Perception Compressor: A Training-Free Prompt Compression Framework in Long Context Scenarios

TL;DR

Perception Compressor is a training-free framework that effectively compresses prompts in long context scenarios, improving LLM performance by retaining key information and reducing redundancy without additional training.

Contribution

It introduces a novel, training-free prompt compression method with dynamic allocation and semi-guided iterative compression for long context scenarios.

Findings

Outperforms existing methods significantly

Achieves state-of-the-art results on benchmarks

Effectively retains key information while removing distractions

Abstract

Large language models (LLMs) demonstrate exceptional capabilities in various scenarios. However, they suffer from much redundant information and are sensitive to the position of key information in long context scenarios. To address these challenges, we present Perception Compressor, a training-free prompt compression framework. It includes a perception retriever that leverages guiding questions and instruction to retrieve the most relevant demonstrations, a dual-slope ratio allocator to dynamically allocate compression ratios and open-book ratios, and a semi-guided iterative compression that retains key information at the token level while removing tokens that distract the LLM. We conduct extensive experiments on long context benchmarks, i.e., NaturalQuestions, LongBench, and MuSiQue. Experiment results show that Perception Compressor outperforms existing methods by a large margin, achieving state-of-the-art performance.