-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathMixCPT.html
More file actions
514 lines (478 loc) · 26.9 KB
/
Copy pathMixCPT.html
File metadata and controls
514 lines (478 loc) · 26.9 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
<!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="Massive Language Adaptation of Large Language Models">
<meta property="og:title" content="MaLA-LM" />
<meta property="og:description" content="LLM" />
<meta property="og:url" content="mala-lm.github.io" />
<!-- 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>MixCPT</title>
<link rel="icon" type="image/x-icon" href="static/images/favicon.ico">
<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-1 publication-title">Rethinking Multilingual Continual Pretraining: Data
Mixing for Adapting LLMs Across Languages and Resources</h1>
<div class="is-size-5 publication-authors">
<span class="author-block">
<a href="http://zihao.cool" target="_blank">Zihao Li</a><sup>1</sup>,
</span>
<span class="author-block">
<a href="https://shaoxiongji.github.io/" target="_blank">Shaoxiong
Ji</a><sup>2,3,1*</sup>,
</span>
<span class="author-block">
<a href="https://researchportal.helsinki.fi/en/persons/hengyu-luo"
target="_blank">Hengyu Luo</a><sup>1</sup>,
</span>
<span class="author-block">
<a href="https://researchportal.helsinki.fi/en/persons/j%C3%B6rg-tiedemann"
target="_blank">Jörg Tiedemann</a><sup>1</sup>,
</span>
</div>
<div class="is-size-5 publication-authors">
<span class="author-block">
<sup>1</sup>University of Helsinki
<sup>2</sup>ELLIS Institute Finland
<sup>3</sup>University of Turku
<br>
</span>
<span class="eql-cntrb">
<small><br><sup>*</sup>Corresponding to shaoxiong.ji@utu.fi</small>
</span>
</div>
<div class="column has-text-centered">
<div class="publication-links">
<!-- Arxiv PDF link -->
<span class="link-block">
<a href="http://arxiv.org/abs/2504.04152" 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>
<!-- Github link -->
<span class="link-block">
<a href="https://github.com/MaLA-LM/MixCPT" target="_blank"
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>
<!-- ArXiv abstract Link -->
<span class="link-block">
<a href="http://arxiv.org/abs/2504.04152" 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>
<!-- HuggingFace Link -->
<span class="link-block">
<a href="https://huggingface.co/collections/Zihao-Li/mixcpt-67ea87fbc68ccdaa83fdc01c"
target="_blank" class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<span role="img" aria-label="Hugging Face Emoji">🤗</span>
</span>
<span>Models</span>
</a>
</span>
<!-- Discord link -->
<span class="link-block">
<a href="https://discord.com/invite/F5mEb7U6we" target="_blank"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="fab fa-discord"></i>
</span>
<span>Discord</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>
Large Language Models (LLMs) exhibit significant disparities in performance across
languages, primarily benefiting high-resource languages while marginalizing underrepresented
ones. Continual Pretraining (CPT) has emerged as a promising approach to address this
imbalance, although the relative effectiveness of monolingual, bilingual, and code-augmented
data strategies remains unclear. This study systematically evaluates 36 CPT configurations
involving three multilingual base models, across 30+ languages categorized as altruistic,
selfish, and stagnant, spanning various resource levels. Our findings reveal three major
insights: (1) Bilingual CPT improves multilingual classification but often causes language
mixing issues during generation. (2) Including programming code data during CPT consistently
enhances multilingual classification accuracy, particularly benefiting low-resource
languages, but introduces a trade-off by slightly degrading generation quality. (3) Contrary
to prior work, we observe substantial deviations from language classifications according to
their impact on cross-lingual transfer: Languages classified as altruistic often negatively
affect related languages, selfish languages show conditional and configuration-dependent
behavior, and stagnant languages demonstrate surprising adaptability under certain CPT
conditions. These nuanced interactions emphasize the complexity of multilingual
representation learning, underscoring the importance of systematic studies on generalizable
language classification to inform future multilingual CPT strategies.
</p>
</div>
</div>
</div>
</div>
</section>
<!-- End paper abstract -->
<style>
table {
width: 100%;
border-collapse: collapse;
margin-bottom: 20px;
box-shadow: 0 2px 5px rgba(0, 0, 0, 0.1);
border-radius: 8px;
overflow: hidden;
}
th,
td {
padding: 12px 15px;
text-align: left;
border-bottom: 1px solid #ddd;
}
th {
background-color: #f2f2f2;
font-weight: bold;
color: #333;
}
tbody tr:nth-child(even) {
background-color: #f9f9f9;
}
tbody tr:hover {
background-color: #e0f7fa;
}
td a {
color: #007bff;
text-decoration: none;
}
td a:hover {
text-decoration: underline;
}
</style>
<section class="section hero ">
<div class="container">
<div class="columns is-centered">
<table>
<thead>
<tr>
<th>🤗 Model Code</th>
<th>📝 Model Description</th>
</tr>
</thead>
<tbody>
<tr>
<td><a href="https://huggingface.co/Zihao-Li/L3-Mono-Alt" target="_blank">🤗 L3-Mono-Alt</a>
</td>
<td>🦙 Llama-3.1-8B CPT on monolingual data for altruistic languages</td>
</tr>
<tr>
<td><a href="https://huggingface.co/Zihao-Li/L3-Mono-Sel" target="_blank">🤗 L3-Mono-Sel</a>
</td>
<td>🦙 Llama-3.1-8B CPT on monolingual data for selfish languages</td>
</tr>
<tr>
<td><a href="https://huggingface.co/Zihao-Li/L3-Mono-Stag" target="_blank">🤗
L3-Mono-Stag</a></td>
<td>🦙 Llama-3.1-8B CPT on monolingual data for stagnant languages</td>
</tr>
<tr>
<td><a href="https://huggingface.co/Zihao-Li/L3-Bi-Alt" target="_blank">🤗 L3-Bi-Alt</a>
</td>
<td>🦙 Llama-3.1-8B CPT on bilingual data for altruistic languages</td>
</tr>
<tr>
<td><a href="https://huggingface.co/Zihao-Li/L3-Bi-Sel" target="_blank">🤗 L3-Bi-Sel</a>
</td>
<td>🦙 Llama-3.1-8B CPT on bilingual data for selfish languages</td>
</tr>
<tr>
<td><a href="https://huggingface.co/Zihao-Li/L3-Bi-Stag" target="_blank">🤗 L3-Bi-Stag</a>
</td>
<td>🦙 Llama-3.1-8B CPT on bilingual data for stagnant languages</td>
</tr>
<tr>
<td><a href="https://huggingface.co/Zihao-Li/L3-Mono-Code-Alt" target="_blank">🤗
L3-Mono-Code-Alt</a></td>
<td>🦙 Llama-3.1-8B CPT on monolingual+code data for altruistic languages</td>
</tr>
<tr>
<td><a href="https://huggingface.co/Zihao-Li/L3-Mono-Code-Sel" target="_blank">🤗
L3-Mono-Code-Sel</a></td>
<td>🦙 Llama-3.1-8B CPT on monolingual+code data for selfish languages</td>
</tr>
<tr>
<td><a href="https://huggingface.co/Zihao-Li/L3-Mono-Code-Stag" target="_blank">🤗
L3-Mono-Code-Stag</a></td>
<td>🦙 Llama-3.1-8B CPT on monolingual+code data for stagnant languages</td>
</tr>
<tr>
<td><a href="https://huggingface.co/Zihao-Li/L3-Bi-Code-Alt" target="_blank">🤗
L3-Bi-Code-Alt</a></td>
<td>🦙 Llama-3.1-8B CPT on bilingual+code data for altruistic languages</td>
</tr>
<tr>
<td><a href="https://huggingface.co/Zihao-Li/L3-Bi-Code-Sel" target="_blank">🤗
L3-Bi-Code-Sel</a></td>
<td>🦙 Llama-3.1-8B CPT on bilingual+code data for selfish languages</td>
</tr>
<tr>
<td><a href="https://huggingface.co/Zihao-Li/L3-Bi-Code-Stag" target="_blank">🤗
L3-Bi-Code-Stag</a></td>
<td>🦙 Llama-3.1-8B CPT on bilingual+code data for stagnant languages</td>
</tr>
<tr>
<td><a href="https://huggingface.co/Zihao-Li/L2-Mono-Alt" target="_blank">🤗 L2-Mono-Alt</a>
</td>
<td>🦙 Llama-2-7B CPT on monolingual data for altruistic languages</td>
</tr>
<tr>
<td><a href="https://huggingface.co/Zihao-Li/L2-Mono-Sel" target="_blank">🤗 L2-Mono-Sel</a>
</td>
<td>🦙 Llama-2-7B CPT on monolingual data for selfish languages</td>
</tr>
<tr>
<td><a href="https://huggingface.co/Zihao-Li/L2-Mono-Stag" target="_blank">🤗
L2-Mono-Stag</a></td>
<td>🦙 Llama-2-7B CPT on monolingual data for stagnant languages</td>
</tr>
<tr>
<td><a href="https://huggingface.co/Zihao-Li/L2-Bi-Alt" target="_blank">🤗 L2-Bi-Alt</a>
</td>
<td>🦙 Llama-2-7B CPT on bilingual data for altruistic languages</td>
</tr>
<tr>
<td><a href="https://huggingface.co/Zihao-Li/L2-Bi-Sel" target="_blank">🤗 L2-Bi-Sel</a>
</td>
<td>🦙 Llama-2-7B CPT on bilingual data for selfish languages</td>
</tr>
<tr>
<td><a href="https://huggingface.co/Zihao-Li/L2-Bi-Stag" target="_blank">🤗 L2-Bi-Stag</a>
</td>
<td>🦙 Llama-2-7B CPT on bilingual data for stagnant languages</td>
</tr>
<tr>
<td><a href="https://huggingface.co/Zihao-Li/L2-Mono-Code-Alt" target="_blank">🤗
L2-Mono-Code-Alt</a></td>
<td>🦙 Llama-2-7B CPT on monolingual+code data for altruistic languages</td>
</tr>
<tr>
<td><a href="https://huggingface.co/Zihao-Li/L2-Mono-Code-Sel" target="_blank">🤗
L2-Mono-Code-Sel</a></td>
<td>🦙 Llama-2-7B CPT on monolingual+code data for selfish languages</td>
</tr>
<tr>
<td><a href="https://huggingface.co/Zihao-Li/L2-Mono-Code-Stag" target="_blank">🤗
L2-Mono-Code-Stag</a></td>
<td>🦙 Llama-2-7B CPT on monolingual+code data for stagnant languages</td>
</tr>
<tr>
<td><a href="https://huggingface.co/Zihao-Li/L2-Bi-Code-Alt" target="_blank">🤗
L2-Bi-Code-Alt</a></td>
<td>🦙 Llama-2-7B CPT on bilingual+code data for altruistic languages</td>
</tr>
<tr>
<td><a href="https://huggingface.co/Zihao-Li/L2-Bi-Code-Sel" target="_blank">🤗
L2-Bi-Code-Sel</a></td>
<td>🦙 Llama-2-7B CPT on bilingual+code data for selfish languages</td>
</tr>
<tr>
<td><a href="https://huggingface.co/Zihao-Li/L2-Bi-Code-Stag" target="_blank">🤗
L2-Bi-Code-Stag</a></td>
<td>🦙 Llama-2-7B CPT on bilingual+code data for stagnant languages</td>
</tr>
<tr>
<td><a href="https://huggingface.co/Zihao-Li/V7-Mono-Alt" target="_blank">🤗 V7-Mono-Alt</a>
</td>
<td> Viking-7B CPT on monolingual data for altruistic languages</td>
</tr>
<tr>
<td><a href="https://huggingface.co/Zihao-Li/V7-Mono-Sel" target="_blank">🤗 V7-Mono-Sel</a>
</td>
<td> Viking-7B CPT on monolingual data for selfish languages</td>
</tr>
<tr>
<td><a href="https://huggingface.co/Zihao-Li/V7-Mono-Stag" target="_blank">🤗
V7-Mono-Stag</a></td>
<td> Viking-7B CPT on monolingual data for stagnant languages</td>
</tr>
<tr>
<td><a href="https://huggingface.co/Zihao-Li/V7-Bi-Alt" target="_blank">🤗 V7-Bi-Alt</a>
</td>
<td> Viking-7B CPT on bilingual data for altruistic languages</td>
</tr>
<tr>
<td><a href="https://huggingface.co/Zihao-Li/V7-Bi-Sel" target="_blank">🤗 V7-Bi-Sel</a>
</td>
<td> Viking-7B CPT on bilingual data for selfish languages</td>
</tr>
<tr>
<td><a href="https://huggingface.co/Zihao-Li/V7-Bi-Stag" target="_blank">🤗 V7-Bi-Stag</a>
</td>
<td> Viking-7B CPT on bilingual data for stagnant languages</td>
</tr>
<tr>
<td><a href="https://huggingface.co/Zihao-Li/V7-Mono-Code-Alt" target="_blank">🤗
V7-Mono-Code-Alt</a></td>
<td> Viking-7B CPT on monolingual+code data for altruistic languages</td>
</tr>
<tr>
<td><a href="https://huggingface.co/Zihao-Li/V7-Mono-Code-Sel" target="_blank">🤗
V7-Mono-Code-Sel</a></td>
<td> Viking-7B CPT on monolingual+code data for selfish languages</td>
</tr>
<tr>
<td><a href="https://huggingface.co/Zihao-Li/V7-Mono-Code-Stag" target="_blank">🤗
V7-Mono-Code-Stag</a></td>
<td> Viking-7B CPT on monolingual+code data for stagnant languages</td>
</tr>
<tr>
<td><a href="https://huggingface.co/Zihao-Li/V7-Bi-Code-Alt" target="_blank">🤗
V7-Bi-Code-Alt</a></td>
<td> Viking-7B CPT on bilingual+code data for altruistic languages</td>
</tr>
<tr>
<td><a href="https://huggingface.co/Zihao-Li/V7-Bi-Code-Sel" target="_blank">🤗
V7-Bi-Code-Sel</a></td>
<td> Viking-7B CPT on bilingual+code data for selfish languages</td>
</tr>
<tr>
<td><a href="https://huggingface.co/Zihao-Li/V7-Bi-Code-Stag" target="_blank">🤗
V7-Bi-Code-Stag</a></td>
<td> Viking-7B CPT on bilingual+code data for stagnant languages</td>
</tr>
</tbody>
</table>
</div>
</div>
</section>
<section class="section hero">
<div class="container">
<div class="columns is-centered">
<table class="datasets-table">
<thead>
<tr>
<th>🤗 Dataset Name</th>
<th>📝 Description</th>
</tr>
</thead>
<tbody>
<tr>
<td><a href="https://huggingface.co/datasets/Zihao-Li/Stagnant" target="_blank">🤗
Stagnant</a></td>
<td>Stagnant language data.</td>
</tr>
<tr>
<td><a href="https://huggingface.co/datasets/Zihao-Li/Selfish" target="_blank">🤗
Selfish</a></td>
<td>Selfish language data.</td>
</tr>
<tr>
<td><a href="https://huggingface.co/datasets/Zihao-Li/Altruistic" target="_blank">🤗
Altruistic</a></td>
<td>Altruistic language data.</td>
</tr>
<tr>
<td><a href="https://huggingface.co/datasets/Zihao-Li/Code" target="_blank">🤗 Code </a>
</td>
<td>Programming code data.</td>
</tr>
</tbody>
</table>
</div>
</div>
</section>
<!--BibTex citation -->
<section class="section" id="BibTeX">
<div class="container is-max-desktop content">
<h2 class="title">BibTeX</h2>
<pre><code>
@inproceedings{MixCPT,
title={Rethinking Multilingual Continual Pretraining: Data Mixing for Adapting LLMs Across Languages and Resources},
author={Zihao Li and Shaoxiong Ji and Hengyu Luo and Jörg Tiedemann},
year={2025},
booktitle={Proceedings of Conference on Language Modeling (COLM)},
url={https://arxiv.org/abs/2504.04152},
}
</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.
You are free to borrow the source code of this website, we just ask that you link back to
this page in the footer. <br> 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>
<!-- Google tag (gtag.js) -->
<script async src="https://www.googletagmanager.com/gtag/js?id=G-HHYRBMV6MN"></script>
<script>
window.dataLayer = window.dataLayer || [];
function gtag() { dataLayer.push(arguments); }
gtag('js', new Date());
gtag('config', 'G-HHYRBMV6MN');
</script>
</body>
</html>