2503.19114

Total: 1

#1 Understanding and Improving Information Preservation in Prompt Compression for LLMs [PDF4] [Copy] [Kimi4] [REL]

Authors: Weronika Łajewska, Momchil Hardalov, Laura Aina, Neha Anna John, Hang Su, Lluís Màrquez

Recent advancements in large language models (LLMs) have enabled their successful application to a broad range of tasks. However, in information-intensive tasks, the prompt length can grow fast, leading to increased computational requirements, performance degradation, and induced biases from irrelevant or redundant information. Recently, various prompt compression techniques have been introduced to optimize the trade-off between reducing input length and retaining performance. We propose a holistic evaluation framework that allows for in-depth analysis of prompt compression methods. We focus on three key aspects, besides compression ratio: (i) downstream task performance, (ii) grounding in the input context, and (iii) information preservation. Through this framework, we investigate state-of-the-art soft and hard compression methods, showing that they struggle to preserve key details from the original prompt, limiting their performance on complex tasks. We demonstrate that modifying soft prompting methods to control better the granularity of the compressed information can significantly improve their effectiveness -- up to +23\% in downstream task performance, more than +8 BERTScore points in grounding, and 2.7x more entities preserved in compression.

Subjects: Computation and Language , Information Retrieval , Machine Learning

Publish: 2025-03-24 20:06:11 UTC