The Asymmetric Harms of LLM Compression

2026-08-20

AI Summary

研究评估了11种压缩方法对3个大型语言模型的影响，揭示了压缩导致的知识保留、模型自信度及社会偏见的变化问题。

Abstract

Large language models (LLMs) compression reduces deployment costs, but standard aggregate metrics like perplexity and accuracy often mask underlying behavioral shifts. In this work, we systematically evaluate 3 LLMs across 11 compression methods to investigate the effects of compression on knowledge retention, model confidence, and social bias. We find that compression disproportionately reduces the relative retention of head knowledge compared to tail knowledge. Furthermore, compressed models often remain substantially confident in their incorrect answers on newly lost knowledge. Finally, we demonstrate that stable aggregate bias scores can conceal substantial, opposing shifts in stereotypical preferences across demographic subgroups. Together, these findings reveal asymmetric behavioral changes that aggregate performance measures fail to capture, highlighting the need for granular evaluation of compressed models before deployment.

Problem

• LLM compression
• knowledge retention
• model confidence
• social bias

Innovation

• LLM Compression
• Knowledge Retention
• Model Confidence
• Social Bias
• Aggregate Metrics