Releases: lucidrains/make-a-video-pytorch
Releases · lucidrains/make-a-video-pytorch
0.4.0
0.3.1
ability to do causal attention for time axis
0.3.0
add flash attention
0.2.1
feedforward should be agnostic to images or video
0.2.0
fix layer norm in feedforward
0.1.2
token shift within feedforward best accompanied by a layernorm, from … …normformer paper
0.1.1
quick fix
0.1.0
follow spatiotemporal attention with a feedforward, and add the highl… …y effective token shift along the time axis in the hidden layer
0.0.10
make continuous positional bias calculations more efficient, while ge… …neralizing to any number of dimensions
0.0.9
allow for customizing number of resnet blocks per stage in space time… … unet