MMCOMPOSITION: Revisiting the Compositionality of Pre-trained Vision-Language Models Paper โข 2410.09733 โข Published Oct 13 โข 8