Jing Zhang, Peng Zhang, Xindian Ma, Yidi Lu
We have not lifted any functions out of this paper's repositories yet, so there is nothing we have run. The repositories linked to it are listed below.
Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.
The integration of visual information into Large Language Models (LLMs) has enabled Multimodal LLMs (MLLMs), but the quadratic memory and computational costs of Transformer architectures remain a bottleneck. Existing KV cache eviction strategies fail to address the heterogeneous attention distributions between visual and text tokens, leading to suboptimal efficiency or degraded performance. In this paper, we propose Hierarchical Adaptive Eviction (HAE), a KV cache eviction framework that optimizes text-visual token interaction in MLLMs by implementing Dual-Attention Pruning during pre-filling (leveraging visual token sparsity and attention variance) and a Dynamic Decoding Eviction Strategy (inspired by OS Recycle Bins) during decoding. HAE minimizes KV cache usage across layers, reduces computational overhead via index broadcasting, and theoretically ensures superior information integrity and lower error bounds compared to greedy strategies, enhancing efficiency in both comprehension and generation tasks. Empirically, HAE reduces KV-Cache memory by 41% with minimal accuracy loss (0.3% drop) in image understanding tasks and accelerates story generation inference by 1.5× while maintaining output quality on Phi3.5-Vision-Instruct model. The source code and related materials are publicly available in the GitHub repository at https://github.com/maxindian/HAE_for_VLMs
The same record, over MCP at https://syntology.ai/mcp:
get_harvested_code_for_paper("2602.02197")
get_code_for_paper("2602.02197")
have("2602.02197")
The run record, dated, one paper per request, free:
curl https://syntology.ai/api/ran/2602.02197.json
A badge for a README (the split and the date, never a ratio):
[](https://syntology.ai/paper/2602.02197)
Connect an agent — have() is free.