DiT-MoE: A New Version of the DiT Architecture for Image Generation – MarkTechPost
Moreover, techniques like classifier-free guidance are applied during training, and a pre-trained variational autoencoder model from Stable Diffusion ...
