0.1.6

allow for generating arbitrary number of video frames after training

0.1.5

allow for generating arbitrary number of video frames after training

0.1.4

add geglu and talking head improvements, two of Shazeers papers

0.1.2

add ability to do any custom dimensions of the kernel size for sparse…

0.1.1

fix bug when chunking queries for sparse 3d nearby attention

0.1.0

add reversible networks from Reformer, achieve depth far beyond what …

0.0.26

better error message, and always use sandwich norm

0.0.25

fix bug where text embedding during generation is not masked properly

0.0.24

add token shifting module for video along spatial dimensions, prepare…

0.0.23

add feedforward chunking