How to train a customized Cogvideox-fun-v1.1-5b-control model based on CogVideoX-Fun-V1.1-5b-InP weights?

[yiboz2001]

Hi! I am looking to train a customized controlled video generation model using specific control data types (e.g., normal maps or optical flow).
The released pre-trained model has been trained on non-target control types. I aim to leverage CogVideoX-Fun-V1.1-5b-InP model's weights and perform training based on the framework of CogVideoX-Fun-V1.1-5b-Control.
How can I achieve this using the provided training code? Any advice would be greatly appreciated.