ColPali is an efficient document retrieval model that leverages Vision Language Models (VLMs) to generate high-quality contextualized embeddings from document images. It outperforms existing document retrieval systems while being faster and end-to-end trainable. The paper introduces ViDoRe, a benchmark for visually rich document retrieval, which includes various page-level tasks across multiple domains, languages, and settings. ViDoRe evaluates systems on their ability to match queries to relevant documents, considering both textual and visual elements. The benchmark highlights the shortcomings of current text-centric systems in visually rich document retrieval settings. ColPali, based on VLMs, simplifies document retrieval while achieving stronger performance with better latencies. It uses a late interaction matching mechanism to enhance retrieval efficiency. The model is trained on a dataset of 127,460 query-page pairs, including academic and synthetic data. ColPali outperforms other retrieval systems on ViDoRe, demonstrating strong performance on tasks involving figures, tables, and infographics. It also shows improved performance on text-centric documents across various languages. ColPali's efficiency and effectiveness make it a promising solution for industrial document retrieval applications. The paper also discusses the limitations of current systems, including the need for efficient preprocessing and the challenges of handling non-English languages. The model's performance is evaluated using metrics such as NDCG@5, and it achieves significant improvements over existing methods. The paper concludes that ColPali has high potential for industrial document retrieval applications and encourages further research in this area.ColPali is an efficient document retrieval model that leverages Vision Language Models (VLMs) to generate high-quality contextualized embeddings from document images. It outperforms existing document retrieval systems while being faster and end-to-end trainable. The paper introduces ViDoRe, a benchmark for visually rich document retrieval, which includes various page-level tasks across multiple domains, languages, and settings. ViDoRe evaluates systems on their ability to match queries to relevant documents, considering both textual and visual elements. The benchmark highlights the shortcomings of current text-centric systems in visually rich document retrieval settings. ColPali, based on VLMs, simplifies document retrieval while achieving stronger performance with better latencies. It uses a late interaction matching mechanism to enhance retrieval efficiency. The model is trained on a dataset of 127,460 query-page pairs, including academic and synthetic data. ColPali outperforms other retrieval systems on ViDoRe, demonstrating strong performance on tasks involving figures, tables, and infographics. It also shows improved performance on text-centric documents across various languages. ColPali's efficiency and effectiveness make it a promising solution for industrial document retrieval applications. The paper also discusses the limitations of current systems, including the need for efficient preprocessing and the challenges of handling non-English languages. The model's performance is evaluated using metrics such as NDCG@5, and it achieves significant improvements over existing methods. The paper concludes that ColPali has high potential for industrial document retrieval applications and encourages further research in this area.