一句话:提出 STARC,通过 KV 对聚类与 PIM bank 内连续共置,将细粒度稀疏注意力映射为行级 PIM 操作,降低 LLM 解码延迟与能耗。
大语言模型服务因频繁访问与随上下文增长的键值(KV)缓存而给内存系统带来巨大压力。PIM 虽提供高内部带宽与近存计算,但现有设计面向稠密注意力,难以应对动态 KV 稀疏的不规则访问。STARC 提出面向 PIM 上高效 LLM 解码的稀疏优化数据映射:将语义相似的 KV 对聚类并在 PIM bank 内连续共置,使查询通过与预计算质心匹配实现簇粒度检索,从而衔接细粒度稀疏注意力与行级 PIM 操作。在模拟 HBM-PIM 系统、受限 KV 预算下,相比 token 级稀疏方法,注意力层延迟和能耗最多降低 78% 与 65%,相比全注意力最多降低 93% 与 92%,并保持模型精度。
Serving large language models (LLMs) places significant pressure on memory systems due to frequent accesses and growing key–value (KV) caches as context lengths increase. Processing-in-memory (PIM) architectures offer high internal bandwidth and near-data compute parallelism, but current designs target dense attention and perform poorly under the irregular access patterns of dynamic KV cache sparsity. To mitigate this limitation, we propose STARC, a sparsity-optimized data mapping scheme for efficient LLM decoding on PIM. STARC clusters semantically similar KV pairs and co-locates them contiguously within PIM banks, enabling retrieval at cluster granularity by matching queries against precomputed centroids. This bridges the gap between fine-grained sparse attention and row-level PIM operations, improving utilization while minimizing overhead. On a simulated HBM-PIM system, under constrained KV budgets, STARC achieves up to 78% and 65% reductions in attention-layer latency and energy over token-wise sparsity methods, and up to 93% and 92% reductions relative to full attention, while preserving model accuracy.
DOI 原文 · arXiv · PDF(开放获取) · DBLP
@inproceedings{abs-2505-05772,
title = {STARC: Selective Token Access with Remapping and Clustering for Efficient LLM Decoding on PIM Systems},
author = {Zehao Fan and Yunzhen Liu and Garrett Gagnon and Zhenyu Liu and Yayue Hou and Hadjer Benmeziane and Kaoutar El Maghraoui and Liu Liu},
booktitle = {ASPLOS},
year = {2026},
doi = {10.1145/3779212.3790226}
}