Beyond Perplexity: Multi-dimensional Safety Evaluation of LLM Compression

TL;DR

This paper systematically evaluates how different model compression techniques affect various safety aspects of large language models, revealing complex impacts on bias, toxicity, and task performance beyond traditional perplexity measures.

Contribution

It introduces a multi-dimensional safety assessment framework for compressed LLMs, highlighting the diverse and sometimes unexpected effects of compression methods on safety and bias.

Findings

Compression can reduce degeneration harm but may increase representational harm.

Different compression methods have distinct safety impacts, e.g., quantization vs. pruning.

Safety impacts vary across protected groups and bias types.

Abstract

Increasingly, model compression techniques enable large language models (LLMs) to be deployed in real-world applications. As a result of this momentum towards local deployment, compressed LLMs will interact with a large population. Prior work on compression typically prioritize preserving perplexity, which is directly analogous to training loss. The impact of compression method on other critical aspects of model behavior\, -- \,particularly safety\, -- \,requires systematic assessment. To this end, we investigate the impact of model compression along four dimensions: (1) degeneration harm, i.e., bias and toxicity in generation; (2) representational harm, i.e., biases in discriminative tasks; (3) dialect bias; and(4) language modeling and downstream task performance. We examine a wide spectrum of LLM compression techniques, including unstructured pruning, semi-structured pruning, and quantization. Our analysis reveals that compression can lead to unexpected consequences. Although compression may unintentionally alleviate LLMs' degeneration harm, it can still exacerbate representational harm. Furthermore, increasing compression produces a divergent impact on different protected groups. Finally, different compression methods have drastically different safety impacts: for example, quantization mostly preserves bias while pruning degrades quickly. Our findings underscore the importance of integrating safety assessments into the development of compressed LLMs to ensure their reliability across real-world applications.\footnote{Our implementation and results are available here: \url{https://github.com/zhichaoxu-shufe/Beyond-Perplexity-Compression-Safety-Eval}}