Question about DMD regression loss

[zhangchi2004]

May I ask why do you omit the regression loss in DMD distillation?

[Turlan]

I also have a similar problem here. Without a regression loss in DMD1 or a GAN loss in DMD2, could the distillation training become unstable?