-
Notifications
You must be signed in to change notification settings - Fork 14
/
index.html
946 lines (899 loc) · 32 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
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
<!DOCTYPE html>
<html>
<head>
<meta charset="utf-8">
<meta name="description"
content="SeaLLMs - Large Language Models for Southeast Asia">
<meta name="keywords" content="SeaLLM, SeaLMMM">
<meta name="viewport" content="width=device-width, initial-scale=1">
<title>SeaLLMs - Large Language Models for Southeast Asia</title>
<!-- Global site tag (gtag.js) - Google Analytics -->
<script async src="https://www.googletagmanager.com/gtag/js?id=G-PYVRSFMDRL"></script>
<script>
window.dataLayer = window.dataLayer || [];
function gtag() {
dataLayer.push(arguments);
}
gtag('js', new Date());
gtag('config', 'G-PYVRSFMDRL');
</script>
<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">
<!-- <link rel="icon" href="./static/images/favicon-32x32.png"> -->
<link rel="icon" href="./static/images/seal_logo.png">
<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/index.js"></script>
</head>
<body>
<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">
<img src="./static/images/seal_logo.png" alt="" style="height: 2.5em;">
<br>
SeaLLMs - Large Language Models for Southeast Asia
</h1>
<div class="is-size-5 publication-authors">
<span class="author-block">
<a href="https://isakzhang.github.io">Wenxuan Zhang</a>,
</span>
<span class="author-block">
<a href="https://kenchan0226.github.io/">Hou Pong Chan</a>,
</span>
<span class="author-block">
<a href="https://zhaoyiran924.github.io//">Yiran Zhao</a>,
</span>
<span class="author-block">
<a href="">Mahani Aljunied</a>,
</span>
<span class="author-block">
<a href="https://jianyu-cs.github.io/">Jianyu Wang</a>,
</span>
<span class="author-block">
<a href="https://www.linkedin.com/in/chaoqun-liu/">Chaoqun Liu</a>,
</span>
<span class="author-block">
<a href="https://ntudy.github.io/">Yue Deng</a>,
</span>
<span class="author-block">
<a href="https://hzq950419.github.io/HomePage/">Zhiqiang Hu</a>,
</span>
<span class="author-block">
<a href="https://wwxu21.github.io">Weiwen Xu</a>,
</span>
<span class="author-block">
<a href="https://chiayewken.com/">Yew Ken Chia</a>,
</span>
<span class="author-block">
<a href="https://lixin4ever.github.io">Xin Li</a>,
</span>
<span class="author-block">
<a href="https://lidongbing.github.io">Lidong Bing*</a> (Corresponding Author)
</span>
</div>
<div class="is-size-5 publication-authors">
<span class="author-block">DAMO Academy, Alibaba Group</span>
</div>
<div class="column has-text-centered">
<div class="publication-links">
<!-- PDF Link. -->
<span class="link-block">
<a href="https://arxiv.org/abs/2312.00738"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<!-- <i class="fas fa-file-pdf"></i> -->
<i class="ai ai-arxiv"></i>
</span>
<span>Paper</span>
</a>
</span>
<!-- Demo Link. -->
<span class="link-block">
<a href="https://huggingface.co/spaces/SeaLLMs/SeaLLM-Chat"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
🤗
</span>
<span>DEMO</span>
</a>
</span>
<!-- Code Link. -->
<span class="link-block">
<a href="https://github.com/DAMO-NLP-SG/SeaLLMs"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="fab fa-github"></i>
</span>
<span>Github</span>
</a>
</span>
<!-- Models Link. -->
<span class="link-block">
<a href="https://huggingface.co/SeaLLMs/SeaLLM3-7B-Chat"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<!-- <i class="far fa-images"></i> -->
🤗
</span>
<span>Models</span>
</a>
</span>
<!-- darasets Link. -->
<!-- <span class="link-block">
<a href="https://huggingface.co/datasets/SeaLLMs/Sea-bench"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="far fa-images"></i>
</span>
<span>Data</span>
</a>
</span> -->
</div>
</div>
</div>
</div>
</div>
</div>
</section>
<section class="hero teaser">
<div class="container is-max-desktop">
<div class="hero-body">
<h4 class="subtitle has-text-centered">
🔥<span style="color: #ff3860">[NEW!]</span>
<a href="https://huggingface.co/SeaLLMs/SeaLLM3-7B-Chat">SeaLLMs-v3</a> is released with SoTA performance in various tasks and specifically enhanced to be more trustworthy.
<br>
<!--
🔥<span style="color: #ff3860">[HOT!]</span>
<a href="https://huggingface.co/SeaLLMs/SeaLMMM-7B-v0.1">SeaLMMM-7B-v0.1</a> is introduced with <b>Multimodal</b> Multilingual capabilities in SEA languages.
-->
</h4>
<script
type="module"
src="https://gradio.s3-us-west-2.amazonaws.com/4.26.0/gradio.js">
</script>
<gradio-app src="https://seallms-seallm-7b-v2-5-simple.hf.space"></gradio-app>
</div>
</div>
</section>
<section class="section">
<div class="container is-max-desktop">
<!-- Abstract. -->
<div class="columns is-centered has-text-centered">
<div class="column is-four-fifths">
<h2 class="title is-3">Abstract</h2>
<div class="content has-text-justified">
<p>
We introduce SeaLLMs-v3, the latest series of the SeaLLMs (Large Language Models for Southeast Asian languages) family. It achieves state-of-the-art performance among models with similar sizes, excelling across a diverse array of tasks such as world knowledge, mathematical reasoning, translation, and instruction following. In the meantime, it was specifically enhanced to be more trustworthy, exhibiting reduced hallucination and providing safe responses, particularly in queries closed related to Southeast Asian culture.
</p>
<p>
<a href="https://huggingface.co/collections/SeaLLMs/seallms-65be16f92e67686440ae29f3">SeaLLMs</a> is a continuously iterated and improved series of language models
that specifically focuses on Southeast Asian (SEA) languages. SeaLLMs are typically continue-pretrained and fine-tuned from strong English models to build outstanding
capabilities in SEA languages without degrading performances in high-resource languages. SeaLLMs are built with focus in prioritizing local cultural and legal norms, customs,
stylistic preferences, as well as cost-effectiveness
</p>
</div>
</div>
</div>
<!--/ Abstract. -->
</div>
</section>
<!-- introduction -->
<!-- <section class="section">
<div class="container is-max-desktop">
<h2 class="title is-3">Introduction</h2>
</div>
</section> -->
<!-- <section class="section">
<div class="container is-max-desktop">
<h2 class="title is-4">SeaLLM-7B-v2.5 DEMO</h2>
<gradio-app src="https://seallms-seallm-7b-v2-5-simple.hf.space"></gradio-app>
</div>
</section> -->
<!-- Evaluation -->
<section class="section">
<div class="container is-max-desktop">
<!-- <h2 class="title is-3">Evaluation</h2> -->
<h2 class="title is-4">Multilingual World Knowledge - M3Exam</h2>
<div class="content has-text-justified">
<p><a href="https://arxiv.org/abs/2306.05179">M3Exam</a> consists of local exam questions collected from each country. It reflects the model's world knowledge (e.g., with language or social science subjects) and reasoning abilities (e.g., with mathematics or natural science subjects).</p>
<div class="table-container">
<table class="table is-bordered is-striped is-narrow is-hoverable is-fullwidth">
<thead>
<tr>
<th>Model</th>
<th>en</th>
<th>zh</th>
<th>id</th>
<th>th</th>
<th>vi</th>
<th>avg</th>
<th>avg_sea</th>
</tr>
</thead>
<tbody>
<tr>
<td>Sailor-7B-Chat</td>
<td>0.66</td>
<td>0.652</td>
<td>0.475</td>
<td>0.462</td>
<td>0.513</td>
<td>0.552</td>
<td>0.483</td>
</tr>
<tr>
<td>gemma-7b</td>
<td>0.732</td>
<td>0.519</td>
<td>0.475</td>
<td>0.46</td>
<td>0.594</td>
<td>0.556</td>
<td>0.510</td>
</tr>
<tr>
<td>SeaLLM-7B-v2.5</td>
<td>0.758</td>
<td>0.581</td>
<td>0.499</td>
<td>0.502</td>
<td>0.622</td>
<td>0.592</td>
<td>0.541</td>
</tr>
<tr>
<td>Qwen2-7B</td>
<td>0.815</td>
<td>0.874</td>
<td>0.53</td>
<td>0.479</td>
<td>0.628</td>
<td>0.665</td>
<td>0.546</td>
</tr>
<tr>
<td>Qwen2-7B-Instruct</td>
<td>0.809</td>
<td>0.88</td>
<td>0.558</td>
<td>0.555</td>
<td>0.624</td>
<td>0.685</td>
<td>0.579</td>
</tr>
<tr>
<td>Sailor-14B</td>
<td>0.748</td>
<td>0.84</td>
<td>0.536</td>
<td>0.528</td>
<td>0.621</td>
<td>0.655</td>
<td>0.562</td>
</tr>
<tr>
<td>Sailor-14B-Chat</td>
<td>0.749</td>
<td>0.843</td>
<td>0.553</td>
<td>0.566</td>
<td>0.637</td>
<td>0.67</td>
<td>0.585</td>
</tr>
<tr>
<td>SeaLLMs-v3-7B</td>
<td>0.814</td>
<td>0.866</td>
<td>0.549</td>
<td>0.52</td>
<td>0.628</td>
<td>0.675</td>
<td>0.566</td>
</tr>
<tr>
<td>SeaLLMs-v3-7B-Chat</td>
<td>0.809</td>
<td>0.874</td>
<td>0.558</td>
<td>0.569</td>
<td>0.649</td>
<td>0.692</td>
<td>0.592</td>
</tr>
</tbody>
</table>
</div>
</div>
<!-- seaexam -->
<!--
<h2 class="title is-4">SeaExam Leaderboard</h2>
<div class="content has-text-justified">
<p>
According to the <a href="https://huggingface.co/spaces/SeaLLMs/SeaExam_leaderboard">SeaExam leaderboard</a>, which evaluates model performance through human-exam style questions in Southeast Asian languages, the latest SeaLLMs-v2.5 is ranked at the top among open-source models of similar size.
<gradio-app src="https://seallms-seaexam-leaderboard.hf.space"></gradio-app>
</p>
</div>
-->
<!-- SeaBench -->
<h2 class="title is-4">Multilingual Instruction-following Capability - SeaBench</h2>
<div class="content has-text-justified">
<p>SeaBench consists of multi-turn human instructions spanning various task types. It evaluates chat-based models on their ability to follow human instructions in both single and multi-turn settings and assesses their performance across different task types. The dataset and corresponding evaluation code will be released soon!</p>
<div class="table-container">
<table class="table is-bordered is-striped is-narrow is-hoverable is-fullwidth">
<thead>
<tr>
<th>Model</th>
<th>id<br>turn1</th>
<th>id<br>turn2</th>
<th>id<br>avg</th>
<th>th<br>turn1</th>
<th>th<br>turn2</th>
<th>th<br>avg</th>
<th>vi<br>turn1</th>
<th>vi<br>turn2</th>
<th>vi<br>avg</th>
<th>avg</th>
</tr>
</thead>
<tbody>
<tr>
<td>Qwen2-7B-Instruct</td>
<td>5.93</td>
<td>5.84</td>
<td>5.89</td>
<td>5.47</td>
<td>5.20</td>
<td>5.34</td>
<td>6.17</td>
<td>5.60</td>
<td>5.89</td>
<td>5.70</td>
</tr>
<tr>
<td>SeaLLM-7B-v2.5</td>
<td>6.27</td>
<td>4.96</td>
<td>5.62</td>
<td>5.79</td>
<td>3.82</td>
<td>4.81</td>
<td>6.02</td>
<td>4.02</td>
<td>5.02</td>
<td>5.15</td>
</tr>
<tr>
<td>Sailor-14B-Chat</td>
<td>5.26</td>
<td>5.53</td>
<td>5.40</td>
<td>4.62</td>
<td>4.36</td>
<td>4.49</td>
<td>5.31</td>
<td>4.74</td>
<td>5.03</td>
<td>4.97</td>
</tr>
<tr>
<td>Sailor-7B-Chat</td>
<td>4.60</td>
<td>4.04</td>
<td>4.32</td>
<td>3.94</td>
<td>3.17</td>
<td>3.56</td>
<td>4.82</td>
<td>3.62</td>
<td>4.22</td>
<td>4.03</td>
</tr>
<tr>
<td>SeaLLMs-v3-7B-Chat</td>
<td>6.73</td>
<td>6.59</td>
<td>6.66</td>
<td>6.48</td>
<td>5.90</td>
<td>6.19</td>
<td>6.34</td>
<td>5.79</td>
<td>6.07</td>
<td>6.31</td>
</tr>
</tbody>
</table>
</div>
</div>
<h2 class="title is-4">Multilingual Math</h2>
<div class="content has-text-justified">
<p>We evaluate the multilingual math capability using the MGSM dataset. MGSM originally contains Chinese and Thai testing sets only, we use Google Translate to translate the same English questions into other SEA languages. Note that we adopt the tradition of each country to represent the number, e.g., in Indonesian and Vietnamese, dots are used as thousands separators and commas as decimal separators, the opposite of the English system.</p>
<div class="table-container">
<table class="table is-bordered is-striped is-narrow is-hoverable is-fullwidth">
<thead>
<tr>
<th>MGSM</th>
<th>en</th>
<th>id</th>
<th>ms</th>
<th>th</th>
<th>vi</th>
<th>zh</th>
<th>avg</th>
</tr>
</thead>
<tbody>
<tr>
<td>Sailor-7B-Chat</td>
<td>33.6</td>
<td>22.4</td>
<td>22.4</td>
<td>21.6</td>
<td>25.2</td>
<td>29.2</td>
<td>25.7</td>
</tr>
<tr>
<td>Meta-Llama-3-8B-Instruct</td>
<td>77.6</td>
<td>48</td>
<td>57.6</td>
<td>56</td>
<td>46.8</td>
<td>58.8</td>
<td>57.5</td>
</tr>
<tr>
<td>glm-4-9b-chat</td>
<td>72.8</td>
<td>53.6</td>
<td>53.6</td>
<td>34.8</td>
<td>52.4</td>
<td>70.8</td>
<td>56.3</td>
</tr>
<tr>
<td>Qwen1.5-7B-Chat</td>
<td>64</td>
<td>34.4</td>
<td>38.4</td>
<td>25.2</td>
<td>36</td>
<td>53.6</td>
<td>41.9</td>
</tr>
<tr>
<td>Qwen2-7B-instruct</td>
<td>82</td>
<td>66.4</td>
<td>62.4</td>
<td>58.4</td>
<td>64.4</td>
<td>76.8</td>
<td>68.4</td>
</tr>
<tr>
<td>aya-23-8B</td>
<td>28.8</td>
<td>16.4</td>
<td>14.4</td>
<td>2</td>
<td>16</td>
<td>12.8</td>
<td>15.1</td>
</tr>
<tr>
<td>gemma-1.1-7b-it</td>
<td>58.8</td>
<td>32.4</td>
<td>34.8</td>
<td>31.2</td>
<td>39.6</td>
<td>35.2</td>
<td>38.7</td>
</tr>
<tr>
<td>SeaLLM-7B-v2.5</td>
<td>79.6</td>
<td>69.2</td>
<td>70.8</td>
<td>61.2</td>
<td>66.8</td>
<td>62.4</td>
<td>68.3</td>
</tr>
<tr>
<td>SeaLLMs-v3-7B-Chat</td>
<td>74.8</td>
<td>71.2</td>
<td>70.8</td>
<td>71.2</td>
<td>71.2</td>
<td>79.6</td>
<td>73.1</td>
</tr>
</tbody>
</table>
</div>
</div>
<h2 class="title is-4">Translation</h2>
<div class="content has-text-justified">
<p>We use the test sets from Flores-200 for evaluation and report the zero-shot chrF scores for translations between every pair of languages. Each row in the table below presents the average results of translating from various source languages into the target languages. The last column displays the overall average results of translating from any language to any other language for each model.</p>
<div class="table-container">
<table class="table is-bordered is-striped is-narrow is-hoverable is-fullwidth">
<thead>
<tr>
<th>Model</th>
<th>en</th>
<th>id</th>
<th>jv</th>
<th>km</th>
<th>lo</th>
<th>ms</th>
<th>my</th>
<th>ta</th>
<th>th</th>
<th>tl</th>
<th>vi</th>
<th>zh</th>
<th>avg</th>
</tr>
</thead>
<tbody>
<tr>
<td>Meta-Llama-3-8B-Instruct</td>
<td>51.54</td>
<td>49.03</td>
<td>22.46</td>
<td>15.34</td>
<td>5.42</td>
<td>46.72</td>
<td>21.24</td>
<td>32.09</td>
<td>35.75</td>
<td>40.80</td>
<td>39.31</td>
<td>14.87</td>
<td>31.22</td>
</tr>
<tr>
<td>Qwen2-7B-Instruct</td>
<td>50.36</td>
<td>47.55</td>
<td>29.36</td>
<td>19.26</td>
<td>11.06</td>
<td>42.43</td>
<td>19.33</td>
<td>20.04</td>
<td>36.07</td>
<td>37.91</td>
<td>39.63</td>
<td>22.87</td>
<td>31.32</td>
</tr>
<tr>
<td>Sailor-7B-Chat</td>
<td>49.40</td>
<td>49.78</td>
<td>28.33</td>
<td>2.68</td>
<td>6.85</td>
<td>47.75</td>
<td>5.35</td>
<td>18.23</td>
<td>38.92</td>
<td>29.00</td>
<td>41.76</td>
<td>20.87</td>
<td>28.24</td>
</tr>
<tr>
<td>SeaLLM-7B-v2.5</td>
<td>55.09</td>
<td>53.71</td>
<td>18.13</td>
<td>18.09</td>
<td>15.53</td>
<td>51.33</td>
<td>19.71</td>
<td>26.10</td>
<td>40.55</td>
<td>45.58</td>
<td>44.56</td>
<td>24.18</td>
<td>34.38</td>
</tr>
<tr>
<td>SeaLLMs-v3-7B-Chat</td>
<td>54.68</td>
<td>52.52</td>
<td>29.86</td>
<td>27.30</td>
<td>26.34</td>
<td>45.04</td>
<td>21.54</td>
<td>31.93</td>
<td>41.52</td>
<td>38.51</td>
<td>43.78</td>
<td>26.10</td>
<td>36.52</td>
</tr>
</tbody>
</table>
</div>
</div>
<h2 class="title is-4">Hallucination</h2>
<div class="content has-text-justified">
<p>Performance of whether a model can refuse questions about the non-existing entity. The following is the F1 score. We use refusal as the positive label. Our test set consists of ~1k test samples per language. Each unanswerable question is generated by GPT4o. The ratio of answerable and unanswerable questions are 1:1. We define keywords to automatically detect whether a model-generated response is a refusal response.</p>
<div class="table-container">
<table class="table is-bordered is-striped is-narrow is-hoverable is-fullwidth">
<thead>
<tr>
<th>Refusal-F1 Scores</th>
<th>en</th>
<th>zh</th>
<th>vi</th>
<th>th</th>
<th>id</th>
<th>avg</th>
</tr>
</thead>
<tbody>
<tr>
<td>Qwen1.5-7B-Instruct</td>
<td>53.85</td>
<td>51.70</td>
<td>52.85</td>
<td>35.50</td>
<td>58.40</td>
<td>50.46</td>
</tr>
<tr>
<td>Qwen2-7B-Instruct</td>
<td>58.79</td>
<td>33.08</td>
<td>56.21</td>
<td>44.60</td>
<td>55.98</td>
<td>49.732</td>
</tr>
<tr>
<td>SeaLLM-7B-v2.5</td>
<td>12.90</td>
<td>0.77</td>
<td>2.45</td>
<td>19.42</td>
<td>0.78</td>
<td>7.26</td>
</tr>
<tr>
<td>Sailor-7B-Chat</td>
<td>33.49</td>
<td>18.82</td>
<td>5.19</td>
<td>9.68</td>
<td>16.42</td>
<td>16.72</td>
</tr>
<tr>
<td>glm-4-9b-chat</td>
<td>44.48</td>
<td>37.89</td>
<td>18.66</td>
<td>4.27</td>
<td>1.97</td>
<td>21.45</td>
</tr>
<tr>
<td>aya-23-8B</td>
<td>6.38</td>
<td>0.79</td>
<td>2.83</td>
<td>1.98</td>
<td>14.80</td>
<td>5.36</td>
</tr>
<tr>
<td>Llama-3-8B-Instruct</td>
<td>72.08</td>
<td>0.00</td>
<td>1.23</td>
<td>0.80</td>
<td>3.91</td>
<td>15.60</td>
</tr>
<tr>
<td>gemma-1.1-7b-it</td>
<td>52.39</td>
<td>27.74</td>
<td>23.96</td>
<td>22.97</td>
<td>31.72</td>
<td>31.76</td>
</tr>
<tr>
<td>SeaLLMs-v3-7B-Chat</td>
<td>71.36</td>
<td>78.39</td>
<td>77.93</td>
<td>61.31</td>
<td>68.95</td>
<td>71.588</td>
</tr>
</tbody>
</table>
</div>
</div>
<h2 class="title is-4">Safety</h2>
<div class="content has-text-justified">
<p>Multijaildataset consists of harmful prompts in multiple languages. We take those relevant prompts in SEA languages here and report their safe rate (the higher the better).</p>
<div class="table-container">
<table class="table is-bordered is-striped is-narrow is-hoverable is-fullwidth">
<thead>
<tr>
<th>Model</th>
<th>en</th>
<th>jv</th>
<th>th</th>
<th>vi</th>
<th>zh</th>
<th>avg</th>
</tr>
</thead>
<tbody>
<tr>
<td>Qwen2-7B-Instruct</td>
<td>0.8857</td>
<td>0.4381</td>
<td>0.6381</td>
<td>0.7302</td>
<td>0.873</td>
<td>0.713</td>
</tr>
<tr>
<td>Sailor-7B-Chat</td>
<td>0.7873</td>
<td>0.5492</td>
<td>0.6222</td>
<td>0.6762</td>
<td>0.7619</td>
<td>0.6794</td>
</tr>
<tr>
<td>Meta-Llama-3-8B-Instruct</td>
<td>0.8825</td>
<td>0.2635</td>
<td>0.7111</td>
<td>0.6984</td>
<td>0.7714</td>
<td>0.6654</td>
</tr>
<tr>
<td>Sailor-14B-Chat</td>
<td>0.8698</td>
<td>0.3048</td>
<td>0.5365</td>
<td>0.6095</td>
<td>0.727</td>
<td>0.6095</td>
</tr>
<tr>
<td>glm-4-9b-chat</td>
<td>0.7714</td>
<td>0.2127</td>
<td>0.3016</td>
<td>0.6063</td>
<td>0.7492</td>
<td>0.52824</td>
</tr>
<tr>
<td>SeaLLMs-v3-7B-Chat</td>
<td>0.8889</td>
<td>0.6000</td>
<td>0.7333</td>
<td>0.8381</td>
<td>0.927</td>
<td>0.7975</td>
</tr>
</tbody>
</table>
</div>
</div>
</div>
</section>
<!-- model information -->
<section class="section" id="BibTeX">
<div class="container is-max-desktop content">
<h2 class="title is-3">Related Links</h2>
<div class="content has-text-justified">
<p>
<a href="https://huggingface.co/SeaLLMs/SeaLLM3-7B-Chat">SeaLLMs-v3</a> was released in July 2024. It achieves SOTA performance of diverse tasks while specifically enhanced to be more trustworthy, exhibiting reduced hallucination and providing safe response.
</p>
<p>
<a href="https://huggingface.co/SeaLLMs/SeaLLM-7B-v2">SeaLLM-7B-v2.5</a> was released in April 2024. It possesses outstanding abilities in world knowledge and math reasoning in both English and SEA languages.
</p>
<p>
<a href="https://huggingface.co/SeaLLMs/SeaLLM-7B-v2">SeaLLM-7B-v2</a> was released in Feb 2024. It possesses outstanding abilities in math and commonsense reasoning in Sea languages.
</p>
<p>
<a href="https://huggingface.co/SeaLLMs/SeaLLM-7B-v1">SeaLLM-7B-v1</a> was released in Nov 2023. It was the first release of SeaLLMs model family, and the first LLM built specifically for Southeast Asia.
</p>
</div>
</div>
<div class="container is-max-desktop content">
<h2 class="title">Contributions</h2>
<p>Contributors of SeaLLMs (of all versions), ranked alphabetically by last name: Mahani Aljunied, Lidong Bing, Guanzheng Chen, Liying Cheng, Yue Deng, Zhiqiang Hu, Yew Ken Chia, Xingxuan Li, Xin Li, Chaoqun Liu, Xuan-Phi Nguyen, Hou Pong Chan, Chenhui Shen, Qingyu Tan, Jianyu Wang, Weiwen Xu, Sen Yang, Wenxuan Zhang, Hang Zhang, Yiran Zhao.</p>
</div>
<div class="container is-max-desktop content">
<h2 class="title">Acknowledgement</h2>
<p>We would like to express our special thanks to our professional and native linguists, Tantong Champaiboon, Nguyen Ngoc Yen Nhi and Tara Devina Putri, who helped build, evaluate, and fact-check our sampled pretraining and SFT dataset as well as evaluating our models across different aspects, especially safety.</p>
</div>
<div class="container is-max-desktop content">
<h2 class="title">BibTeX</h2>
<p>
If you find our project useful, we hope you would kindly star our repo and cite our work as follows.
<br>
Corresponding Author: <a href= "mailto: l.bing@alibaba-inc.com">l.bing@alibaba-inc.com</a>
</p>
<pre><code>@article{damonlp2024seallm3,
author = {Wenxuan Zhang*, Hou Pong Chan*, Yiran Zhao*, Mahani Aljunied*,
Jianyu Wang*, Chaoqun Liu, Yue Deng, Zhiqiang Hu, Weiwen Xu,
Yew Ken Chia, Xin Li, Lidong Bing},
title = {SeaLLMs 3: Open Foundation and Chat Multilingual Large Language Models for Southeast Asian Languages},
year = {2024},
url = {https://arxiv.org/abs/2407.19672}
}
</code></pre>
<pre><code>@article{damonlpsg2023seallm,
author = {Xuan-Phi Nguyen*, Wenxuan Zhang*, Xin Li*, Mahani Aljunied*,
Zhiqiang Hu, Chenhui Shen, Yew Ken Chia, Xingxuan Li, Jianyu Wang,
Qingyu Tan, Liying Cheng, Guanzheng Chen, Yue Deng, Sen Yang,
Chaoqun Liu, Hang Zhang, Lidong Bing},
title = {SeaLLMs - Large Language Models for Southeast Asia},
year = {2024},
booktitle = {ACL 2024 System Demonstrations},
url = {https://arxiv.org/pdf/2312.00738},
</code></pre>
</div>
</section>
<footer class="footer">
<div class="container">
<div class="content has-text-centered">
<a class="icon-link"
href="https://arxiv.org/pdf/2312.00738.pdf">
<i class="fas fa-file-pdf"></i>
</a>
<a class="icon-link" href="https://github.com/DAMO-NLP-SG/SeaLLMs" class="external-link" disabled>
<i class="fab fa-github"></i>
</a>
</div>
<div class="columns is-centered">
<div class="column is-8">
<div class="content">
<p>
This website is adapted from <a href="https://github.com/nerfies/nerfies.github.io">Nerfies</a>, licensed under a <a rel="license" href="http://creativecommons.org/licenses/by-sa/4.0/">Creative
Commons Attribution-ShareAlike 4.0 International License</a>.
</p>
</div>
</div>
</div>
</div>
</footer>
</body>
</html>