Multimodal-Mind2Web is the multimodal version of Mind2Web, a dataset for developing and evaluating generalist agents for the web that can follow language instructions to complete complex tasks on any website. In this dataset, we align each HTML document in the dataset with its corresponding webpage screenshot image from the Mind2Web raw dump. This multimodal version addresses the inconvenience of loading images from the ~300GB Mind2Web Raw Dump.

Dataset… See the full description on the dataset page: https://huggingface.co/datasets/osunlp/Multimodal-Mind2Web.