NVIDIA releases Nemotron 3.5 Embed for chart and diagram search
NVIDIA introduced Nemotron 3.5 Embed, a set of open models designed to search documents containing charts, diagrams, and infographics as images. The release includes three embedding models (8B, 1B BF16, and 1B NVFP4), one reranking model (8B), and a parser (631M parameters) for converting document images into structured text. These models aim to address a common challenge in RAG systems: the…
Key points
- Nemotron 3.5 Embed includes three embedding models (8B, 1B BF16, 1B NVFP4) and one reranking model (8B) for chart/document search
- Models support text, images, or combined inputs, outputting fixed-length vectors (4,096 for 8B, 2,048 for 1B) with adjustable dimensions
- NVIDIA claims backward compatibility with prior-generation models of the same size for text vector reuse
The Embed models use a dense single-vector approach, allowing for easier integration into existing RAG pipelines. NVIDIA claims backward compatibility with previous-generation models of the same size, enabling reuse of text vectors and indices without full rebuilding. Performance benchmarks highlight improvements over prior versions, with the 8B model achieving competitive results on the ViDoRe V3 benchmark. The models are licensed under OpenMDW v1.1 (commercial use allowed) and Apache 2.0 for base components, while inference containers (NIM) follow NVIDIA’s proprietary licensing terms.
Decoding NVIDIA's chart search model 'Nemotron 3.5 Embed' from an AI engineer's perspective
note.com · 31 October 2026
Loading the full article…
This text was published by note.com and written by humble | 20年選手のAIエンジニア | 主に生成AI系の調査やノウハウを投稿. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Research
All →- Researchers release two-stage AI model for neonatal brain MRI analysis · 1 src
- DeepDelivery identifies human proteins for mRNA delivery · 1 src
- Anthropic opens biology lab with AI agents to find uncharacterized enzyme system · 3 src
- Author tests six update strategies on local AI model · 1 src
- NTU Singapore and Google expand AI partnership for education and research · 1 src
Comments
via GitHub Discussions