Hi, thank you for your paper and sharing the code. May I ask some questions?
-
In the code below
Is it possible for r to be exactly 0? it seems it is approaching 0, but not exactly 0. If it cannot get value 0, how the consistency model knows the ground truth x_0? I am not sure if there is some misunderstanding here.
-
In the original consistency models paper, the minimum t is 0.002 and in this paper it can be 0. Would you like to expand more on the difference here? Thanks.
Hi, thank you for your paper and sharing the code. May I ask some questions?
In the code below
ect/training/loss.py
Line 56 in cb35360
Is it possible for r to be exactly 0? it seems it is approaching 0, but not exactly 0. If it cannot get value 0, how the consistency model knows the ground truth x_0? I am not sure if there is some misunderstanding here.
In the original consistency models paper, the minimum t is 0.002 and in this paper it can be 0. Would you like to expand more on the difference here? Thanks.