Question about continue pretrain on Florence large base model with more data

#30
by lucasjin - opened

Hello, wanna consult several questions, florence2 is amazing, however, it limited in identify some foregin images like Chinese OCR etc.

So, I want to ask:

  1. Is the ok, to continue pretrain with Chinese caption (web, noisy) data only without anyother tasks, will it forgotten all knowledge or effect much?
  2. What lr rate it should be and what steps is necessary to do this?

Sign up or log in to comment