0.0.60

biasless layernorm, also chip away at rectangular image size support

0.0.59

use feature map size to denote which layer to add self attention in d…

0.0.58

self attention in the stylegan-esque discriminator

0.0.57

some basic dropouts

0.0.56

add position generation at each layer of the transformer for cvivit, …

0.0.55

just duplicate phenaki trainer for token critic training code for now

0.0.54

add a depthwise conv for position generation at beginning of encoder …

0.0.53

if dataset fields is being passed in, validate there are no duplicate…

0.0.52

automatically infer what to pass to Phenaki from the Dataset, given t…

0.0.51

allow for passing in captions when evaluating as a file path to a tex…