array/

SAT-v2

	SAT-v2 Dataset




	
	Paper

SAT: Dynamic Spatial Aptitude Training for Multimodal Language Models This dataset is part of the SAT (Spatial Aptitude Training) project, which introduces a dynamic benchmark for evaluating and improving spatial reasoning capabilities in multimodal language models.

Project Page: https://arijitray.com/SAT/ Paper: arXiv:2412.07755

SAT-v2 is a comprehensive spatial reasoning benchmark containing over 300,000… See the full description on the dataset page: https://huggingface.co/datasets/array/SAT-v2.