imagen-pytorch

fix ddp and self conditioning in elucidated ddpm

make sure self conditining works in distributed training, cleanup later, bump

Fixing DDP Unet Access when Training in a MultiGPU Setting (#206)

fix a bug with inpainting for non-elucidated ddpm

allow for one to pass in a tuple to use_linear_attn kwargs in unet, to finely specify which layers get sparse linear attention, and to turn off attention altogether