Open moluchase opened 3 years ago
if i have <other language text , image, label> pair data, can i directly use 'distilbert-base-multilingual-cased' to pre-train clip model? Why re-train a model for englist to other languages ?
I guess because the image-text matching task is done on English corpus only.
if i have <other language text , image, label> pair data, can i directly use 'distilbert-base-multilingual-cased' to pre-train clip model? Why re-train a model for englist to other languages ?