The timesteps in inference_wan is fixed

[plorrrrrrr]

I found that in fastvideo/sample/inference_wan.py, the actual inference step is fixed and each timestep is also fixed.
Changing the value of num_inference_steps in the scripts is useless.

So how to modify the cus_timesteps If I want to sample 8 steps,

[cszy98]

During inference, we find that evenly splitting the total steps between the two experts works well. For example, with 8 or 4 steps, each expert gets 4 or 2 steps. These are uniformly sampled within each sub-trajectory. You can manually set the total steps (e.g., 6 or 8) as needed.