mlsanityData to 14 Oct 2026

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

arXiv 2609.32831 ↗First posted 26 Sep 2026cs.LGcited by 0

On the map: Efficient LLM inference and serving → KV cache optimization

Abstract (arXiv)

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× KV cache compression for understanding and editing and 2.5× for generation with negligible quality loss, while increasing throughput by up to 1.78× in long-context settings, significantly improving the practicality of scaling unified multimodal models to longer context.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Placed in KV cache optimization, one of 12 directions in Efficient LLM inference and serving. Builds on 24 of its 26 references that are in the map. First posted on arXiv 26 Sep 2026.

Builds on

24 of 26 references.

Closest ancestors · 4

Same direction, newest first

  1. 2024DiffKV: Differentiated Memory Management for Large Language Models with Parallel KV Compaction4 Dec
  2. 2023Efficient Memory Management for Large Language Model Serving with PagedAttention12 Sep
  3. 2023H2O: Heavy-Hitter Oracle for Efficient Generative Inference of Large Language Models24 Jun
  4. 2023Scissorhands: Exploiting the Persistence of Importance Hypothesis for LLM KV Cache Compression at Test Time26 May

Other references · 20

  1. 2023Efficient Streaming Language Models with Attention Sinkscited by 2,186
  2. 2023MMBench: Is Your Multi-modal Model an All-around Player?cited by 2,141
  3. 2023MMMU: A Massive Multi-discipline Multimodal Understanding and Reasoning Benchmark for Expert AGIcited by 2,080
  4. 2023MME: A Comprehensive Evaluation Benchmark for Multimodal Large Language Modelscited by 1,652
  5. 2023GenEval: An Object-Focused Framework for Evaluating Text-to-Image Alignmentcited by 972
  6. 2024Chameleon: Mixed-Modal Early-Fusion Foundation Modelscited by 927
  7. 2025Emerging Properties in Unified Multimodal Pretrainingcited by 897
  8. 2024Show-o: One Single Transformer to Unify Multimodal Understanding and Generationcited by 752
  9. 2019Fast Transformer Decoding: One Write-Head is All You Needcited by 717
  10. 2024KIVI: A Tuning-Free Asymmetric 2bit Quantization for KV Cachecited by 637
  11. 2024Transfusion: Predict the Next Token and Diffuse Images with One Multi-Modal Modelcited by 483
  12. 2024Janus: Decoupling Visual Encoding for Unified Multimodal Understanding and Generationcited by 457
8 more
  1. 2025Step1X-Edit: A Practical Framework for General Image Editingcited by 429
  2. 2024Mixture-of-Transformers: A Sparse and Scalable Architecture for Multi-Modal Foundation Modelscited by 163
  3. 2023Direct Inversion: Boosting Diffusion-based Editing with 3 Lines of Codecited by 145
  4. 2024ZipVL: Efficient Large Vision-Language Models with Dynamic Token Sparsificationcited by 44
  5. 2024VL-Cache: Sparsity and Modality-Aware KV Cache Compression for Vision-Language Model Inference Accelerationcited by 43
  6. 2025Hyper-Bagel: A Unified Acceleration Framework for Multimodal Understanding and Generationcited by 13
  7. 2026AlphaQ: Calibration-Free Bit Allocation for Mixture-of-Experts Quantizationcited by 7
  8. 2026Flash-Unified: A Training-Free and Task-Aware Acceleration Framework for Native Unified Modelscited by 5

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026HybridKV: Hybrid KV Cache Compression for Efficient Multimodal Large Language Model Inference
  2. 2025Mixing Importance with Diversity: Joint Optimization for KV Cache Compression in Large Vision-Language Models
  3. 2025Enhancing Large Multimodal Models with Adaptive Sparsity and KV Cache Compression
  4. 2026EchoKV: Efficient KV Cache Compression via Similarity-Based Reconstruction
  5. 2025Revisiting Multimodal KV Cache Compression: A Frequency-Domain-Guided Outlier-KV-Aware Approach
  6. 2024DynamicKV: Task-Aware Adaptive KV Cache Compression for Long Context LLMs
  7. 2026Hierarchical Adaptive Eviction for KV Cache Management in Multimodal Language Models
  8. 2026Behavior-Preserving KV Cache Compression