Tag
EMMI proposes a method for communication-efficient multimodal large language model inference on edge devices by compressing representations, reducing communication payload by 32× while maintaining accuracy.