Lune

ACL Findings2026

Preserving Fairness and Safety in Quantized LLMs Through Critical Weight Protection

Muhammad Alif Al Hakim, Alfan Farizki Wicaksono, Fajri Koto

Abstract

Quantization is widely adopted to reduce the computational cost of large language models (LLMs); however, its implications for fairness and safety, particularly in dynamic quantization and multilingual contexts, remain underexplored.In this work, we conduct a systematic study of how static and dynamic quantization methods impact fairness and safety across benchmarks measuring intrinsic and extrinsic bias and safety alignment.For fairness, we evaluate English, French, Dutch, Spanish, and Turkish; for safety, we focus on English, Korean, and Arabic.Our findings reveal that quantization consistently degrades fairness and safety, with dynamic methods demonstrating greater stability than static ones.Moreover, fairness degradation varies across languages, while safety deterioration is especially pronounced in non-English settings.To address these risks, we introduce Critical Weight Protection, a novel technique that identifies and preserves fairnessand safety-critical weights during quantization.This approach effectively mitigates bias and safety deterioration without costly retraining or alignment, maintaining trustworthiness while retaining efficiency. 1 Gemma-7B-Instruct Full Precision -0.021 -0.001 -0.006 0.002 -0.004 0.005 -0.003 0.005 FP-8 -0.021 +0.000 0.000 +0.000 -0.005 +0.001 0.001 +0.000 -0.007 -0.003 0.007 -0.002 -0.002 +0.001 0.006 -0.001 LLM8 -0.025 -0.004 0.000 +0.001 -0.008 -0.003 0.002 -7.881 × 10 -5 -0.004 +0.000 0.007 -0.001 -0.002 +0.001 0.005 -0.000 SmoothQuant -0.037 -0.016 -0.002 -0.001 0.025 -0.019 0.010 -0.008 0.000 +0.004 0.001 +0.004 -0.024 -0.021 -0.010 -0.005 GPTQ -0.018 +0.003 0.000 +0.001 -0.006 -0.000 0.000 +0.001 0.000 +0.004 0.003 +0.002 -0.006 -0.003 0.003 +0.003 AWQ -0.026 -0.005 0.000 +0.000 0.002 +0.003 -0.002 -0.000 0.000 +0.004 0.003 +0.002 -0.004 -0.001 0.005 +0.000 Llama-3.1-8B-InstructFull Precision -0.002 -0.002 0.002 -0.008 0.000 -0.003 -0.001 0.004 FP-8 -0.001 +0.000 -0.003 -0.001 0.004 -0.003 -0.009 -0.001 -0.002 -0.002 -0.004 -0.000 0.000 +0.000 0.003 +0.001 LLM8 -0.001 +0.001 -0.001 +0.001 0.001 +0.001 -0.010 -0.002 0.000 -0.000 -0.001 +0.002 0.000 +0.001 0.000 +0.004 SmoothQuant 0.000 +0.001 -0.001 +0.001 0.002 -0.001 -0.008 -0.000 0.001 -0.000 -0.002 +0.002 0.001 -0.001 0.003 +0.002 GPTQ 0.002 +0.000 -0.001 +0.000 -0.004 -0.002 -0.008 +0.000 0.002 -0.001 -0.006 -0.003 -0.001 -7.943 × 10 -5 0.002 +0.

Ask Lune

Know this paper in minutes, not an afternoon.

Lune can read every page of it and the top-tier papers that cite it, and gives you its claims, its evidence and its limits, each one cited to the paper it comes from. No skimming, no AI slop.

Questions to start from

Lune can draw on

the full paper · related work

Get the answer

Free to start. No credit card required.

Lineage

Builds onCited byLeft to right by year; the more cited, the higher.

Builds on15

  1. Direct Preference Optimization: Your Language Model is Secretly a Reward ModelRafael Rafailov, Archit Sharma, Eric Mitchell, Christopher D. Manning et al.NeurIPS 2023 · 10,924 citations
  2. SmoothQuant: Accurate and Efficient Post-Training Quantization for Large Language ModelsGuangxuan Xiao, Ji Lin, Mickaël Seznec, Hao Wu et al.ICML 2023 · 1,493 citations
  3. Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!Xiangyu Qi, Yi Zeng, Tinghao Xie, Pin-Yu Chen et al.ICLR 2024 · 1,104 citations
  4. HarmBench: A Standardized Evaluation Framework for Automated Red Teaming and Robust RefusalMantas Mazeika, Long Phan, Xuwang Yin, Andy Zou et al.ICML 2024 · 1,031 citations
  5. Catastrophic Jailbreak of Open-source LLMs via Exploiting GenerationYangsibo Huang, Samyak Gupta, Mengzhou Xia, Kai Li et al.ICLR 2024 · 481 citations

Related papers

  1. Assessing Safety Risks and Quantization-aware Safety Patching for Quantized Large Language ModelsKejia Chen, Jiawen Zhang, Jiacong Hu, Yu Wang et al.ICML 2025
  2. Durable Quantization Conditioned Misalignment Attack on Large Language ModelsPeiran Dong, Haowei Li, Song GuoICLR 2025
  3. Toward Safe Quantization-Aware Fine-tuning: Understanding and Mitigating Safety Alignment DegradationYuning Yang, Guowei Peng, Xiurui Xie, Minrui Jiang et al.ICML 2026
  4. How Does Quantization Affect Multilingual LLMs?Kelly Marchisio, Saurabh Dash, Hongyu Chen, Dennis Aumiller et al.EMNLP Findings 2024 · 14 citations
  5. Watermarking LLMs with Weight QuantizationLinyang Li, Botian Jiang, Pengyu Wang, Ke Ren et al.EMNLP Findings 2023 · 13 citations
Dusk over the sea between two cliffs drawn in fine vertical lines