Why is there no Output when IDEFICS based model is run on CUDA?

I am using Huggingface based IDEFICS “idefics-9b” to instruct the model print a caption for the given image. I am using the code as mentioned in the section 1. Which means the model and the processor are on ‘cuda’ device. But When I run this code, the output doesn’t contain anything after “Story” tag. But when I shift the model and processor on cpu, then there is a juicy story after the “Story” tag in the output.

It would be a big help to know what is the problem. Refer to the screenshot below.