-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtrain_pod2.py
1060 lines (906 loc) · 40.7 KB
/
train_pod2.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
"""
======================================================================
TRAIN_POD2 ---
New stealing mechamism.
Author: Zi Liang <zi1415926.liang@connect.polyu.hk>
Copyright © 2024, ZiLiang, all rights reserved.
Created: 27 March 2024
======================================================================
"""
# ------------------------ Code --------------------------------------
import math
import torch
# import json
from torch.utils.tensorboard import SummaryWriter
# from torch.distributions import Categorical
from torch.utils.data import TensorDataset, DataLoader
from tqdm import tqdm
import math
import time
# import argparse
# from transformers import AutoModelForCausalLM
# from transformers import AutoModelForSequenceClassification
# from transformers import AutoModelForTokenClassification
# from transformers import AutoTokenizer, AutoConfig, AutoModel
# from training_data_collecting_openai import load_raw_train_datals
# from training_data_collecting_openai import load_steal_datals
# from glue_process import load_glue_datals
from sequence_utils import my_padding, my_padding_logits
from sequence_utils import my_padding_token_dist
from sequence_utils import my_padding_logit
from sequence_utils import left_pad
from sequence_utils import random_shut
import torch.nn.functional as F
# from rlhf_train import clip, log_clip
import random
from rlhf_train import clip, log_clip
from peft import (
LoraConfig,
# PeftConfig,
# PeftModel,
get_peft_model,
# prepare_model_for_kbit_training,
)
from peft import PeftModel
from transformers import AutoModelForCausalLM
from transformers import AutoTokenizer, AutoConfig, AutoModel
import logging
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)
printt=print
print = logger.info
def random_take(num, ls, seed,):
random.seed(seed)
random.shuffle(ls)
if num >= len(ls):
return ls
return ls[:num]
def train(lm, lm_tokenizer, args,
raw_train_datals, max_new_tokens=16):
sub_stage_num = args.sub_stage_num
steps = sub_stage_num*args.sub_set_num *\
args.period_num
print(f"OVERALL STEPS: {steps}.")
p_i_11_ls = None
p_i_12_ls = None
p_m_11_ls = None
p_m_12_ls = None
p_logits_11_ls = None
p_logits_12_ls = None
p_i2ls = None
pmask2s = None
p_logits2ls = None
p_vic_logits2ls = None
pp11ls=None
pp12ls=None
preset_subset_num=args.sub_set_num
subset_pointer=0
for ssn in tqdm(range(sub_stage_num)):
#### Transform the LLM into a single device.
print(f"stage_num: {ssn+1}.")
if (ssn+1)%args.save_step==0:
print(f" ------>NOW save the ckpt in stage {ssn+1}.")
args.temp_save_path=args.save_path+"___period"+str(ssn+1)
lm_tokenizer.save_pretrained(args.temp_save_path)
lm.save_pretrained(args.temp_save_path)
# #### Transform back.
# lm=None
# print("TO AUTO.")
# lm = AutoModelForCausalLM.from_pretrained(
# args.from_path,
# device_map="auto",
# trust_remote_code=True,
# torch_dtype=torch.bfloat16,
# )
# # if args.use_lora == 1:
# # lm = PeftModel.from_pretrained(lm, args.temp_save_path)
# print("TO AUTO DONE.")
# if ssn<3:
# args.sub_set_num=16
# else:
# args.sub_set_num=preset_subset_num
print(f"subset_num: {args.sub_set_num}.")
lm, p_i_11_ls, p_i_12_ls, p_m_11_ls,\
p_m_12_ls, p_logits_11_ls, p_logits_12_ls,\
p_i2ls, pmask2s, p_logits2ls, p_vic_logits2ls, \
pp11ls, pp12ls = train_pod(
lm, lm_tokenizer,
args, raw_train_datals,
max_new_tokens,
p_i_11_ls, p_i_12_ls, p_m_11_ls,
p_m_12_ls, p_logits_11_ls, p_logits_12_ls,
p_i2ls, pmask2s, p_logits2ls, p_vic_logits2ls,
pp11ls, pp12ls,
subset_pointer,
)
subset_pointer+=1
from accelerate import load_checkpoint_and_dispatch
def train_pod(lm,
lm_tokenizer,
args, raw_train_datals,
max_new_tokens,
p_i_11_ls, p_i_12_ls, p_m_11_ls,
p_m_12_ls, p_logits_11_ls, p_logits_12_ls,
p_i2ls, pmask2s, p_logits2ls, p_vic_logits2ls,
pp11ls, pp12ls,
subset_pointer,
):
# print(">>>> DATA PREPERATION")
tau1 = args.tau1
tau2 = args.tau2
tau_delta = args.tau_delta
print(f" Tau1: {tau1}\t Tau2: {tau2}.")
print(f"MAX NEW TOKENS: {max_new_tokens}.")
# STEP 1: DATA Preperation.
ITER_num = args.period_num
tb_writer = SummaryWriter(log_dir=args.save_path+"___log_writer")
op_ls, oidx2ls, ologits2ls, oidx2_dist = raw_train_datals
tasks_data2text_wrmk=[
"e2e_nlg@wrmk",
"allenai/common_gen@wrmk",
"cs-en@wrmk",
"de-en@wrmk",
"ro-en@wrmk",
]
if args.dataset_task in tasks_data2text_wrmk:
ologits2ls=None
oidxs2_dist=None
subset_num = args.sub_set_num
# 1. in every period, random take a subset.
seed = time.time()
if subset_pointer>= math.floor(len(op_ls)/subset_num)-1:
subset_pointer=subset_pointer%(math.floor(len(op_ls)/subset_num))
p_ls = op_ls[subset_pointer*subset_num:\
(subset_pointer+1)*subset_num]
idx2ls = oidx2ls[subset_pointer*subset_num:\
(subset_pointer+1)*subset_num]
# p_ls = random_take(subset_num, op_ls, seed,)
# idx2ls = random_take(subset_num, oidx2ls, seed)
if ologits2ls is not None:
# vic_logits2ls = random_take(subset_num, ologits2ls, seed)
# idx2_dist = random_take(subset_num, oidx2_dist, seed)
vic_logits2ls = ologits2ls[subset_pointer*subset_num:\
(subset_pointer+1)*subset_num
]
idx2_dist = oidx2_dist[subset_pointer*subset_num:\
(subset_pointer+1)*subset_num
]
# idx2_dist = random_take(subset_num, oidx2_dist, seed)
else:
vic_logits2ls = [None for _ in range(subset_num)]
idx2_dist = [None for _ in range(subset_num)]
need_pre_store = 0
if p_i_11_ls is None:
p_i_11_ls = []
p_i_12_ls = []
p_logits_11_ls = []
p_logits_12_ls = []
p_m_11_ls = []
p_m_12_ls = []
need_pre_store = 1
period_break = 0
p_i2ls = []
pmask2s = []
p_logits2ls = []
p_vic_logits2ls = []
else:
period_break = 1
for iter_idx in range(ITER_num):
tensorboard_name = f"Period {iter_idx}"
idxs11_ls = []
idxs12_ls = []
old_logits11_ls = []
old_logits12_ls = []
old_logits2_ls = []
# 2. generate
with torch.no_grad():
# if True:
# ## =======================================================
# ## New version of the code: chunked generation.
# ## =======================================================
# print(f"device: {idxs11.device}")
for i in range(len(idx2ls)):
idxs2 = torch.tensor(idx2ls[i])\
.unsqueeze(0)
idxs2=idxs2.to("cuda")
# idxs2=idxs2.to("cpu")
old_logits2 = lm(idxs2).logits
old_logits2 = old_logits2[:,:-1]
old_logits2 = F.log_softmax(old_logits2, dim=-1)
bs, sql2 = idxs2.shape
old_logits2 = old_logits2[
torch.arange(1).unsqueeze(1),
torch.arange(sql2-1).unsqueeze(0),
idxs2[:, 1:sql2]
]
old_logits2_ls.append(old_logits2.squeeze(0).to("cpu"))
if args.with_early_shut==1:
print("EXECUTE ERALY SHUT...")
p_ls=random_shut(p_ls)
chunked_size=args.infer_batch_size
num_chunks=math.floor(len(p_ls)/chunked_size)
if num_chunks*chunked_size - len(p_ls)==0.0:
num_range=num_chunks
else:
num_range=num_chunks+1
# 1. first divided the model into chunks.
for i_chunked in range(num_range):
print(f"Chunks: {i_chunked}/{num_chunks}")
if i_chunked == num_chunks:
if i_chunked*chunked_size!=len(p_ls):
prompt=p_ls[i_chunked*chunked_size:]
## left padding
print(f"BOS TOKEN ID: {lm_tokenizer.bos_token_id}")
prompt=left_pad(prompt,lm_tokenizer.bos_token_id)
prompt=prompt.to(args.device)
else:
print(f"BOS TOKEN ID: {lm_tokenizer.bos_token_id}")
prompt=p_ls[i_chunked*chunked_size:\
(i_chunked+1)*chunked_size]
prompt=left_pad(prompt,lm_tokenizer.bos_token_id)
prompt=prompt.to(args.device)
printt("++++++++++++++++++++++++++++++++++++++++++++++++")
printt(f"prompt.shape: {prompt.shape}")
printt(f"prompt: {prompt}")
printt(f"temperature: {args.T}")
gen_idx=lm.generate(
prompt,
do_sample=True,
max_length=args.max_length,
max_new_tokens=max_new_tokens,
num_return_sequences=2,
temperature=args.T,
top_p=0.98,
use_cache=True,
)
print("*********************************************")
print(f"Shape of generated indexes: {gen_idx.shape}")
# gen_idx=lm.generate(
# prompt,
# do_sample=False,
# num_beams=2,
# num_beam_groups=2,
# diversity_penalty=3.3,
# num_return_sequences=2,
# max_length=args.max_length,
# max_new_tokens=max_new_tokens,
# use_cache=True,
# )
# 2. extract idx12 and idx11 from gen_idx
idxs11=gen_idx[0::2,:]
idxs12=gen_idx[1::2,:]
# print(gen_idx.shape)
# print(idxs11.shape)
# print(idxs12.shape)
# shape of gen idx: (2*chunked_size, msl)
bs, sqqql = idxs11.shape
# print(idxs1)
print("========================================")
print(f"idxs11 {lm_tokenizer.decode(idxs11[0])}")
print(f"idxs12 {lm_tokenizer.decode(idxs12[0])}")
old_logits11 = lm(idxs11[:, :-1]).logits
old_logits11 = F.log_softmax(old_logits11, dim=-1)
old_logits11 = old_logits11[
torch.arange(1).unsqueeze(1),
torch.arange(sqqql-1).unsqueeze(0),
idxs11[:, 1:sqqql]
]
bs, sqqql2 = idxs12.shape
old_logits12 = lm(idxs12[:, :-1]).logits
old_logits12 = F.log_softmax(old_logits12, dim=-1)
old_logits12 = old_logits12[
torch.arange(1).unsqueeze(1),
torch.arange(sqqql2-1).unsqueeze(0),
idxs12[:, 1:sqqql2]
]
idxs11_ls.extend([x for x in idxs11.to("cpu")])
idxs12_ls.extend([x for x in idxs12.to("cpu")])
old_logits11_ls.extend([x for x in old_logits11
.to("cpu")])
old_logits12_ls.extend([x for x in old_logits12
.to("cpu")])
# print(idxs11_ls)
# print(idxs12_ls)
# assert len(idxs11_ls)==len(idxs12_ls)
# assert len(idxs11_ls)==len(p_ls)
# ## =======================================================
# ## OLD CODE: NOT FAST ENOUGH MAYBE.
# ## =======================================================
# for i, prompt in tqdm(enumerate(p_ls),
# desc="Data Collecting...",
# total=len(p_ls)):
# prompt = prompt.to(args.device).unsqueeze(0)
# # Generate New Tokens
# idxs12 = lm.generate(prompt,
# do_sample=True,
# max_length=args.max_length,
# max_new_tokens=max_new_tokens,
# # temperature=args.temperature,
# )
# idxs11 = lm.generate(prompt,
# do_sample=True,
# max_length=args.max_length,
# max_new_tokens=max_new_tokens,
# # temperature=args.temperature,
# )
# bs, sqqql = idxs11.shape
# # print(idxs1)
# print(f"idxs11 {lm_tokenizer.decode(idxs11[0])}")
# print(f"idxs12 {lm_tokenizer.decode(idxs12[0])}")
# old_logits11 = lm(idxs11[:, :-1]).logits
# old_logits11 = F.log_softmax(old_logits11, dim=-1)
# old_logits11 = old_logits11[
# torch.arange(1).unsqueeze(1),
# torch.arange(sqqql-1).unsqueeze(0),
# idxs11[:, 1:sqqql]
# ]
# bs, sqqql2 = idxs12.shape
# old_logits12 = lm(idxs12[:, :-1]).logits
# old_logits12 = F.log_softmax(old_logits12, dim=-1)
# old_logits12 = old_logits12[
# torch.arange(1).unsqueeze(1),
# torch.arange(sqqql2-1).unsqueeze(0),
# idxs12[:, 1:sqqql2]
# ]
# idxs2 = torch.tensor(idx2ls[i], dtype=torch.long)\
# .to(args.device).unsqueeze(0)
# print(f"idxs2 {lm_tokenizer.decode(idxs2[0])}")
# old_logits2 = lm(idxs2[:, :-1]).logits
# old_logits2 = F.log_softmax(old_logits2, dim=-1)
# bs, sql2 = idxs2.shape
# old_logits2 = old_logits2[
# torch.arange(1).unsqueeze(1),
# torch.arange(sql2-1).unsqueeze(0),
# idxs2[:, 1:sql2]
# ]
# idxs11_ls.append(idxs11.squeeze(0).to("cpu"))
# idxs12_ls.append(idxs12.squeeze(0).to("cpu"))
# old_logits11_ls.append(old_logits11
# .squeeze(0).to("cpu"))
# old_logits12_ls.append(old_logits12
# .squeeze(0).to("cpu"))
# old_logits2_ls.append(old_logits2.squeeze(0).to("cpu"))
# do truncations and paddings.
# max_token_num_11 = min(args.max_length,
# max([len(x) for x in p_i_11_ls]))
# max_token_num_12 = min(args.max_length,
# max([len(x) for x in p_i_12_ls]))
# max_token_num_2 = min(args.max_length,
# max([len(x) for x in p_i2ls]))
cmax_token_num_2 = min(args.max_length,
max([len(x) for x in idx2ls]))
cmax_token_num_11 = min(args.max_length,
max([len(x) for x in idxs11_ls]))
cmax_token_num_12 = min(args.max_length,
max([len(x) for x in idxs12_ls]))
max_token_num = max(cmax_token_num_2, cmax_token_num_11)
max_token_num = max(max_token_num, cmax_token_num_12)
print(f"max_token_num: {max_token_num}")
pad_idx = lm_tokenizer.pad_token_id
idx2ls, mask2 = my_padding(idx2ls, p_ls,
max_token_num, pad_idx)
idxs11_ls, mask11 = my_padding(idxs11_ls,
p_ls, max_token_num, pad_idx)
idxs12_ls, mask12 = my_padding(idxs12_ls,
p_ls, max_token_num, pad_idx)
if args.with_early_shut==1:
mask11=torch.ones_like(mask11)
mask12=torch.ones_like(mask12)
else:
pass
old_logits11_ls = my_padding_logit(old_logits11_ls,
max_token_num-1, pad_idx)
old_logits12_ls = my_padding_logit(old_logits12_ls,
max_token_num-1, pad_idx)
old_logits2_ls = my_padding_logit(old_logits2_ls,
max_token_num-1, pad_idx)
# p_i_11_ls, p_m_11_ls = my_padding(p_i_11_ls,
# p_ls,
# max_token_num, pad_idx)
# p_i_12_ls, p_m_12_ls = my_padding(p_i_12_ls,
# p_ls,
# max_token_num, pad_idx)
# p_logits_11_ls = my_padding_logit(p_logits_11_ls,
# max_token_num-1,
# pad_idx)
# p_logits_12_ls = my_padding_logit(p_logits_12_ls,
# max_token_num-1,
# pad_idx)
if vic_logits2ls[0] is not None:
newvic_logits2ls = []
for per_data in vic_logits2ls:
# print(f"PER_DATA: {per_data}")
sl = len(per_data)
v = len(per_data[0])
tmp_ts = torch.ones((sl, v), dtype=torch.float)
for jjjj, per_token_logit in enumerate(per_data):
tmp_ts[jjjj] = torch.tensor(per_token_logit,)
newvic_logits2ls.append(tmp_ts)
vic_logits2ls = my_padding_logits(newvic_logits2ls,
max_token_num-1, pad_idx)
idxs2_dist = my_padding_token_dist(idx2_dist,
max_token_num-1, pad_idx)
# print(f"{old_logits11_ls.shape}")
# ---------------------------------------------------------
# now fix the logic of constructive two samples
if need_pre_store == 1:
need_pre_store = 0
# at first stage, we have no previous stage, so we use
# the ground truth in current stage.
p_i_11_ls = idx2ls # absolute positive label
p_i_12_ls = idxs12_ls
p_logits_11_ls = old_logits2_ls
p_logits_12_ls = old_logits12_ls
p_m_11_ls = mask2
p_m_12_ls = mask12
p_i2ls = idx2ls
pmask2s = mask2
p_logits2ls = old_logits2_ls
p_vic_logits2ls = vic_logits2ls
pp11ls=[]
pp12ls=[]
for i, prompt in enumerate(p_ls):
# p11 = float(torch.sum(torch.exp(p_logits_11_ls[i])
# * p_m_11_ls[i, :-1])
# / torch.sum(p_m_11_ls[i, :-1]))
# p12 = float(torch.sum(torch.exp(p_logits_12_ls[i])
# * p_m_12_ls[i, :-1])
# / torch.sum(p_m_12_ls[i, :-1]))
p11=float(torch.exp(
torch.sum(
p_logits_11_ls[i]*p_m_11_ls[i, :-1]
)/torch.sum(p_m_11_ls[i, :-1])))
p12=float(torch.exp(
torch.sum(
p_logits_12_ls[i]*p_m_12_ls[i, :-1]
)/torch.sum(p_m_12_ls[i, :-1])))
pp11ls.append(p11)
pp12ls.append(p12)
elif need_pre_store == 0:
for i, prompt in enumerate(p_ls):
# print("~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~")
# print(f"shape of logits 11: {p_logits_11_ls[i].shape}")
# print(f"value of logits 11: {torch.sum(torch.exp(p_logits_11_ls[i])*p_m_11_ls[i,:-1])/torch.sum(p_m_11_ls[i,:-1])}")
# print(f"value of logits 12: {torch.sum(torch.exp(p_logits_12_ls[i])*p_m_11_ls[i,:-1])/torch.sum(p_m_12_ls[i,:-1])}")
pidx11 = p_i_11_ls[i].unsqueeze(0).to(args.device)
pidx12 = p_i_12_ls[i].unsqueeze(0).to(args.device)
bs, sqqql = pidx11.shape
P_theta_t_logits11=lm(pidx11).logits[:,:-1]
P_theta_t_logits11=F.log_softmax(P_theta_t_logits11,
dim=-1)
P_theta_t_logits11=P_theta_t_logits11[
torch.arange(1).unsqueeze(1),
torch.arange(sqqql-1).unsqueeze(0),
pidx11[:, 1:sqqql]
].to("cpu")
bs, sqqql = pidx12.shape
P_theta_t_logits12=lm(pidx12).logits[:,:-1]
P_theta_t_logits12=F.log_softmax(P_theta_t_logits12,
dim=-1)
P_theta_t_logits12=P_theta_t_logits12[
torch.arange(1).unsqueeze(1),
torch.arange(sqqql-1).unsqueeze(0),
pidx12[:, 1:sqqql]
].to("cpu")
# p11 = float(torch.sum(torch.exp(P_theta_t_logits11)
# * p_m_11_ls[i, :-1])
# / torch.sum(p_m_11_ls[i, :-1]))
# p12 = float(torch.sum(torch.exp(P_theta_t_logits12)
# * p_m_12_ls[i, :-1])
# / torch.sum(p_m_12_ls[i, :-1]))
p11=float(torch.exp(
torch.sum(
P_theta_t_logits11*p_m_11_ls[i, :-1]
)/torch.sum(p_m_11_ls[i, :-1])))
p12=float(torch.exp(
torch.sum(
P_theta_t_logits12*p_m_12_ls[i, :-1]
)/torch.sum(p_m_12_ls[i, :-1])))
# print("LOGITS 11:")
# print(torch.exp(P_theta_t_logits11))
# print("LOGITS 12:")
# print(torch.exp(P_theta_t_logits12))
delta11=p11-pp11ls[i]
delta12=p12-pp12ls[i]
print(">>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>")
print(f"Old Confidence, 1: {pp11ls[i]}, 2: {pp12ls[i]}")
print(f"New Confidence, 1: {p11}, 2: {p12}")
print(f"Delta, 1: {delta11}, 2: {delta12}")
if p12 > p11:
# if delta11<delta12:
print("SWAP.")
p_i_11_ls[i] = pidx12.squeeze(0).to("cpu")
p_i_12_ls[i] = pidx11.squeeze(0).to("cpu")
temppp = p_logits_11_ls[i]
p_logits_11_ls[i] = p_logits_12_ls[i]
p_logits_12_ls[i] = temppp
temppp = p_m_11_ls[i]
p_m_11_ls[i] = p_m_12_ls[i]
p_m_12_ls[i] = temppp
temppp = delta12
delta12=delta11
delta11=temppp
# if max(p11, p12) < tau1 or abs(p11-p12)<0.01:
# if delta11 < tau1:
if max(p11, p12) < tau1 and delta11<tau_delta:
print("RED:->BUT still use the VIC's labels.")
# print(f"shape of 11: {p_i_11_ls.shape}")
# print(f"shape of 2: {idx2ls.shape}")
# print(f"shape of 12: {p_i_12_ls.shape}")
p_i_11_ls[i] = p_i2ls[i]
p_m_11_ls[i] = pmask2s[i]
p_logits_11_ls[i] = p_logits2ls[i]
if min(p11, p12) < tau2:
period_break = 0
## compute the current idx11's probablity on current model
pp11ls=[]
pp12ls=[]
for i, prompt in enumerate(p_ls):
p11 = float(torch.sum(torch.exp(old_logits11_ls[i])
* mask11[i, :-1])
/ torch.sum(mask11[i, :-1]))
p12 = float(torch.sum(torch.exp(old_logits12_ls[i])
* mask12[i, :-1])
/ torch.sum(mask12[i, :-1]))
pp11ls.append(p11)
pp12ls.append(p12)
need_pre_store = 0
# Dataset what we seen is about the last stage,
# not current stage.
# If it is the first stage, then we use the victim's label
# to guide the training, for a better bootstrapping.
if vic_logits2ls[0] is not None:
trainset = TensorDataset(
p_i_11_ls,
p_i_12_ls,
p_i2ls,
p_m_11_ls,
p_m_12_ls,
pmask2s,
p_logits_11_ls,
p_logits_12_ls,
p_logits2ls,
p_vic_logits2ls,
)
else:
trainset = TensorDataset(
p_i_11_ls,
p_i_12_ls,
p_i2ls,
p_m_11_ls,
p_m_12_ls,
pmask2s,
p_logits_11_ls,
p_logits_12_ls,
p_logits2ls,
p_logits2ls,
)
# reuse the tensors in current stage as the training dataset
# for next stage.
p_i_11_ls = idxs11_ls
p_i_12_ls = idxs12_ls
p_i2ls = idx2ls
p_m_11_ls = mask11
p_m_12_ls = mask12
pmask2s = mask2
p_logits_11_ls = old_logits11_ls
p_logits_12_ls = old_logits12_ls
p_logits2ls = old_logits2_ls
p_vic_logits2ls = vic_logits2ls
if period_break == 1:
print("\n\n NOW BREAK SINCE ENOUGH TRAINING\n\n")
return lm, p_i_11_ls, p_i_12_ls, p_m_11_ls,\
p_m_12_ls, p_logits_11_ls, p_logits_12_ls,\
p_i2ls, pmask2s, p_logits2ls, p_vic_logits2ls,\
pp11ls, pp12ls
loader = DataLoader(trainset,
batch_size=args.batch_size,
shuffle=True,
)
# #### Transform back.
# #### Transform back.
# lm=None
# print("TO AUTO.")
# lm = AutoModelForCausalLM.from_pretrained(
# args.from_path,
# device_map="auto",
# trust_remote_code=True,
# torch_dtype=torch.bfloat16,
# )
# if args.use_lora == 1:
# lm = PeftModel.from_pretrained(lm, args.temp_save_path)
# print("TO AUTO DONE.")
print(">>>> Period {}".format(iter_idx))
# print("warning: Set Epoch to 1 without any reason.")
# args.epoch=1
lm = one_period(args, lm,
lm_tokenizer,
loader,
args.epoch, args.device,
tb_writer,
tensorboard_name,
args.save_path,
args.LR,
args.acc_step, args.log_step,
args.save_step,
args.beta,
is_black_box=0,
method=args.task,
)
# lm_tokenizer.save_pretrained(args.save_path+"___finally")
# lm.save_pretrained(args.save_path+"___finally")
return lm, p_i_11_ls, p_i_12_ls, p_m_11_ls,\
p_m_12_ls, p_logits_11_ls, p_logits_12_ls,\
p_i2ls, pmask2s, p_logits2ls, p_vic_logits2ls, \
pp11ls, pp12ls
def one_period(args, lm,
lm_tokenizer,
loader, epoch, device,
tb_writer,
tensorboard_name,
save_path,
LR=3e-5,
acc_step=1,
log_step=100,
save_step=1000,
beta=0.7,
epsln=1e-6,
is_black_box=0,
method="LORD-II",
):
overall_loss = 0.
overall_step = 0
pad_token_id = lm_tokenizer.pad_token_id
kl_loss = torch.nn.KLDivLoss(reduction="none")
sigmoid = torch.nn.Sigmoid()
opt1 = torch.optim.AdamW(lm.parameters(), lr=LR)
for e in tqdm(range(epoch), desc="epoch"):
for item in tqdm(loader, desc="loader"):
overall_step += 1
# print(item)
idxs11, idxs12, idxs2, mask11, mask12,\
mask2, old_logits11, old_logits12,\
old_logits2, vic_logits2 = item
bs, sqlen1 = idxs11.shape
sqlen = sqlen1
idxs11 = idxs11.to(device) # bs, sql
idxs12 = idxs12.to(device) # bs, sql
idxs2 = idxs2.to(device) # bs, sql
mask11 = mask11.to(device)
mask12 = mask12.to(device)
mask2 = mask2.to(device)
# mask11 = mask11 == 0
# mask12 = mask12 == 0
# print(f"idxs11 shape: {idxs11.shape}")
# print(f"idxs12 shape: {idxs12.shape}")
# print(f"idxs2 shape: {idxs2.shape}")
# print(f"MASK11: {mask11}")
# print(f"MASK12: {mask12}")
# print(f"MASK2: {mask2}")
# already normalized by softmax
old_logits11 = old_logits11.to(device) # bs, sql,
old_logits12 = old_logits12.to(device) # bs, sql,
old_logits2 = old_logits2.to(device) # bs, sql,
if args.is_black_box==0:
vic_logits2 = vic_logits2.to(device) # bs, sql, 5
# print("===========================================")
# print(f">>>idx11text: {lm_tokenizer.decode(idxs11[0])}")
# print(f">>>idx12text: {lm_tokenizer.decode(idxs12[0])}")
# print(f">>>idx2text: {lm_tokenizer.decode(idxs2[0])}")
# loss1 = lm(idxs11,
# labels=labels11).loss
# loss2 = lm(idxs12,
# labels=labels12).loss
# # loss = loss1-loss2
# loss = loss1
# hard to say: our method.
logits11 = lm(idxs11).logits[:, :-1, :]
logits11 = F.log_softmax(logits11, dim=-1)
logits11 = logits11[torch.arange(bs).unsqueeze(1),
torch.arange(sqlen-1).unsqueeze(0),
idxs11[:, 1:sqlen]]
logits12 = lm(idxs12).logits[:, :-1, :]
logits12 = F.log_softmax(logits12, dim=-1)
logits12 = logits12[torch.arange(bs).unsqueeze(1),
torch.arange(sqlen-1).unsqueeze(0),
idxs12[:, 1:sqlen]]
logits2 = lm(idxs2).logits[:, :-1, :]
logits2 = torch.log_softmax(logits2, dim=-1)
logits2_cons = logits2[torch.arange(bs).unsqueeze(1),
torch.arange(sqlen-1).unsqueeze(0),
idxs2[:, 1:sqlen]]
# mask = torch.logical_or(mask11, mask12).long()
# term1 = torch.sum(log_clip(-old_logits12+logits12)
# * mask12[:, :-1])
# term2 = torch.sum(log_clip(old_logits11-logits11)
# * mask11[:, :-1])
term1 = torch.mean(log_clip(-old_logits12+logits12))
term2 = torch.mean(log_clip(old_logits11-logits11))
if args.is_black_box == 0:
term3 = \
(vic_logits2[:, :, 0]+old_logits2-2*logits2_cons)
else:
term3 = old_logits2 - logits2_cons
term3 = torch.sum(term3 * mask2[:, :-1])
if method == "LoRD-II":
loss_1 = term1 + term2
loss_2 = term3
loss = loss_1 + loss_2
elif method == "LoRD-V":
if args.is_black_box == 0:
term3 = \
(vic_logits2[:, :, 0]+old_logits2-2*logits2_cons)
else:
term3 = old_logits2 - logits2_cons
term3 = torch.sum(log_clip(term3) * mask2[:, :-1])
loss = 2*term2 - term1 + term3
# elif method == "LoRD-VI":
# if args.is_black_box == 0:
# # term3 = \
# # (vic_logits2[:, :, 0]+old_logits2-2*logits2_cons)
# term3 = \
# (vic_logits2[:, :, 0]-logits2_cons)
# else:
# # term3 = old_logits2 - logits2_cons
# term3 = - logits2_cons
# # term3 = torch.sum(log_clip(term3) * mask2[:, :-1])
# # term3 = torch.sum(log_clip(term3) * mask2[:, :-1])
# term3 = torch.mean(log_clip(term3))
# # loss = -1*torch.mean(logits2_cons-logits12)\
# # -0.5*log_clip(torch.mean(logits11-logits12))
# loss = -1*torch.mean(logits2_cons)\
# -1*log_clip(torch.mean(logits11-logits12))
# # loss = \
# # -1*log_clip(torch.mean(logits11-logits12))
# print(f"TERM1: {term1}\nTERM2: {term2}\nTERM3: {term3}\n")
# print(f"LOSS: {loss}\n\n")
elif method =="w.y_vic":
term3=-logits2_cons.clone().detach()
term3=torch.mean(term3)
# print(term3)
los2=-1*torch.mean(logits2_cons)
loss11=-1*torch.mean(logits11)
loss12=1*torch.mean(logits12)
loss= term3*(loss11+loss12)+(los2+loss12)
loss=sigmoid(loss)
elif method=="usey+":
los2=-1*torch.mean(logits2_cons)
loss11=-1*torch.mean(logits11)
loss12=1*torch.mean(logits12)
loss= (loss11+loss12)+(los2-loss11)
loss=sigmoid(loss)
elif method=="w.KL":
los2=-1*torch.mean(logits2_cons)
loss11=-1*torch.mean(logits11)
loss12=1*torch.mean(logits12)
target = torch.exp(logits2,)
prediction_logsf = logits12.clone().detach()
KL_loss = F.kl_div(prediction_logsf, target)
KL_loss = torch.mean(KL_loss)
loss= (loss11+loss12)+KL_loss
loss=sigmoid(loss)
elif method=="w.o.Sigmoid":
los2=-1*torch.mean(logits2_cons)
loss11=-1*torch.mean(logits11)
loss12=1*torch.mean(logits12)
loss= log_clip(loss11+loss12)+log_clip(los2+loss12)
loss=loss
elif method=="simPO":
beta=2.5
gamma=0.55*beta
loss = beta*torch.mean(logits11)-beta*torch.mean(logits12)
loss = loss - gamma
loss=-torch.log(sigmoid(loss))
elif method=="simPO2":
beta=2.5
gamma=0.55*beta
loss = beta*torch.mean(logits2_cons)-beta*torch.mean(logits12)
loss = loss - gamma
loss=-torch.log(sigmoid(loss))
elif method in ["LoRD-VI", "LoRD-VII", "LoRD-VIII", "LoRD-IX"]:
if args.is_black_box == 0:
term3 = \
(vic_logits2[:, :, 0]-logits2_cons)
else:
# term3 = old_logits2 - logits2_cons
term3 = - logits2_cons
term3 = torch.mean(log_clip(term3))
# loss = -1*torch.mean(logits2_cons)\
# -1*log_clip(torch.mean(logits11-logits12))
# loss = -1*log_clip(torch.mean(logits2_cons-logits12))\
# -1*log_clip(torch.mean(logits11-logits12))
# print(f"logits2_cons: {logits2_cons}")
# print(f"logits11: {logits11}")
# print(f"logits12: {logits12}")
# print(f"mask2: {mask2}")
# print(f"mask11: {mask11}")
# print(f"mask12: {mask12}")
# los2=-1*(torch.sum(logits2_cons*mask2[:,:-1])/torch.sum(mask2[:,:-1]))
# loss11=-1*(torch.sum(logits11*mask11[:,:-1])/torch.sum(mask11[:,:-1]))
# loss12=1*(torch.sum(logits12*mask12[:,:-1])/torch.sum(mask12[:,:-1]))
los2=-1*torch.mean(logits2_cons)
loss11=-1*torch.mean(logits11)
loss12=1*torch.mean(logits12)
loss = los2+loss11+2*loss12
if method=="LoRD-VII":
# loss=sigmoid(loss/(loss11+loss12))
loss=sigmoid(loss/(los2))
# loss=sigmoid(loss/loss11)
elif method=="LoRD-VIII":
# Failed
# loss=sigmoid((los2+loss12)/los2+(loss11+loss12)/loss11)
# In validating
# loss=sigmoid((los2+loss12)/loss11+(loss11+loss12)/los2)
loss=2*sigmoid(args.lambda1*(los2+loss12)/loss11+(1-args.lambda1)*(loss11+loss12)/los2)
elif method=="LoRD-IX":
# loss=sigmoid(args.lambda1*(los2+loss12)*(-1*loss11)+(1-args.lambda1)*(loss11+loss12)*(-1*los2))
# loss=args.lambda1*(logits2_cons-logits12)*(-logits11)+(1-args.lambda1)*(logits11-logits12)*(-logits2_cons)
# loss=args.lambda1*log_clip(logits2_cons-logits12)*(-logits11)+(1-args.lambda1)*log_clip(logits11-logits12)*(-logits2_cons)
loss=args.lambda1*log_clip(logits2_cons-logits12)+(1-args.lambda1)*log_clip(logits11-logits12)
# loss=torch.mean(sigmoid(loss))
loss=torch.mean(loss)
else:
loss=sigmoid(loss)
# loss=sigmoid(loss/loss12)
print(f"term2: {-1*torch.sum(logits2_cons*mask2[:,:-1])/torch.sum(mask2[:,:-1])}")
print(f"term11: {-1*torch.sum(logits11*mask11[:,:-1])/torch.sum(mask11[:,:-1])}")
print(f"term12: {-2*torch.sum(logits12*mask12[:,:-1])/torch.sum(mask12[:,:-1])}")
# print(f"normalized term2: {-1*clip(torch.sum(logits2_cons*mask2[:,:-1])/torch.sum(mask2[:,:-1]))}")
# print(f"normalized term11: {-1*clip(torch.sum(logits11*mask11[:,:-1])/torch.sum(mask11[:,:-1]))}")
# print(f"normalized term12: {-2*clip(torch.sum(logits12*mask12[:,:-1])/torch.sum(mask12[:,:-1]))}")
print(f"LOSS: {loss}\n\n")
else:
print("NO LOSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSS.")
loss=0.0
return -1