Task Routing for Multimodal LLMs

pick different expert encoders based on input (OCR, classification, etc)