0.1.7

add ability to clip the flow during sampling, seems to help much bett…

0.1.6

unsure about clipping, default to off for now

0.1.5

less aggressive clipping

0.1.4

for predict noise objective, clipping still helps a lot

0.1.1

smaller epsilon when deriving flow from predict noise objective

0.1.0

make sure one can do predict epsilon (noise) objective

0.0.29

able to use one nested dictionary to instantiate everything

0.0.28

log the loss breakdown due to consistency fm

0.0.27

prevent reflow training if using consistency fm

0.0.26

add some basic log