Visual Document Retrieval benchmark v3 evaluating ability to retrieve relevant document pages given queries, across 10 datasets, 26,000+ pages, 3,000+ queries in 6 languages. Tests the document retrieval step critical for RAG pipelines.
Only benchmark testing document retrieval for RAG; multilingual (6 languages); enterprise-relevant retrieval scenarios; actively maintained HuggingFace leaderboard.
Tests retrieval models (ColPali, ColQwen) not generative LLMs directly; NDCG@5 metric is retrieval-specific.
Higher is better. NDCG@5 (Normalized Discounted Cumulative Gain at 5) averaged across 10 datasets, 0-100.
Max score: 100
Tests retrieval models (ColPali, ColQwen) not generative LLMs directly.
No scores available for this benchmark yet.