Why use the noise_level instead of timestep t to condition the model

[xuelin-chen]

Hi, thanks for this nice work! But, in the ConditionalDiffusion module, could u help explain the choice of having the noise level, instead of the raw timestep t, as the conditional input to the network?