RaBitQCache: Rotated Binary Quantization for KVCache in Long Context LLM Inference
This work improves inference efficiency for long-context LLMs, a critical bottleneck for deployment in applications like document analysis and chatbots.
RaBitQCache addresses the KV cache bottleneck in long-context LLM inference by using rotated binary quantization for unbiased attention estimation, enabling adaptive Top-p retrieval. It achieves significant speedup and reduced memory I/O while maintaining generation quality.
Long-context Large Language Model inference is severely bottlenecked by the massive Key-Value (KV) cache, yet existing sparse attention methods often suffer from static fixed-budget (Top-k) retrieval or rely on proxy scores that are computationally expensive and biased. To address these limitations, we propose RaBitQCache, a novel sparse attention framework that utilizes randomized rotated binary quantization and high-throughput binary-INT4 arithmetic to efficiently estimate attention weights. Our proxy score serves as an unbiased estimator with a proven error bound, enabling adaptive Top-p retrieval that dynamically adjusts the token budget based on actual attention sparsity. We further implement a hardware-aware system with asynchronous pipelining and lazy updates to mask overhead. Evaluations demonstrate that RaBitQCache significantly accelerates inference and reduces memory I/O while preserving generation quality compared to state-of-the-art baselines. Code is available at https://github.com/Sakuraaa0/RaBitQCache.git.