-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathindex.html
More file actions
1579 lines (1516 loc) · 101 KB
/
index.html
File metadata and controls
1579 lines (1516 loc) · 101 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
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
<!DOCTYPE html>
<html>
<head>
<meta charset="utf-8">
<meta name="description" content="Evaluating Concurrent Robustness of Language Models Across Diverse Challenge Sets">
<meta name="keywords" content="InfoTabs, Info Tabs">
<meta name="viewport" content="width=device-width, initial-scale=1">
<title> NTSEBench: Cognitive Reasoning Benchmark for Vision Language Models </title>
<link rel="icon" href="./static/images/msin-removebg-preview.png">
<link href="https://fonts.googleapis.com/css?family=Google+Sans|Noto+Sans|Castoro" rel="stylesheet">
<link rel="stylesheet" href="./static/css/bulma.min.css">
<link rel="stylesheet" href="./static/css/bulma-carousel.min.css">
<link rel="stylesheet" href="./static/css/bulma-slider.min.css">
<link rel="stylesheet" href="./static/css/fontawesome.all.min.css">
<link rel="stylesheet" href="https://cdn.jsdelivr.net/gh/jpswalsh/academicons@1/css/academicons.min.css">
<link rel="stylesheet" href="./static/css/index.css">
<script src="https://ajax.googleapis.com/ajax/libs/jquery/3.5.1/jquery.min.js"></script>
<script defer src="./static/js/fontawesome.all.min.js"></script>
<script src="./static/js/bulma-carousel.min.js"></script>
<script src="./static/js/bulma-slider.min.js"></script>
<script src="./static/js/explorer-index.js"></script>
<script src="./data/results/output_folders.js" defer></script>
<script src="./data/results/model_scores.js" defer></script>
<script src="./visualizer/data/data_public.js" defer></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>
</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 is-bold">
<img src="static/images/msin-removebg-preview.png" style="width:1em;vertical-align: middle" alt="Logo">
<span class="infotabs" style="vertical-align: middle">NTSEBench: Cognitive Reasoning Benchmark for Vision Language Models</span>
</h1>
<div class="is-size-5 publication-authors">
<span class="author-block">
<a href="https://www.linkedin.com/in/vatsal-gupta-iitg/">Pranshu Pandya</a><b>*</b>
<sup style="color:#6fbf73;">1</sup>
,
</span>
<span class="author-block">
<a href="https://www.linkedin.com/in/pranshu-pandya/">Vatsal Gupta</a><b>*</b>
<sup style="color:#6fbf73;">1</sup>
,
</span>
<br>
<span class="author-block">
<a href="https://tushaarkataria.github.io/">Agney S Talwarr</a>
<sup style="color:#6fbf73;">1</sup>
,
</span>
<span class="author-block">
<a href="https://vgupta123.github.io/">Tushar Kataria</a>
<sup style="color:#ed4b82">2</sup>
,
</span>
<span class="author-block">
<a href="https://www.cis.upenn.edu/~danroth/">Dan Roth</a>
<sup style="color:#ffac33">3</sup>
</span>
<span class="author-block">
<a href="https://search.asu.edu/profile/5176374">Vivek Gupta</a>
<sup style="color:#d8bfd8">4</sup>
</span>
</div>
<div class="is-size-5 publication-authors">
<span class="author-block">
<sup ><b>*</b></sup>
Equal Contribution
</span>
<br>
<span class="author-block">
<sup style="color:#6fbf73;">1</sup>
Indian Institute of Technology-Guwahati,
</span>
<br>
<span class="author-block">
<sup style="color:#ed4b82">2</sup>
University of Utah,
</span>
<span class="author-block">
<sup style="color:#ffac33">3</sup>
University of Pennsylvania,
</span>
<span class="author-block">
<sup style="color:#d8bfd8">4</sup>
Arizona State University
</span>
<br>
<span class="paper-block">
<b style="color:#f41c1c">NAACL 2025 Findings</b>
</span>
</div>
<div class="column has-text-centered">
<div class="publication-links">
<span class="link-block">
<a href="https://aclanthology.org/2025.findings-naacl.204/" class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<svg height="260.090482" viewBox="0 0 68 46" width="384.481582" xmlns="http://www.w3.org/2000/svg" fill="white"><path d="m41.977553 0v3.0158h-34.4906476-7.4869054v7.48499 27.97788 7.52133h7.4869054 42.0138966 7.486906 11.012292v-15.00632h-11.012292v-20.49289-7.48499c0-1.57369 0-1.25402 0-3.0158zm-26.967398 17.98578h26.967398v13.0079h-26.967398z" fill="white" fill-rule="evenodd"/></svg> </span>
<span>ACL Anthology</span>
</a>
</span>
<span class="link-block">
<a href="https://arxiv.org/abs/2407.10380" 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/NTSEBench/NTSEBench" class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="fab fa-github"></i>
</span>
<span>Code</span>
</a>
</span>
<span class="link-block">
<a href="https://ncert.nic.in/national-talent-examination.php?ln=en" class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="fa fa-user"></i>
</span>
<span>NTSE- Official</span>
</a>
</span>
<!-- Dataset Link.
<span class="link-block">
<a href="https://youtu.be/3sT4u5SCeZc" class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="fa fa-play"></i>
</span>
<span>Video</span>
</a>
</span> -->
<!-- Twitter Link. -->
<!-- <span class="link-block">
<a href="" 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>
<style>
.content-table {
border-collapse: collapse;
margin: 25px 0;
font-size: 0.9em;
min-width: 400px;
border-radius: 5px 5px 0 0;
overflow: hidden;
box-shadow: 0 0 20px rgba(0, 0, 0, 0.15);
}
.content-table thead tr {
background-color: #f3e357;
color: #ffffff;
text-align: left;
font-weight: bold;
font-size: 80%;
}
.content-table th,
.content-table td {
padding: 12px 15px;
font-size: 80%;
}
.content-table tbody tr {
border-bottom: 1px solid #dddddd;
font-size: 80%;
}
.content-table tbody tr:nth-of-type(even) {
background-color: #f3f3f3;
}
.content-table tbody tr:last-of-type {
border-bottom: 2px solid #f3e357;
}
.content-table tbody tr.active-row {
font-weight: bold;
color: #f3e357;
}
.section-header {
background-color: #d9d9d9;
font-weight: bold;
}
</style>
<section class="section">
<div class="container" style="margin-top: -150px; margin-bottom: -100px;">
<div class="columns is-centered m-6">
<div class="column is-full has-text-centered content">
<div>
<div class="box m-5">
<div class="content has-text-centered">
<img src="static/images/NTSEBench-example.png" alt="geometric reasoning" width="84%">
<p>
<b>
<img src="static/images/msin-removebg-preview.png" style="width:1.0em;vertical-align: middle" alt="Logo">
Examples from the <span class='infotabs'>NTSEBench</span> dataset
</b>
<br>
Three samples of textual, direction, and spatial reasoning questions from the proposed dataset. Solutions to these questions are not included here but are provided in the dataset
<span class="infotabs">NTSEBench</span>
dataset
</p>
</div>
</div>
</div>
</div>
</div>
</section>
<section class="section">
<div class="container" style="margin-bottom: 2vh;">
<!-- Abstract. -->
<div class="columns is-centered has-text-centered">
<div class="column is-four-fifths">
<h2 class="title is-3">Summary and Abstract</h2>
<div class="content has-text-justified">
<!-- <iframe src="https://drive.google.com/file/d/1ixJTqdV7OGCy5jzu_eP6QUyzz26Vze36/preview" width="100%" height="650"></iframe> -->
<p>
Cognitive textual and visual reasoning tasks, including puzzles, series, and analogies, demand the ability to quickly reason, decipher, and evaluate patterns both textually and spatially. Due to extensive training on vast amounts of human-curated data, large language models (LLMs) and vision language models (VLMs) excel in common-sense reasoning tasks, but still struggle with more complex reasoning that demands deeper cognitive understanding. We introduce <span class='infotabs'>NTSEBench</span>, a new dataset designed to evaluate cognitive multimodal reasoning and problem-solving skills of large models. The dataset contains 2,728 multiple-choice questions, accompanied by a total of 4,642 images, categorized into 26 different types. These questions are drawn from the nationwide NTSE examination in India and feature a mix of visual and textual general aptitude challenges, designed to assess intelligence and critical thinking skills beyond mere rote learning. We establish baselines on the dataset using state-of-the-art LLMs and VLMs. To facilitate a comparison between open-source and propriety models, we propose four distinct modeling strategies to handle different modalities—text and images—in the dataset instances.
</p>
</div>
</div>
</div>
</div>
<!--/ Abstract. -->
</div>
</section>
<!-- DATASET SECTION -->
<section class="hero is-light is-small">
<div class="hero-body has-text-centered">
<!-- <h1 class="title is-1 InfoTabs"><img src="static/images/msin-removebg-preview.png" style="width:1.0em;vertical-align: middle" alt="Logo"/>InfoTabs Dataset</h1> -->
<h1 class="title is-1 InfoTabs">
<img src="static/images/msin-removebg-preview.png" style="width:1em;vertical-align: middle" alt="Logo">
<span class="infotabs" style="vertical-align: middle"><span class='infotabs'>NTSEBench</span> Dataset</span>
</h1>
</div>
</section>
<section class="section">
<div class="container">
<div class="columns is-centered has-text-centered">
<!-- <div class="column is-full-width has-text-centered"> -->
<div class="column is-four-fifths">
<h2 class="title is-3">Overview</h2>
<div class="content has-text-justified">
<p>
The National Talent Search Examination (NTSE), administered by the National Council of Educational Research and Training (NCERT) in India since 1963, is a nationwide exam for secondary-grade students. The exam consists of two sections designed to assess a wide range of analytical skills: the Mental Ability Test (MAT) and the Scholastic Aptitude Test (SAT). The MAT section evaluates students’ general aptitude, critical thinking, logical and spatial reasoning, and analytical problem-solving skills (for both textual and visual problems). In contrast, the SAT assesses their domain-specific knowledge in science and mathematics. All questions in the NTSE are multiplechoice (MCQ) with one correct option. Questions and options can be text/image or a combination of both, i.e., multi-modal. We aim to create a dataset focused on cognitive reasoning abilities (MAT-type questions). Cognitive Reasoning. Cognitive understanding in the context of <span class='infotabs'>NTSEBench</span> refers to the ability to process information, recognize patterns, draw inferences, and solve problems using critical, logical and analytical reasoning. This aligns with fundamental concepts in cognitive science, such as problem-solving, pattern recognition, and inferential reasoning
</p>
<style>
.content-table-2 {
border-collapse: collapse;
margin: 25px 0;
font-size: 0.9em;
min-width: 400px;
border-radius: 5px 5px 0 0;
overflow: hidden;
box-shadow: 0 0 20px rgba(0, 0, 0, 0.15);
}
.content-table-2 thead tr {
background-color: #209ceeb6;
color: #ffffff;
text-align: left;
font-weight: bold;
}
.content-table-2 th,
.content-table-2 td {
padding: 12px 15px;
}
.content-table-2 tbody tr {
border-bottom: 1px solid #dddddd;
}
.content-table-2 tbody tr:nth-of-type(even) {
background-color: #f3f3f3;
}
.content-table-2 tbody tr:last-of-type {
border-bottom: 2px solid #209ceeb6;
}
.content-table-2 tbody tr.active-row {
font-weight: bold;
color: #209ceeb6;
}
</style>
<table class="content-table-2">
<thead>
<tr>
<th>Categories</th>
<th># Samples</th>
<th>Categories</th>
<th># Samples</th>
</tr>
</thead>
<tbody>
<tr>
<td>Series</td>
<td>256</td>
<td>Non-Verbal Series</td>
<td>95</td>
</tr>
<tr>
<td>Alphabet Test</td>
<td>94</td>
<td>Missing Character</td>
<td>127</td>
</tr>
<tr>
<td>Odd one out</td>
<td>170</td>
<td>Embedded Figure</td>
<td>96</td>
</tr>
<tr>
<td>Analogy</td>
<td>151</td>
<td>Non-Verbal odd one out</td>
<td>70</td>
</tr>
<tr>
<td>Coding-Decoding</td>
<td>149</td>
<td>Non-Verbal Analogy</td>
<td>100</td>
</tr>
<tr>
<td>Number and Ranking</td>
<td>139</td>
<td>Paper Folding & Cutting</td>
<td>96</td>
</tr>
<tr>
<td>Blood Relation</td>
<td>126</td>
<td>Incomplete Figure</td>
<td>94</td>
</tr>
<tr>
<td>Mathematical Operations</td>
<td>99</td>
<td>Figure Partition</td>
<td>71</td>
</tr>
<tr>
<td>Puzzle Test</td>
<td>95</td>
<td>Cube and Dice</td>
<td>23</td>
</tr>
<tr>
<td>Syllogisms</td>
<td>44</td>
<td>Dot problem</td>
<td>23</td>
</tr>
<tr>
<td>Statement & Conclusions</td>
<td>143</td>
<td>Direction Sense</td>
<td>36</td>
</tr>
<tr>
<td>Data Sufficiency</td>
<td>90</td>
<td>Time and Clock</td>
<td>51</td>
</tr>
<tr>
<td></td>
<td></td>
<td>Mirror, Water and Images</td>
<td>50</td>
</tr>
<tr>
<td></td>
<td></td>
<td>Venn diagrams</td>
<td>111</td>
</tr>
</tbody>
</table>
<p>
</p>
</div>
</div>
</div>
<!-- <div class="columns is-centered">
<div class="column" style="margin-right: -20rem;">
<div class="content has-text-centered">
<img src="static/" alt="data-overview" style="max-width: 50%;">
<p>
Key statistics of
<img src="static/images/msin-removebg-preview.png" style="width:1.0em;vertical-align: middle" alt="Logo">
<span class="infotabs">InfoTabs</span>
.
<br>
</p>
</div>
</div>
<div class="column">
<div class="content has-text-centered">
<img src="static/images/.png" alt="" style="max-width: 56%;">
<p>
Source dataset distribution of
<img src="static/images/msin-removebg-preview.png" style="width:1.0em;vertical-align: middle" alt="Logo">
<span class="infotabs">InfoTabs</span>
.
</p>
</div>
</div>
</div> -->
<!-- <div class="columns is-centered m-6">
<div class="column is-full has-text-centered content">
<h2 class="title is-3">Examples</h2>
<p>
One example for each
<b>type of perturbation anlaysed</b> within the
<img src="static/images/msin-removebg-preview.png" style="width:1.0em;vertical-align: middle" alt="Logo">
<span class="infotabs">MSIN-InfoTabs</span> dataset.
</p>
<div id="results-carousel" class="carousel results-carousel">
<div class="box m-5">
<div class="content has-text-centered">
<img src="static/images/examples/" alt="exemplar" width="60%">
<p>example</p>
</div>
</div>
<div class="box m-5">
<div class="content has-text-centered">
<img src="static/images/examples/" alt="exemplar" width="60%">
<p>example</p>
</div>
</div>
<div class="box m-5">
<div class="content has-text-centered">
<img src="static/images/examples/" alt="exemplar" width="60%">
<p>example</p>
</div>
</div>
<div class="box m-5">
<div class="content has-text-centered">
<img src="static/images/examples/" alt="exemplar" width="60%">
<p>example</p>
</div>
</div>
</div>
<br>
<br>
</div>
</div> -->
</div>
</section>
<!--Strategies section-->
<section class="hero is-light is-small">
<div class="hero-body has-text-centered">
<h1 class="title is-1 InfoTabs">
<img src="static/images/msin-removebg-preview.png" style="width:1em;vertical-align: middle" alt="Logo">
<span class="infotabs" style="vertical-align: middle">Strategies Proposed for analysis</span>
</h1>
</div>
</section>
<section class="section">
<div class="container">
<div class="columns is-centered">
<div class="column is-full has-text-centered content">
<div class="content has-text-justified">
<p>
Evaluating the reasoning abilities of large language models (LLMs) with text-based questions is straightforward. For vision-language models (VLMs), reasoning with vision-text questions is generally straightforward.
Some API access model models, like GPT-4o and Gemini, support multi-image inputs, while many others do not (open-source models like LLaVA-OneVision and Ovis are emerging with this capability). To address these task-specific and input-related dependencies, we propose four strategies to fairly evaluate the reasoning abilities of both open-source and API-based models.
</p>
<h5>Standard QA</h5>
<p> For instances where question type(<i>J</i>) or questions(<i>Q</i>),options(<i>O</i>) and solutions(<i>S</i>) is text(<i>T</i>),we use a standard text-based QA model like GPT3.5-Turbo or Llama3-70b</p>
<h5>Image-Only</h5>
<p>We propose a modeling approach where questions and all the options are presented to the model as a single image. This image consolidates all relevant textual and visual content exactly as it appears in the examination paper, effectively capturing the entire question, including both textual and visuals. This strategy utilizes the OCR capabilities of VLM models to interpret and analyze the content, enabling them to process both text and visual elements within the same input.</p>
<h5>Interleaved model</h5>
<p>In this approach, we integrate text with multiple images to create an interwoven context. This method involves placing related textual and visual elements in proximity, enhancing the model’s ability to draw connections between them.</p>
<h5>Standard VQA</h5>
<p>
Open-source models typically lack the capability to integrate text and images within a single prompt. To enable fair comparisons, we propose an alternative modeling strategy where the question and option images are combined into a single composite image, labeled as Figure 1, Figure 2, etc. This composite image is accompanied by a structured textual prompt that describes different parts of the image, directing the model’s attention to relevant visual details. The composite image and prompt are then used to evaluate the model’s performance, testing its ability to interpret and respond to questions based on the integrated visual and textual information.</p>
</div>
<!-- <div class="content">
<img class="mt-3">
Analysing Language Models
<b>Sensitivity to Input Perturbations</b><br>
<center><img src="static/images/sensitive.png" style="width:70%"></img></center>
<img src="static/images/msin-removebg-preview.png" style="width:1em;vertical-align: middle" alt="Logo">
<b>Multi-Set Inoculation framework</b>
</p>
</div> -->
</div>
</div>
</div>
</section>
<!-- RESULTS SECTION -->
<section class="hero is-light is-small">
<div class="hero-body has-text-centered">
<h1 class="title is-1 InfoTabs">
<img src="static/images/msin-removebg-preview.png" style="width:1em;vertical-align: middle" alt="Logo">
Experiment Results</h1>
</div>
</section>
<section class="section">
<div class="container">
<div class="content has-text-justified">
<p>The <strong>Analysis</strong> section of the paper presents a comprehensive evaluation of the impact of various strategies on the <span class='infotabs'>NTSEBench</span> dataset. The findings indicate that
<ul>
<li>Interleaving text and images performs better than Standard VQA and Image Only strategy for most categories. </li>
<li>Multi-modal reasoning is significantly harder. </li>
<li>The results reveal that while LLMs generally perform better on the text-only subset of <span class='infotabs'>NTSEBench</span>, the high standard deviation (11.04) indicates significant variability in model performance across different question types. This variability may stem from some overlap between <span class='infotabs'>NTSEBench</span> and other open-source datasets, suggesting that there are still areas where LLMs exhibit limitations in reasoning capabilities </li>
<li><span class='infotabs'>NTSEBench</span> presents a challenging task for SOTA LLMs and VLMs. Based on the findings, it is evident that the proposed dataset presents a challenging task for all state of-the-art LLM and VLM models. None of the open-source models achieve accuracy exceeding 50% on text-only questions and 35% on multimodal questions, with propriety models achieving 62% and 42% accuracy respectively.</li>
</ul>
<style>
table {
width: 75%;
border-collapse: collapse;
border: 1px solid #209cee;
}
th, td {
border: 1px solid #209cee;
padding: 10px;
text-align: center;
}
.diagonal {
position: relative;
background-color: #209cee; /* Background color for the diagonal cell */
color: white; /* Text color */
}
.diagonal::after {
content: "";
position: absolute;
top: 0;
left: 0;
right: 0;
bottom: 0;
background: white; /* Color of the diagonal part */
transform: skew(-30deg);
transform-origin: 0 0;
z-index: 1;
}
.diagonal span {
position: relative;
z-index: 2; /* Ensure the text appears above the diagonal */
}
</style>
<section class="section">
<div class="container" style="margin-top: -150px; margin-bottom: -100px;">
<div class="columns is-centered m-6">
<div class="column is-full has-text-centered content">
<div id="results-carousel" class="carousel results-carousel">
<div class="box m-5">
<div class="content has-text-centered">
<table class="content-table">
<thead>
<tr>
<th><center>Model</center></th>
<th>SER</th>
<th>ALP</th>
<th>ODO</th>
<th>ANA</th>
<th>COD</th>
<th>NUM</th>
<th>BLR</th>
<th>MTO</th>
<th>PUZ</th>
<th>SYL</th>
<th>STC</th>
<th>DAT</th>
<th>Avg. Per</th>
</tr>
</thead>
<tbody>
<tr class="section-header">
<td colspan="15">Image Only - Zero Shot</td>
</tr>
<tr>
<td>CogVLM-2</td>
<td>14.84</td>
<td>17.02</td>
<td>20.00</td>
<td>19.87</td>
<td>24.83</td>
<td>16.55</td>
<td>23.81</td>
<td>20.20</td>
<td>20.00</td>
<td>22.73</td>
<td>22.12</td>
<td>15.56</td>
<td>19.79</td>
</tr>
<tr>
<td>InternLM-XComposer2</td>
<td>18.36</td>
<td>18.09</td>
<td>21.65</td>
<td>19.95</td>
<td>17.42</td>
<td>11.51</td>
<td>15.87</td>
<td>24.24</td>
<td>25.26</td>
<td>31.16</td>
<td>17.31</td>
<td>8.89</td>
<td>17.22</td>
</tr>
<tr>
<td>Qwen-VL-Chat</td>
<td>29.69</td>
<td>23.44</td>
<td>29.44</td>
<td>45.45</td>
<td>39.29</td>
<td>17.97</td>
<td>22.11</td>
<td>45.36</td>
<td>46.97</td>
<td>45.38</td>
<td>34.25</td>
<td>34.18</td>
<td>33.73</td>
</tr>
<tr>
<td>Gemini 1.5 Pro</td>
<td>32.42</td>
<td>31.91</td>
<td>47.64</td>
<td>52.32</td>
<td>52.37</td>
<td>37.41</td>
<td>47.29</td>
<td>49.29</td>
<td>47.37</td>
<td>47.13</td>
<td>38.46</td>
<td>44.44</td>
<td>39.55</td>
</tr>
<tr>
<td>GPT-4o</td>
<td>28.12</td>
<td>31.91</td>
<td>41.94</td>
<td>50.03</td>
<td>37.14</td>
<td>52.38</td>
<td>34.34</td>
<td>46.34</td>
<td>53.85</td>
<td>43.85</td>
<td>35.38</td>
<td>38.17</td>
<td>36.17</td>
</tr>
<tr class="section-header">
<td colspan="14">Image Only - Few Shot</td>
</tr>
<tr>
<td>Gemini 1.5 Pro</td>
<td>23.32</td>
<td>23.08</td>
<td>46.11</td>
<td>47.97</td>
<td>24.66</td>
<td>36.76</td>
<td>36.59</td>
<td>32.29</td>
<td>42.94</td>
<td>31.71</td>
<td>32.67</td>
<td>22.99</td>
<td>33.37</td>
</tr>
<tr>
<td>GPT-4o</td>
<td>32.02</td>
<td>29.67</td>
<td>50.30</td>
<td>40.42</td>
<td>32.19</td>
<td>35.29</td>
<td>43.09</td>
<td>25.00</td>
<td>46.74</td>
<td>41.46</td>
<td>53.47</td>
<td>34.48</td>
<td>38.85</td>
</tr>
<tr class="section-header">
<td colspan="14">Standard QA - Zero Shot</td>
</tr>
<tr>
<td>Mixtral-8x7B</td>
<td>19.76</td>
<td>19.57</td>
<td>24.71</td>
<td>45.52</td>
<td>14.77</td>
<td>26.09</td>
<td>29.37</td>
<td>29.59</td>
<td>32.93</td>
<td>24.32</td>
<td>53.85</td>
<td>33.33</td>
<td>29.48</td>
</tr>
<tr>
<td>Llama-3 70B</td>
<td>35.18</td>
<td>26.09</td>
<td>47.65</td>
<td>57.93</td>
<td>36.36</td>
<td>36.23</td>
<td>50.79</td>
<td>31.63</td>
<td>60.98</td>
<td>54.05</td>
<td>52.88</td>
<td>40.48</td>
<td>44.18</td>
</tr>
<tr>
<td>GPT-3.5 Turbo</td>
<td>35.97</td>
<td>32.61</td>
<td>40.00</td>
<td>51.72</td>
<td>36.36</td>
<td>25.36</td>
<td>36.51</td>
<td>27.55</td>
<td>46.34</td>
<td>35.14</td>
<td>40.38</td>
<td>32.14</td>
<td>36.67</td>
</tr>
<tr>
<td>CogVLM-2</td>
<td>22.27</td>
<td>21.28</td>
<td>27.65</td>
<td>34.44</td>
<td>22.82</td>
<td>18.71</td>
<td>30.95</td>
<td>19.19</td>
<td>29.47</td>
<td>18.18</td>
<td>28.85</td>
<td>27.78</td>
<td>25.13</td>
</tr>
<tr>
<td>InternLM-XComposer2</td>
<td>21.88</td>
<td>24.47</td>
<td>19.41</td>
<td>36.42</td>
<td>25.50</td>
<td>28.78</td>
<td>25.40</td>
<td>27.27</td>
<td>45.26</td>
<td>40.91</td>
<td>34.62</td>
<td>28.89</td>
<td>29.90</td>
</tr>
<tr>
<td>Qwen-VL-Chat</td>
<td>30.08</td>
<td>18.09</td>
<td>23.53</td>
<td>31.13</td>
<td>26.85</td>
<td>15.11</td>
<td>24.60</td>
<td>27.27</td>
<td>28.26</td>
<td>13.64</td>
<td>15.38</td>
<td>24.44</td>
<td>23.19</td>
</tr>
<tr>
<td><i>Gemini 1.5 Pro</i></td>
<td><b>63.67</b></td>
<td>39.36</td>
<td>60.00</td>
<td><b>69.54</b></td>
<td><b>61.07</b></td>
<td>68.35</td>
<td>58.73</td>
<td>45.45</td>
<td><b>81.05</b></td>
<td><b>65.91</b></td>
<td>70.19</td>
<td><b>63.33</b></td>
<td><b>62.22</b></td>
</tr>
<tr>
<td><i>GPT-4o</i></td>
<td>42.58</td>
<td>35.11</td>
<td>55.88</td>
<td>65.56</td>
<td>38.26</td>
<td>42.45</td>
<td>68.25</td>
<td>41.41</td>
<td>69.47</td>
<td>63.64</td>
<td>70.19</td>
<td>43.33</td>
<td>53.01</td>
</tr>
<tr>
<td><i>LLaVA-OneVision</i></td>
<td>42.19</td>
<td>32.98</td>
<td>50.59</td>
<td>57.62</td>
<td>45.64</td>
<td>36.69</td>
<td>57.94</td>
<td>37.37</td>
<td>64.21</td>
<td>50.00</td>
<td>62.50</td>
<td>46.67</td>
<td>48.70</td>
</tr>
<tr>
<td><i>Ovis1.6-Gemma2-9B</i></td>
<td>42.58</td>
<td>31.91</td>
<td>50.00</td>
<td>50.99</td>
<td>42.95</td>
<td>46.04</td>
<td>38.89</td>
<td>31.31</td>
<td>53.26</td>
<td>27.27</td>
<td>55.77</td>
<td>33.33</td>
<td>42.03</td>
</tr>
<tr class="section-header">
<td colspan="14">Standard QA - Few Shot</td>
</tr>
<tr>
<td>Mixtral-8x7B</td>
<td>27.20</td>
<td>24.72</td>
<td>28.14</td>
<td>50.70</td>
<td>29.41</td>
<td>27.41</td>
<td>33.33</td>
<td>29.47</td>
<td>18.99</td>
<td>#</td>
<td>55.45</td>
<td>32.10</td>
<td>32.44</td>
</tr>
<tr>
<td>Llama-3 70B</td>
<td>34.00</td>
<td>16.85</td>
<td>44.91</td>
<td>51.41</td>
<td>36.47</td>
<td>34.81</td>
<td>39.84</td>
<td>32.63</td>
<td>34.18</td>
<td>#</td>
<td>50.50</td>
<td>34.57</td>
<td>37.28</td>
</tr>
<tr>
<td>GPT-3.5 Turbo</td>
<td>30.80</td>
<td>32.58</td>
<td>20.96</td>
<td>47.89</td>
<td>30.59</td>
<td>31.11</td>
<td>30.08</td>
<td>29.47</td>
<td>36.71</td>
<td>#</td>
<td>40.59</td>
<td>34.57</td>
<td>33.21</td>
</tr>
<tr>
<td><i>Gemini 1.5 Pro</i></td>
<td><b>63.24</b></td>
<td>37.36</td>
<td><b>59.28</b></td>
<td>68.92</td>
<td>60.27</td>
<td><b>68.38</b></td>
<td>58.54</td>
<td>43.75</td>
<td>80.43</td>
<td>63.41</td>
<td>70.30</td>
<td>62.07</td>
<td><b>61.32</b></td>
</tr>
<tr>
<td><i>GPT-4o</i></td>
<td>42.29</td>
<td><b>40.66</b></td>
<td>58.08</td>
<td>67.57</td>
<td>44.52</td>
<td>40.44</td>
<td><b>69.92</b></td>
<td><b>46.88</b></td>
<td>72.83</td>
<td>63.41</td>
<td><b>71.29</b></td>
<td>*</td>