How to determine the value of fake_class_label in models_mage.py
xcyuyuyu opened this issue · comments
xcyuyuyu commented
Tianhong Li commented
It can be of any value other than the tokenzier codebook indices range (0-1023).
xcyuyuyu commented
Thanks. BTW, when I do the pretrain with 8 V100 with 512batchsize, why does the loss reduce to 2.6 and no longer reduce, is it really needs a large batchsize for training?
I trained on the custome dataset with 1.4w images
Tianhong Li commented
A loss of 2.6 is already quite low if you use the default masking strategy: in our experiment, the training loss converges around 5.7-5.8.