KV Cache Compression, But What Must We Give in Return? A Comprehensive Benchmark of Long Context Capable Approaches
Abstract: Long context capability is a crucial competency for large language models (LLMs) as it mitigates the human struggle to digest long-form texts. This capability enables complex task-solving scenarios such as book summarization, code assistance, and many more tasks that are traditionally manpower-intensive. However, transformer-based LLMs face significant challenges with long context input due to the growing size of the KV cache and the intrinsic complexity of attending to extended inputs; where multiple schools of efficiency-driven approaches — such as KV cache quantization, token dropping, prompt compression, linear-time sequence models, and hybrid architectures — have been proposed to produce efficient yet long context-capable models. Despite these advancements, no existing work has comprehensively benchmarked these methods in a reasonably aligned environment. In this work, we fill this gap by providing a taxonomy of current methods and evaluating over 10+ state-of-the-art approaches across seven categories of long context tasks. Our work reveals numerous previously unknown phenomena and offers insights — as well as a friendly workbench — for the future development of long context-capable LLMs.
Paper Type: Long
Research Area: Resources and Evaluation
Research Area Keywords: Efficient/Low-Resource Methods for NLP
Contribution Types: Reproduction study, Approaches low compute settings-efficiency
Languages Studied: English
Submission Number: 129
Loading