-
Notifications
You must be signed in to change notification settings - Fork 0
/
index.html
746 lines (673 loc) · 36.5 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
<!DOCTYPE html>
<html>
<head>
<meta charset="utf-8">
<!-- Meta tags for social media banners, these should be filled in appropriatly as they are your "business card" -->
<!-- Replace the content tag with appropriate information -->
<meta name="description" content="LIBMoE: A Library for comprehensive benchmarking Mixture of Experts in Large Language Models">
<meta property="og:title" content="LIBMoE: A Library for comprehensive benchmarking Mixture of Experts in Large Language Models"/>
<meta property="og:description" content="Multiple Choice QA Benchmark for Assessing Code Understanding Capabilities"/>
<meta property="og:url" content="https://fsoft-aic.github.io/fsoft-LibMoE.github.io"/>
<!-- Path to banner image, should be in the path listed below. Optimal dimenssions are 1200X630-->
<meta property="og:image:width" content="1200"/>
<meta property="og:image:height" content="630"/>
<meta name="twitter:title" content="TWITTER BANNER TITLE META TAG">
<meta name="twitter:description" content="TWITTER BANNER DESCRIPTION META TAG">
<!-- Path to banner image, should be in the path listed below. Optimal dimenssions are 1200X600-->
<meta name="twitter:image" content="static/images/your_twitter_banner_image.png">
<meta name="twitter:card" content="summary_large_image">
<!-- Keywords for your paper to be indexed by-->
<meta name="keywords" content="KEYWORDS SHOULD BE PLACED HERE">
<meta name="viewport" content="width=device-width, initial-scale=1">
<title>LIBMoE: A Library for comprehensive benchmarking Mixture of Experts in Large Language Models</title>
<link href="https://fonts.googleapis.com/css?family=Google+Sans|Noto+Sans|Castoro"
rel="stylesheet">
<!-- TODO: replace with CodeMMLU logo -->
<link rel="icon" href="static/images/libmoe-logo.png">
<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 src="https://documentcloud.adobe.com/view-sdk/main.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>
<style>
table {
width: 100%;
border-collapse: collapse;
}
th, td {
border: 1px solid black;
padding: 8px;
text-align: center;
}
th[colspan="2"] {
border-bottom: 2px solid black;
}
.highlight {
font-weight: bold;
}
</style>
</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://fsoft-aic.github.io/fsoft-LibMoE.github.io">
<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://fsoft-aic.github.io/fsoft-LibMoE.github.io">
LibMoE
</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">LIBMoE: A Library for comprehensive benchmarking Mixture of Experts in Large Language Models</h1>
<div class="is-size-5 publication-authors">
<!-- Paper authors -->
<span class="author-block">
<a href="https://fsoft-aic.github.io/fsoft-LibMoE.github.io" target="_blank">Nam V. Nguyen<sup>1</sup></a>,</span>
<span class="author-block">
<a href="https://fsoft-aic.github.io/fsoft-LibMoE.github.io" target="_blank">Thong T. Doan<sup>1</sup></a>,</span>
<span class="author-block">
<a href="https://fsoft-aic.github.io/fsoft-LibMoE.github.io" target="_blank">Luong Tran<sup>1</sup></a>,</span>
<span class="author-block">
<a href="https://fsoft-aic.github.io/fsoft-LibMoE.github.io" target="_blank">Van Nguyen<sup>1</sup></a>,</span>
<span class="author-block">
<a href="https://fsoft-aic.github.io/fsoft-LibMoE.github.io" target="_blank">Quang Pham<sup>1, 2</sup></a>,</span>
</span>
</div>
<div class="is-size-5 publication-authors">
<span class="author-block"><sup>1</sup>FPT Software AI Center, <sup>2</sup>VNU-HCM- University of Science, Viet Nam</span>
</div>
<!-- <div class="column has-text-centered">
<div class="publication-links"> -->
<!-- Arxiv PDF link -->
<!-- <span class="link-block">
<a href="https://fsoft-aic.github.io/fsoft-LibMoE.github.io" target="_blank"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="fas fa-file-pdf"></i>
</span>
<span>Paper</span>
</a>
</span> -->
<!-- Supplementary PDF link -->
<!-- <span class="link-block">
<a href="static/pdfs/supplementary_material.pdf" target="_blank"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="fas fa-file-pdf"></i>
</span>
<span>Supplementary</span>
</a>
</span> -->
<!-- ArXiv abstract Link -->
<span class="link-block">
<a href="https://arxiv.org/abs/2411.00918" 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>
<!-- Github link -->
<span class="link-block">
<a href="https://github.com/Fsoft-AIC/LibMoE" target="_blank"
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>
<!-- Data Link -->
<!-- <span class="link-block">
<a href="https://github.com/Fsoft-AIC/LibMoE/blob/main/docs/dataset_guide.md" target="_blank"
class="external-link button is-normal is-rounded is-dark">
<span>🤗 Data Guide</span>
</a>
</span> -->
<span class="link-block">
<a href="https://github.com/Fsoft-AIC/LibMoE/blob/main/docs/dataset_guide.md" target="_blank"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<img src="static/images/user-guide.png" alt="Mod Icon" style="width: 16px; height: 16px;">
</span>
<span>Dataset Guide</span>
</a>
</span>
<span class="link-block">
<a href="https://github.com/Fsoft-AIC/LibMoE/blob/main/docs/model_guide.md" target="_blank"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<img src="static/images/user-guide.png" alt="Mod Icon" style="width: 16px; height: 16px;">
</span>
<span>Model Guide</span>
</a>
</span>
<span class="link-block">
<a href="https://github.com/Fsoft-AIC/LibMoE/blob/main/docs/checkpoint_list.md" target="_blank"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<img src="static/images/checkpoint.png" alt="Mod Icon" style="width: 16px; height: 16px;">
</span>
<span>Checkpoints List</span>
</a>
</span>
<!-- Leaderboard Link -->
<!-- <span class="link-block">
<a href="https://fsoft-ai4code.github.io/leaderboards/codemmlu/" target="_blank"
class="external-link button is-normal is-rounded is-dark">
<span>🏆 Leaderboard</span>
</a> -->
</span>
</div>
</div>
</div>
</div>
</div>
</div>
</section>
<!-- Teaser video-->
<!-- <section class="hero teaser">
<div class="container is-max-desktop">
<div class="hero-body">
<img src="static/images/demo.gif" alt="MY ALT TEXT"/>
<h2 class="subtitle has-text-centered">
Aliquam vitae elit ullamcorper tellus egestas pellentesque. Ut lacus tellus, maximus vel lectus at, placerat pretium mi. Maecenas dignissim tincidunt vestibulum. Sed consequat hendrerit nisl ut maximus.
</h2>
</div>
</div>
</section> -->
<!-- End teaser video -->
<!-- Paper abstract -->
<section class="section hero is-light">
<div class="container is-max-desktop">
<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>
Mixture of Experts (MoEs) plays an important role in the development of more efficient and effective large language models (LLMs). Due to the enormous resource requirements, studying large scale MoE algorithms remain in-accessible to many researchers. This work develops LibMoE, a comprehensive and modular framework to streamline the research, training, and evaluation of MoE algorithms. Built upon three core principles: (i) modular design, (ii) efficient training; (iii) comprehensive evaluation, LibMoE brings MoE in LLMs more accessible to a wide range of researchers by standardizing the training and evaluation pipelines. Using LibMoE, we extensively benchmarked five state-of-the-art MoE algorithms over three different LLMs and 11 datasets under the zero-shot setting. The results show that despite the unique characteristics, all MoE algorithms perform roughly similar when averaged across a wide range of tasks. With the modular design and extensive evaluation, we believe LibMoE will be invaluable for researchers to make meaningful progress towards the next generation of MoE and LLMs.
</p>
</div>
</div>
</div>
</div>
</section>
<section class="section hero is-light">
<div class="container is-max-desktop">
<div class="columns is-centered has-text-centered">
<div class="column is-four-fifths">
<h2 class="title is-3">Overview</h2>
<div class="content has-text-justified">
<p>
Our work introduces <b>LibMoE</b>, a toolkit designed to simplify MoE research in LLMs by supporting distributed training and comprehensive evaluations across multiple MoE algorithms. With a modular design, LibMoE allows for extensive customization of MoE components (e.g., sparsity, router interactions, balancing losses). Incorporating the latest sparse upcycling techniques, it enables affordable MoE integration into existing dense LLM checkpoints. Our training pipeline, achievable within <b>55 hours</b> using <b>4 x A100 GPUs</b>, while the MoE upcycling step can be finished within <b>32 hours</b> only, offers a cost-effective solution while preserving evaluation fidelity. Our main contributions is summarized below:
</p>
<p>
</p>
<p>
<ul>
<li> <b>First</b>, we present LibMoE, a comprehensive toolkit to streamline the development of MoE in LLMs.</li>
<li> <b>Second</b>, with LibMoE, we implemented a standard benchmark to standardize the evaluation of five state-of-the-art MoE algorithms.</li>
<li> <b>Lastly</b>, LibMoE facilitates research beyond reporting the final performance by allowing researchers to easily explore various factors such as early-stopping, expert assignments, architecture choices, and many more.</li>
</ul>
</p>
<p>
</div>
<figure>
<img src="static/images/design_principle.png", width="100%"></img>
<figcaption><i><b>The detailed design of LiBMoE</b>, which comprises three major modules. First, the MoE module implements various MoE algorithms. Second, the training modules handles the training process and supports various configurations. Lastly, the evaluation module supports almost 100 zero-shot benchmarks and a wide-range of metrics.</i></figcaption>
</figure>
<!-- <embed src="static/images/data_pipeline.pdf" width="100%"/> -->
</div>
</div>
</div>
</section>
<section class="section hero is-light">
<div class="container is-max-desktop">
<div class="columns is-centered has-text-centered">
<div class="column is-four-fifths">
<h2 class="title is-3">Training and Evaluation Pipelines</h2>
<div class="content has-text-justified">
<p>
<b>Training Pipeline:</b> training, we adopt the vision-language pre-training task, which is one of the more challenging problems and only requires a rather small amount of data to start <b>(around 1e9 tokens)</b>. To this end, we follow the CUMO framework to upcycle the LLaVA model, which consists of three modules: a pre-trained visual encoder, a pre-trained LLM, and a randomly initialized MLP connector. Training follows a two-stage process:
</p>
<p>
<ul>
<li> <b>Dense Training</b> The dense training stage initializes the MLP connector and trains it to connect the pre-trained visual encoder to the pre-trained LLM.</li>
<li> <b>MoE Training</b> the MoE training stage upcycles the model to become MoE and also trains all components to obtain the visual instruction following capabilities.</li>
</ul>
</p>
<p>
<b>Evaluation Pipeline:</b> To drive the MoE developments towards real-world scenarios, we implement LibMoE to evaluate the algorithms in the zero-shot setting. To this end, we modify the <b>LMMS-Eval framework</b> to evaluate the final checkpoints of various MoE algorithms. Particularly, we carefully select <b>11 popular benchmarks</b> provided by LMMS-Eval and report the evaluation results. Additionally, we also provide a LibMoE's model loader so that future users can freely explore almost <b>100 benchmarks</b> supported by LMMS-Eval.
</p>
</div>
<figure>
<img src="static/images/training_module.png", width="100%"></img>
<figcaption><i><b>Overview of the LibMoE architecture and training process.</b> In the first stage of Dense Training, only the MLP is trained to improve alignment. In the second stage, all parameters are trained. During MoE Training, the feed-forward networks (FFNs) of the Vision Encoder (VE) and MLP Connector are used to initialize the experts within the MoE framework, and all parameters continue to be trained.</i></figcaption>
</figure>
</div>
</div>
</div>
</section>
<section class="section hero is-light">
<div class="container is-max-desktop">
<div class="columns is-centered has-text-centered">
<div class="column is-four-fifths">
<h2 class="title is-3">Experimental results</h2>
<div class="content has-text-justified">
<p>
<strong>Comparison of MoE algorithms on different models and training data Sizes for visual instruction tuning.</strong> The data set is constructed from LLaVA-665K. We highlight the highest (best) results in bold. Model: We consider five algorithms: SMoE-R (SMoE Router), Cosine-R, Sigmoid-R (Sigmoid Router), Hyper-R (Hyper Router), and Perturbed Cosine-R (Perturbed Cosine Router). We only show the performance of <b>CLIP + Phi3 model</b>, To see the full results, please visit our <a href="https://arxiv.org/abs/2411.00918">👉 <b>Paper</b> 👈</a>.
</p>
</div>
<div class="content has-text-justified" style="display: flex; justify-content: center;">
<style type="text/css">
.tg {border-collapse:collapse;border-color:#ccc;border-spacing:0;}
.tg td{background-color:#fff;border-color:#ccc;border-style:solid;border-width:1px;color:#333;
font-family:Arial, sans-serif;font-size:14px;overflow:hidden;padding:10px 5px;word-break:normal;}
.tg th{background-color:#f0f0f0;border-color:#ccc;border-style:solid;border-width:1px;color:#333;
font-family:Arial, sans-serif;font-size:14px;font-weight:normal;overflow:hidden;padding:10px 5px;word-break:normal;}
.tg .tg-baqh{text-align:center;vertical-align:center;font-weight: bold;}
.tg .tg-buh4{background-color:#f9f9f9;text-align:center;vertical-align:top}
.tg .tg-0lax{text-align:center;vertical-align:top}
.tg .tg-amwm{font-weight:bold;text-align:center;vertical-align:top}
</style>
<table class="tg">
<thead>
<tr>
<th class="tg-amwm">Data</th>
<th class="tg-amwm">Model</th>
<th class="tg-amwm">MoE Method</th>
<th class="tg-amwm">AI2D</th>
<th class="tg-amwm">Text VQA</th>
<th class="tg-amwm">GQA</th>
<th class="tg-amwm">Hallusion Benchmark</th>
<th class="tg-amwm">MathVista Validation</th>
<th class="tg-amwm">MMBenchEN dev</th>
<th class="tg-amwm">MMMU Validation</th>
<th class="tg-amwm">MMStar</th>
<th class="tg-amwm">POPE</th>
<th class="tg-amwm">SQA Full</th>
<th class="tg-amwm">MME</th>
<th class="tg-amwm">AVEGAGE: (w/o MME)</th>
</tr>
</thead>
<tbody>
<tr>
<td class="tg-baqh" rowspan="5">332k</td>
<td class="tg-baqh" rowspan="5">CLIP + Phi3</td>
<td class="tg-buh4">SMoE-R</td>
<td class="tg-buh4">63.67</td>
<td class="tg-buh4">47.47</td>
<td class="tg-buh4">59.46</td>
<td class="tg-buh4">43.32</td>
<td class="tg-buh4">31.60</td>
<td class="tg-buh4">66.67</td>
<td class="tg-buh4">40.11</td>
<td class="tg-buh4">37.94</td>
<td class="tg-buh4">86.87</td>
<td class="tg-buh4">77.23</td>
<td class="tg-buh4">1,608.21</td>
<td class="tg-buh4">55.42</td>
</tr>
<tr>
<td class="tg-buh4">Cosine-R</td>
<td class="tg-buh4">63.31</td>
<td class="tg-buh4">48.83</td>
<td class="tg-buh4">59.25</td>
<td class="tg-buh4">41.54</td>
<td class="tg-buh4">31.80</td>
<td class="tg-buh4">67.96</td>
<td class="tg-buh4">39.56</td>
<td class="tg-buh4">39.09</td>
<td class="tg-buh4">86.81</td>
<td class="tg-buh4">76.96</td>
<td class="tg-buh4"><b>1,637.99</b></td>
<td class="tg-buh4">55.51</td>
</tr>
<tr>
<td class="tg-buh4">Sigmoid-R</td>
<td class="tg-buh4">63.80</td>
<td class="tg-buh4">47.74</td>
<td class="tg-buh4">59.24</td>
<td class="tg-buh4">41.43</td>
<td class="tg-buh4">31.40</td>
<td class="tg-buh4">68.30</td>
<td class="tg-buh4">40.78</td>
<td class="tg-buh4">38.70</td>
<td class="tg-buh4">87.49</td>
<td class="tg-buh4">77.61</td>
<td class="tg-buh4">1,611.36</td>
<td class="tg-buh4">55.65</td>
</tr>
<tr>
<td class="tg-buh4">Hyper-R</td>
<td class="tg-buh4">64.05</td>
<td class="tg-buh4">47.76</td>
<td class="tg-buh4">59.61</td>
<td class="tg-buh4">41.11</td>
<td class="tg-buh4">32.50</td>
<td class="tg-buh4">69.24</td>
<td class="tg-buh4">41.33</td>
<td class="tg-buh4">39.27</td>
<td class="tg-buh4">86.68</td>
<td class="tg-buh4">77.31</td>
<td class="tg-buh4">1,602.59</td>
<td class="tg-buh4"><b>55.89</b></td>
</tr>
<tr>
<td class="tg-buh4">Perturbed Cosine-R</td>
<td class="tg-buh4">64.60</td>
<td class="tg-buh4">47.92</td>
<td class="tg-buh4">59.08</td>
<td class="tg-buh4">41.54</td>
<td class="tg-buh4">30.60</td>
<td class="tg-buh4">67.87</td>
<td class="tg-buh4">40.22</td>
<td class="tg-buh4">38.84</td>
<td class="tg-buh4">86.81</td>
<td class="tg-buh4">77.82</td>
<td class="tg-buh4">1,619.69</td>
<td class="tg-buh4">55.63</td>
</tr>
<tr>
<td class="tg-baqh" rowspan="5">665k</td>
<td class="tg-baqh" rowspan="5">CLIP + Phi3</td>
<td class="tg-buh4">SMoE-R</td>
<td class="tg-buh4">64.25</td>
<td class="tg-buh4">46.57</td>
<td class="tg-buh4">62.12</td>
<td class="tg-buh4">40.48</td>
<td class="tg-buh4">31.00</td>
<td class="tg-buh4">68.12</td>
<td class="tg-buh4">39.89</td>
<td class="tg-buh4">37.13</td>
<td class="tg-buh4">87.50</td>
<td class="tg-buh4">77.74</td>
<td class="tg-buh4">1,700.61</td>
<td class="tg-buh4">55.48</td>
</tr>
<tr>
<td class="tg-buh4">Cosine-R</td>
<td class="tg-buh4">64.51</td>
<td class="tg-buh4">49.79</td>
<td class="tg-buh4">61.38</td>
<td class="tg-buh4">40.80</td>
<td class="tg-buh4">31.30</td>
<td class="tg-buh4">67.01</td>
<td class="tg-buh4">40.67</td>
<td class="tg-buh4">39.36</td>
<td class="tg-buh4">87.52</td>
<td class="tg-buh4">77.48</td>
<td class="tg-buh4">1,687.37</td>
<td class="tg-buh4">55.98</td>
</tr>
<tr>
<td class="tg-buh4">Sigmoid-R</td>
<td class="tg-buh4">64.38</td>
<td class="tg-buh4">47.12</td>
<td class="tg-buh4">61.65</td>
<td class="tg-buh4">40.80</td>
<td class="tg-buh4">31.90</td>
<td class="tg-buh4">67.87</td>
<td class="tg-buh4">40.11</td>
<td class="tg-buh4">39.20</td>
<td class="tg-buh4">86.93</td>
<td class="tg-buh4">77.17</td>
<td class="tg-buh4">1,710.42</td>
<td class="tg-buh4">55.71</td>
</tr>
<tr>
<td class="tg-buh4">Hyper-R</td>
<td class="tg-buh4">64.37</td>
<td class="tg-buh4">47.59</td>
<td class="tg-buh4">59.70</td>
<td class="tg-buh4">40.38</td>
<td class="tg-buh4">31.30</td>
<td class="tg-buh4">68.30</td>
<td class="tg-buh4">40.78</td>
<td class="tg-buh4">38.33</td>
<td class="tg-buh4">85.70</td>
<td class="tg-buh4">80.33</td>
<td class="tg-buh4"><b>1,726.87</b></td>
<td class="tg-buh4">55.68</td>
</tr>
<tr>
<td class="tg-buh4">Perturbed Cosine-R</td>
<td class="tg-buh4">64.70</td>
<td class="tg-buh4">47.16</td>
<td class="tg-buh4">61.90</td>
<td class="tg-buh4">39.43</td>
<td class="tg-buh4">32.80</td>
<td class="tg-buh4">69.50</td>
<td class="tg-buh4">39.89</td>
<td class="tg-buh4">40.33</td>
<td class="tg-buh4">87.42</td>
<td class="tg-buh4">77.64</td>
<td class="tg-buh4">1,672.70</td>
<td class="tg-buh4"><b>56.08</b></td>
</tr>
</tbody>
</table>
</div>
</div>
</div>
</div>
</section>
<section class="section hero is-light">
<div class="container is-max-desktop">
<div class="columns is-centered has-text-centered">
<div class="column is-four-fifths">
<h2 class="title is-3">Analysis: The Performance of different MoE algorithms over time</h2>
<div class="content has-text-justified">
<p>
Figure below offers a detailed view of the time-dependent performance of five MoE algorithms across 11 benchmarks. This figure illustrates the unique behavioral characteristics of each algorithm and supports our observation that, in most cases, the final checkpoints of the MoE algorithms do not necessarily yield the best performance. This finding underscores the potential benefits of applying early stopping to achieve optimal results.
</p>
</div>
<figure>
<img src="static/images/timelearnmetric_all_benchmarks.png", width="100%"></img>
<figcaption><i>Comparison of the performance of different MoE algorithms across 11 benchmarks over time. The experiments were conducted using the LLaVa-332K dataset and the CLIP + Phi3 model.</i></figcaption>
</figure>
</div>
</div>
</div>
</section>
<section class="section hero is-light">
<div class="container is-max-desktop">
<div class="columns is-centered has-text-centered">
<div class="column is-four-fifths">
<h2 class="title is-3">Model Checkpoints</h2>
<div class="content has-text-justified">
<p>
We are making our entire experiment checkpoints publicly available to contribute to the community's research on the topic of Mixture of Experts (MoE). By reusing our checkpoints at the <b>Pre-Training</b> and <b>Pre-FineTuning</b> stages, we hope to help others save time and computational resources in their own experiments. To see the full checkpoints list, please visit our <a href="https://github.com/Fsoft-AIC/LibMoE">👉 <b>GitHub</b> 👈</a>.
</p>
</div>
<div class="content has-text-justified" style="display: flex; justify-content: center;">
<style type="text/css">
.tg {border-collapse:collapse;border-color:#ccc;border-spacing:0;}
.tg td{background-color:#fff;border-color:#ccc;border-style:solid;border-width:1px;color:#333;
font-family:Arial, sans-serif;font-size:14px;overflow:hidden;padding:10px 5px;word-break:normal;
text-align:center;}
/* .tg th{background-color:#f0f0f0;border-color:#ccc;border-style:solid;border-width:1px;color:#333;
font-family:Arial, sans-serif;font-size:14px;font-weight:normal;overflow:hidden;padding:10px 5px;word-break:normal;
text-align:center;} */
.tg .th{background-color:#fff;border-color:#ccc;border-style:solid;border-width:1px;color:#333;
font-family:Arial, sans-serif;font-size:14px;overflow:hidden;padding:10px 5px;word-break:normal;
text-align:center;}
.tg .tg-baqh{text-align:center;vertical-align:top;font-weight: bold;}
.tg .tg-0lax{text-align:center;vertical-align:top;color:blue}
.tg a {color: blue;}
</style>
<table class="tg">
<thead>
<tr>
<td class="tg-baqh">Method</td>
<td class="tg-baqh">Stage</td>
<td class="tg-baqh">Siglip 224 + Phi3.5</td>
<td class="tg-baqh">Siglip 224 + Phi3</td>
<td class="tg-baqh">CLIP 336 + Phi3</td>
</tr>
</thead>
<tbody>
<tr>
<td class="tg-baqh">Pre-Training</td>
<td class="tg-0lax">-</td>
<td class="tg-0lax"><a href="https://huggingface.co/Fsoft-AIC/Phi3.5-Siglip-MoE/tree/main/pretrain">Link</a></td>
<td class="tg-0lax"><a href="https://huggingface.co/Fsoft-AIC/Phi3-SigLiP-MoE/tree/main/pretrain">Link</a></td>
<td class="tg-0lax"><a href="https://huggingface.co/Fsoft-AIC/Phi3-CLIP-MoE/tree/main/pft">Link</a></td>
</tr>
<tr>
<td class="tg-baqh">Pre-FineTuning</td>
<td class="tg-0lax">-</td>
<td class="tg-0lax"><a href="https://huggingface.co/Fsoft-AIC/Phi3.5-Siglip-MoE/tree/main/pft">Link</a></td>
<td class="tg-0lax"><a href="https://huggingface.co/Fsoft-AIC/Phi3-SigLiP-MoE/tree/main/pft">Link</a></td>
<td class="tg-0lax"><a href="https://huggingface.co/Fsoft-AIC/Phi3-CLIP-MoE/tree/main/pretrain">Link</a></td>
</tr>
<tr>
<td class="tg-baqh" rowspan="5">VIT 665K</td>
<td class="tg-baqh">SMoE-R</td>
<td class="tg-0lax"><a href="https://huggingface.co/Fsoft-AIC/Phi3.5-Siglip-MoE/tree/main/sft_full/smoe">Link</a></td>
<td class="tg-0lax"><a href="https://huggingface.co/Fsoft-AIC/Phi3-SigLiP-MoE/tree/main/sft/smoe">Link</a></td>
<td class="tg-0lax"><a href="https://huggingface.co/Fsoft-AIC/Phi3-CLIP-MoE/tree/main/sft_full/smoe">Link</a></td>
</tr>
<tr>
<td class="tg-baqh">Cosine-R</td>
<td class="tg-0lax"><a href="https://huggingface.co/Fsoft-AIC/Phi3.5-Siglip-MoE/tree/main/sft_full/smoe_cosinegating">Link</a></td>
<td class="tg-0lax"><a href="https://huggingface.co/Fsoft-AIC/Phi3-SigLiP-MoE/tree/main/sft/smoe_cosinegating">Link</a></td>
<td class="tg-0lax"><a href="https://huggingface.co/Fsoft-AIC/Phi3-CLIP-MoE/tree/main/sft_full/smoe_cosinegating">Link</a></td>
</tr>
<tr>
<td class="tg-baqh">Sigmoid-R</td>
<td class="tg-0lax"><a href="https://huggingface.co/Fsoft-AIC/Phi3-SigLiP-MoE/tree/main/sft">Link</a></td>
<td class="tg-0lax"><a href="https://huggingface.co/Fsoft-AIC/Phi3-SigLiP-MoE/tree/main/sft">Link</a></td>
<td class="tg-0lax"><a href="https://huggingface.co/Fsoft-AIC/Phi3-SigLiP-MoE/tree/main/sft">Link</a></td>
</tr>
<tr>
<td class="tg-baqh">Hyper-R</td>
<td class="tg-0lax"><a href="https://huggingface.co/Fsoft-AIC/Phi3.5-Siglip-MoE/tree/main/sft_full/hyperrouter">Link</a></td>
<td class="tg-0lax"><a href="https://huggingface.co/Fsoft-AIC/Phi3-SigLiP-MoE/tree/main/sft/hyperrouter">Link</a></td>
<td class="tg-0lax"><a href="https://huggingface.co/Fsoft-AIC/Phi3-CLIP-MoE/tree/main/sft_full/hyperrouter">Link</a></td>
</tr>
<tr>
<td class="tg-baqh">Perturbed Cosine-R</td>
<td class="tg-0lax"><a href="https://huggingface.co/Fsoft-AIC/Phi3.5-Siglip-MoE/tree/main/sft_full/smoe_perturbed">Link</a></td>
<td class="tg-0lax"><a href="https://huggingface.co/Fsoft-AIC/Phi3-SigLiP-MoE/tree/main/sft/smoe_perturbed">Link</a></td>
<td class="tg-0lax"><a href="https://huggingface.co/Fsoft-AIC/Phi3-CLIP-MoE/tree/main/sft_full/smoe_perturbed">Link</a></td>
</tr>
</tbody>
</table>
</div>
</div>
</div>
</div>
</section>
<!-- <section class="section hero is-light">
<div class="container is-max-desktop">
<div class="columns is-centered has-text-centered">
<div class="column is-four-fifths">
<h2 class="title is-3">Enhancing Functional Correctness and Dependency Invocation abilities</h2>
<div class="content has-text-justified">
<p>
Two approaches are investigated to enhance the performance of generated code in terms of both functional correctness and dependency invocation.
<ul>
<li><b>Multi-round Debugging:</b> Leveraging test execution outputs and incorporating self-refinement through multiple rounds can dramatically boost a model's performance in generating accurate code and effectively utilizing dependencies.</li>
<figure>
<img src="static/images/debug_result.png", width="100%">
<figcaption>Figure 2: Improvement of the performance of several models on RepoExec after 3-round debugging process.</figcaption>
</figure>
<li><b>Instruction tuning:</b> RepoExec also comes with a valuable instruction-tuning training dataset. The experimental results, highlighted in the table below, clearly demonstrate the effectiveness of this approach with just a single round of generation.</li>
<figure>
<img src="static/images/instruction_tuning.png", width="100%">
<figcaption>Table 2: Improvement of the performance of several models on RepoExec after instruction tuning.</figcaption>
</figure>
</ul>
</p>
</div>
</div>
</div>
</div>
</section> -->
<!-- End paper abstract -->
<!-- Youtube video -->
<!-- -->
<!-- End youtube video -->
<!-- Video carousel -->
<!-- End video carousel -->
<!-- Paper poster -->
<!--End paper poster -->
<!--BibTex citation -->
<section class="section" id="BibTeX">
<div class="container is-max-desktop content">
<h2 class="title">BibTeX</h2>
<pre>
<code>
@misc{nguyen2024libmoelibrarycomprehensivebenchmarking,
title={LIBMoE: A Library for comprehensive benchmarking Mixture of Experts in Large Language Models},
author={Nam V. Nguyen and Thong T. Doan and Luong Tran and Van Nguyen and Quang Pham},
year={2024},
eprint={2411.00918},
archivePrefix={arXiv},
primaryClass={cs.CL},
url={https://arxiv.org/abs/2411.00918},
}
</code>
</pre>
</div>
</section>
<!--End BibTex citation -->
<footer class="footer">
<div class="container">
<div class="columns is-centered">
<div class="column is-8">
<div class="content", style="text-align: center;">
<img src="static/images/aic_logo.png", width="120px">
<p>
<h5>Contact us</h5>
🌐: <a href="https://www.fpt-aicenter.com/ai-residency/">fpt-aicenter</a>
</ul>
</p>
<p>
<h5>Acknowledgements</h5>
This page was built using the <a href="https://github.com/eliahuhorwitz/Academic-project-page-template" target="_blank">Academic Project Page Template</a> which was adopted from the <a href="https://nerfies.github.io" target="_blank">Nerfies</a> project page.
</p>
</div>
</div>
</div>
</div>
</footer>
<!-- Statcounter tracking code -->
<!-- You can add a tracker to track page visits by creating an account at statcounter.com -->
<!-- End of Statcounter Code -->
</body>
</html>