Abstract

We present the training recipe and results of scaling up PaLI-X, amultilingual vision and language model, both in terms of size of the componentsand the breadth of its training task mixture. Our model achieves new levels ofperformance on a wide-range of varied and complex tasks, including multipleimage-based captioning and question-answering tasks, image-based documentunderstanding and few-shot (in-context) learning, as well as object detection,video question answering, and video captioning. PaLI-X advances thestate-of-the-art on most vision-and-language benchmarks considered (25+ ofthem). Finally, we observe emerging capabilities, such as complex counting andmultilingual object detection, tasks that are not explicitly in the trainingmix.