Problems with setting parameters Cskip and Cout

[wudier483]

What are the design principles? Is it enough to just ensure the boundary conditions?

def scalings_for_boundary_conditions(timestep, sigma_data=0.5, timestep_scaling=10.0):
c_skip = sigma_data2 / ((timestep / 0.1) ** 2 + sigma_data2)
c_out = (timestep / 0.1) / ((timestep / 0.1) ** 2 + sigma_data**2) ** 0.5
return c_skip, c_out

[Junsheng121]

I think a 'model_pred = pred_x_0' is enough rather than 'model_pred = c_skip_start * noisy_model_input + c_out_start * pred_x_0'