Document question-answering benchmark where systems answer natural-language questions grounded in document images.
Anls is the reported evaluation metric for DocVQA. Codesota tracks published model scores on this metric so readers can compare state-of-the-art results across sources and model families.
Higher is better