UniCache: Task- and Type-Aware KV Cache Compression for Unified Multimodal Models

2026-09-26

AI Summary

This study addresses the limitation of unified multimodal models where a single KV cache compression strategy ignores task and cache-type differences, leading to critical information loss. We propose a training-free, task- and type-aware cache compression framework that identifies task-activated cache segments via offline calibration and assigns tailored strategies accordingly. By integrating attention-guided resource allocation with task-aware temporal scheduling to coordinate parallel execution, our approach resolves compression mismatches caused by dynamic cross-task variations. Experimental results demonstrate that the proposed method achieves 5× and 2.5× lossless compression on understanding-and-editing and generation tasks, respectively, while improving long-context throughput by 1.78×.

Abstract

Unified multimodal models combine understanding, generation, and editing within a single network, offering a promising foundation for versatile multimodal applications. However, growing multimodal contexts make KV cache storage and access increasingly costly. Existing KV cache compression methods are typically tailored to specific tasks and single-modality caches, while overlooking changes in cache importance across tasks and timesteps. However, in unified multimodal models, each task involves multiple KV cache types, and both their composition and dynamics differ across tasks. As a result, a single compression policy overlooks task- and type-specific requirements, leading to the loss of critical information and degraded quality across tasks. Based on these findings, we propose UniCache, a training-free framework for task- and type-aware KV cache compression. UniCache identifies the cache segments activated by each task and assigns suitable compression policies through offline calibration. It coordinates their parallel execution under a shared storage budget through attention-guided allocation and task-aware temporal scheduling. Experiments show that UniCache achieves $5\times$ KV cache compression for understanding and editing and $2.5\times$ for generation with negligible quality loss, while increasing throughput by up to $1.78\times$ in long-context settings, significantly improving the practicality of scaling unified multimodal models to longer context.

Problem

• Unified Multimodal Models
• KV Cache Compression
• Task-aware
• Type-aware
• Long-context

Innovation

• KV Cache Compression
• Unified Multimodal Models
• Training-Free Framework
• Task-Aware Scheduling
• Attention-Guided Allocation