Question about train code

[terry2100]

Thanks for your work!
In the code for image2image_loss, why set vit_embeds.requires_grad = True?

[BecauseImBatman0]

+1 , and why "left padding for editing" in collate_func_img2img() ?