Tokenize
ucasyjz opened this issue · comments
HappyYang commented
Is there any difference in tokenizer between 'longclip.tokenize' and original clip tokenize in diffusers, can you give me some guidance, thanks. I change the length from 77 to 248 in original clip tokenizer config, but the output features embedding is different from the 'longclip.tokenize'
Beichen Zhang commented
there's no difference except from the positional embedding. The file ./model/simple_tokenizer.py didn't change. You may refer to ./model/longclip.py for further details.