0.0.16

oops

0.0.15

fix a bug with residual quantize dropout, and also figure out a way t…

0.0.14

fix coarse cross entropy loss weights

0.0.12

complete first pass at unique consecutive issue with semantic token i…

0.0.11

correct weighting of cross entropy losses

0.0.10

follow researcher @eonglints advice and add unique consecutive for se…

0.0.9

add classifier free guidance training logic, cite

0.0.8

add cross attention layers as well as setup t5 and some conditioning …

0.0.7

go for single-headed key / values for all decoding attention networks…

0.0.6

listen to @eonglints and add hubert with kmeans as an option