Questions about the input and output of the Diffusion Decoder Layer

[Hamusukesama]

Hi. Thank you for your brilliant work and look forward to the code!
I have a few questions and hope to get answers:

1. What is the output of the diffusion layer? Suppose the input is noisy_traj (corresponding to anchor + noise of t steps, 0<t<50)
   a. noise of one step (which is the noise added to anchor and is not related to gt)
   b. offset from noisy_traj to gt
   c. something else
2. How many diffusion timesteps are involved in training and inference, respectively?
   a. one for training (such as DDPM, t -> t-1 ), and two steps for inference (DDIM, 50 -> 25 -> 0)
   b. something else
   Thank you again!