What is guidance-distilled?
#17
by
YangHu
- opened
any reference paper or blog?
I know it's very late, but https://arxiv.org/abs/2210.03142 adresses this.
Short answer is that the model was trained to replicate the output of the base model with classifier-free-guidance in a single forward pass (instead of two passes for CFG). The guidance scale is passed as a parameter to the model.