madlad-400-translate / langmap /langid_mapping.py
Daryl Lim
feat: restrict language mapping to 22 Tier 1 production-ready languages
bf4b225
Raw
History Blame
758 Bytes
# Tier 1 production-ready languages: BLEU 35+ in both translation directions
# (en→xx and xx→en) from Section A.10 of the MADLAD-400 paper.
# See: https://arxiv.org/pdf/2309.04662
langid_to_language = {
"<2af>": "Afrikaans",
"<2bg>": "Bulgarian",
"<2bs>": "Bosnian",
"<2ca>": "Catalan",
"<2cs>": "Czech",
"<2da>": "Danish",
"<2de>": "German",
"<2es>": "Spanish",
"<2fil>": "Filipino",
"<2fr>": "French",
"<2hr>": "Croatian",
"<2id>": "Indonesian",
"<2it>": "Italian",
"<2mk>": "Macedonian",
"<2ms>": "Malay",
"<2mt>": "Maltese",
"<2nn>": "Norwegian Nynorsk",
"<2no>": "Norwegian",
"<2pt>": "Portuguese",
"<2ro>": "Romanian",
"<2sk>": "Slovak",
"<2sv>": "Swedish",
}