Fetching the paper…

M3P: Learning Universal Representations via Multitask Multilingual Multimodal Pre-training · Around