1.12.3

always rederive the predicted noise from the clipped x0 for ddim + pr…

1.12.2

bump to newer package of clip-anytorch that allows for text encodings…

1.12.1

fix for self conditioning in diffusion prior network https://github.c…

1.12.0

default ddim sampling eta to 0

1.11.4

extra insurance in case eos id is not there

1.11.2

address https://github.com/lucidrains/DALLE2-pytorch/issues/266

1.11.1

bring in prediction of v objective, combining the findings from progr…

1.11.0

bring in prediction of v objective, combining the findings from progr…

1.10.9

fix openclipadapter to be able to use latest open sourced sota model

0.10.8

fix a dtype conversion issue for the diffusion timesteps in the diffu…