PLUME: Latent Reasoning Based Universal Multimodal Embedding

2026-04-02

AI Summary

This work addresses the high computational cost and information bottlenecks associated with explicit chain-of-thought (CoT) reasoning in general-purpose multimodal embeddings. To overcome these limitations, the authors propose PLUME, a framework that replaces explicit CoT with a short autoregressive expansion of continuous latent states, enabling efficient reasoning within a fixed computational budget. PLUME incorporates semantic-anchor-guided transition adapters and a progressive training curriculum that shifts from explicit to implicit reasoning, thereby establishing a structured implicit inference mechanism. Experimental results demonstrate that PLUME outperforms existing explicit CoT methods on the MMEB-v2 benchmark, compressing reasoning steps from hundreds of tokens to fewer than ten and achieving over a 30-fold speedup.

Abstract

Universal multimodal embedding (UME) maps heterogeneous inputs into a shared retrieval space with a single model. Recent approaches improve UME by generating explicit chain-of-thought (CoT) rationales before extracting embeddings, enabling multimodal large language models to better infer complex query intent. However, explicit CoT incurs substantial inference overhead and can compress rich multimodal evidence into a narrow textual bottleneck. We propose PLUME, a latent reasoning framework that advances UME by replacing verbalized CoT with a short autoregressive rollout of continuous latent states. To support diverse multimodal queries, PLUME further introduces a semantic-anchor-guided transition adapter that steers latent rollout along different reasoning trajectories under the same fixed computation budget. To stabilize training, PLUME adopts a progressive explicit-to-latent curriculum that uses verbalized reasoning only as a temporary training scaffold and gradually transfers this behavior into hidden-state computation, eliminating explicit CoT at inference. On the 78-task MMEB-v2 benchmark, PLUME outperforms strong explicit-CoT UME baselines while reducing reasoning from hundreds of generated tokens to fewer than 10 latent steps, delivering over 30x faster inference. PLUME is especially well suited to retrieval settings where relevant evidence is dense, structurally complex, and difficult to organize through verbalized intermediate rationales, such as video and visual document retrieval. These results show that structured latent computation can preserve the benefits of intermediate reasoning without the overhead of explicit rationale generation, providing a stronger and more efficient paradigm for practical retrieval systems.

Problem

• universal multimodal embedding
• chain-of-thought reasoning
• multimodal retrieval
• inference overhead
• latent reasoning

Innovation

• latent reasoning
• universal multimodal embedding
• chain-of-thought
• autoregressive rollout
• semantic-anchor-guided adapter