generated from eliahuhorwitz/Academic-project-page-template
-
Notifications
You must be signed in to change notification settings - Fork 0
/
index.html
615 lines (583 loc) · 23.3 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
<!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="DESCRIPTION META TAG">
<meta property="og:title" content="SOCIAL MEDIA TITLE TAG"/>
<meta property="og:description" content="SOCIAL MEDIA DESCRIPTION TAG TAG"/>
<meta property="og:url" content="URL OF THE WEBSITE"/>
<!-- Path to banner image, should be in the path listed below. Optimal dimenssions are 1200X630-->
<meta property="og:image" content="static/image/your_banner_image.png" />
<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>BaichuanSEED</title>
<link rel="icon" type="image/x-icon" href="static/images/seeding.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 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>
</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-2 publication-title">BaichuanSEED: <u>S</u>haring the Potential of <u>E</u>xtensiv<u>E</u> <u>D</u>ata Collection and Deduplication by Introducing a Competitive Large Language Model Baseline</h1>
<div class="is-size-5 publication-authors">
<!-- Paper authors -->
<span class="author-block">
Guosheng Dong<sup>1*</sup>,</span>
<span class="author-block">
Da Pan<sup>1</sup>,</span>
<span class="author-block">
<a href="https://emanual20.github.io" target="_blank">Yiding Sun</a><sup>1,2</sup>,
</span>
<span class="author-block">
Shusen Zhang<sup>1</sup>,</span>
<span class="author-block">
Zheng Liang<sup>1</sup>,</span>
<span class="author-block">
Xin Wu<sup>1</sup>,</span>
<span class="author-block">
Yanjun Shen<sup>1</sup>,</span>
<span class="author-block">
Fan Yang<sup>1</sup>,</span>
<span class="author-block">
Haoze Sun<sup>1</sup>,</span>
<span class="author-block">
Tianpeng Li<sup>1</sup>,</span>
<span class="author-block">
Mingan Lin<sup>1</sup>,</span>
<span class="author-block">
Jianhua Xu<sup>1</sup>,</span>
<span class="author-block">
Yufan Zhang<sup>1</sup>,</span>
<span class="author-block">
Xiaonan Nie<sup>1</sup>,</span>
<span class="author-block">
Lei Su<sup>1</sup>,</span>
<span class="author-block">
Bingning Wang<sup>1</sup>,</span>
<span class="author-block">
Wentao Zhang<sup>3</sup>,</span>
<span class="author-block">
Jiaxin Mao<sup>2</sup>,</span>
<span class="author-block">
Zenan Zhou<sup>1*</sup>,</span>
<span class="author-block">
Weipeng Chen<sup>1</sup></span>
</div>
<div class="is-size-5 publication-authors">
<span class="author-block">Baichuan Inc.<sup>1</sup><br> Gaoling School of Artificial Intelligence, Renmin University of China<sup>2</sup><br> Peking University<sup>3</sup></span>
<span class="eql-cntrb"><small><br><sup>*</sup>Corresponding Authors, {<a href="mailto:dongguosheng@baichuan-inc.com">dongguosheng</a>, <a href="mailto:zhouzenan@baichuan-inc.com">zhouzenan</a>} @baichuan-inc.com</small></span>
</div>
<div class="column has-text-centered">
<div class="publication-links">
<!-- ArXiv abstract Link -->
<span class="link-block">
<a href="https://arxiv.org/abs/2408.15079" 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/BaichuanSEED/BaichuanSEED.github.io" 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>
<!-- Huggingface link -->
<span class="link-block">
<a href="" target="_blank"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
🤗
</span>
<span>HuggingFace(TBU)</span>
</a>
</span>
</div>
</div>
</div>
</div>
</div>
</div>
</section>
<!-- 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>
The general capabilities of Large Language Models (LLM) highly rely on the composition and selection on extensive pretraining datasets, treated as commercial secrets by several institutions. To mitigate this issue, we open-source the details of a universally applicable data processing pipeline and validate its effectiveness and potential by introducing a competitive LLM baseline. Specifically, the data processing pipeline consists of broad collection to scale up and reweighting to improve quality. We then pretrain a 7B model BaichuanSEED with 3T tokens processed by our pipeline without any deliberate downstream task-related optimization, followed by an easy but effective supervised fine-tuning stage. The model demonstrates consistency and predictability throughout training. BaichuanSEED achieves comparable performance on comprehensive benchmarks with several commercial advanced large language models, such as Qwen1.5 and Llama3. We also conduct several heuristic experiments to discuss the potential for further optimization on downstream tasks, such as mathematics and coding.
</p>
</div>
</div>
</div>
</div>
</section>
<!-- End paper abstract -->
<!-- Paper abstract -->
<section class="section hero">
<div class="container is-max-desktop">
<div class="columns is-centered has-text-centered">
<div class="column">
<h2 class="title">Universal Data Processing Pipeline</h2>
<div class="content has-text-justified">
<ul>
<li>Broad Collection: broad collection from trusted sources, mainly including web pages, high knowledge density data, code.</li>
<li>Reweighting: deduplication and mixture (The details can be found in our technical report)
<ul>
<li> Deduplication
<ul>
<li>Document-level deduplication globally</li>
<li>Sentence-level deduplication across documents</li>
<li>PII and harmful content filtering</li>
</ul>
<li> Mixture
<ul>
<li>Heauristic mixture experiments</li>
</ul>
</ul>
</li>
</ul>
</div>
</div>
</div>
</div>
</section>
<!-- End paper abstract -->
<!-- Performance -->
<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">Performance</h2>
<div class="content has-text-justified">
<p>
BaichuanSEED achieves comparable performance with cutting-edge commercial LLMs (Qwen1.5-7B and Llama3-8B), and better performance over existing fully transparent LLMs (OLMO-7B and MAP-Neo-7B).
</p>
</div>
<h3 class="content">Comprehensive Benchmarks</h3>
<table style="width:100%; border-collapse:collapse;">
<thead>
<tr>
<th style="width: 26%"></th>
<th>Training Tokens</th>
<th>MMLU (5-shot)</th>
<th>CMMLU (5-shot)</th>
<th>AGIEval (0-shot)</th>
<th>C-Eval (5-shot)</th>
<th>MMLU-Pro (5-shot)</th>
<th>LiveBench (0-shot)</th>
</tr>
</thead>
<tr>
<td colspan="8" style="text-align:center; border-top:1px solid black; border-bottom:1px solid black;"></td>
</tr>
<tbody>
<tr>
<td>Baichuan2-7B</td>
<td>2.6T</td>
<td>54.65</td>
<td>56.95</td>
<td>28.95</td>
<td>56.19</td>
<td>21.65</td>
<td>-</td>
</tr>
<tr>
<td>Baichuan2-13B</td>
<td>2.6T</td>
<td>59.83</td>
<td>61.32</td>
<td>24.07</td>
<td>58.10</td>
<td>26.59</td>
<td>-</td>
</tr>
<tr>
<td>Qwen1.5-7B</td>
<td>3T</td>
<td><u>62.19</u></td>
<td><b>71.84</b></td>
<td><b>39.46</b></td>
<td><b>73.64</b></td>
<td><u>30.30</u></td>
<td>-</td>
</tr>
<tr>
<td>Llama3-8B</td>
<td>15T</td>
<td><b>66.57</b></td>
<td>50.68</td>
<td>26.74</td>
<td>49.89</td>
<td><b>35.30</b></td>
<td>-</td>
</tr>
<tr>
<td>OLMo-7B</td>
<td>2.5T</td>
<td>28.40</td>
<td>25.55</td>
<td>19.89</td>
<td>27.27</td>
<td>13.05</td>
<td>-</td>
</tr>
<tr>
<td>MAP-Neo-7B</td>
<td>4.5T</td>
<td>58.18</td>
<td>55.06</td>
<td><u>33.87</u></td>
<td>57.50</td>
<td>26.89</td>
<td>-</td>
</tr>
<tr>
<td colspan="8" style="text-align:center; border-top:1px solid black; border-bottom:1px solid black;"></td>
</tr>
<tr>
<td><b>BaichuanSEED</b></td>
<td>3T</td>
<td>60.25</td>
<td><u>62.09</u></td>
<td>31.07</td>
<td><u>61.58</u></td>
<td>26.57</td>
<td>-</td>
</tr>
<tr>
<td colspan="8" style="text-align:center; border-top:1px solid black; border-bottom:1px solid black;"></td>
</tr>
<tr>
<td>Baichuan2-7B-Chat</td>
<td>2.6T</td>
<td>54.35</td>
<td>55.36</td>
<td>35.29</td>
<td>55.09</td>
<td>25.11</td>
<td>12.89</td>
</tr>
<tr>
<td>Baichuan2-13B-Chat</td>
<td>2.6T</td>
<td>57.28</td>
<td><u>61.32</u></td>
<td>30.15</td>
<td>58.04</td>
<td>28.03</td>
<td>13.04</td>
</tr>
<tr>
<td>Qwen1.5-7B-Chat</td>
<td>3T</td>
<td><u>61.49</u></td>
<td><b>68.02</b></td>
<td><b>39.29</b></td>
<td><b>68.96</b></td>
<td>16.29</td>
<td>16.78</td>
</tr>
<tr>
<td>Llama3-8B-Instruct</td>
<td>15T</td>
<td><b>67.10</b></td>
<td>51.66</td>
<td><u>38.37</u></td>
<td>50.71</td>
<td><b>41.88</b></td>
<td><b>25.91</b></td>
</tr>
<tr>
<td>OLMo-7B-SFT</td>
<td>2.5T</td>
<td>47.49</td>
<td>35.49</td>
<td>29.12</td>
<td>35.43</td>
<td>17.99</td>
<td>8.80</td>
</tr>
<tr>
<td>MAP-Neo-7B-SFT</td>
<td>4.5T</td>
<td>58.31</td>
<td>55.24</td>
<td>37.98</td>
<td>55.58</td>
<td><u>30.24</u></td>
<td>14.35</td>
</tr>
<tr>
<td colspan="8" style="text-align:center; border-top:1px solid black; border-bottom:1px solid black;"></td>
</tr>
<tr>
<td><b>BaichuanSEED-SFT</b></td>
<td>3T</td>
<td>60.15</td>
<td>60.84</td>
<td>32.62</td>
<td><u>59.41</u></td>
<td>29.63</td>
<td><u>18.32</u></td>
</tr>
</tbody>
<tr>
<td colspan="8" style="text-align:center; border-top:1px solid black; border-bottom:1px solid black;"></td>
</tr>
</table>
<br>
<br>
<div class="content has-text-justified">
<p>
BaichuanSEED-SFT achieves the second best in code (MBPP and HumanEval), best in HellaSwag among all baselines, will underperforms in mathsmatics (MATH and GSM8K). To emphasis, we deliberately exclude downstream-task optimization to make BaichuanSEED "completely pure", such as upsamling on math and code, annealing trianing or introducing synthetic data. We discuss the potential of our model in the Discussion section of our paper.
</p>
</div>
<h3 class="content">Downstream Tasks</h3>
<table style="width:100%; border-collapse:collapse;">
<thead>
<tr>
<th style="width: 26%;"></th>
<th>Training Tokens</th>
<th>MBPP (3-shot)</th>
<th>HumanEval (0-shot)</th>
<th>MATH (4-shot)</th>
<th>GSM8K (4-shot)</th>
<th>TriviaQA (0-shot)</th>
<th>HellaSwag (0-shot)</th>
</tr>
</thead>
<tr>
<td colspan="8" style="text-align:center; border-top:1px solid black; border-bottom:1px solid black;"></td>
</tr>
<tbody>
<tr>
<td>Baichuan2-7B</td>
<td>2.6T</td>
<td>25.40</td>
<td>17.68</td>
<td>5.94</td>
<td>25.02</td>
<td>53.73</td>
<td>67.56</td>
</tr>
<tr>
<td>Baichuan2-13B</td>
<td>2.6T</td>
<td>30.88</td>
<td>17.07</td>
<td>10.68</td>
<td>52.08</td>
<td><u>58.73</u></td>
<td>71.09</td>
</tr>
<tr>
<td>Qwen1.5-7B</td>
<td>3T</td>
<td><u>36.60</u></td>
<td><b>53.05</b></td>
<td><b>21.08</b></td>
<td><b>54.74</b></td>
<td>50.92</td>
<td><u>72.64</u></td>
</tr>
<tr>
<td>Llama3-8B</td>
<td>15T</td>
<td><b>44.60</b></td>
<td><u>26.22</u></td>
<td>13.44</td>
<td>50.11</td>
<td><b>65.23</b></td>
<td><b>74.54</b></td>
</tr>
<tr>
<td>OLMo-7B</td>
<td>2.5T</td>
<td>21.00</td>
<td>11.59</td>
<td>1.72</td>
<td>2.00</td>
<td>49.81</td>
<td>70.31</td>
</tr>
<tr>
<td>MAP-Neo-7B</td>
<td>4.5T</td>
<td>25.90</td>
<td>7.93</td>
<td><u>15.14</u></td>
<td><u>53.90</u></td>
<td>54.80</td>
<td>67.85</td>
</tr>
<tr>
<td colspan="8" style="text-align:center; border-top:1px solid black; border-bottom:1px solid black;"></td>
</tr>
<tr>
<td><b>BaichuanSEED</b></td>
<td>3T</td>
<td>34.12</td>
<td>21.34</td>
<td>9.84</td>
<td>38.81</td>
<td>45.92</td>
<td>70.20</td>
</tr>
<tr>
<td colspan="8" style="text-align:center; border-top:1px solid black; border-bottom:1px solid black;"></td>
</tr>
<tr>
<td>Baichuan2-7B-Chat</td>
<td>2.6T</td>
<td>22.40</td>
<td>15.24</td>
<td>8.70</td>
<td>32.37</td>
<td>44.65</td>
<td>69.18</td>
</tr>
<tr>
<td>Baichuan2-13B-Chat</td>
<td>2.6T</td>
<td>26.30</td>
<td>18.90</td>
<td>8.62</td>
<td>56.79</td>
<td>53.47</td>
<td>72.32</td>
</tr>
<tr>
<td>Qwen1.5-7B-Chat</td>
<td>3T</td>
<td>12.58</td>
<td><b>29.27</b></td>
<td>13.12</td>
<td>56.10</td>
<td>10.22</td>
<td>72.81</td>
</tr>
<tr>
<td>Llama3-8B-Instruct</td>
<td>15T</td>
<td><b>52.17</b></td>
<td>21.34</td>
<td><u>25.62</u></td>
<td><b>78.17</b></td>
<td><b>63.37</b></td>
<td>71.45</td>
</tr>
<tr>
<td>OLMo-7B-SFT</td>
<td>2.5T</td>
<td>25.16</td>
<td>19.51</td>
<td>2.52</td>
<td>17.66</td>
<td>42.87</td>
<td><u>72.62</u></td>
</tr>
<tr>
<td>MAP-Neo-7B-SFT</td>
<td>4.5T</td>
<td>33.66</td>
<td><b>29.27</b></td>
<td><b>30.86</b></td>
<td><u>70.28</u></td>
<td><u>53.82</u></td>
<td>68.48</td>
</tr>
<tr>
<td colspan="8" style="text-align:center; border-top:1px solid black; border-bottom:1px solid black;"></td>
</tr>
<tr>
<td><b>BaichuanSEED-SFT</b></td>
<td>3T</td>
<td><u>37.60</u></td>
<td><u>23.17</u></td>
<td>14.06</td>
<td>53.98</td>
<td>43.92</td>
<td><b>73.03</b></td>
</tr>
<tr>
<td colspan="8" style="text-align:center; border-top:1px solid black; border-bottom:1px solid black;"></td>
</tr>
</tbody>
</table>
</div>
</div>
</div>
</section>
<!-- Performance -->
<!-- BibTex citation -->
<section class="section" id="BibTeX">
<div class="container is-max-desktop content">
<h2 class="title">BibTeX</h2>
<pre><code>@misc{dong2024baichuanseedsharingpotentialextensive,
title={BaichuanSEED: Sharing the Potential of ExtensivE Data Collection and Deduplication by Introducing a Competitive Large Language Model Baseline},
author={Guosheng Dong and Da Pan and Yiding Sun and Shusen Zhang and Zheng Liang and Xin Wu and Yanjun Shen and Fan Yang and Haoze Sun and Tianpeng Li and Mingan Lin and Jianhua Xu and Yufan Zhang and Xiaonan Nie and Lei Su and Bingning Wang and Wentao Zhang and Jiaxin Mao and Zenan Zhou and Weipeng Chen},
year={2024},
eprint={2408.15079},
archivePrefix={arXiv},
primaryClass={cs.CL},
url={https://arxiv.org/abs/2408.15079},
}</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">
<p>
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.
This website is licensed under a <a rel="license" href="http://creativecommons.org/licenses/by-sa/4.0/" target="_blank">Creative
Commons Attribution-ShareAlike 4.0 International License</a>.
</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>