I agree. Even bi-lingual (English+1) small models would be very useful to process localized data, for ex english-french, english-german, etc. Right now the sma

I agree. Even bi-lingual (English+1) small models would be very useful to process localized data, for ex english-french, english-german, etc. Right now the small models (llama 8B) can't handle this type of task, although they could if they were trained for bi lingual data.