search

pluiez / NLLB-inference

56 stars 6 forks source link

undefined mapping: als_Latn #3

Closed fatjoni closed 2 years ago

fatjoni commented 2 years ago

I am experiencing a problem when trying to translate from albania to english bash translate.sh /content/checkpoint.pt als_Latn eng_Latn <<< 'Ku eshte qeni o doktor?'

I receive undefined mapping: als_Latn This happens on my machine with this model as well https://tinyurl.com/nllb200dense3bcheckpoint and only when translating from albanian language to another language, not vice versa. Any idea?

pluiez commented 2 years ago

Hi, this is indeed missing in their configuration file, which is defined in stopes/utils/map_token_lang.tsv. In addition to als_Latn, I also found some other language codes missing. I will open an issue in their repo to confirm about this problem.

pluiez commented 2 years ago

Hi, the missing languages will use English by default. For details please check the related issue on stopes.

fatjoni commented 2 years ago

Thanks a lot, just tested it and it finally works :')