v0.25.3

move accelerator backward outside of autocast context, also calculate…

v0.25.2

do not noise at the last timestep for ddim

v0.25.1

bring in ddim sampling

v0.25.0

bring in ddim sampling

v0.24.4

optimizer should be saved and loaded

v0.24.3

optimizer should be saved and loaded

v0.24.2

add convert_image_to keyword argument, for forcing images being loade…

v0.24.1

also make sure grad scaler actually exists in the saved pt file

v0.24.0

support multi-gpu training using huggingface accelerate, addressing h…

v0.23.4

tqdm auto instead