-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathmodeling_longformer.py
1477 lines (1330 loc) · 62.7 KB
/
modeling_longformer.py
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
import math
import numpy as np
import paddle
import paddlenlp
from paddle.nn import Linear, Dropout, LayerNorm, LayerList, Layer
import paddle.nn.functional as F
import paddle.nn as nn
from paddlenlp.transformers.attention_utils import _convert_param_attr_to_list, Attention, MultiHeadAttention, \
AttentionRegistry
from paddlenlp.transformers import PretrainedModel, register_base_model
"""
由于 Longformer中的attention机制是滑动+全局，与LongFormer相比只少了随机attention，LongFormer在paddlenlp.transformers 已经实现
所以 主要参考了 https://github.com/PaddlePaddle/PaddleNLP/blob/develop/paddlenlp/transformers/LongFormer/modeling.py
以及 https://github.com/huggingface/transformers/blob/master/src/transformers/models/longformer/modeling_longformer.py
"""
__all__ = [
]
def mish(x):
return x * F.tanh(F.softplus(x))
def linear_act(x):
return x
def swish(x):
return x * F.sigmoid(x)
def gelu_new(x):
"""
Implementation of the GELU activation function currently in Google BERT repo (identical to OpenAI GPT). Also see
the Gaussian Error Linear Units paper: https://arxiv.org/abs/1606.08415
"""
return 0.5 * x * (1.0 + paddle.tanh(
math.sqrt(2.0 / math.pi) * (x + 0.044715 * paddle.pow(x, 3.0))))
ACT2FN = {
"relu": F.relu,
"gelu": F.gelu,
"gelu_new": gelu_new,
"tanh": F.tanh,
"sigmoid": F.sigmoid,
"mish": mish,
"linear": linear_act,
"swish": swish,
}
@AttentionRegistry.register("longformer")
class LongFormerSparseAttention(Attention):
def __init__(self,
num_heads=1,
block_size=1,
window_size=512,
num_global_blocks=1):
super(LongFormerSparseAttention,
self).__init__(num_heads, block_size, window_size,
num_global_blocks)
for k, v in locals().items():
if k != "self":
setattr(self, k, v)
self.num_global_blocks_back = num_global_blocks // 2
self.num_global_blocks_front = num_global_blocks // 2 \
if num_global_blocks % 2 == 0 \
else num_global_blocks // 2 + 1
def _get_band_mask(self, blocked_query_mask, blocked_key_mask, batch_size,
sequence_length):
'''
Return second mask: [B, 1, L-G, bs, G+W]
'''
GB = self.num_global_blocks_back
GF = self.num_global_blocks_front
G = self.num_global_blocks
R = self.num_rand_blocks
W = self.window_size
bs = self.block_size
T = sequence_length
L = T // bs # blocked length
B = batch_size
H = self.num_heads
# G+W+R
# query_mask: [B, L, bs]
# key_mask: [B, L, bs]
# [B, L-G, bs, 1] * [B, L-G, 1, G*bs] -> [B, L-G, bs, G*bs]
temp_query_mask = paddle.reshape(blocked_query_mask[:, GF:-GB],
[B, L - G, bs, 1])
temp_key_mask_front = paddle.reshape(blocked_key_mask[:, :GF],
[B, 1, 1, GF * bs])
global_block_mask_front = paddlenlp.ops.einsum(
"blqd,bmdk->blqk", temp_query_mask, temp_key_mask_front)
temp_key_mask_back = paddle.reshape(blocked_key_mask[:, -GB:],
[B, 1, 1, GB * bs])
global_block_mask_back = paddlenlp.ops.einsum(
"blqd,bmdk->blqk", temp_query_mask, temp_key_mask_back)
# create window block mask
key_mask_list = []
for query_block_id in range(GF, GF + W // 2):
left_block_id = query_block_id - W // 2
right_block_id = query_block_id + W // 2
zero_key_mask = paddle.zeros_like(blocked_key_mask[:, -(W - (
right_block_id + 1 - G)):-GB])
temp_key_mask = paddle.concat(
[blocked_key_mask[:, GF:(right_block_id + 1)], zero_key_mask],
axis=1)
temp_key_mask = paddle.unsqueeze(temp_key_mask, 1)
key_mask_list.append(temp_key_mask)
roll_key_mask1 = paddle.concat(key_mask_list, axis=1)
roll_key_mask1 = paddle.reshape(roll_key_mask1, [0, 0, W * bs])
key_mask_list = []
band_length = L - G - W // 2 * 2
for query_block_id in range(GF + W // 2, GF + W // 2 + W):
left_block_id = query_block_id - W // 2
right_block_id = query_block_id + W // 2
key_mask_list.append(blocked_key_mask[:, left_block_id:left_block_id
+ band_length])
window_key_mask = paddle.concat(key_mask_list, axis=2)
window_key_mask = paddle.reshape(window_key_mask, [0, 0, W * bs])
key_mask_list = []
for query_block_id in range((L - GB) - W // 2, L - GB):
left_block_id = query_block_id - W // 2
right_block_id = query_block_id + W // 2
zero_key_mask = paddle.zeros_like(blocked_key_mask[:, GF:GF + W - (
L - left_block_id - GB)])
temp_key_mask = paddle.concat(
[zero_key_mask, blocked_key_mask[:, left_block_id:-GB]], axis=1)
temp_key_mask = paddle.unsqueeze(temp_key_mask, 1)
key_mask_list.append(temp_key_mask)
roll_key_mask2 = paddle.concat(key_mask_list, axis=1)
roll_key_mask2 = paddle.reshape(roll_key_mask2, [0, 0, W * bs])
window_key_mask = paddle.concat(
[roll_key_mask1, window_key_mask, roll_key_mask2], axis=1)
window_key_mask = paddle.unsqueeze(window_key_mask, axis=2)
# [B, L-G, bs, 1] * [B, L-G, 1, W*bs] -> [B, L-G, bs, W*bs]
window_block_mask = paddlenlp.ops.einsum(
"blkd,bldq->blkq", temp_query_mask, window_key_mask)
band_mask = paddle.concat(
[
global_block_mask_front, window_block_mask,
global_block_mask_back
],
axis=3)
band_mask = paddle.unsqueeze(band_mask, 1) # for head
band_mask = paddle.expand(band_mask, [B, H, L - G, bs, -1])
return band_mask
def _get_band_matrix(self, blocked_matrix, B, T):
'''
return global and window matrix: [B, H, L-G, (G+W) * bs, -1]
'''
# blocked_matrix: [B, H, L, bs, -1]
GB = self.num_global_blocks_back
GF = self.num_global_blocks_front
G = self.num_global_blocks
R = self.num_rand_blocks
W = self.window_size
bs = self.block_size
L = T // bs # blocked length
H = self.num_heads
# get roll matrix
blocked_list = []
for query_block_id in range(GF, GF + W // 2):
left_block_id = query_block_id - W // 2
right_block_id = query_block_id + W // 2
temp_blocked_matrix_list = [
blocked_matrix[:, :, 0:(right_block_id + 1)],
blocked_matrix[:, :, -(G + W - right_block_id - 1):]
]
temp_blocked_matrix = paddle.concat(
temp_blocked_matrix_list, axis=2)
temp_blocked_matrix = paddle.unsqueeze(temp_blocked_matrix, axis=2)
blocked_list.append(temp_blocked_matrix)
# get window matrix
band_length = L - G - W // 2 * 2
band_matrix_list = []
for query_block_id in range(GF + W // 2, GF + W // 2 + W):
left_block_id = query_block_id - W // 2
right_block_id = query_block_id + W // 2
band_matrix_list.append(
paddle.unsqueeze(
blocked_matrix[:, :, left_block_id:left_block_id +
band_length],
axis=3))
band_matrix = paddle.concat(band_matrix_list, axis=3)
global_blocked_front_matrix = paddle.unsqueeze(
blocked_matrix[:, :, :GF], axis=2)
global_blocked_front_matrix = paddle.expand(
global_blocked_front_matrix, [B, H, band_length, GF, bs, -1])
global_blocked_back_matrix = paddle.unsqueeze(
blocked_matrix[:, :, -GB:], axis=2)
global_blocked_back_matrix = paddle.expand(
global_blocked_back_matrix, [B, H, band_length, GB, bs, -1])
band_matrix = paddle.concat(
[
global_blocked_front_matrix, band_matrix,
global_blocked_back_matrix
],
axis=3)
blocked_list.append(band_matrix)
for query_block_id in range(L - GB - W // 2, L - GB):
left_block_id = query_block_id - W // 2
right_block_id = query_block_id + W // 2
temp_blocked_matrix_list = [
blocked_matrix[:, :, 0:G + W - (L - left_block_id)],
blocked_matrix[:, :, left_block_id:]
]
temp_blocked_matrix = paddle.concat(
temp_blocked_matrix_list, axis=2)
temp_blocked_matrix = paddle.unsqueeze(temp_blocked_matrix, axis=2)
blocked_list.append(temp_blocked_matrix)
band_matrix = paddle.concat(blocked_list, axis=2)
band_matrix = paddle.reshape(band_matrix,
[B, H, L - G, (G + W) * bs, -1])
return band_matrix
# LongFormer不需要随机的attention块
# def _get_rand_mask(self, blocked_query_mask, blocked_key_mask,
# rand_mask_idx, batch_size, sequence_length):
# '''
# return random mask: [B, H, L-G, bs, R * bs]
# '''
# # rand_mask_idx: [H, T]
# # blocked_query_mask: [B, L, bs]
# # blocked_key_mask: [B, L, bs]
# bs = self.block_size
# B = batch_size
# L = sequence_length // bs
# H = self.num_heads
# G = self.num_global_blocks
# GB = self.num_global_blocks_back
# GF = self.num_global_blocks_front
# R = self.num_rand_blocks
# temp_block_key_mask = paddle.unsqueeze(blocked_key_mask, 1)
# temp_block_key_mask = paddle.expand(temp_block_key_mask, [B, H, L, -1])
# temp_block_key_mask_list = [
# paddle.gather_nd(temp_block_key_mask[b], rand_mask_idx)
# for b in range(B)
# ]
# temp_block_key_mask = paddle.concat(temp_block_key_mask_list, 0)
# temp_block_key_mask = paddle.reshape(temp_block_key_mask, [
# B, temp_block_key_mask.shape[0] // B // (L - GF - GB) // R,
# L - GF - GB, -1
# ])
# rand_mask = paddlenlp.ops.einsum("blq,bhlk->bhlqk",
# blocked_query_mask[:, GF:-GB],
# temp_block_key_mask)
# return rand_mask
# def _gather_random_key_value(self, blocked_matrix, rand_mask_idx, B, T):
# '''
# return random key matrix: [B, H, L-G, R * bs, -1]
# '''
# # blocked_matrix: [B, H, L, bs, -1]
# # rand_mask_idx: [H, T]
# G = self.num_global_blocks
# H = self.num_heads
# bs = self.block_size
# L = T // bs
# R = self.num_rand_blocks
# gathered_matrix = paddle.concat(
# [
# paddle.gather_nd(blocked_matrix[b, :], rand_mask_idx)
# for b in range(B)
# ],
# axis=0)
# gathered_matrix = paddle.reshape(gathered_matrix,
# [B, H, L - G, R * bs, -1])
# return gathered_matrix
def _get_global_out(self,
query_matrix,
key_matrix,
value_matrix,
key_mask,
d_head,
dropout,
is_front=True):
GB = self.num_global_blocks_back
GF = self.num_global_blocks_front
if is_front:
global_query_matrix = query_matrix[:, :, 0:GF * self.block_size]
else:
global_query_matrix = query_matrix[:, :, -GB * self.block_size:]
global_product = paddle.matmul(
global_query_matrix, key_matrix, transpose_y=True)
global_product = global_product * (d_head**-0.5)
global_product += (1 - key_mask) * -1e6
global_weights = F.softmax(global_product)
# [B, H, GF*bs, T] * [B, H, T, D] -> [B, H, GF*bs, D]
global_product = paddle.matmul(global_weights, value_matrix)
return global_product
def _get_splited_matrix(self, matrix):
W = self.window_size // 2
return matrix[:, :, 0:W], matrix[:, :, W:-W], matrix[:, :, -W:]
def forward(self,
query_matrix,
key_matrix,
value_matrix,
d_head,
attn_mask=None,
query_mask=None,
key_mask=None,
dropout=None):
'''
query_matrix: [B, H, T, D]
key_matrix: [B, H, T, D]
value_matrix: [B, H, T, D]
query_mask: [B, 1, T, 1] bool mask
key_mask: [B, 1, 1, T] bool mask
Global Attention
Window Attention
'''
B = query_matrix.shape[0] # batch_size
H = self.num_heads
T = query_matrix.shape[2] # sequence_length
D = query_matrix.shape[3] # size per head
G = self.num_global_blocks
GB = self.num_global_blocks_back
GF = self.num_global_blocks_front
R = self.num_rand_blocks
W = self.window_size
bs = self.block_size
L = T // bs # blocked length
blocked_query_matrix = paddle.reshape(query_matrix, [B, H, L, bs, -1])
blocked_key_matrix = paddle.reshape(key_matrix, [B, H, L, bs, -1])
blocked_value_matrix = paddle.reshape(value_matrix, [B, H, L, bs, -1])
blocked_query_mask = paddle.reshape(query_mask, [B, L, bs])
blocked_key_mask = paddle.reshape(key_mask, [B, L, bs])
# 1. global_front_product
global_front_out = self._get_global_out(
query_matrix, key_matrix, value_matrix, key_mask, d_head, dropout)
# 2. global_back_product
global_back_out = self._get_global_out(query_matrix, key_matrix,
value_matrix, key_mask, d_head,
dropout, False)
# 3. second_product
# create second matrix
# [B, 1, L-G, bs, (G+W)*bs]
band_mask = self._get_band_mask(blocked_query_mask, blocked_key_mask, B,
T)
# # [B, H, L-G, bs, R*bs]
# rand_mask = self._get_rand_mask(blocked_query_mask, blocked_key_mask,
# rand_mask_idx, B, T)
# [B, H, L-G, bs, (G+W+R)*bs]
# second_mask = paddle.concat([band_mask, rand_mask], axis=4)
# [B, H, L-G, R * bs, -1]
# random_keys = self._gather_random_key_value(blocked_key_matrix,
# rand_mask_idx, B, T)
# random_values = self._gather_random_key_value(blocked_value_matrix,
# rand_mask_idx, B, T)
band_keys_matrix = self._get_band_matrix(blocked_key_matrix, B, T)
band_value_matrix = self._get_band_matrix(blocked_value_matrix, B, T)
# [B, H, L - G, bs, -1]
second_query_matrix = blocked_query_matrix[:, :, GF:-GB]
# [B, H, L - G, (G+W+R)*bs, -1]
# second_key_matrix = paddle.concat(
# [band_keys_matrix, random_keys], axis=3)
# [B, H, L - G, (G+W+R)*bs, -1]
# second_value_matrix = paddle.concat(
# [band_value_matrix, random_values], axis=3)
second_top_value_matrix, second_middle_value_matrix, second_bottom_value_matrix = \
self._get_splited_matrix(band_value_matrix)
second_product = paddlenlp.ops.einsum(
"bhlqd,bhlkd->bhlqk", second_query_matrix, band_keys_matrix)
second_product = second_product * (d_head**-0.5)
second_product += (1 - band_mask) * -1e6
second_weights = F.softmax(second_product)
second_top_weights, second_middle_weights, second_bottom_weights = \
self._get_splited_matrix(second_weights)
second_top_out = paddlenlp.ops.einsum(
"bhlqk,bhlkd->bhlqd", second_top_weights, second_top_value_matrix)
second_middle_out = paddlenlp.ops.einsum(
"bhlqk,bhlkd->bhlqd",
second_middle_weights[:, :, :, :, GF * bs:-(GB + R) * bs],
second_middle_value_matrix[:, :, :, GF * bs:-(GB + R) * bs])
# add global block attention
second_middle_out += paddlenlp.ops.einsum(
"bhlqk,bhkd->bhlqd", second_middle_weights[:, :, :, :, :GF * bs],
blocked_value_matrix[:, :, 0])
second_middle_out += paddlenlp.ops.einsum(
"bhlqk,bhkd->bhlqd",
second_middle_weights[:, :, :, :, -(GB + R) * bs:-R * bs],
blocked_value_matrix[:, :, -GB])
# add random block attention
# second_middle_out += paddlenlp.ops.einsum(
# "...qk,...kd->...qd", second_middle_weights[:, :, :, :, -R * bs:],
# random_values[:, :, GF:-GB])
second_bottom_out = paddlenlp.ops.einsum("bhlqk,bhlkd->bhlqd",
second_bottom_weights,
second_bottom_value_matrix)
second_out = paddle.concat(
[second_top_out, second_middle_out, second_bottom_out], axis=2)
second_out = paddle.reshape(second_out, [B, H, (L - G) * bs, -1])
# [B, H, T, D]
out = paddle.concat(
[global_front_out, second_out, global_back_out], axis=2)
out = out * query_mask
return out
class TransformerEncoderLayer(Layer):
def __init__(self,
d_model,
nhead,
dim_feedforward,
dropout=0.1,
activation="relu",
attn_dropout=None,
act_dropout=None,
normalize_before=False,
weight_attr=None,
bias_attr=None,
attention_type="longformer",
block_size=1,
window_size=3,
num_global_blocks=1,
num_rand_blocks=1,
seed=None):
self._config = locals()
self._config.pop("self")
self._config.pop("__class__", None) # py3
super(TransformerEncoderLayer, self).__init__()
attn_dropout = dropout if attn_dropout is None else attn_dropout
act_dropout = dropout if act_dropout is None else act_dropout
self.normalize_before = normalize_before
weight_attrs = _convert_param_attr_to_list(weight_attr, 2)
bias_attrs = _convert_param_attr_to_list(bias_attr, 2)
self.self_attn = MultiHeadAttention(
d_model,
nhead,
dropout=attn_dropout,
weight_attr=weight_attrs[0],
bias_attr=bias_attrs[0],
attention_type=attention_type,
block_size=block_size,
window_size=window_size,
num_global_blocks=num_global_blocks,
num_rand_blocks=num_rand_blocks,
seed=seed)
self.linear1 = Linear(
d_model, dim_feedforward, weight_attrs[1], bias_attr=bias_attrs[1])
self.dropout = Dropout(act_dropout, mode="upscale_in_train")
self.linear2 = Linear(
dim_feedforward, d_model, weight_attrs[1], bias_attr=bias_attrs[1])
self.norm1 = LayerNorm(d_model, epsilon=1e-12)
self.norm2 = LayerNorm(d_model, epsilon=1e-12)
self.dropout1 = Dropout(dropout, mode="upscale_in_train")
self.dropout2 = Dropout(dropout, mode="upscale_in_train")
self.activation = getattr(F, activation)
self.d_model = d_model
def forward(self,
src,
src_mask=None,
rand_mask_idx=None,
query_mask=None,
key_mask=None):
residual = src
if self.normalize_before:
src = self.norm1(src)
src = self.self_attn(src, src, src, src_mask, rand_mask_idx, query_mask,
key_mask)
src = residual + self.dropout1(src)
if not self.normalize_before:
src = self.norm1(src)
residual = src
if self.normalize_before:
src = self.norm2(src)
src = self.linear2(self.dropout(self.activation(self.linear1(src))))
src = residual + self.dropout2(src)
if not self.normalize_before:
src = self.norm2(src)
return src
class TransformerEncoder(Layer):
def __init__(self, encoder_layer, num_layers):
super(TransformerEncoder, self).__init__()
self.layers = LayerList([(encoder_layer if i == 0 else
type(encoder_layer)(**encoder_layer._config))
for i in range(num_layers)])
self.num_layers = num_layers
self.norm = LayerNorm(self.layers[0].d_model, epsilon=1e-12)
self.normalize_before = self.layers[0].normalize_before
def forward(self,
src,
src_mask_list=None,
rand_mask_idx_list=None,
query_mask=None,
key_mask=None):
output = src
if not self.normalize_before:
output = self.norm(output)
for i, mod in enumerate(self.layers):
rand_mask_id = None
if rand_mask_idx_list is not None:
rand_mask_id = rand_mask_idx_list[i]
if src_mask_list is None:
output = mod(output, None, rand_mask_id, query_mask, key_mask)
else:
output = mod(output, src_mask_list[i], rand_mask_id, query_mask,
key_mask)
if self.normalize_before:
output = self.norm(output)
return output
class LongFormerPooler(Layer):
"""
Pool the result of LongFormer Encoder
"""
def __init__(self, hidden_size):
super(LongFormerPooler, self).__init__()
self.dense = nn.Linear(hidden_size, hidden_size)
self.activation = nn.Tanh()
def forward(self, hidden_states):
# We "pool" the model by simply taking the hidden state corresponding
# to the first token.
first_token_tensor = hidden_states[:, 0]
pooled_output = self.dense(first_token_tensor)
pooled_output = self.activation(pooled_output)
return pooled_output
class LongFormerEmbeddings(Layer):
"""
Include embeddings from word, position and token_type embeddings
"""
def __init__(self,
vocab_size,
hidden_size=768,
hidden_dropout_prob=0.1,
max_position_embeddings=512,
type_vocab_size=16,
padding_idx=0):
super(LongFormerEmbeddings, self).__init__()
self.word_embeddings = nn.Embedding(
vocab_size, hidden_size, padding_idx=padding_idx)
self.position_embeddings = nn.Embedding(max_position_embeddings,
hidden_size)
self.token_type_embeddings = nn.Embedding(type_vocab_size, hidden_size)
self.dropout = nn.Dropout(hidden_dropout_prob)
def forward(self, input_ids, token_type_ids=None, position_ids=None):
if position_ids is None:
ones = paddle.ones_like(input_ids, dtype="int64")
seq_length = paddle.cumsum(ones, axis=-1)
position_ids = seq_length - ones
position_ids.stop_gradient = True
if token_type_ids is None:
token_type_ids = paddle.zeros_like(input_ids, dtype="int64")
input_embedings = self.word_embeddings(input_ids)
position_embeddings = self.position_embeddings(position_ids)
token_type_embeddings = self.token_type_embeddings(token_type_ids)
embeddings = input_embedings + position_embeddings + token_type_embeddings
embeddings = self.dropout(embeddings)
return embeddings
class LongFormerPretrainedModel(PretrainedModel):
"""
An abstract class for pretrained LongFormer models. It provides LongFormer related
`model_config_file`, `pretrained_init_configuration`, `resource_files_names`,
`pretrained_resource_files_map`, `base_model_prefix` for downloading and
loading pretrained models.
See :class:`~paddlenlp.transformers.model_utils.PretrainedModel` for more details.
"""
model_config_file = "model_config.json"
pretrained_init_configuration = {
}
resource_files_names = {"model_state": "model_state.pdparams"}
pretrained_resource_files_map = {
}
base_model_prefix = "LongFormer"
def init_weights(self, layer):
# Initialization hook
if isinstance(layer, (nn.Linear, nn.Embedding)):
# In the dygraph mode, use the `set_value` to reset the parameter directly,
# and reset the `state_dict` to update parameter in static mode.
if isinstance(layer.weight, paddle.Tensor):
layer.weight.set_value(
paddle.tensor.normal(
mean=0.0,
std=self.initializer_range
if hasattr(self, "initializer_range") else
self.LongFormer.config["initializer_range"],
shape=layer.weight.shape))
elif isinstance(layer, nn.LayerNorm):
layer._epsilon = 1e-12
@register_base_model
class LongFormerModel(LongFormerPretrainedModel):
"""
The bare LongFormer Model outputting raw hidden-states.
This model inherits from :class:`~paddlenlp.transformers.model_utils.PretrainedModel`.
Refer to the superclass documentation for the generic methods.
This model is also a Paddle `paddle.nn.Layer <https://www.paddlepaddle.org.cn/documentation
/docs/en/api/paddle/fluid/dygraph/layers/Layer_en.html>`__ subclass. Use it as a regular Paddle Layer
and refer to the Paddle documentation for all matter related to general usage and behavior.
Args:
num_layers (int):
Number of hidden layers in the Transformer encoder.
vocab_size (int):
Vocabulary size of `inputs_ids` in `LongFormerModel`. Also is the vocab size of token embedding matrix.
Defines the number of different tokens that can be represented by the `inputs_ids` passed when calling `LongFormerModel`.
nhead (int):
Number of attention heads for each attention layer in the Transformer encoder.
attn_dropout (float, optional):
The dropout probability used in MultiHeadAttention in all encoder layers to drop some attention target.
Defaults to `0.1`.
dim_feedforward (int, optional):
Dimensionality of the feed-forward (ff) layer in the Transformer encoder. Input tensors
to ff layers are firstly projected from `hidden_size` to `intermediate_size`,
and then projected back to `hidden_size`. Typically `intermediate_size` is larger than `hidden_size`.
Defaults to `3072`.
activation (str, optional):
The non-linear activation function in the feed-forward layer.
``"gelu"``, ``"relu"``, ``"silu"`` and ``"gelu_new"`` are supported.
Defaults to `"gelu"`.
normalize_before (bool, optional):
Indicates whether to put layer normalization into preprocessing of MHA and FFN sub-layers.
If True, pre-process is layer normalization and post-precess includes dropout,
residual connection. Otherwise, no pre-process and post-precess includes dropout,
residual connection, layer normalization.
Defaults to `False`.
block_size (int, optional):
The block size for the attention mask.
Defaults to `1`.
window_size (int, optional):
The number of block in a window.
Defaults to `3`.
num_global_blocks (int, optional):
Number of global blocks per sequence.
Defaults to `1`.
num_rand_blocks (int, optional):
Number of random blocks per row.
Defaults to `2`.
seed (int, optional):
The random seed for generating random block id.
Defaults to ``None``.
pad_token_id (int, optional):
The index of padding token for LongFormer embedding.
Defaults to ``0``.
hidden_size (int, optional):
Dimensionality of the embedding layer, encoder layer and pooler layer.
Defaults to `768`.
hidden_dropout_prob (float, optional):
The dropout probability for all fully connected layers in the embeddings and encoder.
Defaults to `0.1`.
max_position_embeddings (int, optional):
The maximum value of the dimensionality of position encoding, which dictates the maximum supported length of an input
sequence. Defaults to `512`.
type_vocab_size (int, optional):
The vocabulary size of the `token_type_ids`.
Defaults to `2`.
"""
def __init__(self,
num_layers,
vocab_size,
nhead,
attn_dropout=0.1,
dim_feedforward=3072,
activation="gelu",
normalize_before=False,
block_size=1,
window_size=3,
num_global_blocks=1,
num_rand_blocks=2,
seed=None,
pad_token_id=0,
hidden_size=768,
hidden_dropout_prob=0.1,
max_position_embeddings=512,
type_vocab_size=2,
**kwargs):
super(LongFormerModel, self).__init__()
# embedding
self.embeddings = LongFormerEmbeddings(
vocab_size, hidden_size, hidden_dropout_prob,
max_position_embeddings, type_vocab_size, pad_token_id)
# encoder
encoder_layer = TransformerEncoderLayer(
hidden_size,
nhead,
dim_feedforward,
attn_dropout,
activation,
normalize_before=normalize_before,
attention_type="LongFormer",
block_size=block_size,
window_size=window_size,
num_global_blocks=num_global_blocks,
num_rand_blocks=num_rand_blocks,
seed=seed)
self.encoder = TransformerEncoder(encoder_layer, num_layers)
# pooler
self.pooler = LongFormerPooler(hidden_size)
self.pad_token_id = pad_token_id
self.num_layers = num_layers
def _process_mask(self, input_ids, attention_mask_list=None):
# [B, T]
attention_mask = (input_ids == self.pad_token_id
).astype(self.pooler.dense.weight.dtype)
# [B, 1, T, 1]
query_mask = paddle.unsqueeze(attention_mask, axis=[1, 3])
# [B, 1, 1, T]
key_mask = paddle.unsqueeze(attention_mask, axis=[1, 2])
query_mask = 1 - query_mask
key_mask = 1 - key_mask
return attention_mask_list, query_mask, key_mask
def forward(self,
input_ids,
token_type_ids=None,
attention_mask_list=None,
rand_mask_idx_list=None):
r"""
The LongFormerModel forward method, overrides the __call__() special method.
Args:
input_ids (`Tensor`):
Indices of input sequence tokens in the vocabulary.
Its data type should be `int64` and it has a shape of [batch_size, sequence_length].
token_type_ids (`Tensor`, optional):
Segment token indices to indicate first and second portions of the inputs.
Indices can either be 0 or 1:
- 0 corresponds to a *sentence A* token,
- 1 corresponds to a *sentence B* token.
Its data type should be `int64` and it has a shape of [batch_size, sequence_length].
Defaults to ``None``, which means we don't add segment embeddings.
attention_mask_list (list, optional):
A list which contains some tensors used in multi-head attention
to prevents attention to some unwanted positions, usually the
paddings or the subsequent positions.
Its data type can be int, float and bool.
When the data type is bool, the `masked` tokens have `False` values and the others have `True` values.
When the data type is int, the `masked` tokens have `0` values and the others have `1` values.
When the data type is float, the `masked` tokens have `-INF` values and the others have `0` values.
It is a tensor with shape broadcasted to `[batch_size, n_head, sequence_length, sequence_length]`.
For example, its shape can be [batch_size, sequence_length], [batch_size, sequence_length, sequence_length],
[batch_size, num_attention_heads, sequence_length, sequence_length].
Defaults to `None`, which means nothing needed to be prevented attention to.
rand_mask_idx_list (`list`, optional):
A list which contains some tensors used in LongFormer random block.
Returns:
tuple: Returns tuple (`encoder_output`, `pooled_output`).
With the fields:
- encoder_output (Tensor):
Sequence of output at the last layer of the model.
Its data type should be float32 and has a shape of [batch_size, sequence_length, hidden_size].
- pooled_output (Tensor):
The output of first token (`[CLS]`) in sequence.
We "pool" the model by simply taking the hidden state corresponding to the first token.
Its data type should be float32 and its shape is [batch_size, hidden_size].
Examples:
.. code-block::
import paddle
from paddlenlp.transformers import LongFormerModel, LongFormerTokenizer
from paddlenlp.transformers import create_LongFormer_rand_mask_idx_list
tokenizer = LongFormerTokenizer.from_pretrained('LongFormer-base-uncased')
model = LongFormerModel.from_pretrained('LongFormer-base-uncased')
config = model.config
max_seq_len = 512
input_ids = tokenizer.convert_tokens_to_ids(
tokenizer(
"This is a docudrama story on the Lindy Chamberlain case and a look at "
"its impact on Australian society It especially looks at the problem of "
"innuendo gossip and expectation when dealing with reallife dramasbr br "
"One issue the story deals with is the way it is expected people will all "
"give the same emotional response to similar situations Not everyone goes "
"into wild melodramatic hysterics to every major crisis Just because the "
"characters in the movies and on TV act in a certain way is no reason to "
"expect real people to do so"
))
input_ids.extend([0] * (max_seq_len - len(input_ids)))
seq_len = len(input_ids)
input_ids = paddle.to_tensor([input_ids])
rand_mask_idx_list = create_LongFormer_rand_mask_idx_list(
config["num_layers"], seq_len, seq_len, config["nhead"],
config["block_size"], config["window_size"], config["num_global_blocks"],
config["num_rand_blocks"], config["seed"])
rand_mask_idx_list = [
paddle.to_tensor(rand_mask_idx) for rand_mask_idx in rand_mask_idx_list
]
output = model(input_ids, rand_mask_idx_list=rand_mask_idx_list)
"""
embedding_output = self.embeddings(input_ids, token_type_ids)
attention_mask_list, query_mask, key_mask = self._process_mask(
input_ids, attention_mask_list)
encoder_output = self.encoder(embedding_output, attention_mask_list,
rand_mask_idx_list, query_mask, key_mask)
pooled_output = self.pooler(encoder_output)
return encoder_output, pooled_output
''' this is early version based on huggingface not use paddlenlp.transformers.attention_utils
class LongformerEmbeddings(Layer):
def __init__(
self,
vocab_size,
hidden_size=768, hidden_dropout_prob=0.1,
max_position_embeddings=512,
type_vocab_size=16,
pad_token_id=0
):
super().__init__()
self.word_embeddings = nn.Embedding(
vocab_size, hidden_size, padding_idx=pad_token_id)
self.position_embeddings = nn.Embedding(
max_position_embeddings, hidden_size, padding_idx=pad_token_id)
self.token_type_embeddings = nn.Embedding(type_vocab_size, hidden_size)
self.LayerNorm = nn.LayerNorm(hidden_size)
self.dropout = nn.Dropout(hidden_dropout_prob)
def forward(self, input_ids, token_type_ids=None, position_ids=None):
if position_ids is None:
# maybe need use shape op to unify static graph and dynamic graph
ones = paddle.ones_like(input_ids, dtype="int64")
seq_length = paddle.cumsum(ones, axis=-1)
position_ids = seq_length - ones
position_ids.stop_gradient = True
if token_type_ids is None:
token_type_ids = paddle.zeros_like(input_ids, dtype="int64")
input_embedings = self.word_embeddings(input_ids)
position_embeddings = self.position_embeddings(position_ids)
token_type_embeddings = self.token_type_embeddings(token_type_ids)
embeddings = input_embedings + position_embeddings + token_type_embeddings
embeddings = self.layer_norm(embeddings)
embeddings = self.dropout(embeddings)
return embeddings
class LongformerSelfAttention(Layer):
"""
Longformer SelfAttention Layer With Sliding Window Attention And Global Attention
"""
def __init__(
self,
hidden_size=768,
num_attention_heads=3,
attention_probs_dropout_probs=0.1,
attention_window: Union[List[int], int] = 512,
layer_id=None
):
super().__init__()
if hidden_size % num_attention_heads != 0:
raise ValueError(
f"The hidden size ({hidden_size}) is not a multiple of the number of attention "
f"heads ({num_attention_heads})"
)
self.num_heads = num_attention_heads
self.head_dim = int(hidden_size / num_attention_heads)
self.embed_dim = hidden_size
self.query = nn.Linear(hidden_size, self.embed_dim)
self.key = nn.Linear(hidden_size, self.embed_dim)
self.value = nn.Linear(hidden_size, self.embed_dim)
# separate projection layers for tokens with global attention
self.query_global = nn.Linear(hidden_size, self.embed_dim)
self.key_global = nn.Linear(hidden_size, self.embed_dim)
self.value_global = nn.Linear(hidden_size, self.embed_dim)
self.dropout = attention_probs_dropout_probs
self.layer_id = layer_id
attention_window = attention_window[self.layer_id]
assert (
attention_window % 2 == 0
), f"`attention_window` for layer {self.layer_id} has to be an even value. Given {attention_window}"
assert (
attention_window > 0
), f"`attention_window` for layer {self.layer_id} has to be positive. Given {attention_window}"
self.one_sided_attn_window_size = attention_window // 2
def forward(
self,
hidden_states,
attention_mask=None,
layer_head_mask=None,
is_index_masked=None,
is_index_global_attn=None,
is_global_attn=None,
output_attentions=False,
):
"""
:class:`LongformerSelfAttention` expects `len(hidden_states)` to be multiple of `attention_window`. Padding to
`attention_window` happens in :meth:`LongformerModel.forward` to avoid redoing the padding on each layer.
The `attention_mask` is changed in :meth:`LongformerModel.forward` from 0, 1, 2 to:
* -10000: no attention
* 0: local attention
* +10000: global attention
"""
class LongformerSelfOutput(Layer):
def __init__(
self,
hidden_size,
hidden_dropout_prob
):
super().__init__()
self.dense = nn.Linear(hidden_size, hidden_size)
self.LayerNorm = nn.LayerNorm(hidden_size)
self.dropout = nn.Dropout(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
class LongformerAttention(Layer):
def __init__(
self,
hidden_size,
num_attention_heads,
attention_probs_dropout_probs,
attention_window,
hidden_dropout_prob,
layer_id
):
super().__init__()
self.self = LongformerSelfAttention(hidden_size=hidden_size,
num_attention_heads=num_attention_heads,
attention_probs_dropout_prob=attention_probs_dropout_probs,
attention_window=attention_window,
layer_id=layer_id)
self.output = LongformerSelfOutput(
hidden_size=hidden_size, hidden_dropout_prob=hidden_dropout_prob)
# self.pruned_heads = set() 指定层的注意力头剪枝 不实现
def forward(
self,
hidden_states,
attention_mask=None,
layer_head_mask=None,
is_index_masked=None,
is_index_global_attn=None,
is_global_attn=None,