Support for llamaindex/vdr-2b-multi-v1: Multilingual Visual Document Retrieval Model

vdr-2b-multi-v1 is a cutting-edge multilingual embedding model designed for visual document retrieval across various languages and domains. The model encodes document page screenshots into dense single-vector representations, allowing efficient search and querying of visually rich multilingual documents without OCR or data extraction pipelines. https://huggingface.co/llamaindex/vdr-2b-multi-v1 https://huggingface.co/blog/vdr-2b-multilingual Highlights: - Multilingual Training: Trained on Italian, Spanish, English, French, and German, forming a dataset of 500k high-quality samples. - Low VRAM and Faster Inference: 3x faster inference with only 30% of the image tokens used by its base model. - Cross-Lingual Retrieval: Search German documents using Italian queries with superior accuracy. - Matryoshka Representation Learning (MRL): Enables dimensional reduction while maintaining embedding quality, optimizing both retrieval speed and storage. Why Include This Model? - Multilingual Applications: Especially beneficial for regions like Europe, where multilingual documents are prevalent. - Performance and Efficiency: Outperforms previous benchmarks in terms of speed, memory efficiency, and retrieval accuracy. - Open Source Contributions: Accompanied by the largest open-source multilingual dataset for visual document retrieval (vdr-multilingual-train).