0.1.29

zero grad before, always detach fakes during discriminator training

0.1.28

0.1.28

0.1.27

start adopting associative scan based circuits for time

0.1.26

just do all constant zero padding

0.1.25

cleanup

0.1.24

also handle output conv separately for first frame, depending on whet…

…her separate_first_frame_encoding is set to True

0.1.23

0.1.23

0.1.22

autocast context for accelerator

0.1.21

remove a gotcha for ddp training

0.1.20

allow for encoding the first frame separate from the rest, like in cv…

…ivit