Saved in:
Bibliographic Details
Main Authors: Sun, Siyu, Ren, Jing, Liao, Zhaohe, Mao, Dongxiao, Ren, Xiangyuan, Zhang, Yiyi, Zhao, Haohua, Lin, Weixiong, Shaohua, Jiang, Zhang, Liqing, Zheng, Yuchao
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2604.11095
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908959720341504
author Sun, Siyu
Ren, Jing
Liao, Zhaohe
Mao, Dongxiao
Ren, Xiangyuan
Zhang, Yiyi
Zhao, Haohua
Lin, Weixiong
Shaohua, Jiang
Zhang, Liqing
Zheng, Yuchao
author_facet Sun, Siyu
Ren, Jing
Liao, Zhaohe
Mao, Dongxiao
Ren, Xiangyuan
Zhang, Yiyi
Zhao, Haohua
Lin, Weixiong
Shaohua, Jiang
Zhang, Liqing
Zheng, Yuchao
contents Adapting decoder-only multimodal large language models (MLLMs) for unified multimodal retrieval faces two structural gaps. First, existing methods rely on implicit pooling, which overloads the hidden state of a standard vocabulary token (e.g., <EOS>) as the sequence-level representation, a mechanism never designed for information aggregation. Second, contrastive fine-tuning specifies what the embedding should match but provides no token-level guidance on how information should be compressed into it. We address both gaps with two complementary components. Architecturally, we introduce Bottleneck Tokens (BToks), a small set of learnable tokens that serve as a fixed-capacity explicit pooling mechanism. For training, we propose Generative Information Condensation: a next-token prediction objective coupled with a Condensation Mask that severs the direct attention path from target tokens to query tokens. All predictive signals are thereby forced through the BToks, converting the generative loss into dense, token-level supervision for semantic compression. At inference time, only the input and BToks are processed in a single forward pass with negligible overhead over conventional last-token pooling. On MMEB-V2 (78 datasets, 3 modalities, 9 meta-tasks), our approach achieves state-of-the-art among 2B-scale methods under comparable data conditions, attaining an Overall score of 59.0 (+3.6 over VLM2Vec-V2) with substantial gains on semantically demanding tasks (e.g., +12.6 on Video-QA).
format Preprint
id arxiv_https___arxiv_org_abs_2604_11095
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Bottleneck Tokens for Unified Multimodal Retrieval
Sun, Siyu
Ren, Jing
Liao, Zhaohe
Mao, Dongxiao
Ren, Xiangyuan
Zhang, Yiyi
Zhao, Haohua
Lin, Weixiong
Shaohua, Jiang
Zhang, Liqing
Zheng, Yuchao
Machine Learning
Artificial Intelligence
Adapting decoder-only multimodal large language models (MLLMs) for unified multimodal retrieval faces two structural gaps. First, existing methods rely on implicit pooling, which overloads the hidden state of a standard vocabulary token (e.g., <EOS>) as the sequence-level representation, a mechanism never designed for information aggregation. Second, contrastive fine-tuning specifies what the embedding should match but provides no token-level guidance on how information should be compressed into it. We address both gaps with two complementary components. Architecturally, we introduce Bottleneck Tokens (BToks), a small set of learnable tokens that serve as a fixed-capacity explicit pooling mechanism. For training, we propose Generative Information Condensation: a next-token prediction objective coupled with a Condensation Mask that severs the direct attention path from target tokens to query tokens. All predictive signals are thereby forced through the BToks, converting the generative loss into dense, token-level supervision for semantic compression. At inference time, only the input and BToks are processed in a single forward pass with negligible overhead over conventional last-token pooling. On MMEB-V2 (78 datasets, 3 modalities, 9 meta-tasks), our approach achieves state-of-the-art among 2B-scale methods under comparable data conditions, attaining an Overall score of 59.0 (+3.6 over VLM2Vec-V2) with substantial gains on semantically demanding tasks (e.g., +12.6 on Video-QA).
title Bottleneck Tokens for Unified Multimodal Retrieval
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2604.11095