There is work done for Vision RWKV, and audio RWKV, an example paper is here: https://arxiv.org/abs/2403.02308 https://arxiv.org/abs/2403.02308 Its the same pr

There is work done for Vision RWKV, and audio RWKV, an example paper is here: https://arxiv.org/abs/2403.02308 https://arxiv.org/abs/2403.02308 Its the same principle as open transformer models where an adapter is used to generate the embedding However currently the core team focus is in scaling the core text model, as this would be the key performance driver, before adapting multi-modal. The tech is there, the base model needs to be better