Pushing parts of a model to different devices

My model is too big to fit in a single GPU. How do I push parts of my model to different GPUs?