Nomic Embed Vision: Expanding the Latent Space

About

This technical report describes the training of nomic-embed-vision, a highly performant, open-code, open-weights image embedding model that shares the same latent space as nomic-embed-text. Together, nomic-embed-vision and nomic-embed-text form the first unified latent space to achieve high performance across vision, language, and multimodal tasks.

Zach Nussbaum, Brandon Duderstadt, Andriy Mulyar · 2024

Related benchmarks

Task	Dataset	Result
Handwriting Retrieval	Handwriting In-Domain Set	Accuracy@159.1	30
Handwriting Retrieval	Handwriting Spanish synthetic disjoint fonts (Out-of-Domain (OOD))	Top-1 Accuracy38.82	30
Multimodal-to-text retrieval	MM-BRIGHT	Acad Score22.6	24
Document Retrieval	DocHaystack 200	Recall@113.76	15
Document Retrieval	DocHaystack-1000	Recall@11.83	15
Document Retrieval	InfoHaystack 100	Recall@134.84	15
Document Retrieval	InfoHaystack-1000	Recall@120.65	15
Visual document retrieval	ViDoRe	ArxivQ Score17.1	12
Document Retrieval	DocHaystack 100	Recall@116.51	7
Document Retrieval	InfoHaystack 200	Recall@130.97	7