MM-Vet v2: A challenging benchmark for evaluating large multimodal models (LMMs) for integrated capabilities 08/09/2024