DirectML was released with support for PyTorch 2.0 which allows for SDP attention and clearer device naming. I've also removed a few patches that are no longer needed and reverted a change in image_to_image.py since the decode_latents function was reverted to returning an ndarray within diffusers.
DirectML was released with support for PyTorch 2.0 which allows for SDP attention and clearer device naming. I've also removed a few patches that are no longer needed and reverted a change in
image_to_image.pysince thedecode_latentsfunction was reverted to returning an ndarray within diffusers.