-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathindex.html
More file actions
973 lines (887 loc) · 33.2 KB
/
Copy pathindex.html
File metadata and controls
973 lines (887 loc) · 33.2 KB
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
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="utf-8" />
<meta content="width=device-width, initial-scale=1.0" name="viewport" />
<title>CapaBench: Modular Attribution Benchmark</title>
<link href="https://cdnjs.cloudflare.com/ajax/libs/bootstrap/5.3.0/css/bootstrap.min.css" rel="stylesheet" />
<link href="https://cdnjs.cloudflare.com/ajax/libs/aos/2.3.4/aos.css" rel="stylesheet" />
<style>
:root {
--primary-color: #4b6cb7;
--secondary-color: #182848;
--accent-color: #8e44ad;
--text-color: #2c3e50;
--light-bg: #f8f9fa;
}
body {
font-family: 'Inter', -apple-system, BlinkMacSystemFont, sans-serif;
line-height: 1.8;
color: var(--text-color);
overflow-x: hidden;
}
header {
background: linear-gradient(135deg, var(--primary-color), var(--secondary-color));
color: white;
padding: 60px 0;
position: relative;
overflow: hidden;
}
header::before {
content: '';
position: absolute;
top: 0;
left: 0;
right: 0;
bottom: 0;
background: url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 200 200"><path fill="%23ffffff10" d="M45,-52.9C57.5,-39.3,66.1,-23.1,69.9,-4.8C73.7,13.5,72.8,33.9,62.1,47.2C51.4,60.5,31.1,66.7,11.4,67.5C-8.2,68.3,-27.2,63.7,-43.5,52.9C-59.8,42,-73.3,25,-76.3,5.7C-79.3,-13.6,-71.7,-35.2,-57.3,-49.4C-42.9,-63.5,-21.4,-70.2,-2.2,-68.3C17,-66.4,34,-66.4,45,-52.9Z" transform="translate(100 100)"/></svg>') no-repeat center center;
opacity: 0.1;
transform: scale(5);
}
header h1 {
font-size: 4rem;
font-weight: 700;
margin-bottom: 20px;
text-shadow: 2px 2px 4px rgba(0, 0, 0, 0.2);
}
.subtitle {
font-size: 1.8rem;
font-weight: 300;
max-width: 800px;
margin: 0 auto 30px;
}
.authors,
.institutions {
font-size: 1.1rem;
opacity: 0.9;
max-width: 1000px;
margin: 15px auto;
line-height: 1.6;
}
section {
padding: 60px 0;
position: relative;
}
section:nth-child(even) {
background-color: var(--light-bg);
}
.section-title {
font-size: 2.5rem;
color: var(--secondary-color);
margin-bottom: 40px;
position: relative;
display: inline-block;
}
.section-title::after {
content: '';
position: absolute;
bottom: -10px;
left: 0;
width: 60%;
height: 4px;
background: var(--primary-color);
border-radius: 2px;
}
.card {
border: none;
border-radius: 15px;
box-shadow: 0 10px 30px rgba(0, 0, 0, 0.1);
transition: transform 0.3s ease;
height: 100%;
}
.card:hover {
transform: translateY(-5px);
}
table {
border-radius: 10px;
overflow: hidden;
box-shadow: 0 5px 15px rgba(0, 0, 0, 0.1);
}
table th {
background: var(--secondary-color);
color: white;
font-weight: 500;
padding: 15px;
}
table td {
padding: 12px;
transition: background-color 0.3s ease;
}
tr:hover td {
background-color: rgba(75, 108, 183, 0.1);
}
.checkmark {
color: var(--accent-color);
font-size: 1.5rem;
}
.btn-primary {
background: var(--primary-color);
border: none;
padding: 10px 25px;
border-radius: 25px;
transition: all 0.3s ease;
}
.btn-primary:hover {
background: var(--secondary-color);
transform: translateY(-2px);
box-shadow: 0 5px 15px rgba(0, 0, 0, 0.2);
}
.btn-secondary {
background: transparent;
border: 2px solid white;
padding: 10px 25px;
border-radius: 25px;
transition: all 0.3s ease;
}
.btn-secondary:hover {
background: white;
color: var(--primary-color);
}
img {
border-radius: 10px;
box-shadow: 0 10px 30px rgba(0, 0, 0, 0.1);
}
pre {
background: #f8f9fa;
padding: 20px;
border-radius: 10px;
box-shadow: inset 0 2px 4px rgba(0, 0, 0, 0.1);
}
footer {
background: var(--secondary-color);
color: white;
padding: 30px 0;
text-align: center;
}
@media (max-width: 768px) {
header h1 {
font-size: 2.5rem;
}
.subtitle {
font-size: 1.4rem;
}
.section-title {
font-size: 2rem;
}
table {
font-size: 0.9rem;
}
}
</style>
<style>
:root {
--primary-color: #4b6cb7;
--secondary-color: #182848;
--accent-color: #8e44ad;
--text-color: #2c3e50;
--light-bg: #f8f9fa;
}
/* 新增：平滑滚动效果 */
html {
scroll-behavior: smooth;
}
/* 升级：导航栏样式 */
.navbar {
background: rgba(24, 40, 72, 0.85);
backdrop-filter: blur(12px);
-webkit-backdrop-filter: blur(12px);
transition: all 0.4s cubic-bezier(0.4, 0, 0.2, 1);
border-bottom: 1px solid rgba(255, 255, 255, 0.1);
}
.navbar.scrolled {
padding: 5px 0;
background: rgba(24, 40, 72, 0.95);
box-shadow: 0 4px 20px rgba(0, 0, 0, 0.15);
}
.navbar-brand {
font-weight: 600;
background: linear-gradient(135deg, #fff, #e0e0e0);
-webkit-background-clip: text;
-webkit-text-fill-color: transparent;
position: relative;
}
.navbar-brand::after {
content: '';
position: absolute;
bottom: -2px;
left: 0;
width: 0;
height: 2px;
background: linear-gradient(90deg, var(--primary-color), var(--accent-color));
transition: width 0.3s ease;
}
.navbar-brand:hover::after {
width: 100%;
}
.nav-link {
position: relative;
padding: 0.5rem 1rem !important;
margin: 0 0.2rem;
transition: color 0.3s ease;
}
.nav-link::before {
content: '';
position: absolute;
bottom: 0;
left: 50%;
transform: translateX(-50%);
width: 0;
height: 2px;
background: var(--accent-color);
transition: width 0.3s ease;
}
.nav-link:hover::before {
width: 80%;
}
.nav-link.active::before {
width: 80%;
}
/* 升级：进度条 */
.scroll-progress {
position: fixed;
top: 0;
left: 0;
width: 0%;
height: 4px;
background: linear-gradient(90deg,
var(--primary-color),
var(--accent-color),
var(--primary-color));
background-size: 200% 100%;
animation: gradientMove 3s linear infinite;
z-index: 1000;
transition: width 0.3s cubic-bezier(0.4, 0, 0.2, 1);
box-shadow: 0 0 10px rgba(75, 108, 183, 0.5);
}
@keyframes gradientMove {
0% {
background-position: 100% 0;
}
100% {
background-position: -100% 0;
}
}
/* 新增：导航项悬停动画 */
@keyframes navItemGlow {
0% {
box-shadow: 0 0 5px rgba(142, 68, 173, 0);
}
50% {
box-shadow: 0 0 10px rgba(142, 68, 173, 0.3);
}
100% {
box-shadow: 0 0 5px rgba(142, 68, 173, 0);
}
}
.nav-item:hover .nav-link {
animation: navItemGlow 2s infinite;
}
/* 新增：滚动指示器 */
.scroll-indicator {
position: fixed;
right: 20px;
top: 50%;
transform: translateY(-50%);
z-index: 1000;
display: flex;
flex-direction: column;
align-items: center;
gap: 10px;
}
.scroll-dot {
width: 8px;
height: 8px;
border-radius: 50%;
background: rgba(255, 255, 255, 0.3);
transition: all 0.3s ease;
cursor: pointer;
}
.scroll-dot.active {
background: var(--accent-color);
transform: scale(1.5);
box-shadow: 0 0 10px var(--accent-color);
}
/* 新增：文字渐变效果 */
.gradient-text {
background: linear-gradient(135deg, var(--primary-color), var(--accent-color));
-webkit-background-clip: text;
-webkit-text-fill-color: transparent;
background-clip: text;
}
/* 新增：卡片悬停效果 */
.feature-card {
border: none;
border-radius: 15px;
overflow: hidden;
transition: all 0.3s ease;
}
.feature-card:hover {
transform: translateY(-10px);
box-shadow: 0 15px 30px rgba(0, 0, 0, 0.1);
}
.feature-card::before {
content: '';
position: absolute;
top: 0;
left: 0;
width: 100%;
height: 100%;
background: linear-gradient(135deg, rgba(75, 108, 183, 0.1), rgba(142, 68, 173, 0.1));
opacity: 0;
transition: opacity 0.3s ease;
}
.feature-card:hover::before {
opacity: 1;
}
/* 新增：表格动画效果 */
.dataset-table tr {
transition: all 0.3s ease;
}
.dataset-table tr:hover {
background: rgba(75, 108, 183, 0.05);
transform: scale(1.01);
}
/* 新增：返回顶部按钮 */
.back-to-top {
position: fixed;
bottom: 20px;
right: 20px;
width: 40px;
height: 40px;
background: var(--primary-color);
color: white;
border-radius: 50%;
display: flex;
align-items: center;
justify-content: center;
cursor: pointer;
opacity: 0;
transition: all 0.3s ease;
z-index: 1000;
}
.back-to-top.visible {
opacity: 1;
}
/* 新增：响应式调整 */
@media (max-width: 768px) {
.feature-card {
margin-bottom: 20px;
}
header h1 {
font-size: 2.5rem;
}
}
</style>
</head>
<body>
<div class="scroll-progress"></div>
<nav class="navbar navbar-expand-lg navbar-dark fixed-top">
<div class="container">
<a class="navbar-brand" href="#">
<i class="fas fa-cube me-2"></i>CapaBench
</a>
<button class="navbar-toggler" data-bs-target="#navbarNav" data-bs-toggle="collapse" type="button">
<span class="navbar-toggler-icon"></span>
</button>
<div class="collapse navbar-collapse" id="navbarNav">
<ul class="navbar-nav ms-auto">
<li class="nav-item">
<a class="nav-link" href="#intro">
<i class="fas fa-info-circle me-1"></i>Introduction
</a>
</li>
<li class="nav-item">
<a class="nav-link" href="#framework">
<i class="fas fa-project-diagram me-1"></i>Framework
</a>
</li>
<li class="nav-item">
<a class="nav-link" href="#dataset">
<i class="fas fa-database me-1"></i>Dataset
</a>
</li>
<li class="nav-item">
<a class="nav-link" href="#results">
<i class="fas fa-chart-bar me-1"></i>Results
</a>
</li>
</ul>
</div>
</div>
</nav>
<div class="scroll-indicator">
<div class="scroll-dot" data-section="intro"></div>
<div class="scroll-dot" data-section="framework"></div>
<div class="scroll-dot" data-section="dataset"></div>
<div class="scroll-dot" data-section="results"></div>
</div>
<!-- Header -->
<header class="text-center" style="padding-top: 80px;; padding-top: 120px;">
<div class="container">
<h1 data-aos="fade-down">CapaBench</h1>
<p class="subtitle" data-aos="fade-up">A Game-Theoretic Evaluation Benchmark for Modular Attribution in LLM Agents
</p>
<div class="authors" data-aos="fade-up" data-aos-delay="100">
Yingxuan Yang<sup>1</sup>, Bo Huang<sup>1</sup>, Siyuan Qi<sup>1</sup>, Chao Feng<sup>1</sup>,
Haoyi Hu<sup>1</sup>, Yuxuan Zhu<sup>2</sup>, Jinbo Hu<sup>1</sup>, Haoran Zhao<sup>1</sup>,
Ziyi He<sup>3</sup>, Xiao Liu<sup>4</sup>, Zongyu Wang<sup>4</sup>, Lin Qiu<sup>4</sup>,
Xuezhi Cao<sup>4</sup>, Xunliang Cai<sup>4</sup>, Yong Yu<sup>1</sup>, Weinan Zhang<sup>1</sup>
</div>
<div class="institutions" data-aos="fade-up" data-aos-delay="200">
<sup>1</sup>Shanghai Jiao Tong University,
<sup>2</sup>University of Chicago,
<sup>3</sup>University of Toronto,
<sup>4</sup>Meituan
</div>
<p class="mt-4" data-aos="fade-up" data-aos-delay="300">
Contact:
<a class="text-light" href="mailto:zoeyyx@sjtu.edu.cn">zoeyyx@sjtu.edu.cn</a>,
<a class="text-light" href="mailto:wnzhang@sjtu.edu.cn">wnzhang@sjtu.edu.cn</a>
</p>
<div class="mt-4" data-aos="fade-up" data-aos-delay="400">
<a class="btn btn-primary mx-2 mb-2" href="https://arxiv.org/pdf/2502.00510">Download Paper</a>
<a class="btn btn-secondary mx-2 mb-2" href="https://github.com/zoe-yyx/CapaBench/tree/main">View on GitHub</a>
</div>
</div>
</header>
<!-- Main Content -->
<main>
<!-- Introduction -->
<section id="intro">
<div class="container">
<h2 class="section-title" data-aos="fade-right">Introduction</h2>
<div class="row">
<div class="col-lg-12" data-aos="fade-up">
<div class="card p-4">
<p>
CapaBench is a novel evaluation framework leveraging Shapley Value from cooperative game theory to
measure
the contributions of individual modules within modular Large Language Models (LLMs). By quantifying
these
contributions, CapaBench facilitates systematic optimization and enhances the interpretability of agent
architectures.
</p>
<p class="mb-0">
The framework evaluates Planning, Reasoning, Action, and Reflection capabilities by analyzing their
individual and synergistic impacts on task performance. With a comprehensive dataset of over 1,000
multi-round,
practical task scenarios, CapaBench enables robust and generalizable assessments tailored to real-world
applications.
</p>
<p>
</p>
<p>
</p>
<div class="text-center mb-5" data-aos="zoom-in">
<img alt="Agent Workflow" src="images/rader_chart.png" style="width: 95%;" />
</div>
</div>
</div>
</div>
</div>
</section>
<!-- Framework Overview -->
<section id="framework">
<div class="container">
<h2 class="section-title" data-aos="fade-right">Framework Overview</h2>
<h3 class="mb-4" data-aos="fade-up">Agent Workflow</h3>
<div class="text-center mb-5" data-aos="zoom-in">
<img alt="Agent Workflow" src="images/Agent workfolw.png" style="width: 85%;" />
</div>
<div class="row">
<div class="col-lg-12" data-aos="fade-up">
<div class="card p-4">
<h4 class="mb-3">Key Components</h4>
<div class="row">
<div class="col-md-6 mb-4">
<div class="h-100 p-4" style="background: rgba(75,108,183,0.1); border-radius: 10px;">
<h5 class="text-primary">Planning</h5>
<p class="mb-0">Decomposes complex tasks into structured subtasks, enabling efficient resource
allocation.</p>
</div>
</div>
<div class="col-md-6 mb-4">
<div class="h-100 p-4" style="background: rgba(75,108,183,0.1); border-radius: 10px;">
<h5 class="text-primary">Reasoning</h5>
<p class="mb-0">Uses logical inference and contextual understanding to determine appropriate
actions.</p>
</div>
</div>
<div class="col-md-6 mb-4">
<div class="h-100 p-4" style="background: rgba(75,108,183,0.1); border-radius: 10px;">
<h5 class="text-primary">Action</h5>
<p class="mb-0">Translates cognitive processes into executable operations, ensuring effective task
execution.</p>
</div>
</div>
<div class="col-md-6 mb-4">
<div class="h-100 p-4" style="background: rgba(75,108,183,0.1); border-radius: 10px;">
<h5 class="text-primary">Reflection</h5>
<p class="mb-0">Analyzes outcomes to iteratively improve performance through feedback and
adjustments.</p>
</div>
</div>
</div>
</div>
</div>
</div>
</div>
</section>
<!DOCTYPE html>
<!-- Shapley Value Illustration -->
<section id="shapley">
<div class="container">
<h2 class="section-title" data-aos="fade-right">Shapley Value Illustration</h2>
<div class="row">
<div class="col-lg-12" data-aos="fade-up">
<div class="card p-4">
<div class="text-center mb-4">
<img alt="Shapley Value Illustration" src="images/shapley-value-illustration.png" style="width: 85%;" />
</div>
<p>
The Shapley Value, a cornerstone of cooperative game theory, provides a mathematically rigorous method
for
quantifying the marginal contributions of individual modules in an agent's architecture. This ensures
fair
attribution of credit based on all possible permutations of module combinations.
</p>
<p class="mb-0">
For example, in a task requiring Planning, Reasoning, and Action, the Shapley Value captures the unique
contribution of each module and their interactions. This enables a deeper understanding of how modules
work
together to drive performance.
</p>
</div>
</div>
</div>
</div>
</section>
<!-- Dataset Construction -->
<section id="dataset">
<div class="container">
<h2 class="section-title" data-aos="fade-right">Dataset Construction</h2>
<div class="row">
<div class="col-lg-12" data-aos="fade-up">
<div class="card p-4">
<p>
To ensure that our evaluation reflects realistic, multi-faceted application scenarios, we build a
large-scale dataset of over 1,000 multi-round tasks spanning a diverse range of categories (e.g.,
shopping, operation system, robot control, math, and theorem proving).
</p>
<p class="mb-0">
These tasks integrate various capabilities such as planning, tool usage, and reflection, thereby
requiring holistic agent performance rather than isolated skill assessments. Our dataset will be
open-sourced in the future to support further research and development, and we are actively adding more
scenarios to broaden its coverage and applicability.
</p>
<p>
</p>
<div class="table-responsive">
<style>
.table {
border-collapse: collapse;
width: 100%;
}
.table th,
.table td {
border: 1px solid #ddd;
padding: 8px;
text-align: center;
}
.table thead tr {
border-top: 2px solid #000;
border-bottom: 2px solid #000;
}
.table tbody tr:last-child {
border-bottom: 2px solid #000;
}
.check {
color: purple;
}
</style>
<table class="table dataset-table">
<thead>
<tr>
<th></th>
<th></th>
<th>Daily Activities</th>
<th colspan="3">Computation</th>
<th>Role Control</th>
</tr>
<tr>
<th></th>
<th></th>
<th>Shopping</th>
<th>Math</th>
<th>Automatic Theorem Proving</th>
<th>OS</th>
<th>Robot Cooperation</th>
</tr>
</thead>
<tbody>
<tr>
<td rowspan="2">Planning</td>
<td>Task Steps</td>
<td><span class="check">✔</span></td>
<td><span class="check">✔</span></td>
<td><span class="check">✔</span></td>
<td></td>
<td><span class="check">✔</span></td>
</tr>
<tr>
<td>Resource Constraints</td>
<td></td>
<td></td>
<td></td>
<td><span class="check">✔</span></td>
<td></td>
</tr>
<tr>
<td rowspan="3">Reasoning</td>
<td>Logical Validation</td>
<td></td>
<td><span class="check">✔</span></td>
<td><span class="check">✔</span></td>
<td></td>
<td></td>
</tr>
<tr>
<td>Knowledge Inference</td>
<td><span class="check">✔</span></td>
<td></td>
<td></td>
<td><span class="check">✔</span></td>
<td></td>
</tr>
<tr>
<td>Text Understanding</td>
<td></td>
<td></td>
<td></td>
<td></td>
<td><span class="check">✔</span></td>
</tr>
<tr>
<td rowspan="2">Action</td>
<td>Environmental Actions</td>
<td></td>
<td><span class="check">✔</span></td>
<td><span class="check">✔</span></td>
<td><span class="check">✔</span></td>
<td></td>
</tr>
<tr>
<td>Interactive Actions</td>
<td><span class="check">✔</span></td>
<td></td>
<td></td>
<td></td>
<td><span class="check">✔</span></td>
</tr>
<tr>
<td>Reflection</td>
<td>Failure Analysis</td>
<td><span class="check">✔</span></td>
<td><span class="check">✔</span></td>
<td><span class="check">✔</span></td>
<td><span class="check">✔</span></td>
<td><span class="check">✔</span></td>
</tr>
</tbody>
</table>
</div>
</div>
</div>
</div>
</div>
</section>
<!-- Experiment Results -->
<section id="results">
<div class="container">
<h2 class="section-title" data-aos="fade-right">Experiment Results</h2>
<div class="row">
<div class="col-lg-12" data-aos="fade-up">
<div class="card p-4">
<h3>Selected Results Across Datasets</h3>
<p>
The image above highlights results for selected models and datasets. Metrics for baseline models are
highlighted in blue. The evaluation covers nine
models across five primary tasks, showcasing notable performance variations and unique module
contributions. Results marked with
‘*‘ below each dataset indicate the best-performing model combinations computed based on Shapley Value.
</p>
<div class="text-center mb-4">
<img alt="Experimental Results Across Datasets" class="img-fluid" src="images/main_results2.png" />
</div>
</div>
</div>
</div>
<div class="mt-4">
<div class="row">
<div class="col-lg-12" data-aos="fade-up">
<div class="card p-4">
<h3>Key Findings</h3>
<h4>Cross-Task Model Performance Comparison</h4>
<p>
A high-level comparison of model performance across diverse tasks reveals distinct strengths and
weaknesses. Notably, <strong>Claude-3.5</strong> outperforms other models in most categories, showing
particular prowess in formal verification (e.g., Coq, Lean 4, Isabelle) and robot cooperation tasks.
This advantage suggests that <strong>Claude-3.5</strong> has a robust underlying chain-of-thought
reasoning mechanism and effective multi-agent collaboration strategies—capabilities essential for
tasks that demand precise logical proof structures and synchronized actions.
On the other hand, open-source models like <strong>Qwen-2.5</strong> and <strong>Mistral-8X7B</strong>
exhibit moderate gains in more straightforward domains, such as shopping or basic Algebra, but
underperform in cognitive-heavy tasks. Their lag in automatic theorem proving and robot cooperation
implies that while these models may be adept at handling routine queries and procedural
problem-solving, they lack the deeper reasoning, advanced planning, or specialized modules needed for
high-stakes coordination and rigorous proof validation. Strengthening these areas—possibly through
fine-tuning on specialized corpora or integrating more advanced tool usage—could help bridge the gap
between open-source and proprietary models in complex, multi-stage tasks.
</p>
<h4>Module Contribution Patterns</h4>
<p>Our findings highlight that module contributions vary according to task demands, reflecting the
distinct cognitive processes involved. Specifically:</p>
<ul>
<li><strong>Tasks with High Cognitive Complexity (e.g., Online Shopping, Robot Cooperation, and
OS):</strong> Reasoning and Planning play pivotal roles. Online shopping requires balancing
constraints (e.g., budget and preferences) and sequencing decisions effectively. In robot
cooperation, Reasoning enables dynamic information updates and efficient task distribution among
agents. Operation system tasks, involving troubleshooting and resource management, rely heavily on
real-time problem-solving and feedback interpretation. Across these tasks, robust Reasoning ensures
logical inference and decision-making under uncertainty.</li>
<li><strong>Tasks Requiring Precision (e.g., Math Solvers and ATP):</strong> Action is the dominant
module. In math solvers, particularly geometry, precise procedural execution, such as applying
theorems or constructing diagrams, outweighs strategic planning. Similarly, in formal verification
tasks (e.g., Coq or Lean), strict adherence to syntactic and semantic correctness is critical. Both
scenarios demand meticulous step-by-step actions to ensure reliability and prevent errors. </li>
\paragraph{Low Reflection Contribution}
</ul>
<h4>Low Reflection Contribution</h4>
<p>
We conclude the seemingly low contribution of the Reflection module to overall task performance
through two main considerations. First, whether or not the reflection directly translates into a
higher success rate does not necessarily reflect the true quality or efficacy of the reflection
itself. In other words, task success alone may not be the best measure of how well the model is
“thinking about” its own mistakes. Second, when the model reflects on its own errors without extra
information or guidance from a more capable model, it may fail to pinpoint the actual causes behind
its mistakes. As a result, the lack of deeper insights into error sources means reflection often does
not generate meaningful improvements in task outcomes. Consequently, while the Reflection module is
present, its practical impact on success rates remains limited.
</p>
</div>
</div>
</div>
</div>
</div>
</section>
<!-- Citation -->
<section id="citation">
<div class="container">
<h2 class="section-title" data-aos="fade-right">Citation</h2>
<div class="card p-4" data-aos="fade-up">
<pre class="citation mb-0">@misc{yang2025whosmvpgametheoreticevaluation,
title={Who's the MVP? A Game-Theoretic Evaluation Benchmark for Modular Attribution in LLM Agents},
author={Yingxuan Yang and Bo Huang and Siyuan Qi and Chao Feng and Haoyi Hu and Yuxuan Zhu and Jinbo Hu and Haoran Zhao and Ziyi He and Xiao Liu and Zongyu Wang and Lin Qiu and Xuezhi Cao and Xunliang Cai and Yong Yu and Weinan Zhang},
year={2025},
eprint={2502.00510},
archivePrefix={arXiv},
primaryClass={cs.AI},
url={https://arxiv.org/abs/2502.00510},
}</pre>
</div>
</div>
</section>
<!-- Footer -->
<footer>
<div class="container">
<p class="mb-0">© 2025 CapaBench Team. All Rights Reserved.</p>
</div>
</footer>
</main>
<script src="https://cdnjs.cloudflare.com/ajax/libs/aos/2.3.4/aos.js"></script>
<script>
document.addEventListener('DOMContentLoaded', function () {
AOS.init({
duration: 800,
once: true
});
});
</script>
<div class="back-to-top">
<i class="fas fa-arrow-up"></i>
</div>
<script src="https://cdnjs.cloudflare.com/ajax/libs/bootstrap/5.3.0/js/bootstrap.bundle.min.js"></script>
<script src="https://cdnjs.cloudflare.com/ajax/libs/aos/2.3.4/aos.js"></script>
<script src="https://cdnjs.cloudflare.com/ajax/libs/font-awesome/6.0.0/js/all.min.js"></script>
<script>
// 初始化 AOS
AOS.init({
duration: 800,
once: true
});
// 处理导航栏和滚动效果
window.addEventListener('scroll', function () {
const navbar = document.querySelector('.navbar');
const scrollProgress = document.querySelector('.scroll-progress');
const backToTop = document.querySelector('.back-to-top');
const scrollDots = document.querySelectorAll('.scroll-dot');
// 更新导航栏样式
if (window.scrollY > 50) {
navbar.classList.add('scrolled');
} else {
navbar.classList.remove('scrolled');
}
// 更新进度条
const winScroll = document.body.scrollTop || document.documentElement.scrollTop;
const height = document.documentElement.scrollHeight - document.documentElement.clientHeight;
const scrolled = (winScroll / height) * 100;
scrollProgress.style.width = scrolled + '%';
// 显示/隐藏返回顶部按钮
if (window.scrollY > 300) {
backToTop.classList.add('visible');
} else {
backToTop.classList.remove('visible');
}
// 更新导航项激活状态
document.querySelectorAll('section').forEach((section, index) => {
const rect = section.getBoundingClientRect();
if (rect.top <= 100 && rect.bottom >= 100) {
document.querySelectorAll('.nav-link').forEach(link => link.classList.remove('active'));
document.querySelector(`a[href="#${section.id}"]`)?.classList.add('active');
// 更新滚动指示器
scrollDots.forEach(dot => dot.classList.remove('active'));
scrollDots[index]?.classList.add('active');
}
});
});
// 滚动指示器点击事件
document.querySelectorAll('.scroll-dot').forEach((dot, index) => {
dot.addEventListener('click', () => {
const sections = document.querySelectorAll('section');
if (sections[index]) {
sections[index].scrollIntoView({ behavior: 'smooth' });
}
});
});
// 返回顶部功能
document.querySelector('.back-to-top').addEventListener('click', function () {
window.scrollTo({
top: 0,
behavior: 'smooth'
});
});
// 平滑滚动到锚点
document.querySelectorAll('a[href^="#"]').forEach(anchor => {
anchor.addEventListener('click', function (e) {
e.preventDefault();
const target = document.querySelector(this.getAttribute('href'));
if (target) {
target.scrollIntoView({
behavior: 'smooth'
});
}
});
});
</script>
</body>
</html>