="k">def</span> <span class="nf">_validate_epoch</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">epoch</span><span class="p">:</span> <span class="nb">int</span><span class="p">,</span> <span class="n">silent_mode</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="kc">False</span><span class="p">)</span> <span class="o">-></span> <span class="nb">tuple</span><span class="p">:</span>
- <span class="sd">"""</span>
- <span class="sd"> Runs evaluation on self.valid_loader, with self.valid_metrics.</span>
- <span class="sd"> :param epoch: (int) epoch idx</span>
- <span class="sd"> :param silent_mode: (bool) controls verbosity</span>
- <span class="sd"> :return: results tuple (tuple) containing the loss items and metric values.</span>
- <span class="sd"> """</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">net</span><span class="o">.</span><span class="n">eval</span><span class="p">()</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">valid_metrics</span><span class="o">.</span><span class="n">reset</span><span class="p">()</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">valid_metrics</span><span class="o">.</span><span class="n">to</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">device</span><span class="p">)</span>
- <span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">evaluate</span><span class="p">(</span><span class="n">data_loader</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">valid_loader</span><span class="p">,</span> <span class="n">metrics</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">valid_metrics</span><span class="p">,</span>
- <span class="n">evaluation_type</span><span class="o">=</span><span class="n">EvaluationType</span><span class="o">.</span><span class="n">VALIDATION</span><span class="p">,</span> <span class="n">epoch</span><span class="o">=</span><span class="n">epoch</span><span class="p">,</span> <span class="n">silent_mode</span><span class="o">=</span><span class="n">silent_mode</span><span class="p">)</span>
- <div class="viewcode-block" id="SgModel.evaluate"><a class="viewcode-back" href="../../../../super_gradients.training.sg_model.html#super_gradients.training.SgModel.evaluate">[docs]</a> <span class="k">def</span> <span class="nf">evaluate</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">data_loader</span><span class="p">:</span> <span class="n">torch</span><span class="o">.</span><span class="n">utils</span><span class="o">.</span><span class="n">data</span><span class="o">.</span><span class="n">DataLoader</span><span class="p">,</span> <span class="n">metrics</span><span class="p">:</span> <span class="n">MetricCollection</span><span class="p">,</span>
- <span class="n">evaluation_type</span><span class="p">:</span> <span class="n">EvaluationType</span><span class="p">,</span> <span class="n">epoch</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span> <span class="n">silent_mode</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="kc">False</span><span class="p">,</span>
- <span class="n">metrics_progress_verbose</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="kc">False</span><span class="p">):</span>
- <span class="sd">"""</span>
- <span class="sd"> Evaluates the model on given dataloader and metrics.</span>
- <span class="sd"> :param data_loader: dataloader to perform evaluataion on</span>
- <span class="sd"> :param metrics: (MetricCollection) metrics for evaluation</span>
- <span class="sd"> :param evaluation_type: (EvaluationType) controls which phase callbacks will be used (for example, on batch end,</span>
- <span class="sd"> when evaluation_type=EvaluationType.VALIDATION the Phase.VALIDATION_BATCH_END callbacks will be triggered)</span>
- <span class="sd"> :param epoch: (int) epoch idx</span>
- <span class="sd"> :param silent_mode: (bool) controls verbosity</span>
- <span class="sd"> :param metrics_progress_verbose: (bool) controls the verbosity of metrics progress (default=False).</span>
- <span class="sd"> Slows down the program significantly.</span>
- <span class="sd"> :return: results tuple (tuple) containing the loss items and metric values.</span>
- <span class="sd"> """</span>
- <span class="c1"># THE DISABLE FLAG CONTROLS WHETHER THE PROGRESS BAR IS SILENT OR PRINTS THE LOGS</span>
- <span class="n">progress_bar_data_loader</span> <span class="o">=</span> <span class="n">tqdm</span><span class="p">(</span><span class="n">data_loader</span><span class="p">,</span> <span class="n">bar_format</span><span class="o">=</span><span class="s2">"</span><span class="si">{l_bar}{bar:10}{r_bar}</span><span class="s2">"</span><span class="p">,</span> <span class="n">dynamic_ncols</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span>
- <span class="n">disable</span><span class="o">=</span><span class="n">silent_mode</span><span class="p">)</span>
- <span class="n">loss_avg_meter</span> <span class="o">=</span> <span class="n">core_utils</span><span class="o">.</span><span class="n">utils</span><span class="o">.</span><span class="n">AverageMeter</span><span class="p">()</span>
- <span class="n">logging_values</span> <span class="o">=</span> <span class="kc">None</span>
- <span class="n">loss_tuple</span> <span class="o">=</span> <span class="kc">None</span>
- <span class="n">lr_warmup_epochs</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">training_params</span><span class="o">.</span><span class="n">lr_warmup_epochs</span> <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">training_params</span> <span class="k">else</span> <span class="kc">None</span>
- <span class="n">context</span> <span class="o">=</span> <span class="n">PhaseContext</span><span class="p">(</span><span class="n">epoch</span><span class="o">=</span><span class="n">epoch</span><span class="p">,</span>
- <span class="n">metrics_compute_fn</span><span class="o">=</span><span class="n">metrics</span><span class="p">,</span>
- <span class="n">loss_avg_meter</span><span class="o">=</span><span class="n">loss_avg_meter</span><span class="p">,</span>
- <span class="n">criterion</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">criterion</span><span class="p">,</span>
- <span class="n">device</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">device</span><span class="p">,</span>
- <span class="n">lr_warmup_epochs</span><span class="o">=</span><span class="n">lr_warmup_epochs</span><span class="p">,</span>
- <span class="n">sg_logger</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">sg_logger</span><span class="p">)</span>
- <span class="k">if</span> <span class="ow">not</span> <span class="n">silent_mode</span><span class="p">:</span>
- <span class="c1"># PRINT TITLES</span>
- <span class="n">pbar_start_msg</span> <span class="o">=</span> <span class="sa">f</span><span class="s2">"Validation epoch </span><span class="si">{</span><span class="n">epoch</span><span class="si">}</span><span class="s2">"</span> <span class="k">if</span> <span class="n">evaluation_type</span> <span class="o">==</span> <span class="n">EvaluationType</span><span class="o">.</span><span class="n">VALIDATION</span> <span class="k">else</span> <span class="s2">"Test"</span>
- <span class="n">progress_bar_data_loader</span><span class="o">.</span><span class="n">set_description</span><span class="p">(</span><span class="n">pbar_start_msg</span><span class="p">)</span>
- <span class="k">with</span> <span class="n">torch</span><span class="o">.</span><span class="n">no_grad</span><span class="p">():</span>
- <span class="k">for</span> <span class="n">batch_idx</span><span class="p">,</span> <span class="n">batch_items</span> <span class="ow">in</span> <span class="nb">enumerate</span><span class="p">(</span><span class="n">progress_bar_data_loader</span><span class="p">):</span>
- <span class="n">batch_items</span> <span class="o">=</span> <span class="n">core_utils</span><span class="o">.</span><span class="n">tensor_container_to_device</span><span class="p">(</span><span class="n">batch_items</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">device</span><span class="p">,</span> <span class="n">non_blocking</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span>
- <span class="n">inputs</span><span class="p">,</span> <span class="n">targets</span><span class="p">,</span> <span class="n">additional_batch_items</span> <span class="o">=</span> <span class="n">sg_model_utils</span><span class="o">.</span><span class="n">unpack_batch_items</span><span class="p">(</span><span class="n">batch_items</span><span class="p">)</span>
- <span class="n">output</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">net</span><span class="p">(</span><span class="n">inputs</span><span class="p">)</span>
- <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">criterion</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
- <span class="c1"># STORE THE loss_items ONLY, THE 1ST RETURNED VALUE IS THE loss FOR BACKPROP DURING TRAINING</span>
- <span class="n">loss_tuple</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_get_losses</span><span class="p">(</span><span class="n">output</span><span class="p">,</span> <span class="n">targets</span><span class="p">)[</span><span class="mi">1</span><span class="p">]</span><span class="o">.</span><span class="n">cpu</span><span class="p">()</span>
- <span class="n">context</span><span class="o">.</span><span class="n">update_context</span><span class="p">(</span><span class="n">batch_idx</span><span class="o">=</span><span class="n">batch_idx</span><span class="p">,</span>
- <span class="n">inputs</span><span class="o">=</span><span class="n">inputs</span><span class="p">,</span>
- <span class="n">preds</span><span class="o">=</span><span class="n">output</span><span class="p">,</span>
- <span class="n">target</span><span class="o">=</span><span class="n">targets</span><span class="p">,</span>
- <span class="n">loss_log_items</span><span class="o">=</span><span class="n">loss_tuple</span><span class="p">,</span>
- <span class="o">**</span><span class="n">additional_batch_items</span><span class="p">)</span>
- <span class="c1"># TRIGGER PHASE CALLBACKS CORRESPONDING TO THE EVALUATION TYPE</span>
- <span class="k">if</span> <span class="n">evaluation_type</span> <span class="o">==</span> <span class="n">EvaluationType</span><span class="o">.</span><span class="n">VALIDATION</span><span class="p">:</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">phase_callback_handler</span><span class="p">(</span><span class="n">Phase</span><span class="o">.</span><span class="n">VALIDATION_BATCH_END</span><span class="p">,</span> <span class="n">context</span><span class="p">)</span>
- <span class="k">else</span><span class="p">:</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">phase_callback_handler</span><span class="p">(</span><span class="n">Phase</span><span class="o">.</span><span class="n">TEST_BATCH_END</span><span class="p">,</span> <span class="n">context</span><span class="p">)</span>
- <span class="c1"># COMPUTE METRICS IF PROGRESS VERBOSITY IS SET</span>
- <span class="k">if</span> <span class="n">metrics_progress_verbose</span> <span class="ow">and</span> <span class="ow">not</span> <span class="n">silent_mode</span><span class="p">:</span>
- <span class="c1"># COMPUTE THE RUNNING USER METRICS AND LOSS RUNNING ITEMS. RESULT TUPLE IS THEIR CONCATENATION.</span>
- <span class="n">logging_values</span> <span class="o">=</span> <span class="n">get_logging_values</span><span class="p">(</span><span class="n">loss_avg_meter</span><span class="p">,</span> <span class="n">metrics</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">criterion</span><span class="p">)</span>
- <span class="n">pbar_message_dict</span> <span class="o">=</span> <span class="n">get_train_loop_description_dict</span><span class="p">(</span><span class="n">logging_values</span><span class="p">,</span>
- <span class="n">metrics</span><span class="p">,</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">loss_logging_items_names</span><span class="p">)</span>
- <span class="n">progress_bar_data_loader</span><span class="o">.</span><span class="n">set_postfix</span><span class="p">(</span><span class="o">**</span><span class="n">pbar_message_dict</span><span class="p">)</span>
- <span class="c1"># NEED TO COMPUTE METRICS FOR THE FIRST TIME IF PROGRESS VERBOSITY IS NOT SET</span>
- <span class="k">if</span> <span class="ow">not</span> <span class="n">metrics_progress_verbose</span><span class="p">:</span>
- <span class="c1"># COMPUTE THE RUNNING USER METRICS AND LOSS RUNNING ITEMS. RESULT TUPLE IS THEIR CONCATENATION.</span>
- <span class="n">logging_values</span> <span class="o">=</span> <span class="n">get_logging_values</span><span class="p">(</span><span class="n">loss_avg_meter</span><span class="p">,</span> <span class="n">metrics</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">criterion</span><span class="p">)</span>
- <span class="n">pbar_message_dict</span> <span class="o">=</span> <span class="n">get_train_loop_description_dict</span><span class="p">(</span><span class="n">logging_values</span><span class="p">,</span>
- <span class="n">metrics</span><span class="p">,</span>
- <span class="bp">self</span><span class="o">.</span><span class="n">loss_logging_items_names</span><span class="p">)</span>
- <span class="n">progress_bar_data_loader</span><span class="o">.</span><span class="n">set_postfix</span><span class="p">(</span><span class="o">**</span><span class="n">pbar_message_dict</span><span class="p">)</span>
- <span class="c1"># TODO: SUPPORT PRINTING AP PER CLASS- SINCE THE METRICS ARE NOT HARD CODED ANYMORE (as done in</span>
- <span class="c1"># calc_batch_prediction_accuracy_per_class in metric_utils.py), THIS IS ONLY RELEVANT WHEN CHOOSING</span>
- <span class="c1"># DETECTIONMETRICS, WHICH ALREADY RETURN THE METRICS VALUEST HEMSELVES AND NOT THE ITEMS REQUIRED FOR SUCH</span>
- <span class="c1"># COMPUTATION. ALSO REMOVE THE BELOW LINES BY IMPLEMENTING CRITERION AS A TORCHMETRIC.</span>
- <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">multi_gpu</span> <span class="o">==</span> <span class="n">MultiGPUMode</span><span class="o">.</span><span class="n">DISTRIBUTED_DATA_PARALLEL</span><span class="p">:</span>
- <span class="n">logging_values</span> <span class="o">=</span> <span class="n">reduce_results_tuple_for_ddp</span><span class="p">(</span><span class="n">logging_values</span><span class="p">,</span> <span class="nb">next</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">net</span><span class="o">.</span><span class="n">parameters</span><span class="p">())</span><span class="o">.</span><span class="n">device</span><span class="p">)</span>
- <span class="k">return</span> <span class="n">logging_values</span></div></div>
- </pre></div>
- </div>
- </div>
- <footer>
- <hr/>
- <div role="contentinfo">
- <p>© Copyright 2021, SuperGradients team.</p>
- </div>
- Built with <a href="https://www.sphinx-doc.org/">Sphinx</a> using a
- <a href="https://github.com/readthedocs/sphinx_rtd_theme">theme</a>
- provided by <a href="https://readthedocs.org">Read the Docs</a>.
-
- </footer>
- </div>
- </div>
- </section>
- </div>
- <script>
- jQuery(function () {
- SphinxRtdTheme.Navigation.enable(true);
- });
- </script>
- </body>
- </html>
|
Models
Annotations
Connect to our Discord channel