xmod-base

X-MOD is a multilingual masked language model trained on filtered CommonCrawl data containing 81 languages. It was introduced in the paper Lifting the Curse of Multilinguality by Pre-training Modular Transformers (Pfeiffer et al., NAACL 2022) and first released in this repository.

Because it has been pre-trained with language-specific modular components (language adapters), X-MOD differs from previous multilingual models like XLM-R. For fine-tuning, the language adapters in each transformer layer are frozen.

Usage

Tokenizer

This model reuses the tokenizer of XLM-R.

Input Language

Because this model uses language adapters, you need to specify the language of your input so that the correct adapter can be activated:

from transformers import XmodModel

model = XmodModel.from_pretrained("facebook/xmod-base")
model.set_default_language("en_XX")

A directory of the language adapters in this model is found at the bottom of this model card.

Fine-tuning

In the experiments in the original paper, the embedding layer and the language adapters are frozen during fine-tuning. A method for doing this is provided in the code:

model.freeze_embeddings_and_language_adapters()
# Fine-tune the model ...

Cross-lingual Transfer

After fine-tuning, zero-shot cross-lingual transfer can be tested by activating the language adapter of the target language:

model.set_default_language("de_DE")
# Evaluate the model on German examples ...

Bias, Risks, and Limitations

Please refer to the model card of XLM-R, because X-MOD has a similar architecture and has been trained on similar training data.

Citation

BibTeX:

@inproceedings{pfeiffer-etal-2022-lifting,
    title = "Lifting the Curse of Multilinguality by Pre-training Modular Transformers",
    author = "Pfeiffer, Jonas  and
      Goyal, Naman  and
      Lin, Xi  and
      Li, Xian  and
      Cross, James  and
      Riedel, Sebastian  and
      Artetxe, Mikel",
    booktitle = "Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies",
    month = jul,
    year = "2022",
    address = "Seattle, United States",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2022.naacl-main.255",
    doi = "10.18653/v1/2022.naacl-main.255",
    pages = "3479--3495"
}

Languages

This model contains the following language adapters:

lang_id (Adapter index)	Language code	Language
0	en_XX	English
1	id_ID	Indonesian
2	vi_VN	Vietnamese
3	ru_RU	Russian
4	fa_IR	Persian
5	sv_SE	Swedish
6	ja_XX	Japanese
7	fr_XX	French
8	de_DE	German
9	ro_RO	Romanian
10	ko_KR	Korean
11	hu_HU	Hungarian
12	es_XX	Spanish
13	fi_FI	Finnish
14	uk_UA	Ukrainian
15	da_DK	Danish
16	pt_XX	Portuguese
17	no_XX	Norwegian
18	th_TH	Thai
19	pl_PL	Polish
20	bg_BG	Bulgarian
21	nl_XX	Dutch
22	zh_CN	Chinese (simplified)
23	he_IL	Hebrew
24	el_GR	Greek
25	it_IT	Italian
26	sk_SK	Slovak
27	hr_HR	Croatian
28	tr_TR	Turkish
29	ar_AR	Arabic
30	cs_CZ	Czech
31	lt_LT	Lithuanian
32	hi_IN	Hindi
33	zh_TW	Chinese (traditional)
34	ca_ES	Catalan
35	ms_MY	Malay
36	sl_SI	Slovenian
37	lv_LV	Latvian
38	ta_IN	Tamil
39	bn_IN	Bengali
40	et_EE	Estonian
41	az_AZ	Azerbaijani
42	sq_AL	Albanian
43	sr_RS	Serbian
44	kk_KZ	Kazakh
45	ka_GE	Georgian
46	tl_XX	Tagalog
47	ur_PK	Urdu
48	is_IS	Icelandic
49	hy_AM	Armenian
50	ml_IN	Malayalam
51	mk_MK	Macedonian
52	be_BY	Belarusian
53	la_VA	Latin
54	te_IN	Telugu
55	eu_ES	Basque
56	gl_ES	Galician
57	mn_MN	Mongolian
58	kn_IN	Kannada
59	ne_NP	Nepali
60	sw_KE	Swahili
61	si_LK	Sinhala
62	mr_IN	Marathi
63	af_ZA	Afrikaans
64	gu_IN	Gujarati
65	cy_GB	Welsh
66	eo_EO	Esperanto
67	km_KH	Central Khmer
68	ky_KG	Kirghiz
69	uz_UZ	Uzbek
70	ps_AF	Pashto
71	pa_IN	Punjabi
72	ga_IE	Irish
73	ha_NG	Hausa
74	am_ET	Amharic
75	lo_LA	Lao
76	ku_TR	Kurdish
77	so_SO	Somali
78	my_MM	Burmese
79	or_IN	Oriya
80	sa_IN	Sanskrit