File size: 72,551 Bytes
f1069cc
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
# coding=utf-8
# Copyright 2020 Microsoft and the Hugging Face Inc. team.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
""" PyTorch DeBERTa-v2 model."""

from collections.abc import Sequence
from typing import Optional, Tuple, Union

import torch
import torch.utils.checkpoint
from torch import nn
from torch.nn import BCEWithLogitsLoss, CrossEntropyLoss, LayerNorm, MSELoss

from transformers.activations import ACT2FN
from transformers.modeling_outputs import (
    ModelOutput,
    BaseModelOutput,
    MaskedLMOutput,
    MultipleChoiceModelOutput,
    QuestionAnsweringModelOutput,
    SequenceClassifierOutput,
    TokenClassifierOutput,
)
from transformers.modeling_utils import PreTrainedModel
from transformers.pytorch_utils import softmax_backward_data
from transformers.utils import add_code_sample_docstrings, add_start_docstrings, add_start_docstrings_to_model_forward, logging
from transformers.models.deberta_v2.configuration_deberta_v2 import DebertaV2Config


logger = logging.get_logger(__name__)

_CONFIG_FOR_DOC = "DebertaV2Config"
_CHECKPOINT_FOR_DOC = "microsoft/deberta-v2-xlarge"
_QA_TARGET_START_INDEX = 2
_QA_TARGET_END_INDEX = 9

DEBERTA_V2_PRETRAINED_MODEL_ARCHIVE_LIST = [
    "microsoft/deberta-v2-xlarge",
    "microsoft/deberta-v2-xxlarge",
    "microsoft/deberta-v2-xlarge-mnli",
    "microsoft/deberta-v2-xxlarge-mnli",
]


# Copied from transformers.models.deberta.modeling_deberta.ContextPooler
class ContextPooler(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.dense = nn.Linear(config.pooler_hidden_size, config.pooler_hidden_size)
        self.dropout = StableDropout(config.pooler_dropout)
        self.config = config

    def forward(self, hidden_states):
        # We "pool" the model by simply taking the hidden state corresponding
        # to the first token.

        context_token = hidden_states[:, 0]
        context_token = self.dropout(context_token)
        pooled_output = self.dense(context_token)
        pooled_output = ACT2FN[self.config.pooler_hidden_act](pooled_output)
        return pooled_output

    @property
    def output_dim(self):
        return self.config.hidden_size


# Copied from transformers.models.deberta.modeling_deberta.XSoftmax with deberta->deberta_v2
class XSoftmax(torch.autograd.Function):
    """
    Masked Softmax which is optimized for saving memory

    Args:
        input (`torch.tensor`): The input tensor that will apply softmax.
        mask (`torch.IntTensor`):
            The mask matrix where 0 indicate that element will be ignored in the softmax calculation.
        dim (int): The dimension that will apply softmax

    Example:

    ```python
    >>> import torch
    >>> from transformers.models.deberta_v2.modeling_deberta_v2 import XSoftmax

    >>> # Make a tensor
    >>> x = torch.randn([4, 20, 100])

    >>> # Create a mask
    >>> mask = (x > 0).int()

    >>> # Specify the dimension to apply softmax
    >>> dim = -1

    >>> y = XSoftmax.apply(x, mask, dim)
    ```"""

    @staticmethod
    def forward(self, input, mask, dim):
        self.dim = dim
        rmask = ~(mask.to(torch.bool))

        output = input.masked_fill(rmask, torch.tensor(torch.finfo(input.dtype).min))
        output = torch.softmax(output, self.dim)
        output.masked_fill_(rmask, 0)
        self.save_for_backward(output)
        return output

    @staticmethod
    def backward(self, grad_output):
        (output,) = self.saved_tensors
        inputGrad = softmax_backward_data(self, grad_output, output, self.dim, output)
        return inputGrad, None, None

    @staticmethod
    def symbolic(g, self, mask, dim):
        import torch.onnx.symbolic_helper as sym_help
        from torch.onnx.symbolic_opset9 import masked_fill, softmax

        mask_cast_value = g.op("Cast", mask, to_i=sym_help.cast_pytorch_to_onnx["Long"])
        r_mask = g.op(
            "Cast",
            g.op("Sub", g.op("Constant", value_t=torch.tensor(1, dtype=torch.int64)), mask_cast_value),
            to_i=sym_help.cast_pytorch_to_onnx["Bool"],
        )
        output = masked_fill(
            g, self, r_mask, g.op("Constant", value_t=torch.tensor(torch.finfo(self.type().dtype()).min))
        )
        output = softmax(g, output, dim)
        return masked_fill(g, output, r_mask, g.op("Constant", value_t=torch.tensor(0, dtype=torch.bool)))


# Copied from transformers.models.deberta.modeling_deberta.DropoutContext
class DropoutContext(object):
    def __init__(self):
        self.dropout = 0
        self.mask = None
        self.scale = 1
        self.reuse_mask = True


# Copied from transformers.models.deberta.modeling_deberta.get_mask
def get_mask(input, local_context):
    if not isinstance(local_context, DropoutContext):
        dropout = local_context
        mask = None
    else:
        dropout = local_context.dropout
        dropout *= local_context.scale
        mask = local_context.mask if local_context.reuse_mask else None

    if dropout > 0 and mask is None:
        mask = (1 - torch.empty_like(input).bernoulli_(1 - dropout)).to(torch.bool)

    if isinstance(local_context, DropoutContext):
        if local_context.mask is None:
            local_context.mask = mask

    return mask, dropout


# Copied from transformers.models.deberta.modeling_deberta.XDropout
class XDropout(torch.autograd.Function):
    """Optimized dropout function to save computation and memory by using mask operation instead of multiplication."""

    @staticmethod
    def forward(ctx, input, local_ctx):
        mask, dropout = get_mask(input, local_ctx)
        ctx.scale = 1.0 / (1 - dropout)
        if dropout > 0:
            ctx.save_for_backward(mask)
            return input.masked_fill(mask, 0) * ctx.scale
        else:
            return input

    @staticmethod
    def backward(ctx, grad_output):
        if ctx.scale > 1:
            (mask,) = ctx.saved_tensors
            return grad_output.masked_fill(mask, 0) * ctx.scale, None
        else:
            return grad_output, None

    @staticmethod
    def symbolic(g: torch._C.Graph, input: torch._C.Value, local_ctx: Union[float, DropoutContext]) -> torch._C.Value:
        from torch.onnx import symbolic_opset12

        dropout_p = local_ctx
        if isinstance(local_ctx, DropoutContext):
            dropout_p = local_ctx.dropout
        # StableDropout only calls this function when training.
        train = True
        # TODO: We should check if the opset_version being used to export
        # is > 12 here, but there's no good way to do that. As-is, if the
        # opset_version < 12, export will fail with a CheckerError.
        # Once https://github.com/pytorch/pytorch/issues/78391 is fixed, do something like:
        # if opset_version < 12:
        #   return torch.onnx.symbolic_opset9.dropout(g, input, dropout_p, train)
        return symbolic_opset12.dropout(g, input, dropout_p, train)


# Copied from transformers.models.deberta.modeling_deberta.StableDropout
class StableDropout(nn.Module):
    """
    Optimized dropout module for stabilizing the training

    Args:
        drop_prob (float): the dropout probabilities
    """

    def __init__(self, drop_prob):
        super().__init__()
        self.drop_prob = drop_prob
        self.count = 0
        self.context_stack = None

    def forward(self, x):
        """
        Call the module

        Args:
            x (`torch.tensor`): The input tensor to apply dropout
        """
        if self.training and self.drop_prob > 0:
            return XDropout.apply(x, self.get_context())
        return x

    def clear_context(self):
        self.count = 0
        self.context_stack = None

    def init_context(self, reuse_mask=True, scale=1):
        if self.context_stack is None:
            self.context_stack = []
        self.count = 0
        for c in self.context_stack:
            c.reuse_mask = reuse_mask
            c.scale = scale

    def get_context(self):
        if self.context_stack is not None:
            if self.count >= len(self.context_stack):
                self.context_stack.append(DropoutContext())
            ctx = self.context_stack[self.count]
            ctx.dropout = self.drop_prob
            self.count += 1
            return ctx
        else:
            return self.drop_prob


# Copied from transformers.models.deberta.modeling_deberta.DebertaSelfOutput with DebertaLayerNorm->LayerNorm
class DebertaV2SelfOutput(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.dense = nn.Linear(config.hidden_size, config.hidden_size)
        self.LayerNorm = LayerNorm(config.hidden_size, config.layer_norm_eps)
        self.dropout = StableDropout(config.hidden_dropout_prob)

    def forward(self, hidden_states, input_tensor):
        hidden_states = self.dense(hidden_states)
        hidden_states = self.dropout(hidden_states)
        hidden_states = self.LayerNorm(hidden_states + input_tensor)
        return hidden_states


# Copied from transformers.models.deberta.modeling_deberta.DebertaAttention with Deberta->DebertaV2
class DebertaV2Attention(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.self = DisentangledSelfAttention(config)
        self.output = DebertaV2SelfOutput(config)
        self.config = config

    def forward(
        self,
        hidden_states,
        attention_mask,
        output_attentions=False,
        query_states=None,
        relative_pos=None,
        rel_embeddings=None,
    ):
        self_output = self.self(
            hidden_states,
            attention_mask,
            output_attentions,
            query_states=query_states,
            relative_pos=relative_pos,
            rel_embeddings=rel_embeddings,
        )
        if output_attentions:
            self_output, att_matrix = self_output
        if query_states is None:
            query_states = hidden_states
        attention_output = self.output(self_output, query_states)

        if output_attentions:
            return (attention_output, att_matrix)
        else:
            return attention_output


# Copied from transformers.models.bert.modeling_bert.BertIntermediate with Bert->DebertaV2
class DebertaV2Intermediate(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.dense = nn.Linear(config.hidden_size, config.intermediate_size)
        if isinstance(config.hidden_act, str):
            self.intermediate_act_fn = ACT2FN[config.hidden_act]
        else:
            self.intermediate_act_fn = config.hidden_act

    def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
        hidden_states = self.dense(hidden_states)
        hidden_states = self.intermediate_act_fn(hidden_states)
        return hidden_states


# Copied from transformers.models.deberta.modeling_deberta.DebertaOutput with DebertaLayerNorm->LayerNorm
class DebertaV2Output(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.dense = nn.Linear(config.intermediate_size, config.hidden_size)
        self.LayerNorm = LayerNorm(config.hidden_size, config.layer_norm_eps)
        self.dropout = StableDropout(config.hidden_dropout_prob)
        self.config = config

    def forward(self, hidden_states, input_tensor):
        hidden_states = self.dense(hidden_states)
        hidden_states = self.dropout(hidden_states)
        hidden_states = self.LayerNorm(hidden_states + input_tensor)
        return hidden_states


# Copied from transformers.models.deberta.modeling_deberta.DebertaLayer with Deberta->DebertaV2
class DebertaV2Layer(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.attention = DebertaV2Attention(config)
        self.intermediate = DebertaV2Intermediate(config)
        self.output = DebertaV2Output(config)

    def forward(
        self,
        hidden_states,
        attention_mask,
        query_states=None,
        relative_pos=None,
        rel_embeddings=None,
        output_attentions=False,
    ):
        attention_output = self.attention(
            hidden_states,
            attention_mask,
            output_attentions=output_attentions,
            query_states=query_states,
            relative_pos=relative_pos,
            rel_embeddings=rel_embeddings,
        )
        if output_attentions:
            attention_output, att_matrix = attention_output
        intermediate_output = self.intermediate(attention_output)
        layer_output = self.output(intermediate_output, attention_output)
        if output_attentions:
            return (layer_output, att_matrix)
        else:
            return layer_output


class ConvLayer(nn.Module):
    def __init__(self, config):
        super().__init__()
        kernel_size = getattr(config, "conv_kernel_size", 3)
        groups = getattr(config, "conv_groups", 1)
        self.conv_act = getattr(config, "conv_act", "tanh")
        self.conv = nn.Conv1d(
            config.hidden_size, config.hidden_size, kernel_size, padding=(kernel_size - 1) // 2, groups=groups
        )
        self.LayerNorm = LayerNorm(config.hidden_size, config.layer_norm_eps)
        self.dropout = StableDropout(config.hidden_dropout_prob)
        self.config = config

    def forward(self, hidden_states, residual_states, input_mask):
        out = self.conv(hidden_states.permute(0, 2, 1).contiguous()).permute(0, 2, 1).contiguous()
        rmask = (1 - input_mask).bool()
        out.masked_fill_(rmask.unsqueeze(-1).expand(out.size()), 0)
        out = ACT2FN[self.conv_act](self.dropout(out))

        layer_norm_input = residual_states + out
        output = self.LayerNorm(layer_norm_input).to(layer_norm_input)

        if input_mask is None:
            output_states = output
        else:
            if input_mask.dim() != layer_norm_input.dim():
                if input_mask.dim() == 4:
                    input_mask = input_mask.squeeze(1).squeeze(1)
                input_mask = input_mask.unsqueeze(2)

            input_mask = input_mask.to(output.dtype)
            output_states = output * input_mask

        return output_states


class DebertaV2Encoder(nn.Module):
    """Modified BertEncoder with relative position bias support"""

    def __init__(self, config):
        super().__init__()

        self.layer = nn.ModuleList([DebertaV2Layer(config) for _ in range(config.num_hidden_layers)])
        self.relative_attention = getattr(config, "relative_attention", False)

        if self.relative_attention:
            self.max_relative_positions = getattr(config, "max_relative_positions", -1)
            if self.max_relative_positions < 1:
                self.max_relative_positions = config.max_position_embeddings

            self.position_buckets = getattr(config, "position_buckets", -1)
            pos_ebd_size = self.max_relative_positions * 2

            if self.position_buckets > 0:
                pos_ebd_size = self.position_buckets * 2

            self.rel_embeddings = nn.Embedding(pos_ebd_size, config.hidden_size)

        self.norm_rel_ebd = [x.strip() for x in getattr(config, "norm_rel_ebd", "none").lower().split("|")]

        if "layer_norm" in self.norm_rel_ebd:
            self.LayerNorm = LayerNorm(config.hidden_size, config.layer_norm_eps, elementwise_affine=True)

        self.conv = ConvLayer(config) if getattr(config, "conv_kernel_size", 0) > 0 else None
        self.gradient_checkpointing = False

    def get_rel_embedding(self):
        rel_embeddings = self.rel_embeddings.weight if self.relative_attention else None
        if rel_embeddings is not None and ("layer_norm" in self.norm_rel_ebd):
            rel_embeddings = self.LayerNorm(rel_embeddings)
        return rel_embeddings

    def get_attention_mask(self, attention_mask):
        if attention_mask.dim() <= 2:
            extended_attention_mask = attention_mask.unsqueeze(1).unsqueeze(2)
            attention_mask = extended_attention_mask * extended_attention_mask.squeeze(-2).unsqueeze(-1)
        elif attention_mask.dim() == 3:
            attention_mask = attention_mask.unsqueeze(1)

        return attention_mask

    def get_rel_pos(self, hidden_states, query_states=None, relative_pos=None):
        if self.relative_attention and relative_pos is None:
            q = query_states.size(-2) if query_states is not None else hidden_states.size(-2)
            relative_pos = build_relative_position(
                q,
                hidden_states.size(-2),
                bucket_size=self.position_buckets,
                max_position=self.max_relative_positions,
                device=hidden_states.device,
            )
        return relative_pos

    def forward(
        self,
        hidden_states,
        attention_mask,
        output_hidden_states=True,
        output_attentions=False,
        query_states=None,
        relative_pos=None,
        return_dict=True,
    ):
        if attention_mask.dim() <= 2:
            input_mask = attention_mask
        else:
            input_mask = attention_mask.sum(-2) > 0
        attention_mask = self.get_attention_mask(attention_mask)
        relative_pos = self.get_rel_pos(hidden_states, query_states, relative_pos)

        all_hidden_states = () if output_hidden_states else None
        all_attentions = () if output_attentions else None

        if isinstance(hidden_states, Sequence):
            next_kv = hidden_states[0]
        else:
            next_kv = hidden_states
        rel_embeddings = self.get_rel_embedding()
        output_states = next_kv
        for i, layer_module in enumerate(self.layer):
            if output_hidden_states:
                all_hidden_states = all_hidden_states + (output_states,)

            if self.gradient_checkpointing and self.training:

                def create_custom_forward(module):
                    def custom_forward(*inputs):
                        return module(*inputs, output_attentions)

                    return custom_forward

                output_states = torch.utils.checkpoint.checkpoint(
                    create_custom_forward(layer_module),
                    next_kv,
                    attention_mask,
                    query_states,
                    relative_pos,
                    rel_embeddings,
                )
            else:
                output_states = layer_module(
                    next_kv,
                    attention_mask,
                    query_states=query_states,
                    relative_pos=relative_pos,
                    rel_embeddings=rel_embeddings,
                    output_attentions=output_attentions,
                )

            if output_attentions:
                output_states, att_m = output_states

            if i == 0 and self.conv is not None:
                output_states = self.conv(hidden_states, output_states, input_mask)

            if query_states is not None:
                query_states = output_states
                if isinstance(hidden_states, Sequence):
                    next_kv = hidden_states[i + 1] if i + 1 < len(self.layer) else None
            else:
                next_kv = output_states

            if output_attentions:
                all_attentions = all_attentions + (att_m,)

        if output_hidden_states:
            all_hidden_states = all_hidden_states + (output_states,)

        if not return_dict:
            return tuple(v for v in [output_states, all_hidden_states, all_attentions] if v is not None)
        return BaseModelOutput(
            last_hidden_state=output_states, hidden_states=all_hidden_states, attentions=all_attentions
        )


def make_log_bucket_position(relative_pos, bucket_size, max_position):
    sign = torch.sign(relative_pos)
    mid = bucket_size // 2
    abs_pos = torch.where(
        (relative_pos < mid) & (relative_pos > -mid),
        torch.tensor(mid - 1).type_as(relative_pos),
        torch.abs(relative_pos),
    )
    log_pos = (
        torch.ceil(torch.log(abs_pos / mid) / torch.log(torch.tensor((max_position - 1) / mid)) * (mid - 1)) + mid
    )
    bucket_pos = torch.where(abs_pos <= mid, relative_pos.type_as(log_pos), log_pos * sign)
    return bucket_pos


def build_relative_position(query_size, key_size, bucket_size=-1, max_position=-1, device=None):
    """
    Build relative position according to the query and key

    We assume the absolute position of query \\(P_q\\) is range from (0, query_size) and the absolute position of key
    \\(P_k\\) is range from (0, key_size), The relative positions from query to key is \\(R_{q \\rightarrow k} = P_q -
    P_k\\)

    Args:
        query_size (int): the length of query
        key_size (int): the length of key
        bucket_size (int): the size of position bucket
        max_position (int): the maximum allowed absolute position
        device (`torch.device`): the device on which tensors will be created.

    Return:
        `torch.LongTensor`: A tensor with shape [1, query_size, key_size]
    """

    q_ids = torch.arange(0, query_size, device=device)
    k_ids = torch.arange(0, key_size, device=device)
    rel_pos_ids = q_ids[:, None] - k_ids[None, :]
    if bucket_size > 0 and max_position > 0:
        rel_pos_ids = make_log_bucket_position(rel_pos_ids, bucket_size, max_position)
    rel_pos_ids = rel_pos_ids.to(torch.long)
    rel_pos_ids = rel_pos_ids[:query_size, :]
    rel_pos_ids = rel_pos_ids.unsqueeze(0)
    return rel_pos_ids


@torch.jit.script
# Copied from transformers.models.deberta.modeling_deberta.c2p_dynamic_expand
def c2p_dynamic_expand(c2p_pos, query_layer, relative_pos):
    return c2p_pos.expand([query_layer.size(0), query_layer.size(1), query_layer.size(2), relative_pos.size(-1)])


@torch.jit.script
# Copied from transformers.models.deberta.modeling_deberta.p2c_dynamic_expand
def p2c_dynamic_expand(c2p_pos, query_layer, key_layer):
    return c2p_pos.expand([query_layer.size(0), query_layer.size(1), key_layer.size(-2), key_layer.size(-2)])


@torch.jit.script
# Copied from transformers.models.deberta.modeling_deberta.pos_dynamic_expand
def pos_dynamic_expand(pos_index, p2c_att, key_layer):
    return pos_index.expand(p2c_att.size()[:2] + (pos_index.size(-2), key_layer.size(-2)))


class DisentangledSelfAttention(nn.Module):
    """
    Disentangled self-attention module

    Parameters:
        config (`DebertaV2Config`):
            A model config class instance with the configuration to build a new model. The schema is similar to
            *BertConfig*, for more details, please refer [`DebertaV2Config`]

    """

    def __init__(self, config):
        super().__init__()
        if config.hidden_size % config.num_attention_heads != 0:
            raise ValueError(
                f"The hidden size ({config.hidden_size}) is not a multiple of the number of attention "
                f"heads ({config.num_attention_heads})"
            )
        self.num_attention_heads = config.num_attention_heads
        _attention_head_size = config.hidden_size // config.num_attention_heads
        self.attention_head_size = getattr(config, "attention_head_size", _attention_head_size)
        self.all_head_size = self.num_attention_heads * self.attention_head_size
        self.query_proj = nn.Linear(config.hidden_size, self.all_head_size, bias=True)
        self.key_proj = nn.Linear(config.hidden_size, self.all_head_size, bias=True)
        self.value_proj = nn.Linear(config.hidden_size, self.all_head_size, bias=True)

        self.share_att_key = getattr(config, "share_att_key", False)
        self.pos_att_type = config.pos_att_type if config.pos_att_type is not None else []
        self.relative_attention = getattr(config, "relative_attention", False)

        if self.relative_attention:
            self.position_buckets = getattr(config, "position_buckets", -1)
            self.max_relative_positions = getattr(config, "max_relative_positions", -1)
            if self.max_relative_positions < 1:
                self.max_relative_positions = config.max_position_embeddings
            self.pos_ebd_size = self.max_relative_positions
            if self.position_buckets > 0:
                self.pos_ebd_size = self.position_buckets

            self.pos_dropout = StableDropout(config.hidden_dropout_prob)

            if not self.share_att_key:
                if "c2p" in self.pos_att_type:
                    self.pos_key_proj = nn.Linear(config.hidden_size, self.all_head_size, bias=True)
                if "p2c" in self.pos_att_type:
                    self.pos_query_proj = nn.Linear(config.hidden_size, self.all_head_size)

        self.dropout = StableDropout(config.attention_probs_dropout_prob)

    def transpose_for_scores(self, x, attention_heads):
        new_x_shape = x.size()[:-1] + (attention_heads, -1)
        x = x.view(new_x_shape)
        return x.permute(0, 2, 1, 3).contiguous().view(-1, x.size(1), x.size(-1))

    def forward(
        self,
        hidden_states,
        attention_mask,
        output_attentions=False,
        query_states=None,
        relative_pos=None,
        rel_embeddings=None,
    ):
        """
        Call the module

        Args:
            hidden_states (`torch.FloatTensor`):
                Input states to the module usually the output from previous layer, it will be the Q,K and V in
                *Attention(Q,K,V)*

            attention_mask (`torch.BoolTensor`):
                An attention mask matrix of shape [*B*, *N*, *N*] where *B* is the batch size, *N* is the maximum
                sequence length in which element [i,j] = *1* means the *i* th token in the input can attend to the *j*
                th token.

            output_attentions (`bool`, optional):
                Whether return the attention matrix.

            query_states (`torch.FloatTensor`, optional):
                The *Q* state in *Attention(Q,K,V)*.

            relative_pos (`torch.LongTensor`):
                The relative position encoding between the tokens in the sequence. It's of shape [*B*, *N*, *N*] with
                values ranging in [*-max_relative_positions*, *max_relative_positions*].

            rel_embeddings (`torch.FloatTensor`):
                The embedding of relative distances. It's a tensor of shape [\\(2 \\times
                \\text{max_relative_positions}\\), *hidden_size*].


        """
        if query_states is None:
            query_states = hidden_states
        query_layer = self.transpose_for_scores(self.query_proj(query_states), self.num_attention_heads)
        key_layer = self.transpose_for_scores(self.key_proj(hidden_states), self.num_attention_heads)
        value_layer = self.transpose_for_scores(self.value_proj(hidden_states), self.num_attention_heads)

        rel_att = None
        # Take the dot product between "query" and "key" to get the raw attention scores.
        scale_factor = 1
        if "c2p" in self.pos_att_type:
            scale_factor += 1
        if "p2c" in self.pos_att_type:
            scale_factor += 1
        scale = torch.sqrt(torch.tensor(query_layer.size(-1), dtype=torch.float) * scale_factor)
        attention_scores = torch.bmm(query_layer, key_layer.transpose(-1, -2)) / scale.to(dtype=query_layer.dtype)
        if self.relative_attention:
            rel_embeddings = self.pos_dropout(rel_embeddings)
            rel_att = self.disentangled_attention_bias(
                query_layer, key_layer, relative_pos, rel_embeddings, scale_factor
            )

        if rel_att is not None:
            attention_scores = attention_scores + rel_att
        attention_scores = attention_scores
        attention_scores = attention_scores.view(
            -1, self.num_attention_heads, attention_scores.size(-2), attention_scores.size(-1)
        )

        # bsz x height x length x dimension
        attention_probs = XSoftmax.apply(attention_scores, attention_mask, -1)
        attention_probs = self.dropout(attention_probs)
        context_layer = torch.bmm(
            attention_probs.view(-1, attention_probs.size(-2), attention_probs.size(-1)), value_layer
        )
        context_layer = (
            context_layer.view(-1, self.num_attention_heads, context_layer.size(-2), context_layer.size(-1))
            .permute(0, 2, 1, 3)
            .contiguous()
        )
        new_context_layer_shape = context_layer.size()[:-2] + (-1,)
        context_layer = context_layer.view(new_context_layer_shape)
        if output_attentions:
            return (context_layer, attention_probs)
        else:
            return context_layer

    def disentangled_attention_bias(self, query_layer, key_layer, relative_pos, rel_embeddings, scale_factor):
        if relative_pos is None:
            q = query_layer.size(-2)
            relative_pos = build_relative_position(
                q,
                key_layer.size(-2),
                bucket_size=self.position_buckets,
                max_position=self.max_relative_positions,
                device=query_layer.device,
            )
        if relative_pos.dim() == 2:
            relative_pos = relative_pos.unsqueeze(0).unsqueeze(0)
        elif relative_pos.dim() == 3:
            relative_pos = relative_pos.unsqueeze(1)
        # bsz x height x query x key
        elif relative_pos.dim() != 4:
            raise ValueError(f"Relative position ids must be of dim 2 or 3 or 4. {relative_pos.dim()}")

        att_span = self.pos_ebd_size
        relative_pos = relative_pos.long().to(query_layer.device)

        rel_embeddings = rel_embeddings[0 : att_span * 2, :].unsqueeze(0)
        if self.share_att_key:
            pos_query_layer = self.transpose_for_scores(
                self.query_proj(rel_embeddings), self.num_attention_heads
            ).repeat(query_layer.size(0) // self.num_attention_heads, 1, 1)
            pos_key_layer = self.transpose_for_scores(self.key_proj(rel_embeddings), self.num_attention_heads).repeat(
                query_layer.size(0) // self.num_attention_heads, 1, 1
            )
        else:
            if "c2p" in self.pos_att_type:
                pos_key_layer = self.transpose_for_scores(
                    self.pos_key_proj(rel_embeddings), self.num_attention_heads
                ).repeat(
                    query_layer.size(0) // self.num_attention_heads, 1, 1
                )  # .split(self.all_head_size, dim=-1)
            if "p2c" in self.pos_att_type:
                pos_query_layer = self.transpose_for_scores(
                    self.pos_query_proj(rel_embeddings), self.num_attention_heads
                ).repeat(
                    query_layer.size(0) // self.num_attention_heads, 1, 1
                )  # .split(self.all_head_size, dim=-1)

        score = 0
        # content->position
        if "c2p" in self.pos_att_type:
            scale = torch.sqrt(torch.tensor(pos_key_layer.size(-1), dtype=torch.float) * scale_factor)
            c2p_att = torch.bmm(query_layer, pos_key_layer.transpose(-1, -2))
            c2p_pos = torch.clamp(relative_pos + att_span, 0, att_span * 2 - 1)
            c2p_att = torch.gather(
                c2p_att,
                dim=-1,
                index=c2p_pos.squeeze(0).expand([query_layer.size(0), query_layer.size(1), relative_pos.size(-1)]),
            )
            score += c2p_att / scale.to(dtype=c2p_att.dtype)

        # position->content
        if "p2c" in self.pos_att_type:
            scale = torch.sqrt(torch.tensor(pos_query_layer.size(-1), dtype=torch.float) * scale_factor)
            if key_layer.size(-2) != query_layer.size(-2):
                r_pos = build_relative_position(
                    key_layer.size(-2),
                    key_layer.size(-2),
                    bucket_size=self.position_buckets,
                    max_position=self.max_relative_positions,
                    device=query_layer.device,
                )
                r_pos = r_pos.unsqueeze(0)
            else:
                r_pos = relative_pos

            p2c_pos = torch.clamp(-r_pos + att_span, 0, att_span * 2 - 1)
            p2c_att = torch.bmm(key_layer, pos_query_layer.transpose(-1, -2))
            p2c_att = torch.gather(
                p2c_att,
                dim=-1,
                index=p2c_pos.squeeze(0).expand([query_layer.size(0), key_layer.size(-2), key_layer.size(-2)]),
            ).transpose(-1, -2)
            score += p2c_att / scale.to(dtype=p2c_att.dtype)

        return score


# Copied from transformers.models.deberta.modeling_deberta.DebertaEmbeddings with DebertaLayerNorm->LayerNorm
class DebertaV2Embeddings(nn.Module):
    """Construct the embeddings from word, position and token_type embeddings."""

    def __init__(self, config):
        super().__init__()
        pad_token_id = getattr(config, "pad_token_id", 0)
        self.embedding_size = getattr(config, "embedding_size", config.hidden_size)
        self.word_embeddings = nn.Embedding(config.vocab_size, self.embedding_size, padding_idx=pad_token_id)

        self.position_biased_input = getattr(config, "position_biased_input", True)
        if not self.position_biased_input:
            self.position_embeddings = None
        else:
            self.position_embeddings = nn.Embedding(config.max_position_embeddings, self.embedding_size)

        if config.type_vocab_size > 0:
            self.token_type_embeddings = nn.Embedding(config.type_vocab_size, self.embedding_size)

        if self.embedding_size != config.hidden_size:
            self.embed_proj = nn.Linear(self.embedding_size, config.hidden_size, bias=False)
        self.LayerNorm = LayerNorm(config.hidden_size, config.layer_norm_eps)
        self.dropout = StableDropout(config.hidden_dropout_prob)
        self.config = config

        # position_ids (1, len position emb) is contiguous in memory and exported when serialized
        self.register_buffer("position_ids", torch.arange(config.max_position_embeddings).expand((1, -1)))

    def forward(self, input_ids=None, token_type_ids=None, position_ids=None, mask=None, inputs_embeds=None):
        if input_ids is not None:
            input_shape = input_ids.size()
        else:
            input_shape = inputs_embeds.size()[:-1]

        seq_length = input_shape[1]

        if position_ids is None:
            position_ids = self.position_ids[:, :seq_length]

        if token_type_ids is None:
            token_type_ids = torch.zeros(input_shape, dtype=torch.long, device=self.position_ids.device)

        if inputs_embeds is None:
            inputs_embeds = self.word_embeddings(input_ids)

        if self.position_embeddings is not None:
            position_embeddings = self.position_embeddings(position_ids.long())
        else:
            position_embeddings = torch.zeros_like(inputs_embeds)

        embeddings = inputs_embeds
        if self.position_biased_input:
            embeddings += position_embeddings
        if self.config.type_vocab_size > 0:
            token_type_embeddings = self.token_type_embeddings(token_type_ids)
            embeddings += token_type_embeddings

        if self.embedding_size != self.config.hidden_size:
            embeddings = self.embed_proj(embeddings)

        embeddings = self.LayerNorm(embeddings)

        if mask is not None:
            if mask.dim() != embeddings.dim():
                if mask.dim() == 4:
                    mask = mask.squeeze(1).squeeze(1)
                mask = mask.unsqueeze(2)
            mask = mask.to(embeddings.dtype)

            embeddings = embeddings * mask

        embeddings = self.dropout(embeddings)
        return embeddings


# Copied from transformers.models.deberta.modeling_deberta.DebertaPreTrainedModel with Deberta->DebertaV2
class DebertaV2PreTrainedModel(PreTrainedModel):
    """
    An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained
    models.
    """

    config_class = DebertaV2Config
    base_model_prefix = "deberta"
    _keys_to_ignore_on_load_missing = ["position_ids"]
    _keys_to_ignore_on_load_unexpected = ["position_embeddings"]
    supports_gradient_checkpointing = True

    def _init_weights(self, module):
        """Initialize the weights."""
        if isinstance(module, nn.Linear):
            # Slightly different from the TF version which uses truncated_normal for initialization
            # cf https://github.com/pytorch/pytorch/pull/5617
            module.weight.data.normal_(mean=0.0, std=self.config.initializer_range)
            if module.bias is not None:
                module.bias.data.zero_()
        elif isinstance(module, nn.Embedding):
            module.weight.data.normal_(mean=0.0, std=self.config.initializer_range)
            if module.padding_idx is not None:
                module.weight.data[module.padding_idx].zero_()

    def _set_gradient_checkpointing(self, module, value=False):
        if isinstance(module, DebertaV2Encoder):
            module.gradient_checkpointing = value


DEBERTA_START_DOCSTRING = r"""
    The DeBERTa model was proposed in [DeBERTa: Decoding-enhanced BERT with Disentangled
    Attention](https://arxiv.org/abs/2006.03654) by Pengcheng He, Xiaodong Liu, Jianfeng Gao, Weizhu Chen. It's build
    on top of BERT/RoBERTa with two improvements, i.e. disentangled attention and enhanced mask decoder. With those two
    improvements, it out perform BERT/RoBERTa on a majority of tasks with 80GB pretraining data.

    This model is also a PyTorch [torch.nn.Module](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) subclass.
    Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage
    and behavior.


    Parameters:
        config ([`DebertaV2Config`]): Model configuration class with all the parameters of the model.
            Initializing with a config file does not load the weights associated with the model, only the
            configuration. Check out the [`~PreTrainedModel.from_pretrained`] method to load the model weights.
"""

DEBERTA_INPUTS_DOCSTRING = r"""
    Args:
        input_ids (`torch.LongTensor` of shape `({0})`):
            Indices of input sequence tokens in the vocabulary.

            Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and
            [`PreTrainedTokenizer.__call__`] for details.

            [What are input IDs?](../glossary#input-ids)
        attention_mask (`torch.FloatTensor` of shape `({0})`, *optional*):
            Mask to avoid performing attention on padding token indices. Mask values selected in `[0, 1]`:

            - 1 for tokens that are **not masked**,
            - 0 for tokens that are **masked**.

            [What are attention masks?](../glossary#attention-mask)
        token_type_ids (`torch.LongTensor` of shape `({0})`, *optional*):
            Segment token indices to indicate first and second portions of the inputs. Indices are selected in `[0,
            1]`:

            - 0 corresponds to a *sentence A* token,
            - 1 corresponds to a *sentence B* token.

            [What are token type IDs?](../glossary#token-type-ids)
        position_ids (`torch.LongTensor` of shape `({0})`, *optional*):
            Indices of positions of each input sequence tokens in the position embeddings. Selected in the range `[0,
            config.max_position_embeddings - 1]`.

            [What are position IDs?](../glossary#position-ids)
        inputs_embeds (`torch.FloatTensor` of shape `({0}, hidden_size)`, *optional*):
            Optionally, instead of passing `input_ids` you can choose to directly pass an embedded representation. This
            is useful if you want more control over how to convert *input_ids* indices into associated vectors than the
            model's internal embedding lookup matrix.
        output_attentions (`bool`, *optional*):
            Whether or not to return the attentions tensors of all attention layers. See `attentions` under returned
            tensors for more detail.
        output_hidden_states (`bool`, *optional*):
            Whether or not to return the hidden states of all layers. See `hidden_states` under returned tensors for
            more detail.
        return_dict (`bool`, *optional*):
            Whether or not to return a [`~utils.ModelOutput`] instead of a plain tuple.
"""


@add_start_docstrings(
    "The bare DeBERTa Model transformer outputting raw hidden-states without any specific head on top.",
    DEBERTA_START_DOCSTRING,
)
# Copied from transformers.models.deberta.modeling_deberta.DebertaModel with Deberta->DebertaV2
class DebertaV2Model(DebertaV2PreTrainedModel):
    def __init__(self, config):
        super().__init__(config)

        self.embeddings = DebertaV2Embeddings(config)
        self.encoder = DebertaV2Encoder(config)
        self.z_steps = 0
        self.config = config
        # Initialize weights and apply final processing
        self.post_init()

    def get_input_embeddings(self):
        return self.embeddings.word_embeddings

    def set_input_embeddings(self, new_embeddings):
        self.embeddings.word_embeddings = new_embeddings

    def _prune_heads(self, heads_to_prune):
        """
        Prunes heads of the model. heads_to_prune: dict of {layer_num: list of heads to prune in this layer} See base
        class PreTrainedModel
        """
        raise NotImplementedError("The prune function is not implemented in DeBERTa model.")

    @add_start_docstrings_to_model_forward(DEBERTA_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
    @add_code_sample_docstrings(
        checkpoint=_CHECKPOINT_FOR_DOC,
        output_type=BaseModelOutput,
        config_class=_CONFIG_FOR_DOC,
    )
    def forward(
        self,
        input_ids: Optional[torch.Tensor] = None,
        attention_mask: Optional[torch.Tensor] = None,
        token_type_ids: Optional[torch.Tensor] = None,
        position_ids: Optional[torch.Tensor] = None,
        inputs_embeds: Optional[torch.Tensor] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        return_dict: Optional[bool] = None,
    ) -> Union[Tuple, BaseModelOutput]:
        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
        output_hidden_states = (
            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
        )
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict

        if input_ids is not None and inputs_embeds is not None:
            raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
        elif input_ids is not None:
            input_shape = input_ids.size()
        elif inputs_embeds is not None:
            input_shape = inputs_embeds.size()[:-1]
        else:
            raise ValueError("You have to specify either input_ids or inputs_embeds")

        device = input_ids.device if input_ids is not None else inputs_embeds.device

        if attention_mask is None:
            attention_mask = torch.ones(input_shape, device=device)
        if token_type_ids is None:
            token_type_ids = torch.zeros(input_shape, dtype=torch.long, device=device)

        embedding_output = self.embeddings(
            input_ids=input_ids,
            token_type_ids=token_type_ids,
            position_ids=position_ids,
            mask=attention_mask,
            inputs_embeds=inputs_embeds,
        )

        encoder_outputs = self.encoder(
            embedding_output,
            attention_mask,
            output_hidden_states=True,
            output_attentions=output_attentions,
            return_dict=return_dict,
        )
        encoded_layers = encoder_outputs[1]

        if self.z_steps > 1:
            hidden_states = encoded_layers[-2]
            layers = [self.encoder.layer[-1] for _ in range(self.z_steps)]
            query_states = encoded_layers[-1]
            rel_embeddings = self.encoder.get_rel_embedding()
            attention_mask = self.encoder.get_attention_mask(attention_mask)
            rel_pos = self.encoder.get_rel_pos(embedding_output)
            for layer in layers[1:]:
                query_states = layer(
                    hidden_states,
                    attention_mask,
                    output_attentions=False,
                    query_states=query_states,
                    relative_pos=rel_pos,
                    rel_embeddings=rel_embeddings,
                )
                encoded_layers.append(query_states)

        sequence_output = encoded_layers[-1]

        if not return_dict:
            return (sequence_output,) + encoder_outputs[(1 if output_hidden_states else 2) :]

        return BaseModelOutput(
            last_hidden_state=sequence_output,
            hidden_states=encoder_outputs.hidden_states if output_hidden_states else None,
            attentions=encoder_outputs.attentions,
        )


@add_start_docstrings("""DeBERTa Model with a `language modeling` head on top.""", DEBERTA_START_DOCSTRING)
class DebertaV2ForMaskedLM(DebertaV2PreTrainedModel):
    _keys_to_ignore_on_load_unexpected = [r"pooler"]
    _keys_to_ignore_on_load_missing = [r"position_ids", r"predictions.decoder.bias", "cls.predictions.decoder.weight"]

    def __init__(self, config):
        super().__init__(config)

        self.deberta = DebertaV2Model(config)
        self.cls = DebertaV2OnlyMLMHead(config)

        # Initialize weights and apply final processing
        self.post_init()

    def get_output_embeddings(self):
        return self.cls.predictions.decoder

    def set_output_embeddings(self, new_embeddings):
        self.cls.predictions.decoder = new_embeddings

    @add_start_docstrings_to_model_forward(DEBERTA_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
    @add_code_sample_docstrings(
        checkpoint=_CHECKPOINT_FOR_DOC,
        output_type=MaskedLMOutput,
        config_class=_CONFIG_FOR_DOC,
        mask="[MASK]",
    )
    # Copied from transformers.models.deberta.modeling_deberta.DebertaForMaskedLM.forward with Deberta->DebertaV2
    def forward(
        self,
        input_ids: Optional[torch.Tensor] = None,
        attention_mask: Optional[torch.Tensor] = None,
        token_type_ids: Optional[torch.Tensor] = None,
        position_ids: Optional[torch.Tensor] = None,
        inputs_embeds: Optional[torch.Tensor] = None,
        labels: Optional[torch.Tensor] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        return_dict: Optional[bool] = None,
    ) -> Union[Tuple, MaskedLMOutput]:
        r"""
        labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
            Labels for computing the masked language modeling loss. Indices should be in `[-100, 0, ...,
            config.vocab_size]` (see `input_ids` docstring) Tokens with indices set to `-100` are ignored (masked), the
            loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]`
        """

        return_dict = return_dict if return_dict is not None else self.config.use_return_dict

        outputs = self.deberta(
            input_ids,
            attention_mask=attention_mask,
            token_type_ids=token_type_ids,
            position_ids=position_ids,
            inputs_embeds=inputs_embeds,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
            return_dict=return_dict,
        )

        sequence_output = outputs[0]
        prediction_scores = self.cls(sequence_output)

        masked_lm_loss = None
        if labels is not None:
            loss_fct = CrossEntropyLoss()  # -100 index = padding token
            masked_lm_loss = loss_fct(prediction_scores.view(-1, self.config.vocab_size), labels.view(-1))

        if not return_dict:
            output = (prediction_scores,) + outputs[1:]
            return ((masked_lm_loss,) + output) if masked_lm_loss is not None else output

        return MaskedLMOutput(
            loss=masked_lm_loss,
            logits=prediction_scores,
            hidden_states=outputs.hidden_states,
            attentions=outputs.attentions,
        )


# copied from transformers.models.bert.BertPredictionHeadTransform with bert -> deberta
class DebertaV2PredictionHeadTransform(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.dense = nn.Linear(config.hidden_size, config.hidden_size)
        if isinstance(config.hidden_act, str):
            self.transform_act_fn = ACT2FN[config.hidden_act]
        else:
            self.transform_act_fn = config.hidden_act
        self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)

    def forward(self, hidden_states):
        hidden_states = self.dense(hidden_states)
        hidden_states = self.transform_act_fn(hidden_states)
        hidden_states = self.LayerNorm(hidden_states)
        return hidden_states


# copied from transformers.models.bert.BertLMPredictionHead with bert -> deberta
class DebertaV2LMPredictionHead(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.transform = DebertaV2PredictionHeadTransform(config)

        # The output weights are the same as the input embeddings, but there is
        # an output-only bias for each token.
        self.decoder = nn.Linear(config.hidden_size, config.vocab_size, bias=False)

        self.bias = nn.Parameter(torch.zeros(config.vocab_size))

        # Need a link between the two variables so that the bias is correctly resized with `resize_token_embeddings`
        self.decoder.bias = self.bias

    def forward(self, hidden_states):
        hidden_states = self.transform(hidden_states)
        hidden_states = self.decoder(hidden_states)
        return hidden_states


# copied from transformers.models.bert.BertOnlyMLMHead with bert -> deberta
class DebertaV2OnlyMLMHead(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.predictions = DebertaV2LMPredictionHead(config)

    def forward(self, sequence_output):
        prediction_scores = self.predictions(sequence_output)
        return prediction_scores


@add_start_docstrings(
    """
    DeBERTa Model transformer with a sequence classification/regression head on top (a linear layer on top of the
    pooled output) e.g. for GLUE tasks.
    """,
    DEBERTA_START_DOCSTRING,
)
class DebertaV2ForSequenceClassification(DebertaV2PreTrainedModel):
    def __init__(self, config):
        super().__init__(config)

        num_labels = getattr(config, "num_labels", 2)
        self.num_labels = num_labels

        self.deberta = DebertaV2Model(config)
        self.pooler = ContextPooler(config)
        output_dim = self.pooler.output_dim

        self.classifier = nn.Linear(output_dim, num_labels)
        drop_out = getattr(config, "cls_dropout", None)
        drop_out = self.config.hidden_dropout_prob if drop_out is None else drop_out
        self.dropout = StableDropout(drop_out)

        # Initialize weights and apply final processing
        self.post_init()

    def get_input_embeddings(self):
        return self.deberta.get_input_embeddings()

    def set_input_embeddings(self, new_embeddings):
        self.deberta.set_input_embeddings(new_embeddings)

    @add_start_docstrings_to_model_forward(DEBERTA_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
    @add_code_sample_docstrings(
        checkpoint=_CHECKPOINT_FOR_DOC,
        output_type=SequenceClassifierOutput,
        config_class=_CONFIG_FOR_DOC,
    )
    # Copied from transformers.models.deberta.modeling_deberta.DebertaForSequenceClassification.forward with Deberta->DebertaV2
    def forward(
        self,
        input_ids: Optional[torch.Tensor] = None,
        attention_mask: Optional[torch.Tensor] = None,
        token_type_ids: Optional[torch.Tensor] = None,
        position_ids: Optional[torch.Tensor] = None,
        inputs_embeds: Optional[torch.Tensor] = None,
        labels: Optional[torch.Tensor] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        return_dict: Optional[bool] = None,
    ) -> Union[Tuple, SequenceClassifierOutput]:
        r"""
        labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
            Labels for computing the sequence classification/regression loss. Indices should be in `[0, ...,
            config.num_labels - 1]`. If `config.num_labels == 1` a regression loss is computed (Mean-Square loss), If
            `config.num_labels > 1` a classification loss is computed (Cross-Entropy).
        """
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict

        outputs = self.deberta(
            input_ids,
            token_type_ids=token_type_ids,
            attention_mask=attention_mask,
            position_ids=position_ids,
            inputs_embeds=inputs_embeds,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
            return_dict=return_dict,
        )

        encoder_layer = outputs[0]
        pooled_output = self.pooler(encoder_layer)
        pooled_output = self.dropout(pooled_output)
        logits = self.classifier(pooled_output)

        loss = None
        if labels is not None:
            if self.config.problem_type is None:
                if self.num_labels == 1:
                    # regression task
                    loss_fn = nn.MSELoss()
                    logits = logits.view(-1).to(labels.dtype)
                    loss = loss_fn(logits, labels.view(-1))
                elif labels.dim() == 1 or labels.size(-1) == 1:
                    label_index = (labels >= 0).nonzero()
                    labels = labels.long()
                    if label_index.size(0) > 0:
                        labeled_logits = torch.gather(
                            logits, 0, label_index.expand(label_index.size(0), logits.size(1))
                        )
                        labels = torch.gather(labels, 0, label_index.view(-1))
                        loss_fct = CrossEntropyLoss()
                        loss = loss_fct(labeled_logits.view(-1, self.num_labels).float(), labels.view(-1))
                    else:
                        loss = torch.tensor(0).to(logits)
                else:
                    log_softmax = nn.LogSoftmax(-1)
                    loss = -((log_softmax(logits) * labels).sum(-1)).mean()
            elif self.config.problem_type == "regression":
                loss_fct = MSELoss()
                if self.num_labels == 1:
                    loss = loss_fct(logits.squeeze(), labels.squeeze())
                else:
                    loss = loss_fct(logits, labels)
            elif self.config.problem_type == "single_label_classification":
                loss_fct = CrossEntropyLoss()
                loss = loss_fct(logits.view(-1, self.num_labels), labels.view(-1))
            elif self.config.problem_type == "multi_label_classification":
                loss_fct = BCEWithLogitsLoss()
                loss = loss_fct(logits, labels)
        if not return_dict:
            output = (logits,) + outputs[1:]
            return ((loss,) + output) if loss is not None else output

        return SequenceClassifierOutput(
            loss=loss, logits=logits, hidden_states=outputs.hidden_states, attentions=outputs.attentions
        )


@add_start_docstrings(
    """
    DeBERTa Model with a token classification head on top (a linear layer on top of the hidden-states output) e.g. for
    Named-Entity-Recognition (NER) tasks.
    """,
    DEBERTA_START_DOCSTRING,
)
# Copied from transformers.models.deberta.modeling_deberta.DebertaForTokenClassification with Deberta->DebertaV2
class DebertaV2ForTokenClassification(DebertaV2PreTrainedModel):
    _keys_to_ignore_on_load_unexpected = [r"pooler"]

    def __init__(self, config):
        super().__init__(config)
        self.num_labels = config.num_labels

        self.deberta = DebertaV2Model(config)
        self.dropout = nn.Dropout(config.hidden_dropout_prob)
        self.classifier = nn.Linear(config.hidden_size, config.num_labels)

        # Initialize weights and apply final processing
        self.post_init()

    @add_start_docstrings_to_model_forward(DEBERTA_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
    @add_code_sample_docstrings(
        checkpoint=_CHECKPOINT_FOR_DOC,
        output_type=TokenClassifierOutput,
        config_class=_CONFIG_FOR_DOC,
    )
    def forward(
        self,
        input_ids: Optional[torch.Tensor] = None,
        attention_mask: Optional[torch.Tensor] = None,
        token_type_ids: Optional[torch.Tensor] = None,
        position_ids: Optional[torch.Tensor] = None,
        inputs_embeds: Optional[torch.Tensor] = None,
        labels: Optional[torch.Tensor] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        return_dict: Optional[bool] = None,
    ) -> Union[Tuple, TokenClassifierOutput]:
        r"""
        labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
            Labels for computing the token classification loss. Indices should be in `[0, ..., config.num_labels - 1]`.
        """
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict

        outputs = self.deberta(
            input_ids,
            attention_mask=attention_mask,
            token_type_ids=token_type_ids,
            position_ids=position_ids,
            inputs_embeds=inputs_embeds,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
            return_dict=return_dict,
        )

        sequence_output = outputs[0]

        sequence_output = self.dropout(sequence_output)
        logits = self.classifier(sequence_output)

        loss = None
        if labels is not None:
            loss_fct = CrossEntropyLoss()
            loss = loss_fct(logits.view(-1, self.num_labels), labels.view(-1))

        if not return_dict:
            output = (logits,) + outputs[1:]
            return ((loss,) + output) if loss is not None else output

        return TokenClassifierOutput(
            loss=loss, logits=logits, hidden_states=outputs.hidden_states, attentions=outputs.attentions
        )
    
class TokenClassifierRegressionOutput(ModelOutput):
    """
    Base class for outputs of token classification models.

    Args:
        loss (`torch.FloatTensor` of shape `(1,)`, *optional*, returned when `labels` is provided) :
            Classification loss.
        logits (`torch.FloatTensor` of shape `(batch_size, sequence_length, config.num_labels)`):
            Classification scores (before SoftMax).
        hidden_states (`tuple(torch.FloatTensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
            Tuple of `torch.FloatTensor` (one for the output of the embeddings, if the model has an embedding layer, +
            one for the output of each layer) of shape `(batch_size, sequence_length, hidden_size)`.

            Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.
        attentions (`tuple(torch.FloatTensor)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
            Tuple of `torch.FloatTensor` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
            sequence_length)`.

            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
            heads.
    """

    loss: Optional[torch.FloatTensor] = None
    logits: torch.FloatTensor = None
    values: torch.FloatTensor = None
    hidden_states: Optional[Tuple[torch.FloatTensor]] = None
    attentions: Optional[Tuple[torch.FloatTensor]] = None
    
class DebertaV2ForTokenClassificationRegression(DebertaV2PreTrainedModel):
    _keys_to_ignore_on_load_unexpected = [r"pooler"]

    def __init__(self, config):
        super().__init__(config)
        self.num_labels = 4

        self.deberta = DebertaV2Model(config)
        self.dropout = nn.Dropout(config.hidden_dropout_prob)
        
        self.hidden1 = nn.Linear(config.hidden_size, config.hidden_size)
        self.classifier = nn.Linear(config.hidden_size, self.num_labels)
        
        self.hidden2 = nn.Linear(config.hidden_size, config.hidden_size)
        self.regressor = nn.Linear(config.hidden_size, 1)

        # Initialize weights and apply final processing
        self.post_init()

    @add_start_docstrings_to_model_forward(DEBERTA_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
    @add_code_sample_docstrings(
        checkpoint=_CHECKPOINT_FOR_DOC,
        output_type=TokenClassifierOutput,
        config_class=_CONFIG_FOR_DOC,
    )
    def forward(
        self,
        input_ids: Optional[torch.Tensor] = None,
        attention_mask: Optional[torch.Tensor] = None,
        token_type_ids: Optional[torch.Tensor] = None,
        position_ids: Optional[torch.Tensor] = None,
        inputs_embeds: Optional[torch.Tensor] = None,
        labels: Optional[torch.Tensor] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        return_dict: Optional[bool] = None,
    ) -> Union[Tuple, TokenClassifierOutput]:
        r"""
        labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
            Labels for computing the token classification loss. Indices should be in `[0, ..., config.num_labels - 1]`.
        """
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict

        outputs = self.deberta(
            input_ids,
            attention_mask=attention_mask,
            token_type_ids=token_type_ids,
            position_ids=position_ids,
            inputs_embeds=inputs_embeds,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
            return_dict=return_dict,
        )

        sequence_output = outputs[0]

        sequence_output = self.dropout(sequence_output)
        
        logits = self.classifier(self.hidden1(sequence_output))
        values = self.regressor(self.hidden2(sequence_output))

        loss = None
        if labels is not None:
            loss_fct = CrossEntropyLoss()
            loss = loss_fct(logits.view(-1, self.num_labels), labels.view(-1))

        if not return_dict:
            output = (logits,) + outputs[1:]
            return ((loss,) + output) if loss is not None else output

        return TokenClassifierRegressionOutput(
            loss=loss, logits=logits, values=values, hidden_states=outputs.hidden_states, attentions=outputs.attentions
        )


@add_start_docstrings(
    """
    DeBERTa Model with a span classification head on top for extractive question-answering tasks like SQuAD (a linear
    layers on top of the hidden-states output to compute `span start logits` and `span end logits`).
    """,
    DEBERTA_START_DOCSTRING,
)
class DebertaV2ForQuestionAnswering(DebertaV2PreTrainedModel):
    _keys_to_ignore_on_load_unexpected = [r"pooler"]

    def __init__(self, config):
        super().__init__(config)
        self.num_labels = config.num_labels

        self.deberta = DebertaV2Model(config)
        self.qa_outputs = nn.Linear(config.hidden_size, config.num_labels)

        # Initialize weights and apply final processing
        self.post_init()

    @add_start_docstrings_to_model_forward(DEBERTA_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
    @add_code_sample_docstrings(
        checkpoint=_CHECKPOINT_FOR_DOC,
        output_type=QuestionAnsweringModelOutput,
        config_class=_CONFIG_FOR_DOC,
        qa_target_start_index=_QA_TARGET_START_INDEX,
        qa_target_end_index=_QA_TARGET_END_INDEX,
    )
    # Copied from transformers.models.deberta.modeling_deberta.DebertaForQuestionAnswering.forward with Deberta->DebertaV2
    def forward(
        self,
        input_ids: Optional[torch.Tensor] = None,
        attention_mask: Optional[torch.Tensor] = None,
        token_type_ids: Optional[torch.Tensor] = None,
        position_ids: Optional[torch.Tensor] = None,
        inputs_embeds: Optional[torch.Tensor] = None,
        start_positions: Optional[torch.Tensor] = None,
        end_positions: Optional[torch.Tensor] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        return_dict: Optional[bool] = None,
    ) -> Union[Tuple, QuestionAnsweringModelOutput]:
        r"""
        start_positions (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
            Labels for position (index) of the start of the labelled span for computing the token classification loss.
            Positions are clamped to the length of the sequence (`sequence_length`). Position outside of the sequence
            are not taken into account for computing the loss.
        end_positions (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
            Labels for position (index) of the end of the labelled span for computing the token classification loss.
            Positions are clamped to the length of the sequence (`sequence_length`). Position outside of the sequence
            are not taken into account for computing the loss.
        """
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict

        outputs = self.deberta(
            input_ids,
            attention_mask=attention_mask,
            token_type_ids=token_type_ids,
            position_ids=position_ids,
            inputs_embeds=inputs_embeds,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
            return_dict=return_dict,
        )

        sequence_output = outputs[0]

        logits = self.qa_outputs(sequence_output)
        start_logits, end_logits = logits.split(1, dim=-1)
        start_logits = start_logits.squeeze(-1).contiguous()
        end_logits = end_logits.squeeze(-1).contiguous()

        total_loss = None
        if start_positions is not None and end_positions is not None:
            # If we are on multi-GPU, split add a dimension
            if len(start_positions.size()) > 1:
                start_positions = start_positions.squeeze(-1)
            if len(end_positions.size()) > 1:
                end_positions = end_positions.squeeze(-1)
            # sometimes the start/end positions are outside our model inputs, we ignore these terms
            ignored_index = start_logits.size(1)
            start_positions = start_positions.clamp(0, ignored_index)
            end_positions = end_positions.clamp(0, ignored_index)

            loss_fct = CrossEntropyLoss(ignore_index=ignored_index)
            start_loss = loss_fct(start_logits, start_positions)
            end_loss = loss_fct(end_logits, end_positions)
            total_loss = (start_loss + end_loss) / 2

        if not return_dict:
            output = (start_logits, end_logits) + outputs[1:]
            return ((total_loss,) + output) if total_loss is not None else output

        return QuestionAnsweringModelOutput(
            loss=total_loss,
            start_logits=start_logits,
            end_logits=end_logits,
            hidden_states=outputs.hidden_states,
            attentions=outputs.attentions,
        )


@add_start_docstrings(
    """
    DeBERTa Model with a multiple choice classification head on top (a linear layer on top of the pooled output and a
    softmax) e.g. for RocStories/SWAG tasks.
    """,
    DEBERTA_START_DOCSTRING,
)
class DebertaV2ForMultipleChoice(DebertaV2PreTrainedModel):
    def __init__(self, config):
        super().__init__(config)

        num_labels = getattr(config, "num_labels", 2)
        self.num_labels = num_labels

        self.deberta = DebertaV2Model(config)
        self.pooler = ContextPooler(config)
        output_dim = self.pooler.output_dim

        self.classifier = nn.Linear(output_dim, 1)
        drop_out = getattr(config, "cls_dropout", None)
        drop_out = self.config.hidden_dropout_prob if drop_out is None else drop_out
        self.dropout = StableDropout(drop_out)

        self.init_weights()

    def get_input_embeddings(self):
        return self.deberta.get_input_embeddings()

    def set_input_embeddings(self, new_embeddings):
        self.deberta.set_input_embeddings(new_embeddings)

    @add_start_docstrings_to_model_forward(DEBERTA_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
    @add_code_sample_docstrings(
        checkpoint=_CHECKPOINT_FOR_DOC,
        output_type=MultipleChoiceModelOutput,
        config_class=_CONFIG_FOR_DOC,
    )
    def forward(
        self,
        input_ids: Optional[torch.Tensor] = None,
        attention_mask: Optional[torch.Tensor] = None,
        token_type_ids: Optional[torch.Tensor] = None,
        position_ids: Optional[torch.Tensor] = None,
        inputs_embeds: Optional[torch.Tensor] = None,
        labels: Optional[torch.Tensor] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        return_dict: Optional[bool] = None,
    ) -> Union[Tuple, MultipleChoiceModelOutput]:
        r"""
        labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
            Labels for computing the multiple choice classification loss. Indices should be in `[0, ...,
            num_choices-1]` where `num_choices` is the size of the second dimension of the input tensors. (See
            `input_ids` above)
        """
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
        num_choices = input_ids.shape[1] if input_ids is not None else inputs_embeds.shape[1]

        flat_input_ids = input_ids.view(-1, input_ids.size(-1)) if input_ids is not None else None
        flat_position_ids = position_ids.view(-1, position_ids.size(-1)) if position_ids is not None else None
        flat_token_type_ids = token_type_ids.view(-1, token_type_ids.size(-1)) if token_type_ids is not None else None
        flat_attention_mask = attention_mask.view(-1, attention_mask.size(-1)) if attention_mask is not None else None
        flat_inputs_embeds = (
            inputs_embeds.view(-1, inputs_embeds.size(-2), inputs_embeds.size(-1))
            if inputs_embeds is not None
            else None
        )

        outputs = self.deberta(
            flat_input_ids,
            position_ids=flat_position_ids,
            token_type_ids=flat_token_type_ids,
            attention_mask=flat_attention_mask,
            inputs_embeds=flat_inputs_embeds,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
            return_dict=return_dict,
        )

        encoder_layer = outputs[0]
        pooled_output = self.pooler(encoder_layer)
        pooled_output = self.dropout(pooled_output)
        logits = self.classifier(pooled_output)
        reshaped_logits = logits.view(-1, num_choices)

        loss = None
        if labels is not None:
            loss_fct = CrossEntropyLoss()
            loss = loss_fct(reshaped_logits, labels)

        if not return_dict:
            output = (reshaped_logits,) + outputs[1:]
            return ((loss,) + output) if loss is not None else output

        return MultipleChoiceModelOutput(
            loss=loss,
            logits=reshaped_logits,
            hidden_states=outputs.hidden_states,
            attentions=outputs.attentions,
        )