What is the role of 'unconditional_guidance_vex' in 'rdm768x768/model.ckpt'

[jwliu-cc]

Hello, I found that there is an unexpected parameter 'unconditional_guidance_vex' with the dimension of 768, Could you please explain its purpose？

if i want to finetune RDM on my own dataset, is it possible to just resume from 'rdm768x768/model.ckpt'? I m not sure about this cause i found the unexpected parameter and no optimizer_states or lr_schedulers in the .ckpt file.