Yuchen Lin comments

Repositories
Issues
Comments

Results 1 comments of


                                            Yuchen Lin

The density_for_timestep_sampling and loss_weighting for SD3 Training！！！

> Thanks a lot. And for my question3: "when we use logit_normal, it based on the RF-setting. So the weight of the loss should be t/(1-t), but the code doesn't...