-
Notifications
You must be signed in to change notification settings - Fork 3
/
index.html
922 lines (792 loc) · 32.7 KB
/
index.html
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
<!DOCTYPE html>
<html class="fontawesome-i2svg-active fontawesome-i2svg-complete">
<head>
<meta http-equiv="Content-Type" content="text/html; charset=UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1">
<style>
.content {
font-size: 22px; /* Adjust the font size as desired */
}
</style>
<style type="text/css">svg:not(:root).svg-inline--fa {
overflow: visible
}
.svg-inline--fa {
display: inline-block;
font-size: inherit;
height: 1em;
overflow: visible;
vertical-align: -.125em
}
.svg-inline--fa.fa-lg {
vertical-align: -.225em
}
.svg-inline--fa.fa-w-1 {
width: .0625em
}
.svg-inline--fa.fa-w-2 {
width: .125em
}
.svg-inline--fa.fa-w-3 {
width: .1875em
}
.svg-inline--fa.fa-w-4 {
width: .25em
}
.svg-inline--fa.fa-w-5 {
width: .3125em
}
.svg-inline--fa.fa-w-6 {
width: .375em
}
.svg-inline--fa.fa-w-7 {
width: .4375em
}
.svg-inline--fa.fa-w-8 {
width: .5em
}
.svg-inline--fa.fa-w-9 {
width: .5625em
}
.svg-inline--fa.fa-w-10 {
width: .625em
}
.svg-inline--fa.fa-w-11 {
width: .6875em
}
.svg-inline--fa.fa-w-12 {
width: .75em
}
.svg-inline--fa.fa-w-13 {
width: .8125em
}
.svg-inline--fa.fa-w-14 {
width: .875em
}
.svg-inline--fa.fa-w-15 {
width: .9375em
}
.svg-inline--fa.fa-w-16 {
width: 1em
}
.svg-inline--fa.fa-w-17 {
width: 1.0625em
}
.svg-inline--fa.fa-w-18 {
width: 1.125em
}
.svg-inline--fa.fa-w-19 {
width: 1.1875em
}
.svg-inline--fa.fa-w-20 {
width: 1.25em
}
.svg-inline--fa.fa-pull-left {
margin-right: .3em;
width: auto
}
.svg-inline--fa.fa-pull-right {
margin-left: .3em;
width: auto
}
.svg-inline--fa.fa-border {
height: 1.5em
}
.svg-inline--fa.fa-li {
width: 2em
}
.svg-inline--fa.fa-fw {
width: 1.25em
}
.fa-layers svg.svg-inline--fa {
bottom: 0;
left: 0;
margin: auto;
position: absolute;
right: 0;
top: 0
}
.fa-layers {
display: inline-block;
height: 1em;
position: relative;
text-align: center;
vertical-align: -.125em;
width: 1em
}
.fa-layers svg.svg-inline--fa {
-webkit-transform-origin: center center;
transform-origin: center center
}
.fa-layers-counter, .fa-layers-text {
display: inline-block;
position: absolute;
text-align: center
}
.fa-layers-text {
left: 50%;
top: 50%;
-webkit-transform: translate(-50%, -50%);
transform: translate(-50%, -50%);
-webkit-transform-origin: center center;
transform-origin: center center
}
.fa-layers-counter {
background-color: #ff253a;
border-radius: 1em;
-webkit-box-sizing: border-box;
box-sizing: border-box;
color: #fff;
height: 1.5em;
line-height: 1;
max-width: 5em;
min-width: 1.5em;
overflow: hidden;
padding: .25em;
right: 0;
text-overflow: ellipsis;
top: 0;
-webkit-transform: scale(.25);
transform: scale(.25);
-webkit-transform-origin: top right;
transform-origin: top right
}
.fa-layers-bottom-right {
bottom: 0;
right: 0;
top: auto;
-webkit-transform: scale(.25);
transform: scale(.25);
-webkit-transform-origin: bottom right;
transform-origin: bottom right
}
.fa-layers-bottom-left {
bottom: 0;
left: 0;
right: auto;
top: auto;
-webkit-transform: scale(.25);
transform: scale(.25);
-webkit-transform-origin: bottom left;
transform-origin: bottom left
}
.fa-layers-top-right {
right: 0;
top: 0;
-webkit-transform: scale(.25);
transform: scale(.25);
-webkit-transform-origin: top right;
transform-origin: top right
}
.fa-layers-top-left {
left: 0;
right: auto;
top: 0;
-webkit-transform: scale(.25);
transform: scale(.25);
-webkit-transform-origin: top left;
transform-origin: top left
}
.fa-lg {
font-size: 1.3333333333em;
line-height: .75em;
vertical-align: -.0667em
}
.fa-xs {
font-size: .75em
}
.fa-sm {
font-size: .875em
}
.fa-1x {
font-size: 1em
}
.fa-2x {
font-size: 2em
}
.fa-3x {
font-size: 3em
}
.fa-4x {
font-size: 4em
}
.fa-5x {
font-size: 5em
}
.fa-6x {
font-size: 6em
}
.fa-7x {
font-size: 7em
}
.fa-8x {
font-size: 8em
}
.fa-9x {
font-size: 9em
}
.fa-10x {
font-size: 10em
}
.fa-fw {
text-align: center;
width: 1.25em
}
.fa-ul {
list-style-type: none;
margin-left: 2.5em;
padding-left: 0
}
.fa-ul > li {
position: relative
}
.fa-li {
left: -2em;
position: absolute;
text-align: center;
width: 2em;
line-height: inherit
}
.fa-border {
border: solid .08em #eee;
border-radius: .1em;
padding: .2em .25em .15em
}
.fa-pull-left {
float: left
}
.fa-pull-right {
float: right
}
.fa.fa-pull-left, .fab.fa-pull-left, .fal.fa-pull-left, .far.fa-pull-left, .fas.fa-pull-left {
margin-right: .3em
}
.fa.fa-pull-right, .fab.fa-pull-right, .fal.fa-pull-right, .far.fa-pull-right, .fas.fa-pull-right {
margin-left: .3em
}
.fa-spin {
-webkit-animation: fa-spin 2s infinite linear;
animation: fa-spin 2s infinite linear
}
.fa-pulse {
-webkit-animation: fa-spin 1s infinite steps(8);
animation: fa-spin 1s infinite steps(8)
}
@-webkit-keyframes fa-spin {
0% {
-webkit-transform: rotate(0);
transform: rotate(0)
}
100% {
-webkit-transform: rotate(360deg);
transform: rotate(360deg)
}
}
@keyframes fa-spin {
0% {
-webkit-transform: rotate(0);
transform: rotate(0)
}
100% {
-webkit-transform: rotate(360deg);
transform: rotate(360deg)
}
}
.fa-rotate-90 {
-webkit-transform: rotate(90deg);
transform: rotate(90deg)
}
.fa-rotate-180 {
-webkit-transform: rotate(180deg);
transform: rotate(180deg)
}
.fa-rotate-270 {
-webkit-transform: rotate(270deg);
transform: rotate(270deg)
}
.fa-flip-horizontal {
-webkit-transform: scale(-1, 1);
transform: scale(-1, 1)
}
.fa-flip-vertical {
-webkit-transform: scale(1, -1);
transform: scale(1, -1)
}
.fa-flip-both, .fa-flip-horizontal.fa-flip-vertical {
-webkit-transform: scale(-1, -1);
transform: scale(-1, -1)
}
:root .fa-flip-both, :root .fa-flip-horizontal, :root .fa-flip-vertical, :root .fa-rotate-180, :root .fa-rotate-270, :root .fa-rotate-90 {
-webkit-filter: none;
filter: none
}
.fa-stack {
display: inline-block;
height: 2em;
position: relative;
width: 2.5em
}
.fa-stack-1x, .fa-stack-2x {
bottom: 0;
left: 0;
margin: auto;
position: absolute;
right: 0;
top: 0
}
.svg-inline--fa.fa-stack-1x {
height: 1em;
width: 1.25em
}
.svg-inline--fa.fa-stack-2x {
height: 2em;
width: 2.5em
}
.fa-inverse {
color: #fff
}
.sr-only {
border: 0;
clip: rect(0, 0, 0, 0);
height: 1px;
margin: -1px;
overflow: hidden;
padding: 0;
position: absolute;
width: 1px
}
.sr-only-focusable:active, .sr-only-focusable:focus {
clip: auto;
height: auto;
margin: 0;
overflow: visible;
position: static;
width: auto
}
.svg-inline--fa .fa-primary {
fill: var(--fa-primary-color, currentColor);
opacity: 1;
opacity: var(--fa-primary-opacity, 1)
}
.svg-inline--fa .fa-secondary {
fill: var(--fa-secondary-color, currentColor);
opacity: .4;
opacity: var(--fa-secondary-opacity, .4)
}
.svg-inline--fa.fa-swap-opacity .fa-primary {
opacity: .4;
opacity: var(--fa-secondary-opacity, .4)
}
.svg-inline--fa.fa-swap-opacity .fa-secondary {
opacity: 1;
opacity: var(--fa-primary-opacity, 1)
}
.svg-inline--fa mask .fa-primary, .svg-inline--fa mask .fa-secondary {
fill: #000
}
.fad.fa-inverse {
color: #fff
}</style>
<link rel="stylesheet" href="src/codemirror.min.css">
<script src="src/runmode-standalone.min.js"></script>
<script src="src/python.min.js"></script>
<title>Model Editing Harms LLMs</title>
<!-- Google tag (gtag.js)
<script async src="https://www.googletagmanager.com/gtag/js?id=G-S3G8BGY5W5"></script>
<script>
window.dataLayer = window.dataLayer || [];
function gtag(){dataLayer.push(arguments);}
gtag('js', new Date());
gtag('config', 'G-S3G8BGY5W5');
</script>
-->
<link href="src/css" rel="stylesheet">
<link rel="stylesheet" href="src/bulma.min.css">
<link rel="stylesheet" href="src/bulma-carousel.min.css">
<link rel="stylesheet" href="src/bulma-slider.min.css">
<link rel="stylesheet" href="src/fontawesome.all.min.css">
<link rel="stylesheet" href="src/academicons.min.css">
<link rel="stylesheet" href="src/index.css">
<link rel="icon" href="src/pluslab.png">
<script src="src/jquery.min.js"></script>
<script defer="" src="src/fontawesome.all.min.js"></script>
<script src="src/bulma-carousel.min.js"></script>
<script src="src/bulma-slider.min.js"></script>
<script src="src/index.js"></script>
</head>
<body>
<nav class="navbar" role="navigation" aria-label="main navigation">
<div class="navbar-brand">
<a role="button" class="navbar-burger" aria-label="menu" aria-expanded="false">
<span aria-hidden="true"></span>
<span aria-hidden="true"></span>
<span aria-hidden="true"></span>
</a>
</div>
<div class="navbar-menu">
<div class="navbar-start" style="flex-grow: 1; justify-content: center;">
<a class="navbar-item" href="https://jasonforjoy.github.io/Model-Editing-Hurt/">
<span class="icon">
<i class="fas fa-home"></i>
</span>
</a>
<div class="navbar-item has-dropdown is-hoverable">
<a class="navbar-link">
More Research
</a>
<div class="navbar-dropdown">
<a class="navbar-item" href="https://jasonforjoy.github.io/BRIEF/">
BRIEF
</a>
<a class="navbar-item" href="https://mragbench.github.io/">
MRAG-Bench
</a>
<a class="navbar-item" href="https://xiaowu0162.github.io/syncheck/">
SynCheck
</a>
</div>
</div>
</div>
</div>
</nav>
<section class="hero">
<div class="hero-body">
<div class="container is-max-desktop">
<div class="columns is-centered">
<div class="column has-text-centered">
<h1 class="title is-1 publication-title">
<!-- <span class="contextual" style="vertical-align: middle">Repoformer</span> -->
</h1>
<h2 class="subtitle is-3 publication-subtitle" >
<b>Model Editing Harms General Abilities of Large Language Models: Regularization to the Rescue</b></h2>
<h3><strong>Accepted by EMNLP 2024 (Main Track)</strong>🎉</h3>
<div class="is-size-5 publication-authors">
<span class="author-block">
<a href="https://jasonforjoy.github.io/">Jia-Chen Gu</a><sup>1</sup>,
</span>
<span class="author-block">
Hao-Xiang Xu<sup>2</sup>,
</span>
<span class="author-block">
<a href="http://home.ustc.edu.cn/~mjy1999/">Jun-Yu Ma</a><sup>2</sup>,
</span>
<span class="author-block">
<a href="https://lupantech.github.io/">Pan Lu</a><sup>1</sup>,
</span>
<span class="author-block">
<a href="http://staff.ustc.edu.cn/~zhling/">Zhen-Hua Ling</a><sup>2</sup>,
</span>
<span class="author-block">
<a href="https://web.cs.ucla.edu/~kwchang/">Kai-Wei Chang</a><sup>1</sup>,
</span>
<span class="author-block">
<a href="https://violetpeng.github.io/">Nanyun Peng</a><sup>1</sup>
</span>
</div>
<div class="is-size-5 publication-authors">
<sup>1</sup><span class="author-block">University of California, Los Angeles</span>
<sup>2</sup><span class="author-block">University of Science and Technology of China</span>
</div>
<div class="column has-text-centered">
<div class="publication-links">
<!-- ArXiv abstract Link -->
<span class="link-block">
<a href="https://arxiv.org/pdf/2401.04700" target="_blank"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="ai ai-arxiv"></i>
</span>
<span>arXiv</span>
</a>
</span>
<!-- Code Link. -->
<span class="link-block">
<a href="https://github.com/JasonForJoy/Model-Editing-Hurt"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<svg class="svg-inline--fa fa-github fa-w-16" aria-hidden="true" focusable="false"
data-prefix="fab" data-icon="github" role="img" xmlns="http://www.w3.org/2000/svg"
viewBox="0 0 496 512" data-fa-i2svg=""><path fill="currentColor"
d="M165.9 397.4c0 2-2.3 3.6-5.2 3.6-3.3.3-5.6-1.3-5.6-3.6 0-2 2.3-3.6 5.2-3.6 3-.3 5.6 1.3 5.6 3.6zm-31.1-4.5c-.7 2 1.3 4.3 4.3 4.9 2.6 1 5.6 0 6.2-2s-1.3-4.3-4.3-5.2c-2.6-.7-5.5.3-6.2 2.3zm44.2-1.7c-2.9.7-4.9 2.6-4.6 4.9.3 2 2.9 3.3 5.9 2.6 2.9-.7 4.9-2.6 4.6-4.6-.3-1.9-3-3.2-5.9-2.9zM244.8 8C106.1 8 0 113.3 0 252c0 110.9 69.8 205.8 169.5 239.2 12.8 2.3 17.3-5.6 17.3-12.1 0-6.2-.3-40.4-.3-61.4 0 0-70 15-84.7-29.8 0 0-11.4-29.1-27.8-36.6 0 0-22.9-15.7 1.6-15.4 0 0 24.9 2 38.6 25.8 21.9 38.6 58.6 27.5 72.9 20.9 2.3-16 8.8-27.1 16-33.7-55.9-6.2-112.3-14.3-112.3-110.5 0-27.5 7.6-41.3 23.6-58.9-2.6-6.5-11.1-33.3 2.6-67.9 20.9-6.5 69 27 69 27 20-5.6 41.5-8.5 62.8-8.5s42.8 2.9 62.8 8.5c0 0 48.1-33.6 69-27 13.7 34.7 5.2 61.4 2.6 67.9 16 17.7 25.8 31.5 25.8 58.9 0 96.5-58.9 104.2-114.8 110.5 9.2 7.9 17 22.9 17 46.4 0 33.7-.3 75.4-.3 83.6 0 6.5 4.6 14.4 17.3 12.1C428.2 457.8 496 362.9 496 252 496 113.3 383.5 8 244.8 8zM97.2 352.9c-1.3 1-1 3.3.7 5.2 1.6 1.6 3.9 2.3 5.2 1 1.3-1 1-3.3-.7-5.2-1.6-1.6-3.9-2.3-5.2-1zm-10.8-8.1c-.7 1.3.3 2.9 2.3 3.9 1.6 1 3.6.7 4.3-.7.7-1.3-.3-2.9-2.3-3.9-2-.6-3.6-.3-4.3.7zm32.4 35.6c-1.6 1.3-1 4.3 1.3 6.2 2.3 2.3 5.2 2.6 6.5 1 1.3-1.3.7-4.3-1.3-6.2-2.2-2.3-5.2-2.6-6.5-1zm-11.4-14.7c-1.6 1-1.6 3.6 0 5.9 1.6 2.3 4.3 3.3 5.6 2.3 1.6-1.3 1.6-3.9 0-6.2-1.4-2.3-4-3.3-5.6-2z"></path></svg>
<!-- <i class="fab fa-github"></i> Font Awesome fontawesome.com -->
</span>
<span>Code</span>
</a>
</span>
<!-- Model Link. -->
<!-- Twitter Link -->
<span class="link-block">
<a href="https://x.com/Jiachen_Gu/status/1744932437006885060" target="_blank" class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<p style="font-size:18px">🌐</p>
</span>
<span>Twitter</span>
</a>
</span>
</div>
</div>
</div>
</div>
</div>
</div>
</section>
<section class="section hero is-small">
<div class="container is-max-desktop">
<div class="columns is-centered has-text-centered">
<div class="content has-text-justified">
<h2 class="subtitle is-size-3-tablet has-text-weight-bold has-text-centered has-background-info-light mr-0 pt-3 pb-3">
Abstract
</h2>
<p>
Model editing is a technique that edits the large language models (LLMs) with updated knowledge to alleviate hallucinations without resource-intensive retraining. While current model editing methods can effectively modify a model’s behavior within a specific area of interest, they often overlook the potential unintended side effects on the general abilities of LLMs such as reasoning, natural language inference, and question answering. In this paper, we raise concerns that model editing’s improvements on factuality may come at the cost of a significant degradation of the model’s general abilities. We systematically analyze the side effects by evaluating four popular editing methods on three LLMs across eight representative tasks. Our extensive empirical experiments show that it is challenging for current editing methods to simultaneously improve factuality of LLMs and maintain their general abilities. Our analysis reveals that the side effects are caused by model editing altering the original model weights excessively, leading to overfitting to the edited facts. To mitigate this, a method named RECT is proposed to regularize the edit update weights by imposing constraints on their complexity based on the <b><span class="dnerf">RE</span></b>lative <b><span class="dnerf">C</span></b>hange in weigh<b><span class="dnerf">T</span></b>. Evaluation results show that RECT can significantly mitigate the side effects of editing while still maintaining over 94% editing performance.
</p>
<br>
</div>
</div>
</div>
</section>
<section class="section">
<div class="container is-max-desktop">
<h2 class="subtitle is-size-3-tablet has-text-weight-bold has-text-centered has-background-info-light mr-0 pt-3 pb-3">
Model Editing: Definition and Evaluation
</h2>
<div class="content has-text-justified">
<p>
Model editing aims to efficiently change model behavior without affecting other inputs. The effectiveness is evaluated by:
</p>
<p>
<b><span class="dnerf">Generalization</span></b>: recall the fact under the in-scope paraphrase prompts. E.g., Who currently holds the office of President of the United States?
</p>
<p>
<b><span class="dnerf">Locality</span></b>: remain unchanged for the prompts out of the editing scope. E.g., Who is the president of France?
</p>
<img src="src/editing_definition.png" class="center">
</div>
</div>
<!--/ Abstract. -->
</section>
<section class="section">
<div class="container is-max-desktop">
<h2 class="subtitle is-size-3-tablet has-text-weight-bold has-text-centered has-background-info-light mr-0 pt-3 pb-3">
Side Effects of Model Editing
</h2>
<div class="content has-text-justified">
<p>
In this work, we ask:
<center><b><em>Does model editing's improvements on factuality come at the cost of a significant degradation of the model's general abilities?</em></b></center>
</p>
<img src="src/editing_side_effect.png" class="center", style="width: 60%; display: block; margin: auto;">
<p>
Our research shows that LLMs are not robust to weight perturbations.
</p>
</div>
</div>
</section>
<section class="section">
<div class="container is-max-desktop">
<h2 class="subtitle is-size-3-tablet has-text-weight-bold has-text-centered has-background-info-light mr-0 pt-3 pb-3">
Evaluation Paradigm
</h2>
<div class="content has-text-justified">
<p>
<b><em>Single- vs. Sequential-editing</em></b>: Make a single editing operation or multiple editing operations successively.
</p>
<p>
<b><em>Instance- vs. Batch-editing</em></b>: Use only one instance or multiple instances per editing operation.
</p>
<p>
<b><em>Zero-shot</em></b> performance of unedited and edited models on a variety of tasks.
</p>
<img src="src/editing_paradigm.png" class="center", style="width: 60%; display: block; margin: auto;">
</div>
</div>
</section>
<section class="section">
<div class="container is-max-desktop">
<h2 class="subtitle is-size-3-tablet has-text-weight-bold has-text-centered has-background-info-light mr-0 pt-3 pb-3">
Evaluation Setup
</h2>
<div class="content has-text-justified">
<p>
<b><em>Editing Methods</em></b>: KN, MEND, ROME and MEMIT (only MEND and MEMIT support batch-editing).
</p>
<p>
<b><em>Editing Dataset</em></b>: Zero-Shot Relation Extraction (ZSRE).
</p>
<p>
<b><em>Selected LLMs</em></b>: GPT-2 XL (1.5B), LLaMA-1 (7B), and LLaMA-2 (7B).
</p>
<p>
<b><em>Downstream Tasks, Datasets and Metrics</em></b>:
<ul>
<li>Reasoning: GSM8K (solve rate)</li>
<li>NLI: RTE (accuracy)</li>
<li>Open-domain QA: Natural Question (EM) </li>
<li>Closed-domain QA: BoolQ (EM)</li>
<li>Dialogue: MuTual (Recall4@1) </li>
<li>Summarization: SAMSum (ROUGE)</li>
<li>NER: CoNLL03 (entity-level F1)</li>
<li>Sentiment analysis: SST2 (accuracy)</li>
</ul>
</p>
</div>
</div>
</section>
<section class="section">
<div class="container is-max-desktop">
<h2 class="subtitle is-size-3-tablet has-text-weight-bold has-text-centered has-background-info-light mr-0 pt-3 pb-3">
Evaluation Results
</h2>
<div class="content has-text-justified">
<b><em>Impact of Sequential-editing</em></b>
<img src="src/result_sequential_editing.png" class="center", style="width: 100%; display: block; margin: auto;">
<p>
<ul>
<li>Although there is only one instance per editing operation, the performance of edited models on various tasks fluctuates significantly and shows a downward trend as the number of edits increases. Strikingly, the use of KN resulted in a drastic performance degradation to nearly zero on all selected tasks with just a single edit.</li>
<li>The selected LLMs are not robust to weight perturbations even if less than 1% of the parameters are edited, whereby slight perturbations may significantly affect their general abilities.</li>
<li>The difficulty of effectively coupling current editing algorithms with LLMs lies in the dual objective of improving model factuality while simultaneously maintaining their general abilities.</li>
</ul>
</p>
<b><em>Impact of Batch-editing</em></b>
<img src="src/result_batch_editing.png" class="center", style="width: 100%; display: block; margin: auto;">
<p>
<ul>
<li>Even with only one single editing operation, edited models exhibited a trend of performance degradation as the batch size increases in most cases.</li>
<li>Underline the sensitivity of the models to increases in batch size, calling for more research work on scalable editing to facilitate efficient editing of multiple instances.</li>
</ul>
</p>
<b><em>Impact of Batch- and Sequential-editing</em></b>
<img src="src/result_batch_sequential_editing_1.png" class="center", style="width: 100%; display: block; margin: auto;">
<img src="src/result_batch_sequential_editing_2.png" class="center", style="width: 100%; display: block; margin: auto;">
<p>
<ul>
<li>A joint setting to understand how these two factors collaboratively influence, also echoing our previous observations.</li>
</ul>
</p>
</div>
</div>
</section>
<section class="section">
<div class="container is-max-desktop">
<h2 class="subtitle is-size-3-tablet has-text-weight-bold has-text-centered has-background-info-light mr-0 pt-3 pb-3">
Analysis of Causes of Side Effects
</h2>
<div class="content has-text-justified">
<p>
The side effects of editing come from changing the original model weights too much, resulting in <b><em>overfitting</em></b> to the editing facts.
</p>
<p>
Define the absolute value of the relative change in weight δ to characterize the degree of change of each element in ∆W.
</p>
<img src="src/RECT_definition.png" class="center", style="width: 20%; display: block; margin: auto;">
<img src="src/analysis_statistics.png" class="center", style="width: 60%; display: block; margin: auto;">
<p>
<ul>
<li>∆W might be quite sparse, while most elements in ∆W are minor: only 20% (10%) elements have δ greater than 0.077 (0.171).</li>
<li>Accumulation of overfitting across multiple edits can amplify changes to the original weights: the proportion of elements whose δ is greater than a certain threshold increases significantly as the number of edits increases.</li>
</ul>
</p>
<img src="src/analysis_visualize.png" class="center", style="width: 60%; display: block; margin: auto;">
<p>
<ul>
<li>Visualizing the weight change |∆W| as distinction between the final edited weight and the original unedited weight: consistent findings that the update weight ∆W might be quite sparse, while the accumulation across multiple edits can amplify changes.</li>
</ul>
</p>
</div>
</div>
</section>
<section class="section">
<div class="container is-max-desktop">
<h2 class="subtitle is-size-3-tablet has-text-weight-bold has-text-centered has-background-info-light mr-0 pt-3 pb-3">
RECT: RElative Change in weighT
</h2>
<div class="content has-text-justified">
<p>
A regularization method named RECT to <b><em>discourage overly complex editing updates</em></b> that are more likely to overfit. δ can be used to indicate the importance of each element in ∆W when inserting new editing facts.
<ul>
<li><b><em>Principal editing</b></em>: top-k% elements in ∆W that change the most according to δ, keep their original values.</li>
<li><b><em>Minor editing</b></em>: for the remaining elements in ∆W, they are treated as minor contributions to editing and set to zero for regularization.</li>
</ul>
</p>
<img src="src/RECT_formula.png" class="center", style="width: 40%; display: block; margin: auto;">
<img src="src/RECT_overview.png" class="center", style="width: 70%; display: block; margin: auto;">
</div>
</div>
</section>
<section class="section">
<div class="container is-max-desktop">
<h2 class="subtitle is-size-3-tablet has-text-weight-bold has-text-centered has-background-info-light mr-0 pt-3 pb-3">
Regularization Results
</h2>
<div class="content has-text-justified">
<b><em>Editing Performance</em></b>
<img src="src/RECT_editing.png" class="center", style="width: 100%; display: block; margin: auto;">
<p>
<ul>
<li>RECT that keeps an appropriate top-40% elements in ∆W can help maintain over 94% majority of reliability and generalization, and even improve locality.</li>
<li>Setting excessive elements in ∆W to zero (top-20%) might hurt editing as partial important editing information is accidentally removed.</li>
<li>Compared with Random 40% and PCA 40%, RECT top-40% achieves the best, indicating its effectiveness in selecting the principal editing information.</li>
<li>RECT also exhibits advantages in efficiency, since it eliminates the complex calculations required in PCA.</li>
</ul>
</p>
<b><em>General Downstream Task Performance</em></b>
<img src="src/RECT_general.png" class="center", style="width: 100%; display: block; margin: auto;">
<p>
<ul>
<li>As the proportion of elements in ∆W set to 0 increases, the more editing overfitting is regularized, the smaller the change to the original weight, so the general abilities can be more preserved compared with unregularized ones in most tasks.</li>
<li>It still poses a challenge for some tasks such as sentiment analysis, and remains unclear whether it works for larger number of edits, which will be left to future work.</li>
</ul>
</p>
</div>
</div>
</section>
<section class="section">
<div class="container is-max-desktop">
<h2 class="subtitle is-size-3-tablet has-text-weight-bold has-text-centered has-background-info-light mr-0 pt-3 pb-3">
Takeaway
</h2>
<div class="content has-text-justified">
<p>
<ul>
<li>Existing editing algorithms can cause perturbations to LLMs.</li>
<li>How to mitigate the side effects of editing to prevent harm to general abilities of LLMs? Regularization to the rescue.</li>
<li>Model editing is a good solution for upgrading models in terms of specificity and cost, but is currently unstable and uncontrollable which is calling for more investigation.</li>
</ul>
</p>
</div>
</div>
</section>
<section class="section" id="BibTeX">
<div class="container is-max-desktop" content>
<h2 class="title">BibTeX</h2>
<pre><code>@inproceedings{gu-etal-2024-model,
title = "Model Editing Harms General Abilities of Large Language Models: Regularization to the Rescue",
author = "Gu, Jia-Chen and
Xu, Hao-Xiang and
Ma, Jun-Yu and
Lu, Pan and
Ling, Zhen-Hua and
Chang, Kai-Wei and
Peng, Nanyun",
booktitle = "Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing",
year = "2024",
publisher = "Association for Computational Linguistics"
}</code></pre>
</div>
</section>
<footer class="footer">
<div class="container">
<div class="columns is-centered">
<div class="column is-8">
<div class="content">
<p style="color:gray;font-size:9.9px;">
This page was built using the <a href="https://github.com/eliahuhorwitz/Academic-project-page-template" target="_blank">Academic Project Page Template</a>.
</p>
</div>
</div>
</div>
</div>
</footer>
</body>
</html>