0.0.32

0.0.32

0.0.31

using negative euclidean distances is far better than previous dot pr…

0.0.30

add ability for dot product attention to use cdist for degrees > 0

0.0.29

the way it was before is better

0.0.27

switch to depthwise tensor product (DTP) as in paper

0.0.26a

residualize a section of the mlp within the radial function

0.0.25

tweak

0.0.24

cleanup

0.0.23

follow the way unet is conditioned with the relative distances to ten…

0.0.22

move self interaction logic into TP