Fine-tuning CogVideo5B-I2V

Hi, I have been trying to fine-tune CogVideo5B-I2V both using LoRA weights and fully but independently of the number of epochs I train for, when I run the new checkpoints on inference the videos generated are always absolutely identical. Is it possible the model is not really training fully or could I be missing something?