0.2.0

add an autoregressive loss for the hierarchical tokens, following wha…

0.1.2

fix prophet loss using embeds

0.1.1

it was working better the way before

0.1.0

use built-in cross entropy loss to distance matrix for the random pro…

0.0.28

some cleanup, better naming, comments

0.0.27

revert back to no prophet loss on hierarchy of 1, and also allow rand…

0.0.26

only allow random proj quantized predictions for prophet loss in hier…

0.0.25

allow for prediction of random quantized ids even for hierarchy of 1x…

0.0.24

just make it work for now, clean up later

0.0.23

able to turn off attention in hierarchies by setting window_size to be 0