|
|
from .base import * |
|
|
import ctranslate2, sentencepiece as spm |
|
|
import transformers |
|
|
|
|
|
CT_MODEL_PATH = 'data/models/m2m100-1.2B-ctranslate2' |
|
|
|
|
|
@register_translator('m2m100') |
|
|
class M2M100Translator(BaseTranslator): |
|
|
|
|
|
concate_text = False |
|
|
params: Dict = { |
|
|
'device': DEVICE_SELECTOR() |
|
|
} |
|
|
|
|
|
def _setup_translator(self): |
|
|
self.lang_map['Afrikaans'] = 'af' |
|
|
self.lang_map['Albanian'] = 'sq' |
|
|
self.lang_map['Amharic'] = 'am' |
|
|
self.lang_map['Arabic'] = 'ar' |
|
|
self.lang_map['Armenian'] = 'hy' |
|
|
self.lang_map['Asturian'] = 'ast' |
|
|
self.lang_map['Azerbaijani'] = 'az' |
|
|
self.lang_map['Bashkir'] = 'ba' |
|
|
self.lang_map['Belarusian'] = 'be' |
|
|
self.lang_map['Bengali'] = 'bn' |
|
|
self.lang_map['Bosnian'] = 'bs' |
|
|
self.lang_map['Breton'] = 'br' |
|
|
self.lang_map['Bulgarian'] = 'bg' |
|
|
self.lang_map['Burmese'] = 'my' |
|
|
self.lang_map['Catalan'] = 'ca' |
|
|
self.lang_map['Cebuano'] = 'ceb' |
|
|
self.lang_map['Central Khmer'] = 'km' |
|
|
self.lang_map['Chinese'] = 'zh' |
|
|
self.lang_map['Croatian'] = 'hr' |
|
|
self.lang_map['Czech'] = 'cs' |
|
|
self.lang_map['Danish'] = 'da' |
|
|
self.lang_map['Dutch'] = 'nl' |
|
|
self.lang_map['English'] = 'en' |
|
|
self.lang_map['Estonian'] = 'et' |
|
|
self.lang_map['Finnish'] = 'fi' |
|
|
self.lang_map['French'] = 'fr' |
|
|
self.lang_map['Fulah'] = 'ff' |
|
|
self.lang_map['Gaelic'] = 'gd' |
|
|
self.lang_map['Galician'] = 'gl' |
|
|
self.lang_map['Ganda'] = 'lg' |
|
|
self.lang_map['Georgian'] = 'ka' |
|
|
self.lang_map['German'] = 'de' |
|
|
self.lang_map['Greeek'] = 'el' |
|
|
self.lang_map['Gujarati'] = 'gu' |
|
|
self.lang_map['Haitian'] = 'ht' |
|
|
self.lang_map['Hausa'] = 'ha' |
|
|
self.lang_map['Hebrew'] = 'he' |
|
|
self.lang_map['Hindi'] = 'hi' |
|
|
self.lang_map['Hungarian'] = 'hu' |
|
|
self.lang_map['Icelandic'] = 'is' |
|
|
self.lang_map['Igbo'] = 'ig' |
|
|
self.lang_map['Iloko'] = 'ilo' |
|
|
self.lang_map['Indonesian'] = 'id' |
|
|
self.lang_map['Irish'] = 'ga' |
|
|
self.lang_map['Italian'] = 'it' |
|
|
self.lang_map['Japanese'] = 'ja' |
|
|
self.lang_map['Javanese'] = 'jv' |
|
|
self.lang_map['Kannada'] = 'kn' |
|
|
self.lang_map['Kazakh'] = 'kk' |
|
|
self.lang_map['Korean'] = 'ko' |
|
|
self.lang_map['Lao'] = 'lo' |
|
|
self.lang_map['Latvian'] = 'lv' |
|
|
self.lang_map['Lingala'] = 'ln' |
|
|
self.lang_map['Lithuanian'] = 'lt' |
|
|
self.lang_map['Luxembourgish'] = 'lb' |
|
|
self.lang_map['Macedonian'] = 'mk' |
|
|
self.lang_map['Malagasy'] = 'mg' |
|
|
self.lang_map['Malay'] = 'ms' |
|
|
self.lang_map['Malayalam'] = 'ml' |
|
|
self.lang_map['Marathi'] = 'mr' |
|
|
self.lang_map['Mongolian'] = 'mn' |
|
|
self.lang_map['Nepali'] = 'ne' |
|
|
self.lang_map['Northern Sotho'] = 'ns' |
|
|
self.lang_map['Norwegian'] = 'no' |
|
|
self.lang_map['Occitan (post 1500)'] = 'oc' |
|
|
self.lang_map['Oriya'] = 'or' |
|
|
self.lang_map['Panjabi'] = 'pa' |
|
|
self.lang_map['Persian'] = 'fa' |
|
|
self.lang_map['Polish'] = 'pl' |
|
|
self.lang_map['Portuguese'] = 'pt' |
|
|
self.lang_map['Pushto'] = 'ps' |
|
|
self.lang_map['Romanian'] = 'ro' |
|
|
self.lang_map['Russian'] = 'ru' |
|
|
self.lang_map['Serbian'] = 'sr' |
|
|
self.lang_map['Sindhi'] = 'sd' |
|
|
self.lang_map['Sinhala'] = 'si' |
|
|
self.lang_map['Slovak'] = 'sk' |
|
|
self.lang_map['Slovenian'] = 'sl' |
|
|
self.lang_map['Somali'] = 'so' |
|
|
self.lang_map['Spanish'] = 'es' |
|
|
self.lang_map['Sundanese'] = 'su' |
|
|
self.lang_map['Swahili'] = 'sw' |
|
|
self.lang_map['Swati'] = 'ss' |
|
|
self.lang_map['Swedish'] = 'sv' |
|
|
self.lang_map['Tagalog'] = 'tl' |
|
|
self.lang_map['Tamil'] = 'ta' |
|
|
self.lang_map['Thai'] = 'th' |
|
|
self.lang_map['Tswana'] = 'tn' |
|
|
self.lang_map['Turkish'] = 'tr' |
|
|
self.lang_map['Ukrainian'] = 'uk' |
|
|
self.lang_map['Urdu'] = 'ur' |
|
|
self.lang_map['Uzbek'] = 'uz' |
|
|
self.lang_map['Vietnamese'] = 'vi' |
|
|
self.lang_map['Welsh'] = 'cy' |
|
|
self.lang_map['Western Frisian'] = 'fy' |
|
|
self.lang_map['Wolof'] = 'wo' |
|
|
self.lang_map['Xhosa'] = 'xh' |
|
|
self.lang_map['Yiddish'] = 'yi' |
|
|
self.lang_map['Yoruba'] = 'yo' |
|
|
self.lang_map['Zulu'] = 'zu' |
|
|
|
|
|
self.translator = ctranslate2.Translator(CT_MODEL_PATH, device=self.params['device']['value']) |
|
|
self.tokenizer = transformers.AutoTokenizer.from_pretrained(CT_MODEL_PATH, clean_up_tokenization_spaces=True) |
|
|
|
|
|
def _translate(self, src_list: List[str]) -> List[str]: |
|
|
self.tokenizer.src_lang = self.lang_map[self.lang_source] |
|
|
|
|
|
text = [self.tokenizer.convert_ids_to_tokens(self.tokenizer.encode(i)) for i in src_list] |
|
|
target_prefix = [self.tokenizer.lang_code_to_token[self.lang_map[self.lang_target]]] |
|
|
|
|
|
results = self.translator.translate_batch(text, target_prefix=[target_prefix]*len(src_list)) |
|
|
text_translated = [self.tokenizer.decode(self.tokenizer.convert_tokens_to_ids(i.hypotheses[0][1:])) for i in results] |
|
|
|
|
|
return text_translated |
|
|
|
|
|
def updateParam(self, param_key: str, param_content): |
|
|
super().updateParam(param_key, param_content) |
|
|
if param_key == 'device': |
|
|
if hasattr(self, 'translator'): |
|
|
delattr(self, 'translator') |
|
|
self.translator = ctranslate2.Translator(CT_MODEL_PATH, device=self.params['device']['value']) |
|
|
@property |
|
|
def supported_tgt_list(self) -> List[str]: |
|
|
return ['Afrikaans', 'Amharic', 'Arabic', 'Asturian', 'Azerbaijani', 'Bashkir', 'Belarusian', 'Bulgarian', 'Bengali', 'Breton', 'Bosnian', 'Catalan', 'Cebuano', 'Czech', 'Welsh', 'Danish', 'German', 'Greeek', 'English', 'Spanish', 'Estonian', 'Persian', 'Fulah', 'Finnish', 'French', 'Western Frisian', 'Irish', 'Gaelic', 'Galician', 'Gujarati', 'Hausa', 'Hebrew', 'Hindi', 'Croatian', 'Haitian', 'Hungarian', 'Armenian', 'Indonesian', 'Igbo', 'Iloko', 'Icelandic', 'Italian', 'Japanese', 'Javanese', 'Georgian', 'Kazakh', 'Central Khmer', 'Kannada', 'Korean', 'Luxembourgish', 'Ganda', 'Lingala', 'Lao', 'Lithuanian', 'Latvian', 'Malagasy', 'Macedonian', 'Malayalam', 'Mongolian', 'Marathi', 'Malay', 'Burmese', 'Nepali', 'Dutch', 'Norwegian', 'Northern Sotho', 'Occitan (post 1500)', 'Oriya', 'Panjabi', 'Polish', 'Pushto', 'Portuguese', 'Romanian', 'Russian', 'Sindhi', 'Sinhala', 'Slovak', 'Slovenian', 'Somali', 'Albanian', 'Serbian', 'Swati', 'Sundanese', 'Swedish', 'Swahili', 'Tamil', 'Thai', 'Tagalog', 'Tswana', 'Turkish', 'Ukrainian', 'Urdu', 'Uzbek', 'Vietnamese', 'Wolof', 'Xhosa', 'Yiddish', 'Yoruba', 'Chinese', 'Zulu'] |
|
|
|
|
|
@property |
|
|
def supported_src_list(self) -> List[str]: |
|
|
return ['Afrikaans', 'Amharic', 'Arabic', 'Asturian', 'Azerbaijani', 'Bashkir', 'Belarusian', 'Bulgarian', 'Bengali', 'Breton', 'Bosnian', 'Catalan', 'Cebuano', 'Czech', 'Welsh', 'Danish', 'German', 'Greeek', 'English', 'Spanish', 'Estonian', 'Persian', 'Fulah', 'Finnish', 'French', 'Western Frisian', 'Irish', 'Gaelic', 'Galician', 'Gujarati', 'Hausa', 'Hebrew', 'Hindi', 'Croatian', 'Haitian', 'Hungarian', 'Armenian', 'Indonesian', 'Igbo', 'Iloko', 'Icelandic', 'Italian', 'Japanese', 'Javanese', 'Georgian', 'Kazakh', 'Central Khmer', 'Kannada', 'Korean', 'Luxembourgish', 'Ganda', 'Lingala', 'Lao', 'Lithuanian', 'Latvian', 'Malagasy', 'Macedonian', 'Malayalam', 'Mongolian', 'Marathi', 'Malay', 'Burmese', 'Nepali', 'Dutch', 'Norwegian', 'Northern Sotho', 'Occitan (post 1500)', 'Oriya', 'Panjabi', 'Polish', 'Pushto', 'Portuguese', 'Romanian', 'Russian', 'Sindhi', 'Sinhala', 'Slovak', 'Slovenian', 'Somali', 'Albanian', 'Serbian', 'Swati', 'Sundanese', 'Swedish', 'Swahili', 'Tamil', 'Thai', 'Tagalog', 'Tswana', 'Turkish', 'Ukrainian', 'Urdu', 'Uzbek', 'Vietnamese', 'Wolof', 'Xhosa', 'Yiddish', 'Yoruba', 'Chinese', 'Zulu'] |
