# miniReranker: Efficient Multimodal Reranking through Visual Cache Reuse and Interaction Sparsity

Source: [miniReranker: Efficient Multimodal Reranking through Visual Cache Reuse and Interaction Sparsity](<https://www.sota2.com/research/paper/6a28f37890966ade919c19ba>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-06-18T05:00:06.089Z

Authors: Yingqi Fan, Xuan Lu, Anhao Zhao, Junlong Tong, Ping Nie, Kai Zou, Yunpu Ma, Wei Zhang, Xiaoyu Shen

Published: 2026-06-09T12:11:02.000Z

## Abstract

Multimodal large language models (MLLMs) have recently shown strong potential as point-wise rerankers by directly modeling query--document relevance through next-token prediction. However, point-wise reranking suffers from substantial repeated computation across query--document pairs, while the causal structure of transformers allows only prefix segments to be reused via pre-caching. To address the misalignment of existing query-first and document-first formats with both VQA-style prompting and computation-aware reuse, we propose a $\\textit{vision-first}$ formulation that improves both cache reuse efficiency and reranking performance. However, the remaining cost is still considerable and stems from three main sources: (1) $\\textit{model depth}$, for which we reduce active parameters via early exit; (2) $\\textit{cross-segment attention}$, which we restrict to a narrow interaction band across a few layers; and (3) $\\textit{visual tokens}$, where we reduce the number of tokens via embedder-guided pruning. Together, these designs form miniReranker, which reduces reranking runtime to \<1% of the dense implementation under high-reuse settings for a single query, while preserving \>96% of the dense model performance.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Multimodal Reranking on MMEB Video v2](<https://www.sota2.com/research/sota/multimodal-reranking-on-mmeb-video-v2.md>): Multimodal Reranking; MMEB Video v2; Overall Average Score: 62.9.

- [Multimodal Reranking on MMEB VisDoc v2](<https://www.sota2.com/research/sota/multimodal-reranking-on-mmeb-visdoc-v2.md>): Multimodal Reranking; MMEB VisDoc v2; Overall Score: 87.1.

- [Multimodal Retrieval on MMEB Image V2](<https://www.sota2.com/research/sota/multimodal-retrieval-on-mmeb-image-v2.md>): Multimodal Retrieval; MMEB Image V2; Overall Score: 76.2.

- [Multimodal Reranking on MMEB Summary v2 (Overall)](<https://www.sota2.com/research/sota/multimodal-reranking-on-mmeb-summary-v2-overall.md>): Multimodal Reranking; MMEB Summary v2 (Overall); Overall Score: 76.5.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2606.10759>)

[PDF](<https://arxiv.org/pdf/2606.10759.pdf>)