v0.4.1

move to film-like conditioning for time

v0.4.0

bring in dynamic thresholding, from Imagen paper, which purportedly i…

…mproves classifier free guidance

0.3.19

higher default lr

0.3.18

take care of normalization and inverse normalization of image for ddp…

…m, so researcher does not have to worry about it

0.3.17

for numerical accuracy

0.3.16

allow for mask to be passed in for directly specifying which batch el…

…ement needs to focus on the present

0.3.15

attempt to fix long description problem

0.3.14

fix focus-present masking, thanks to @CiaoHe yet again!

0.3.12

allow for non-strict load_state_dict

0.3.11

initial temporal attention should be prenorm residual