Nemotron ColEmbed V2: Top-Performing Late Interaction Embedding Models for Visual Document Retrieval

TL;DR

Nemotron ColEmbed V2 introduces top-performing visual document retrieval models leveraging late interaction embeddings, achieving state-of-the-art results on ViDoRe benchmarks with various parameter sizes.

Contribution

The paper presents Nemotron ColEmbed V2, a family of models based on pre-trained VLMs that set new performance standards for visual document retrieval.

Findings

8B model ranks first on ViDoRe V3 leaderboard

Achieves an average NDCG@10 of 63.42

Techniques like cluster-based sampling and hard-negative mining improve performance

Abstract

Retrieval-Augmented Generation (RAG) systems have been popular for generative applications, powering language models by injecting external knowledge. Companies have been trying to leverage their large catalog of documents (e.g. PDFs, presentation slides) in such RAG pipelines, whose first step is the retrieval component. Dense retrieval has been a popular approach, where embedding models are used to generate a dense representation of the user query that is closer to relevant content embeddings. More recently, VLM-based embedding models have become popular for visual document retrieval, as they preserve visual information and simplify the indexing pipeline compared to OCR text extraction. Motivated by the growing demand for visual document retrieval, we introduce Nemotron ColEmbed V2, a family of models that achieve state-of-the-art performance on the ViDoRe benchmarks. We release three variants - with 3B, 4B, and 8B parameters - based on pre-trained VLMs: NVIDIA Eagle 2 with Llama 3.2 3B backbone, Qwen3-VL-4B-Instruct and Qwen3-VL-8B-Instruct, respectively. The 8B model ranks first on the ViDoRe V3 leaderboard as of February 03, 2026, achieving an average NDCG@10 of 63.42. We describe the main techniques used across data processing, training, and post-training - such as cluster-based sampling, hard-negative mining, bidirectional attention, late interaction, and model merging - that helped us build our top-performing models. We also discuss compute and storage engineering challenges posed by the late interaction mechanism and present experiments on how to balance accuracy and storage with lower dimension embeddings.