Hi everyone! Share some good/bad cases you found playing GENIUS models here ๐Ÿ˜

#2
by beyond - opened

Hi dear friends, here you can share some cases that surprise you or disappoint you. These cases may be very helpful for our future research to further improve our model:)

Hi @beyond , the result looks nice. It would be nicer, if the text generated for a mask is not only a few words, but also a few sentences. Or if we could define how many words for a mask. Thanks

Hi @cahya , yes you are right! The current training mechanism doesn't support defining how many words for a mask. But I do have a trick to let the model generate more words for a given place โ€“ just inserting more mask tokens at that place. It's ugly, but indeed effective, lol.

Sign up or log in to comment