2025/02/11 by Han, Insu, Kapralov, Michael, Kochetkova, Ekaterina +2 · 1 citation
Computer Science · #Advanced Data Compression Techniques #Advanced Data Storage Technologies #Artificial Intelligence (cs.AI) #Data Structures and Algorithms (cs.DS) #FOS: Computer and information sciences #Machine Learning (cs.LG) #Video Coding and Compression Technologies
paper · pdf · doi:10.48550/arxiv.2502.07861
openalex publication_date 2025/02/11 · openalex created_date 2025/02/15 · openalex updated_date 2026/07/30
Large language models (LLMs) have achieved impressive success, but their high memory requirements present challenges for long-context token generation. In this paper we study the streaming complexity of attention approximation, a key computational primitive underlying token generation. Our main contribution is BalanceKV, a streaming algorithm for ε-approximating attention computations based on geometric process for selecting a balanced collection of Key and Value tokens as per Banaszczyk's vector balancing theory. We complement our algorithm with space lower bounds for streaming attention computation. Besides strong theoretical guarantees, BalanceKV exhibits empirically validated performance improvements over existing methods, both for attention approximation and end-to-end performance on various long context benchmarks.