0.3.12

fix, cleanup, make more efficient generation code

0.3.10

fix bugs

0.3.9

fix bug with mask padding

0.3.8

fix an issue with gradient accumulation and memory

0.3.7

remove autopadding, and fix bugs

0.3.6

fix priming and update example

0.3.5

update examples to use gradient accumulation

0.3.4

fix auto-segmenting mask as well

0.3.3

multiply by proper scale when doing full attention for compressed mem…

0.3.2

add a bunch of protective asserts