1.2.1

1.2.1

1.2.0

add a norm post-attention layers

1.1.5

fix autopadder for reformer class as well

1.1.4

fix bug with memory key/values

1.1.3

do not assert that dimension is divisible by heads if dim_head is sup…

1.1.2

remove custom unsort logits function

1.1.1

simplify the usage of undo_sort, from @L-Hugh

1.1.0

Add dim_head keyword for fixing dimension of each head

1.0.5

move local attention to separate repository

1.0.4

fix bug with attention recording when local attention heads are present