Spaces:

cstr

/

translate_datasets

like 0

Sleeping

This translates given english dataset parquet files of typical structures used in model training. It applies the lightweight but well performing wmt21 model per ctranslate2. Target languages can be Hausa (ha), Icelandic (is), Japanese (ja), Czech (cs), Russian (ru), Chinese (zh), or German (de). Other dataset structures could easily be supported too. Without GPU the translation is very very slooow though.