Open L-Hugh opened 1 year ago
Have you been able to reproduce stage 1 and stage 2 of BLIP-2 ?
Thanks for your questions.
Thanks @LiJunnan1992 for providing the details !
One more question I have is that in the paper you mention that the model is trained for 250k steps in stage 1 and 80k steps in stage2.
In order to properly configure the learning rate scheduler, we also need the values for the following config parameters which are not discussed in the paper:
max_epoch iters_per_epoch
Thanks
@Swetha5 Is there any update on max_epoch and iters_per_epoch?
Hello, I'm currently working on replicating BLIP-2 and I have some questions regarding the learning rate settings in the two pretraining stages.
I would appreciate your assistance in understanding these settings. Thank you for your time and effort in developing this great project!