As the sequence length of large language models increases, key-value (KV) caches face significant challenges. Existing state-of-the-art cache eviction methods primarily analyze the inference behavior of attention heads in successful retrieval-reasoning cases, often overlooking diverse behaviors in failure cases, such as bias and distraction. This oversight limits the potential to leverage heterogeneous head behaviors for improved eviction performance.
Inspired by the confusion matrix, we introduce an Attention Behavior Matrix to comprehensively analyze attention head behaviors in both success and failure scenarios. By maximizing the signal-to-noise ratio—strengthening valid reasoning pathways in success cases while inhibiting noise from bias and distraction in failure cases—we propose REtrieval-reAsoning and Logic-constructed (REAL) KV cache eviction, the first method to leverage multi-behavior analysis.
Comprehensive evaluations show that REAL achieves remarkable performance across various models and benchmarks; notably, on LongBench v2, it achieves comparable accuracy to the strongest baseline, HeadKV-R2, while requiring 32x less space. By offering a novel perspective on behavior analysis, we pave the way for a shift from success-only to comprehensive, failure-aware methods in long-context modeling. Our code is available at GitHub.
Blogger's Review: The REAL method significantly enhances the performance of KV cache by focusing on the behavioral analysis of failure cases, showcasing innovative potential in long-context processing. This approach not only improves efficiency but also provides new insights for future research, making it a worthwhile area for attention and exploration.