# Beyond Perplexity: Multi-dimensional Safety Evaluation of LLM Compression

arXiv: https://arxiv.org/abs/2407.04965 · HTML: https://mlsanity.com/paper/2407.04965/

## Key facts

- Cited by 33 papers in the map (data through 14 Oct 2026).
- Builds on 48 of its 73 references that are in the map.
- First posted on arXiv 6 Jul 2024.

## Builds on (most cited first, 30 of 48)

- [GPT-4 Technical Report](https://mlsanity.com/paper/2303.08774/) (2023)
- [Training language models to follow instructions with human feedback](https://mlsanity.com/paper/2203.02155/) (2022)
- [Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer](https://mlsanity.com/paper/1910.10683/) (2019)
- [Llama 2: Open Foundation and Fine-Tuned Chat Models](https://mlsanity.com/paper/2307.09288/) (2023)
- [Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena](https://mlsanity.com/paper/2306.05685/) (2023)
- [Measuring Massive Multitask Language Understanding](https://mlsanity.com/paper/2009.03300/) (2020)
- [DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter](https://mlsanity.com/paper/1910.01108/) (2019)
- [QLoRA: Efficient Finetuning of Quantized LLMs](https://mlsanity.com/paper/2305.14314/) (2023)
- [Pointer Sentinel Mixture Models](https://mlsanity.com/paper/1609.07843/) (2016)
- [TruthfulQA: Measuring How Models Mimic Human Falsehoods](https://mlsanity.com/paper/2109.07958/) (2021)
- [ALBERT: A Lite BERT for Self-supervised Learning of Language Representations](https://mlsanity.com/paper/1909.11942/) (2019)
- [GPTQ: Accurate Post-Training Quantization for Generative Pre-trained Transformers](https://mlsanity.com/paper/2210.17323/) (2022)
- [Don't Give Me the Details, Just the Summary! Topic-Aware Convolutional Neural Networks for Extreme Summarization](https://mlsanity.com/paper/1808.08745/) (2018)
- [RealToxicityPrompts: Evaluating Neural Toxic Degeneration in Language Models](https://mlsanity.com/paper/2009.11462/) (2020)
- [Fantastically Ordered Prompts and Where to Find Them: Overcoming Few-Shot Prompt Order Sensitivity](https://mlsanity.com/paper/2104.08786/) (2021)
- [SparseGPT: Massive Language Models Can Be Accurately Pruned in One-Shot](https://mlsanity.com/paper/2301.00774/) (2023)
- [Language (Technology) is Power: A Critical Survey of "Bias" in NLP](https://mlsanity.com/paper/2005.14050/) (2020)
- [BBQ: A Hand-Built Bias Benchmark for Question Answering](https://mlsanity.com/paper/2110.08193/) (2021)
- [KIVI: A Tuning-Free Asymmetric 2bit Quantization for KV Cache](https://mlsanity.com/paper/2402.02750/) (2024)
- [The Woman Worked as a Babysitter: On Biases in Language Generation](https://mlsanity.com/paper/1909.01326/) (2019)
- [Dolma: an Open Corpus of Three Trillion Tokens for Language Model Pretraining Research](https://mlsanity.com/paper/2402.00159/) (2024)
- [Sheared LLaMA: Accelerating Language Model Pre-training via Structured Pruning](https://mlsanity.com/paper/2310.06694/) (2023)
- [BOLD: Dataset and Metrics for Measuring Biases in Open-Ended Language Generation](https://mlsanity.com/paper/2101.11718/) (2021)
- [Capabilities of Gemini Models in Medicine](https://mlsanity.com/paper/2404.18416/) (2024)
- [Demographic Dialectal Variation in Social Media: A Case Study of African-American English](https://mlsanity.com/paper/1608.08868/) (2016)
- ["I'm sorry to hear that": Finding New Biases in Language Models with a Holistic Descriptor Dataset](https://mlsanity.com/paper/2205.09209/) (2022)
- [Sociotechnical Safety Evaluation of Generative AI Systems](https://mlsanity.com/paper/2310.11986/) (2023)
- [UnQovering Stereotyping Biases via Underspecified Questions](https://mlsanity.com/paper/2010.02428/) (2020)
- [KV Cache is 1 Bit Per Channel: Efficient Large Language Model Inference with Coupled Quantization](https://mlsanity.com/paper/2405.03917/) (2024)
- [Compressing LLMs: The Truth is Rarely Pure and Never Simple](https://mlsanity.com/paper/2310.01382/) (2023)

## Built on by (most cited first, 30 of 33)

- [Investigating the Impact of Quantization Methods on the Safety and Reliability of Large Language Models](https://mlsanity.com/paper/2502.15799/) (2025)
- [Exploring the Trade-Offs: Quantization Methods, Task Difficulty, and Model Size in Large Language Models From Edge to Giant](https://mlsanity.com/paper/2409.11055/) (2024)
- [State Space Models are Strong Text Rerankers](https://mlsanity.com/paper/2412.14354/) (2024)
- [CSPLADE: Learned Sparse Retrieval with Causal Language Models](https://mlsanity.com/paper/2504.10816/) (2025)
- [Does quantization affect models' performance on long-context tasks?](https://mlsanity.com/paper/2505.20276/) (2025)
- [Quantized but Deceptive? A Multi-Dimensional Truthfulness Evaluation of Quantized LLMs](https://mlsanity.com/paper/2508.19432/) (2025)
- [Activation Approximations Can Incur Safety Vulnerabilities Even in Aligned LLMs: Comprehensive Analysis and Defense](https://mlsanity.com/paper/2502.00840/) (2025)
- [Towards Understanding and Improving Refusal in Compressed Models via Mechanistic Interpretability](https://mlsanity.com/paper/2504.04215/) (2025)
- [Beyond Correctness: Rewarding Faithful Reasoning in Retrieval-Augmented Generation](https://mlsanity.com/paper/2510.13272/) (2025)
- [Quantization Undoes Alignment: Bias Emergence in Compressed LLMs Across Models and Precision Levels](https://mlsanity.com/paper/2605.15208/) (2026)
- [LACONIC: Dense-Level Effectiveness for Scalable Sparse Retrieval via a Two-Phase Training Curriculum](https://mlsanity.com/paper/2601.01684/) (2026)
- [Investigating Social Bias Changes in Quantized Language Models](https://mlsanity.com/paper/2602.06181/) (2026)
- [How Quantization Shapes Bias in Large Language Models](https://mlsanity.com/paper/2508.18088/) (2025)
- [Efficient LLM Inference with Activation Checkpointing and Hybrid Caching](https://mlsanity.com/paper/2501.01792/) (2025)
- [Q-resafe: Assessing Safety Risks and Quantization-aware Safety Patching for Quantized Large Language Models](https://mlsanity.com/paper/2506.20251/) (2025)
- [Pruning Weights but Not Truth: Safeguarding Truthfulness While Pruning LLMs](https://mlsanity.com/paper/2509.00096/) (2025)
- [Weight Pruning Amplifies Bias: A Multi-Method Study of Compressed LLMs for Edge AI](https://mlsanity.com/paper/2605.08137/) (2026)
- [From Raw Corpora to Domain Benchmarks: Automated Evaluation of LLM Domain Expertise](https://mlsanity.com/paper/2506.07658/) (2025)
- [Beyond Bias Scores: Unmasking Vacuous Neutrality in Small Language Models](https://mlsanity.com/paper/2506.08487/) (2025)
- [SHIELD: An Auto-Healing Agentic Defense Framework for LLM Resource Exhaustion Attacks](https://mlsanity.com/paper/2601.19174/) (2026)
- [Quality Is Not a Safety Proxy Under Quantization](https://mlsanity.com/paper/2606.10154/) (2026)
- [QuantiBias: Benchmarking Quantization-Induced Bias in LLMs](https://mlsanity.com/paper/2607.21063/) (2026)
- [The Asymmetric Harms of LLM Compression](https://mlsanity.com/paper/2608.19670/) (2026)
- [Fair-GPTQ: Bias-Aware Quantization for Large Language Models](https://mlsanity.com/paper/2509.15206/) (2025)
- [Rethinking Perplexity: Revealing the Impact of Input Length on Perplexity Evaluation in LLMs](https://mlsanity.com/paper/2602.04099/) (2026)
- [The Hrunting of AI: Where and How to Improve English Dialectal Fairness](https://mlsanity.com/paper/2603.15187/) (2026)
- [Mechanistically Interpreting Compression in Vision-Language Models](https://mlsanity.com/paper/2603.25035/) (2026)
- [K-Quantization and its Impact on Output Performance](https://mlsanity.com/paper/2605.19645/) (2026)
- [Reliability Scaling Laws for Quantized Large Language Models](https://mlsanity.com/paper/2607.10855/) (2026)
- [When Personality Meets Quantization: A Layer-wise MBTI Analysis of Quantized LLMs](https://mlsanity.com/paper/2608.25977/) (2026)

## Abstract (arXiv)

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.footnoteOur implementation and results are available here: urlhttps://github.com/zhichaoxu-shufe/Beyond-Perplexity-Compression-Safety-Eval