Convergence issues during training Phase One
apple-love opened this issue · comments
Hello, I encountered convergence issues during training Phase One. I have looked at the details in readme, but despite several attempts, I still haven't achieved convergence. Should I modify
https://github.com/CRIPAC-DIG/tgm-dlm/blob/main/improved-diffusion/scripts/train.py:177
by changing the seed number, and then continue trying?
After lowering the learning rate, the convergence problem is solved. (Code Updated)