A Silver Bullet or a Compromise for Full Attention? A Comprehensive Study of Gist Token-based Context Compression

TL;DR

This paper thoroughly investigates gist token-based context compression for large language models, evaluating its effectiveness, failure modes, and proposing strategies to enhance compression performance in long-context tasks.

Contribution

It provides a comprehensive analysis of gist-based compression, identifies key failure patterns, and introduces two strategies to improve compression quality in large language models.

Findings

Gist compression achieves near-lossless performance in retrieval-augmented generation.

Challenges remain in synthetic recall tasks due to compression failures.

Proposed strategies effectively mitigate key failure patterns.

Abstract

In this work, we provide a thorough investigation of gist-based context compression methods to improve long-context processing in large language models. We focus on two key questions: (1) How well can these methods replace full attention models? and (2) What potential failure patterns arise due to compression? Through extensive experiments, we show that while gist-based compression can achieve near-lossless performance on tasks like retrieval-augmented generation and long-document QA, it faces challenges in tasks like synthetic recall. Furthermore, we identify three key failure patterns: lost by the boundary, lost if surprise, and lost along the way. To mitigate these issues, we propose two effective strategies: fine-grained autoencoding, which enhances the reconstruction of original token information, and segment-wise token importance estimation, which adjusts optimization based on token dependencies. Our work provides valuable insights into the understanding of gist token-based context compression and offers practical strategies for improving compression capabilities.