SCOPE: Optimizing Key-Value Cache Compression in Long-context Generation

TL;DR

SCOPE introduces a framework to optimize Key-Value cache compression during both prefill and decoding phases in long-context generation, improving efficiency and maintaining reasoning accuracy in large language models.

Contribution

The paper proposes a novel two-phase KV cache optimization framework that enhances long-context generation by preserving essential information and selecting heavy hitters during decoding.

Findings

Reduces memory usage and transfer during long-output generation.

Maintains reasoning accuracy with optimized KV cache.

Compatible with existing KV compression methods.

Abstract

Key-Value (KV) cache has become a bottleneck of LLMs for long-context generation. Despite the numerous efforts in this area, the optimization for the decoding phase is generally ignored. However, we believe such optimization is crucial, especially for long-output generation tasks based on the following two observations: (i) Excessive compression during the prefill phase, which requires specific full context impairs the comprehension of the reasoning task; (ii) Deviation of heavy hitters occurs in the reasoning tasks with long outputs. Therefore, SCOPE, a simple yet efficient framework that separately performs KV cache optimization during the prefill and decoding phases, is introduced. Specifically, the KV cache during the prefill phase is preserved to maintain the essential information, while a novel strategy based on sliding is proposed to select essential heavy hitters for the decoding phase. Memory usage and memory transfer are further optimized using adaptive and discontinuous strategies. Extensive experiments on LongGenBench show the effectiveness and generalization of SCOPE and its compatibility as a plug-in to other prefill-only KV compression methods.