0.3.0

support keeping the EMA model on cpu, autoconverting device

0.2.6

use beartype for validation of types on init, use torch clamp

0.2.5

offer a method for copying params and buffers from EMA back to online…

0.2.4

only eval the ema model, but not the online model

0.2.3

0.2.3

0.2.2

catch a bug with updating buffers, thanks to @sbl1996

0.2.1

start using inplace lerp

0.2.0

address https://github.com/lucidrains/ema-pytorch/issues/7

0.1.4

add ability to ignore any parameters whose name starts with some prefix

0.1.2

cleanup