About the design of core code

[lifeofstudy]

Thank you for your excellent work. I have a question after reading. The paper does not explain the hyperparameter n in detail. Could you please explain why n is applied when hidden_states = 1280, but half of the channels are modulated directly when hidden_states = 640?
https://github.com/FreeStyleFreeLunch/FreeStyle/blame/main/diffusers/src/diffusers/models/unet_2d_blocks.py#L2192
https://github.com/FreeStyleFreeLunch/FreeStyle/blame/main/diffusers/src/diffusers/models/unet_2d_blocks.py#L2209
And is there any difference between this modulation using ((self.b - 1 ) * hidden_mean + 1) and directly using self.b?

[CharlesGong12]

@ucasligang I wonder too