Releases: lucidrains/CoLT5-attention
Releases · lucidrains/CoLT5-attention
0.3.3
fix for routed cross attention
0.3.2
make sure one can also route multiple kv heads for the autoregressive…
0.3.1
improvise an autoregressive version of the routing attention, by doin…
0.3.0
improvise an autoregressive version of the routing attention, by doin…
0.2.5
some refactoring
0.2.4
give some more gradients to the routers
0.2.3
num routed key / value sets feature can also work for self attention
0.2.2
yet another step closer to better routing of memories to queries
0.2.1
make sure the routed cross attention can handle either or both cases …
0.2.0
add a variant of conditionally routed attention for cross attention, …