Finetuning on custom video datasets

[yankeee-0411]

Hi, thanks for the great work!

I’d like to fine-tune the released RollingForcingModel checkpoint on my own video dataset.

Before trying, I want to confirm:

If I set config.trainer = "diffusion" and use

trainer = DiffusionTrainer(config)

will this work correctly with a RollingForcingModel, or is a different trainer required?

Thanks!