Implementation of the "the first large-scale multimodal mixture of experts models." from the paper: "Multimodal Contrastive Learning with LIMoE: the Language-Image Mixture of Experts"