TSEmbed: Unlocking Task Scaling in Universal Multimodal Embeddings

TL;DR

TSEmbed introduces a novel framework combining Mixture-of-Experts and Low-Rank Adaptation, along with Expert-Aware Negative Sampling, to improve task scaling and discrimination in universal multimodal embeddings, achieving state-of-the-art results.

Contribution

The paper presents TSEmbed, a new universal multimodal embedding method that explicitly disentangles task conflicts and enhances discriminative power using expert routing and negative sampling.

Findings

Achieves state-of-the-art on MMEB benchmark.

Improves discriminative power with expert-aware negative sampling.

Demonstrates effectiveness on industrial datasets.

Abstract

Despite the exceptional reasoning capabilities of Multimodal Large Language Models (MLLMs), their adaptation into universal embedding models is significantly impeded by task conflict. To address this, we propose TSEmbed, a universal multimodal embedding framework that synergizes Mixture-of-Experts (MoE) with Low-Rank Adaptation (LoRA) to explicitly disentangle conflicting task objectives. Moreover, we introduce Expert-Aware Negative Sampling (EANS), a novel strategy that leverages expert routing distributions as an intrinsic proxy for semantic similarity. By dynamically prioritizing informative hard negatives that share expert activation patterns with the query, EANS effectively sharpens the model's discriminative power and refines embedding boundaries. To ensure training stability, we further devise a two-stage learning paradigm that solidifies expert specialization before optimizing representations via EANS. TSEmbed achieves state-of-the-art performance on both the Massive Multimodal Embedding Benchmark (MMEB) and real-world industrial production datasets, laying a foundation for task-level scaling in universal multimodal embeddings.