UEmbed: Unified Sparse and Dense Multimodal Embeddings

2026-08-03

AI Summary

This work proposes UEmbed, a decoder-only multimodal embedding model that unifies dense semantic and sparse lexical representations within a single causal forward pass, overcoming limitations of existing learning-based sparse retrieval methods constrained by bi-encoder architectures and reliance on auxiliary cross-modal modules. UEmbed is the first to integrate both embedding types within a sole decoder framework, leveraging learnable special tokens, vocabulary subset partitioning, and causal hidden-state prediction to generate sparse weights without additional components. Evaluated on MMEB-v2, UEmbed-9B achieves 71.8 (dense) and 71.0 (sparse), substantially outperforming current public multimodal embedding models, while also demonstrating competitive performance on the BEIR benchmark.

Abstract

Sparse retrieval underpins modern search systems, from web search to retrieval-augmented generation. Existing work has introduced Learned Sparse Retrieval (LSR) to push beyond exact lexical matching toward richer semantics. Yet LSR has so far remained tied to encoder-style bidirectional architectures, and its extension to multimodal settings still relies heavily on auxiliary cross-modal modules. To address these limitations, we introduce UEmbed (Unified Embedding), a decoder-only multimodal embedding model that produces both sparse lexical and dense representations in one causal forward pass. UEmbed appends N learnable special tokens to the input and partitions the vocabulary into N disjoint subsets. Each token's causal hidden state predicts sparse weights over its assigned subset, and the N subsets are concatenated into the full sparse vector. Trained on public data, we release UEmbed at 2B, 4B, and 9B scales. UEmbed-9B reaches 71.8 (dense) and 71.0 (sparse) on MMEB-v2, outperforming multimodal embedding models trained on publicly available data (e.g., RzenEmbed). On BEIR, UEmbed also remains competitive with strong dense and sparse baselines. Furthermore, we demonstrate the practical utility of UEmbed across three dimensions: effectiveness, efficiency, and agentic applications. Overall, UEmbed offers a new paradigm: it unifies dense and sparse embeddings in one model, while further extending sparse retrieval to unify text and multimodal inputs.

Problem

• sparse retrieval
• multimodal embeddings
• learned sparse retrieval
• unified embedding
• decoder-only model

Innovation

• Unified Embedding
• Sparse Retrieval
• Multimodal Embeddings
• Decoder-only Architecture
• Learned Sparse Retrieval